Federated learning (FL) enables privacy-preserving model training across distributed Electronic Health Records (EHRs), but its deployment remains limited by data-view heterogeneity, where institutions maintain incompatible local schemas. Most existing methods address this by enforcing flat, aligned data views, which require extensive cross-site preprocessing and manual harmonisation that often discards client-specific features, or by projecting inputs into a shared latent space, which sacrifices interpretability. We propose a modelling shift from conventional FL with vectorised inputs to a symbolic, relation-centric framework, where each client organises its EHR data as a structured, type-aware relational graph. This enables client-specific inference without requiring schema alignment and supports FL across heterogeneous data views. To model over these symbolic structures, we introduce an architecture that combines relation-aware message passing with a learnable feature relevance mechanism, jointly enabling accurate local predictions and client-specific interpretability while supporting parameter sharing across clients. Beyond strong performance on three real-world EHR datasets exhibiting data-view heterogeneity, we further show that our framework supports multimodal FL under modality-level heterogeneity. Using MC-MED, a publicly available multimodal emergency department dataset, we demonstrate that our method accommodates clients with partially missing modalities, highlighting its robustness and scalability in real-world clinical settings.

Joaquin Polonuer, Lucas Vittor, Iñaki Arango, Ayush Noori, David A. Clifton, Luciano Del Corro, Marinka Zitnik. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Federated Learning (FL) enables collaborative clinical modelling across distributed electronic health records (EHRs) without sharing sensitive patient data. However, variations in medical practice, documentation standards, and data collection across institutions create data-view heterogeneity, where clients possess different or only partially overlapping clinical feature sets. This misalignment hinders the use of standard FL methods. Existing approaches rely on complex preprocessing and manual harmonisation, which can cause information loss, reduce data utility, limit scalability, and restrict client-specific personalisation. To address these limitations, we propose Personalised Attention-based Federated Graph Network (PAFNet), a scalable FL framework that enables meaningful parameter exchange across heterogeneous clients by mapping their distinct data-views into a shared latent space through client-specific projection layers. It then applies a personalised adaptation mechanism using trainable parameter masks, allowing each client to selectively incorporate global model parameters relevant to its own feature set. This design preserves local specificity, improves generalisation, and removes the need for heavy manual preprocessing common in existing approaches. Across CURIAL, eICU, and MIMIC-III datasets, PAFNet consistently outperformed state-of-the-art data-view heterogeneity FL baselines, demonstrating strong generalisation under substantial differences in client feature sets. By enabling effective personalisation and cross-institutional knowledge sharing without extensive harmonisation, PAFNet offers a robust and scalable solution for the federated training of clinical models in data-view heterogeneous environments.

This paper presents Adaptive Parameter Optimisation (APO), a novel framework for optimising shared models across multiple clinical tasks, addressing the challenges of balancing strict parameter sharing—often leading to task conflicts—and soft parameter sharing, which may limit effective cross-task information exchange. The proposed APO framework leverages insights from the lazy behaviour observed in over-parameterised neural networks, where only a small subset of parameters undergo any substantial updates during training. APO dynamically identifies and updates task-specific parameters while treating parameters associated with other tasks as protected, limiting their modification to prevent interference. The remaining unassigned parameters remain unchanged, embodying the lazy training phenomenon. This dynamic management of task-specific, protected, and unclaimed parameters across tasks enables effective information sharing, preserves task-specific adaptability, and mitigates gradient conflicts without enforcing a uniform representation. Experimental results across diverse healthcare datasets demonstrate that APO surpasses traditional information-sharing approaches, such as multi-task learning and model-agnostic meta-learning, in improving task performance.

Federated Learning (FL) involves a server aggregating local models from clients to compute a global model. However, this process can struggle to position the global model in low-loss regions of the parameter space for all clients, resulting in subpar convergence and inequitable performance across clients. This issue is particularly pronounced in non-IID settings, common in clinical contexts, where variations in data distribution, class imbalance, and training sample sizes result in client heterogeneity. To address this issue, we propose a mode connectivity-based FL framework that ensures the global model resides within the overlapping low-loss regions of all clients in the parameter space. This framework models the low-loss regions as non-linear mode connections between the current global and local models, and optimises to identify an intersection among these mode connections to define the new global model. This approach enhances training stability and convergence, yielding better and more equitable performance compared to standard FL frameworks like federated averaging. Empirical evaluations across multiple healthcare datasets demonstrate the benefits of the proposed framework.

Online continual learning (OCL) seeks to learn new tasks from data streams that appear only once, while retaining knowledge of previously learned tasks. Most existing methods rely on replay, focusing on enhancing memory retention through regularization or distillation. However, they often overlook the adaptability of the model, limiting the ability to learn generalizable and discriminative features incrementally from online training data. To address this, we introduce a plug-and-play module, S6MOD, which can be integrated into most existing methods and directly improve adaptability. Specifically, S6MOD introduces an extra branch after the backbone, where a mixture of discretization selectively adjusts parameters in a selective state space model, enriching selective scan patterns such that the model can adaptively select the most sensitive discretization method for current dynamics. We further design a class-conditional routing algorithm for dynamic, uncertainty-based adjustment and implement a contrastive discretization loss to optimize it. Extensive experiments combining our module with various models demonstrate that S6MOD significantly enhances model adaptability, leading to substantial performance gains and achieving the state-of-the-art results. The code is available at https://github.com/MyToumaKazusa/S6MOD.

Effective training of large Vision-Language Models (VLMs) on resource-constrained client devices in Federated Learning (FL) requires the usage of parameter-efficient finetuning (PEFT) strategies. To this end, we demonstrate the impact of two factors viz., client-specific layer importance score that selects the most important VLM layers for finetuning and inter-client layer diversity score that encourages diverse layer selection across clients for optimal VLM layer selection. We first theoretically motivate and leverage the principal eigenvalue magnitude of layerwise Neural Tangent Kernels and show its effectiveness as client-specific layer importance score. Next, we propose a novel layer updating strategy dubbed F3OCUS that jointly optimizes the layer importance and diversity factors by employing a data-free, multi-objective, meta-heuristic optimization on the server. We explore 5 different meta-heuristic algorithms and compare their effectiveness for selecting model layers and adapter layers towards PEFT-FL. Furthermore, we release a new MedVQA-FL dataset involving overall 707,962 VQA triplets and 9 modality-specific clients and utilize it to train and evaluate our method. Overall, we conduct more than 10,000 client-level experiments on 6 Vision-Language FL task settings involving 58 medical image datasets and 4 different VLM architectures of varying sizes to demonstrate the effectiveness of the proposed method. Project Page: https://pramitsaha.github.io/FOCUS/

Antibiotic Susceptibility Testing (AST) based on microorganism culturing is the gold-standard technique to determine whether a pathogen is susceptible or resistant to available antibiotics. While broth microdilution offers a potential high-throughput method for AST, reading and interpreting microtitre plates can be challenging, even for experienced clinical microbiologists. Machine learning models trained on images of microtitre plates obtained during AST could potentially accelerate and even automate this process. However, these image sets are highly imbalanced since each drug on the plate may exhibit different growth distributions due to varying resistance prevalence and mechanisms, which negatively impacts the performance of trained models. To address this problem, we propose a Generative Adversarial Network (GAN)-based framework, named CulplateGAN, to augment the dataset with images of plates displaying specific growth levels for particular drugs. The adversarial loss and weight-controlled content loss are introduced to achieve image transformation and content preservation. Moreover, a Multi-Culplate-GAN architecture is designed to generate multilevel outputs with one single input, which are optimized by the proposed domain-based adversarial loss and domain classification loss. We evaluate Culplate-GAN and Multi-Culplate-GAN by training a classifier on an AST Mycobacterium Tuberculosis dataset. Comprehensive results indicate that our method outperforms existing representative augmentation methods and can be generalized to plates containing other bacterial cultures.

Recent advancements in large language models (LLMs) with billions of parameters have improved performance in various applications, but their inference processes demand significant energy and computational resources. In contrast, the human brain, with approximately 86 billion neurons, is much more energy-efficient than LLMs with similar parameters. Inspired by this, we redesign 7$\sim$70 billion parameter LLMs using bio-plausible spiking mechanisms, emulating the efficient behavior of the human brain. We propose the first spiking large language model, SpikeLLM. Coupled with the proposed model, two essential approaches are proposed to improve spike training efficiency: Generalized Integrate-and-Fire (GIF) neurons to compress spike length from $T$ to $\frac{T}{L} \log_2 L$ bits, and an Optimal Brain Spiking framework to divide outlier channels and allocate different $T$ for GIF neurons, which further compresses spike length to approximate $log_2T$ bits. The necessity of spike-driven LLM is proved by comparison with quantized LLMs with similar operations. In the OmniQuant pipeline, SpikeLLM reduces 11.01\% WikiText2 perplexity and improves 2.55\% accuracy of common scene reasoning on a LLAMA-7B W4A4 model. In the GPTQ pipeline, SpikeLLM achieves direct additive in linear layers, significantly exceeding PB-LLMs. Our code is publicly available at https://github.com/Xingrun-Xing2/SpikeLLM.

Mixture-of-Experts (MoE) is widely adopted to deploy Large Language Models (LLMs) on edge devices with limited memory budgets. Although MoE is, in theory, an inborn memory-friendly architecture requiring only a few activated experts to reside in the memory for inference, current MoE architectures cannot effectively fulfill this advantage and will yield intolerable inference latencies of LLMs on memory-constrained devices. Our investigation pinpoints the essential cause as the remarkable temporal inconsistencies of inter-token expert activations, which generate overly frequent expert swapping demands dominating the latencies. To this end, we propose a novel MoE architecture, Oracle-MoE, to fulfill the real on-device potential of MoE-based LLMs. Oracle-MoE route tokens in a highly compact space suggested by attention scores, termed the oracle space, to effectively maintain the semantic locality across consecutive tokens to reduce expert activation variations, eliminating massive swapping demands. Theoretical analysis proves that Oracle-MoE is bound to provide routing decisions with better semantic locality and, therefore, better expert activation consistencies. Experiments on the pretrained GPT-2 architectures of different sizes (200M, 350M, 790M, and 2B) and downstream tasks demonstrate that without compromising task performance, our Oracle-MoE has achieved state-of-the-art inference speeds across varying memory budgets, revealing its substantial potential for LLM deployments in industry.

Large Language Models (LLMs) have demonstrated remarkable performance in real-world applications. However, adapting LLMs to novel tasks via fine-tuning often requires substantial training data and computational resources that are impractical in few-shot scenarios. Existing approaches, such as In-context learning and Parameter-Efficient Fine-Tuning (PEFT), face key limitations: In-context learning introduces additional inference computational overhead with limited performance gains, while PEFT models are prone to overfitting on the few demonstration examples. In this work, we reinterpret the forward pass of LLMs as an optimization process, a sequence of preconditioned gradient descent steps refining internal representations. Based on this connection, we propose Optimization-Inspired Few-Shot Adaptation (OFA), integrating a parameterization that learns preconditioners without introducing additional trainable parameters, and an objective that improves optimization efficiency by learning preconditioners based on a convergence bound, while simultaneously steering the optimization path toward the flat local minimum. Our method overcomes both issues of ICL-based and PEFT-based methods, and demonstrates superior performance over the existing methods on a variety of few-shot adaptation tasks in experiments.

Stress haunts people in modern society, which may cause severe health issues if left unattended. With social media becoming an integral part of daily life, leveraging social media to detect stress has gained increasing attention. While the majority of the work focuses on classifying stress states and stress categories, this study introduce a new task aimed at estimating more specific stressors (like exam, writing paper, etc.) through users' posts on social media. Unfortunately, the diversity of stressors with many different classes but a few examples per class, combined with the consistent arising of new stressors over time, hinders the machine understanding of stressors. To this end, we cast the stressor estimation problem within a practical scenario few-shot learning setting, and propose a novel meta-learning based stressor estimation framework that is enhanced by a meta-knowledge inheritance mechanism. This model can not only learn generic stressor context through meta-learning, but also has a good generalization ability to estimate new stressors with little labeled data. A fundamental breakthrough in our approach lies in the inclusion of the meta-knowledge inheritance mechanism, which equips our model with the ability to prevent catastrophic forgetting when adapting to new stressors. The experimental results show that our model achieves state-of-the-art performance compared with the baselines. Additionally, we construct a social media-based stressor estimation dataset that can help train artificial intelligence models to facilitate human well-being.

While machine learning algorithms hold promise for personalised medicine, their clinical adoption remains limited, partly due to biases that can compromise the reliability of predictions. In this article, we focus on sample selection bias (SSB), a specific type of bias where the study population is less representative of the target population, leading to biased and potentially harmful decisions. Despite being well-known in the literature, SSB remains scarcely studied in machine learning for healthcare. Moreover, the existing machine learning techniques try to correct the bias mostly by balancing distributions between the study and the target populations, which may result in a loss of predictive performance. To address these problems, our study illustrates the potential risks associated with SSB by examining SSB’s impact on the performance of machine learning algorithms. Most importantly, we propose a new research direction for addressing SSB, based on the target population identification rather than the bias correction. Specifically, we propose two independent networks (T-Net) and a multitasking network (MT-Net) for addressing SSB, where one network/task identifies the target subpopulation which is representative of the study population and the second makes predictions for the identified subpopulation. Our empirical results with synthetic and semi-synthetic datasets highlight that SSB can lead to a large drop in the performance of an algorithm for the target population as compared with the study population, as well as a substantial difference in the performance for the target subpopulations that are representative of the selected and the non-selected patients from the study population. Furthermore, our proposed techniques demonstrate robustness across various settings, including different dataset sizes, event rates and selection rates, outperforming the existing bias correction techniques.

OBJECTIVE: This study aims to tackle the critical challenge of adapting deep learning (DL) models for deployment in real-world healthcare settings, specifically focusing on catastrophic forgetting due to distribution shifts between hospital and non-hospital environments. Metabolic syndrome (MetS) is susceptible to misdiagnosis by DL models due to distribution shifts. This work demonstrates the potential of continual learning (CL) to enhance model performance in MetS identification across diverse settings. MATERIALS AND METHODS: We utilized 3 healthcare datasets from 2 distinct settings: Medical Information Mart for Intensive Care (MIMIC; from a hospital setting), National Health and Nutrition Examination Survey (NHANES), and a proprietary dataset (both from non-hospital settings). We proposed a robust MetS identification pipeline with CL strategies and evaluated its effectiveness in mitigating catastrophic forgetting while maintaining high predictive performance under distribution shift. RESULTS: The CL method outperformed the control (sequential training without any strategies) method. The CL method reached a cumulative area under the ROC curve (AUROC) of 0.85 and area under the precision-recall curve of 0.65 on the combined test set. Moreover, training order proved critical: models trained from hospital to non-hospital settings achieved a 7.6% improvement in AUROC, increasing from 0.79 to 0.85, compared to the reverse order. DISCUSSION: Our results demonstrate the potential of CL for applications across healthcare settings, particularly between hospital and non-hospital settings. We also discuss the impact of training order on the results. CONCLUSION: The proposed CL model effectively mitigates catastrophic forgetting, enhancing the overall performance of DL models. Our results underscore the prospect of CL methods in developing medical DL models and maintaining scalability across diverse healthcare settings.

Text-to-image generation (TTI) refers to the usage of models that could process text input and generate high fidelity images based on text descriptions. Text-to-image generation using neural networks could be traced back to the emergence of Generative Adversial Network (GAN), followed by the autoregressive Transformer. Diffusion models are one prominent type of generative model used for the generation of images through the systematic introduction of noises with repeating steps. As an effect of the impressive results of diffusion models on image synthesis, it has been cemented as the major image decoder used by text-to-image models and brought text-to-image generation to the forefront of machine-learning (ML) research. In the era of large models, scaling up model size and the integration with large language models have further improved the performance of TTI models, resulting the generation result nearly indistinguishable from real-world images, revolutionizing the way we retrieval images. Our explorative study has incentivised us to think that there are further ways of scaling text-to-image models with the combination of innovative model architectures and prediction enhancement techniques. We have divided the work of this survey into five main sections wherein we detail the frameworks of major literature in order to delve into the different types of text-to-image generation methods. Following this we provide a detailed comparison and critique of these methods and offer possible pathways of improvement for future work. In the future work, we argue that TTI development could yield impressive productivity improvements for creation, particularly in the context of the AIGC era, and could be extended to more complex tasks such as video generation and 3D generation.

Given radiology images, automatic radiology report generation aims to produce informative text that reports diseases. It can benefit current clinical practice in diagnostic radiology. Existing methods typically rely on large-scale medical datasets annotated by clinicians to train desirable models. However, for novel diseases, sufficient training data are typically not available. We propose a prompt-based deep learning framework, i.e., PromptLLM, to align, autoencode, and prompt the (large) language model to generate reports for novel diseases accurately and efficiently. Our method includes three major steps: 1) aligning visual images and textual reports to learn general knowledge across modalities from diseases where labeled data are sufficient, 2) autoencoding the LLM using unlabeled data of novel diseases to learn the specific knowledge and writing styles of the novel disease, and 3) prompting the LLM with learned knowledge and writing styles to report the novel diseases contained in the radiology images. Through the above three steps, with limited labels on novel diseases, we show that PromptLLM can rapidly learn the corresponding knowledge for accurate novel disease reporting. The experiments on COVID-19 and diverse thorax diseases show that our approach, utilizing 1% of the training data, achieves desirable performance compared to previous methods. It shows that our approach allows us to relax the reliance on labeled data that is common to existing methods. It could have a real-world impact on data analysis during the early stages of novel diseases.

Shared training approaches, such as multi-task learning (MTL) and gradient-based meta-learning, are widely used in various machine learning applications, but they often suffer from negative transfer, leading to performance degradation in specific tasks. While several optimisation techniques have been developed to mitigate this issue for pre-selected task cohorts, identifying optimal task combinations for joint learning-known as task grouping-remains underexplored and computationally challenging due to the exponential growth in task combinations and the need for extensive training and evaluation cycles. This paper introduces an efficient task grouping framework designed to reduce these overwhelming computational demands of the existing methods. The proposed framework infers pairwise task similarities through a sample-wise optimisation landscape analysis, eliminating the need for the shared model training required to infer task similarities in existing methods. With task similarities acquired, a graph-based clustering algorithm is employed to pinpoint near-optimal task groups, providing an approximate yet efficient and effective solution to the originally NP-hard problem. Empirical assessments conducted on 9 different datasets highlight the effectiveness of the proposed framework, revealing a five-fold speed enhancement compared to previous state-of-the-art methods. Moreover, the framework consistently demonstrates comparable performance, confirming its remarkable efficiency and effectiveness in task grouping.

Denoising-based diffusion models have attained impressive image synthesis; however, their applications on videos can lead to unaffordable computational costs due to the per-frame denoising operations. In pursuit of efficient video generation, we present a Diffusion Reuse MOtion (Dr. Mo) network to accelerate the video-based denoising process. Our crucial observation is that the latent representations in early denoising steps between adjacent video frames exhibit high consistencies with motion clues. Inspired by the discovery, we propose to accelerate the video denoising process by incorporating lightweight, learnable motion features. Specifically, Dr. Mo will only compute all denoising steps for base frames. For a non-based frame, Dr. Mo will propagate the pre-computed based latents of a particular step with inter-frame motions to obtain a fast estimation of its coarse-grained latent representation, from which the denoising will continue to obtain more sensitive and fine-grained representations. On top of this, Dr. Mo employs a meta-network named Denoising Step Selector (DSS) to dynamically determine the step to perform motion-based propagations for each frame, ensuring the correct transformation of multi-granularity visual features. Extensive evaluations on video generation and editing tasks indicate that Dr. Mo delivers widely applicable acceleration for diffusion-based video generations while effectively retaining the visual quality and style. Video generation and visualization results can be found athttps://drmo-denoising-reuse.github.io.

Over the past few decades, osmotherapy has commonly been employed to reduce intracranial pressure in post-stroke oedema. However, evaluating the effectiveness of osmotherapy has been challenging due to the difficulties in clinical intracranial pressure measurement. As a result, there are no established guidelines regarding the selection of administration protocol parameters. Considering that the infusion of osmotic agents can also give rise to various side effects, the effectiveness of osmotherapy has remained a subject of debate. In previous studies, we proposed the first mathematical model for the investigation of osmotherapy and validated the model with clinical intracranial pressure data. The physiological parameters vary among patients and such variations can result in the failure of osmotherapy. Here, we propose an AI-assisted in silico trial for further investigation of the optimisation of administration protocols. The proposed deep neural network predicts intracranial pressure evolution over osmotherapy episodes. The effects of the parameters and the choice of dose of osmotic agents are investigated using the model. In addition, clinical stratifications of patients are related to a brain model for the first time for the optimisation of treatment of different patient groups. This provides an alternative approach to tackle clinical challenges with in silico trials supported by both mathematical/physical laws and patient-specific biomedical information.

Wearable cuffless blood pressure (BP) technology is emerging as a critical tool for monitoring hypertension, the leading risk factor of most cardiovascular diseases. However, current cuffless BP methods are not accurate enough for clinical use, because they mainly use single or dual modalities/features as inputs for estimation. To address this challenge, we propose multimodal McBP-Net, built with hybrid CNN-LSTM architecture combing two-layer convolution operations with four-layer LSTMs to capture both local signal features and temporal dependencies for continuous dynamic beat-to-beat BP estimation. The McBP-Net includes photoplethysmographic, electrocardiographic, impedanceplethysmographic (IPG), and skin temperature (ST) signals as inputs. Validated on 23 subjects undergoing cold pressor test to induce large BP variability, the McBP-Net achieves the mean absolute errors of 4.19 and 2.98 mmHg for systolic BP (SBP) and diastolic BP (DBP), respectively, which fall within the accuracy range required by the Grade A of IEEE standard. The integration of four multimodal signals improves performance by 16.20%, 37.37%, and 49.52% over three-, dual-, and single-modality approaches, respectively, with significant contributions from IPG and ST signals. Notably, ST shows a strong nonlinear relationship with BP with high mutual information of 0.9056 for SBP. Furthermore, McBP-Net achieves a reasonable balance between accuracy and computational efficiency, offering inference speed of 36.7% faster and reducing computational demands by 78% compared to transformer-based models tested. Importantly, it maintains robust performance, with only a 0.21 mmHg degradation in dynamic SBP estimation when trained on rest-stage data. McBP-Net demonstrates promising potential in medical-grade wearable cuffless dynamic BP measurements.

Estimation of individualized treatment effects (ITE) from observational studies is a fundamental problem in causal inference and holds significant importance across domains, including healthcare. However, limited observational datasets pose challenges in reliable ITE estimation as data have to be split among treatment groups to train an ITE learner. While information sharing among treatment groups can partially alleviate the problem, there is currently no general framework for end-to-end information sharing in ITE estimation. To tackle this problem, we propose a deep learning framework based on ‘\textit{soft weight sharing}’ to train ITE learners, enabling \textit{dynamic end-to-end} information sharing among treatment groups. The proposed framework complements existing ITE learners, and introduces a new class of ITE learners, referred to as \textit{HyperITE}. We extend state-of-the-art ITE learners with \textit{HyperITE} versions and evaluate them on IHDP, ACIC-2016, and Twins benchmarks. Our experimental results show that the proposed framework improves ITE estimation error, with increasing effectiveness for smaller datasets.

The proliferation of decentralised electronic healthcare records (EHRs) across medical institutions requires innovative federated learning strategies for collaborative data analysis and global model training, prioritising data privacy. A prevalent issue during decentralised model training is the data-view discrepancies across medical institutions that arises from differences or availability of healthcare services, such as blood test panels. The prevailing way to handle this issue is to select a common subset of features across institutions to make data-views consistent. This approach, however, constrains some institutions to shed some critical features that may play a significant role in improving the model performance. This paper introduces a federated learning framework that relies on augmented graph attention networks to address data-view heterogeneity. The proposed framework utilises an alignment augmentation layer over self-attention mechanisms to weigh the importance of neighbouring nodes when updating a node’s embedding irrespective of the data-views. Furthermore, our framework adeptly addresses both the temporal nuances and structural intricacies of EHR datasets. This dual capability not only offers deeper insights but also effectively encapsulates EHR graphs’ time-evolving nature. Using diverse real-world datasets, we show that the proposed framework significantly outperforms conventional FL methodology for dealing with heterogeneous data-views.

Fenglin Liu, Zheng Li, Hongjian Zhou, Qingyu Yin, Jingfeng Yang, Xianfeng Tang, Chen Luo, Ming Zeng, Haoming Jiang, Yifan Gao, Priyanka Nigam, Sreyashi Nag, Bing Yin, Yining Hua, Xuan Zhou, Omid Rohanian, Anshul Thakur, Lei Clifton, David A. Clifton. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Utilizing large pre-trained models for specific tasks has yielded impressive results. However, fully fine-tuning these increasingly large models is becoming prohibitively resource-intensive. This has led to a focus on more parameter-efficient transfer learning, primarily within the same modality. But this approach has limitations, particularly in video understanding where suitable pre-trained models are less common. Addressing this, our study introduces a novel cross-modality transfer learning approach from images to videos, which we call parameter-efficient image-to-video transfer learning. We present the Facial-Emotion Adapter (FE-Adapter), designed for efficient fine-tuning in video tasks. This adapter allows pre-trained image models, which traditionally lack temporal processing capabilities, to analyze dynamic video content efficiently. Notably, it uses about 15 times fewer parameters than previous methods, while improving accuracy. Our experiments in video emotion recognition demonstrate that the FE-Adapter can match or even surpass existing fine-tuning and video emotion models in both performance and efficiency. This breakthrough highlights the potential for cross-modality approaches in enhancing the capabilities of AI models, particularly in fields like video emotion analysis where the demand for efficiency and accuracy is constantly rising.

.

The entry of large language models (LLMs) into research and commercial spaces has led to a trend of ever-larger models, with initial promises of generalisability. This was followed by a widespread desire to downsize and create specialised models without the need for complete fine-tuning, using Parameter Efficient Fine-tuning (PEFT) methods. We present an investigation into the suitability of different PEFT methods to clinical decision-making tasks, across a range of model sizes, including extremely small models with as few as 25 million parameters. Our analysis shows that the performance of most PEFT approaches varies significantly from one task to another, with the exception of LoRA, which maintains relatively high performance across all model sizes and tasks, typically approaching or matching full fine-tuned performance. The effectiveness of PEFT methods in the clinical domain is evident, particularly for specialised models which can operate on low-cost, in-house computing infrastructure. The advantages of these models, in terms of speed and reduced training costs, dramatically outweighs any performance gain from large foundation LLMs. Furthermore, we highlight how domain-specific pre-training interacts with PEFT methods and model size, finding the domain pre-training to be particularly important in smaller models and discuss how these factors interplay to provide the best efficiency-performance trade-off. Full code available at: https://github.com/nlpie-research/efficient-ml . • State of the art performance in Clinical NLP using efficient fine-tuning methods. • 25 million parameter LLMs benefit from LoRA fine-tuning. • Classification performance matched with 98% fewer trained parameters. • Trade-off in performance in tiny LLMs outweighs cost of much larger Language Models.

This research introduces the Variational Graph Attention Dynamics (VarGATDyn), addressing the complexities of dynamic graph representation learning, where existing models, tailored for static graphs, prove inadequate. VarGATDyn melds attention mechanisms with a Markovian assumption to surpass the challenges of maintaining temporal consistency and the extensive dataset requirements typical of RNN-based frameworks. It harnesses the strengths of the Variational Graph Auto-Encoder (VGAE) framework, Graph Attention Networks (GAT), and Gaussian Mixture Models (GMM) to adeptly navigate the temporal and structural intricacies of dynamic graphs. Through the strategic application of GMMs, the model handles multimodal patterns, thereby rectifying misalignments between prior and estimated posterior distributions. An innovative multiple-learning methodology bolsters the model's adaptability, leading to an encompassing and effective learning process. Empirical tests underscore VarGATDyn's dominance in dynamic link prediction across various datasets, highlighting its proficiency in capturing multimodal distributions and temporal dynamics.

With the rapid growth of memory and computing power, datasets are becoming increasingly complex and imbalanced. This is especially severe in the context of clinical data, where there may be one rare event for many cases in the majority class. We introduce an imbalanced classification framework, based on reinforcement learning, for training extremely imbalanced data sets, and extend it for use in multi-class settings. We combine dueling and double deep Q-learning architectures, and formulate a custom reward function and episode-training procedure, specifically with the capability of handling multi-class imbalanced training. Using real-world clinical case studies, we demonstrate that our proposed framework outperforms current state-of-the-art imbalanced learning methods, achieving more fair and balanced classification, while also significantly improving the prediction of minority classes. Supplementary Information: The online version contains supplementary material available at 10.1007/s10994-023-06481-z.

Abstract Specialised pre-trained language models are becoming more frequent in Natural language Processing (NLP) since they can potentially outperform models trained on generic texts. BioBERT (Sanh et al., Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv: 1910.01108, 2019) and BioClinicalBERT (Alsentzer et al., Publicly available clinical bert embeddings. In Proceedings of the 2nd Clinical Natural Language Processing Workshop, pp. 72–78, 2019) are two examples of such models that have shown promise in medical NLP tasks. Many of these models are overparametrised and resource-intensive, but thanks to techniques like knowledge distillation, it is possible to create smaller versions that perform almost as well as their larger counterparts. In this work, we specifically focus on development of compact language models for processing clinical texts (i.e. progress notes, discharge summaries, etc). We developed a number of efficient lightweight clinical transformers using knowledge distillation and continual learning, with the number of parameters ranging from $15$ million to $65$ million. These models performed comparably to larger models such as BioBERT and ClinicalBioBERT and significantly outperformed other compact models trained on general or biomedical data. Our extensive evaluation was done across several standard datasets and covered a wide range of clinical text-mining tasks, including natural language inference, relation extraction, named entity recognition and sequence classification. To our knowledge, this is the first comprehensive study specifically focused on creating efficient and compact transformers for clinical NLP tasks. The models and code used in this study can be found on our Huggingface profile at https://huggingface.co/nlpie and Github page at https://github.com/nlpie-research/Lightweight-Clinical-Transformers , respectively, promoting reproducibility of our results.

The design of neural networks typically involves trial-and-error, a time-consuming process for obtaining an optimal architecture, even for experienced researchers. Additionally, it is widely accepted that loss functions of deep neural networks are generally non-convex with respect to the parameters to be optimised. We propose the Layer-wise Convex Theorem to ensure that the loss is convex with respect to the parameters of a given layer, achieved by constraining each layer to be an overdetermined system of non-linear equations. Based on this theorem, we developed an end-to-end algorithm (the AutoNet) to automatically generate layer-wise convex networks (LCNs) for any given training set. We then demonstrate the performance of the AutoNet-generated LCNs (AutoNet-LCNs) compared to state-of-the-art models on three electrocardiogram (ECG) classification benchmark datasets, with further validation on two non-ECG benchmark datasets for more general tasks. The AutoNet-LCN was able to find networks customised for each dataset without manual fine-tuning under 2 GPU-hours, and the resulting networks outperformed the state-of-the-art models with fewer than 5% parameters on all the above five benchmark datasets. The efficiency and robustness of the AutoNet-LCN markedly reduce model discovery costs and enable efficient training of deep learning models in resource-constrained settings.

Natural Language Generation (NLG) accepts input data in the form of images, videos, or text and generates corresponding natural language text as output. Existing NLG methods mainly adopt a supervised approach and rely heavily on coupled data-to-text pairs. However, for many targeted scenarios and for non-English languages, sufficient quantities of labeled data are often not available. As a result, it is necessary to collect and label data-text pairs for training, which is both costly and time-consuming. To relax the dependency on labeled data of downstream tasks, we propose an intuitive and effective zero-shot learning framework, ZeroNLG, which can deal with multiple NLG tasks, including image-to-text (image captioning), video-to-text (video captioning), and text-to-text (neural machine translation), across English, Chinese, German, and French within a unified framework. ZeroNLG does not require any labeled downstream pairs for training. During training, ZeroNLG (i) projects different domains (across modalities and languages) to corresponding coordinates in a shared common latent space; (ii) bridges different domains by aligning their corresponding coordinates in this space; and (iii) builds an unsupervised multilingual auto-encoder to learn to generate text by reconstructing the input text given its coordinate in shared latent space. Consequently, during inference, based on the data-to-text pipeline, ZeroNLG can generate target sentences across different languages given the coordinate of input data in the common space. Within this unified framework, given visual (imaging or video) data as input, ZeroNLG can perform zero-shot visual captioning; given textual sentences as input, ZeroNLG can perform zero-shot machine translation. We present the results of extensive experiments on twelve NLG tasks, showing that, without using any labeled downstream pairs for training, ZeroNLG generates high-quality and "believable" outputs and significantly outperforms existing zero-shot methods.

Recent studies on contrastive learning have achieved remarkable performance solely by leveraging few labels in the context of medical image segmentation. Existing methods mainly focus on instance discrimination and invariant mapping (i.e., pulling positive samples closer and negative samples apart in the feature space). However, they face three common pitfalls: (1) tailness: medical image data usually follows an implicit long-tail class distribution. Blindly leveraging all pixels in training hence can lead to the data imbalance issues, and cause deteriorated performance; (2) consistency: it remains unclear whether a segmentation model has learned meaningful and yet consistent anatomical features due to the intra-class variations between different anatomical features; and (3) diversity: the intra-slice correlations within the entire dataset have received significantly less attention. This motivates us to seek a principled approach for strategically making use of the dataset itself to discover similar yet distinct samples from different anatomical views. In this paper, we introduce a novel semi-supervised 2D medical image segmentation framework termed Mine yOur owNAnatomy (MONA), and make three contributions. First, prior work argues that every pixel equally matters to the model training; we observe empirically that this alone is unlikely to define meaningful anatomical features, mainly due to lacking the supervision signal. We show two simple solutions towards learning invariances-through the use of stronger data augmentations and nearest neighbors. Second, we construct a set of objectives that encourage the model to be capable of decomposing medical images into a collection of anatomical features in an unsupervised manner. Lastly, we both empirically and theoretically, demonstrate the efficacy of our MONA on three benchmark datasets with different labeled settings, achieving new state-of-the-art under different labeled semi-supervised settings. MONA makes minimal assumptions on domain expertise, and hence constitutes a practical and versatile solution in medical image analysis. We provide the PyTorch-like pseudo-code in supplementary.

Electrocardiography (ECG) is a non-invasive tool for predicting cardiovascular diseases (CVDs). Current ECG-based diagnosis systems show promising performance owing to the rapid development of deep learning techniques. However, the label scarcity problem, the co-occurrence of multiple CVDs and the poor performance on unseen datasets greatly hinder the widespread application of deep learning-based models. Addressing them in a unified framework remains a significant challenge. To this end, we propose a multi-label semi-supervised model (ECGMatch) to recognize multiple CVDs simultaneously with limited supervision. In the ECGMatch, an ECGAugment module is developed for weak and strong ECG data augmentation, which generates diverse samples for model training. Subsequently, a hyperparameter-efficient framework with neighbor agreement modeling and knowledge distillation is designed for pseudo-label generation and refinement, which mitigates the label scarcity problem. Finally, a label correlation alignment module is proposed to capture the co-occurrence information of different CVDs within labeled samples and propagate this information to unlabeled samples. Extensive experiments on four datasets and three protocols demonstrate the effectiveness and stability of the proposed model, especially on unseen datasets. As such, this model can pave the way for diagnostic systems that achieve robust performance on multi-label CVDs prediction with limited supervision.

Recent advances in machine learning, particularly deep neural network architectures, have shown substantial promise in classifying and predicting cardiac abnormalities from electrocardiogram (ECG) data. Such data are rich in information content, typically in morphology and timing, due to the close correlation between cardiac function and the ECG. However, the ECG is usually not measured ubiquitously in a passive manner from consumer devices, and generally requires 'active' sampling whereby the user prompts a device to take an ECG measurement. Conversely, photoplethysmography (PPG) data are typically measured passively by consumer devices, and therefore available for long-period monitoring and suitable in duration for identifying transient cardiac events. However, classifying or predicting cardiac abnormalities from the PPG is very difficult, because it is a peripherally-measured signal. Hence, the use of the PPG for predictive inference is often limited to deriving physiological parameters (heart rate, breathing rate, etc.) or for obvious abnormalities in cardiac timing, such as atrial fibrillation/flutter ("palpitations"). This work aims to combine the best of both worlds: using continuously-monitored, near-ubiquitous PPG to identify periods of sufficient abnormality in the PPG such that prompting the user to take an ECG would be informative of cardiac risk. We propose a dual-convolutional-attention network (DCA-Net) to achieve this ECG-based PPG classification. With DCA-Net, we prove the plausibility of this concept on MIMIC Waveform Database with high performance level (AUROC 0.9 and AUPRC 0.7) and receive satisfactory result when testing the model on an independent dataset (AUROC 0.7 and AUPRC 0.6) which it is not perfectly-matched to the MIMIC dataset.

This article explores the utilization of the effective degree-of-freedom (DoF) of a deep learning model to regularize its stochastic gradient descent (SGD)-based training. The effective DoF of a deep learning model is defined only by a subset of its total parameters. This subset is highly responsive or sensitive toward the training loss, and its cardinality can be used to govern the effective DoF of a model during training. To this aim, the incremental trainable parameter selection (ITPS) algorithm is introduced in this article. The proposed ITPS algorithm acts as a wrapper over SGD and incrementally selects the parameters for updation that exhibit the maximum sensitivity toward the training loss. Hence, it gradually increases the DoF of the model during training. In ideal cases, the proposed algorithm arrives at a model configuration (i.e., DoF) optimum for the task at hand. This whole process results in a regularization-like behavior induced by a gradual increment of the DoF. Since the selection and updation of parameters is a function of the training loss, the proposed algorithm can be seen as a task and data-dependent regularization mechanism. This article exhibits the general utility of ITPS by evaluating it on various prominent neural network architectures such as CNNs, transformers, recurrent neural networks (RNNs), and multilayer perceptrons. These models are trained for image classification and healthcare tasks using the publicly available CIFAR-10, SLT-10, and MIMIC-III datasets.

Observational studies have recently received significant attention from the machine learning community due to the increasingly available non-experimental observational data and the limitations of the experimental studies, such as considerable cost, impracticality, small and less representative sample sizes, etc. In observational studies, de-confounding is a fundamental problem of individualised treatment effects (ITE) estimation. This paper proposes disentangled representations with adversarial training to selectively balance the confounders in the binary treatment setting for the ITE estimation. The adversarial training of treatment policy selectively encourages treatment-agnostic balanced representations for the confounders and helps to estimate the ITE in the observational studies via counterfactual inference. Empirical results on synthetic and real-world datasets, with varying degrees of confounding, prove that our proposed approach improves the state-of-the-art methods in achieving lower error in the ITE estimation.

Pre-trained Language Models (LMs) have become an integral part of Natural Language Processing (NLP) in recent years, due to their superior performance in downstream applications.In spite of this resounding success, the usability of LMs is constrained by computational and time complexity, along with their increasing size; an issue that has been referred to as 'overparameterisation'.Different strategies have been proposed in the literature to alleviate these problems, with the aim to create effective compact models that nearly match the performance of their bloated counterparts with negligible performance losses.One of the most popular techniques in this area of research is model distillation.Another potent but underutilised technique is cross-layer parameter sharing.In this work, we combine these two strategies and present MiniALBERT, a technique for converting the knowledge of fully parameterised LMs (such as BERT) into a compact recursive student.In addition, we investigate the application of bottleneck adapters for layer-wise adaptation of our recursive student, and also explore the efficacy of adapter tuning for fine-tuning of compact models.We test our proposed models on a number of general and biomedical NLP tasks to demonstrate their viability and compare them with the state-of-the-art and other existing compact models.

Mutual knowledge distillation (MKD) is a technique used to transfer knowledge between multiple models in a collaborative manner. However, it is important to note that not all knowledge is accurate or reliable, particularly under challenging conditions such as label noise, which can lead to models that memorize undesired information. This problem can be addressed by improving the reliability of the knowledge source, as well as selectively selecting reliable knowledge for distillation. While making a model more reliable is a widely studied topic, selective MKD has received less attention. To address this, we propose a new framework called selective mutual knowledge distillation (SMKD). The key component of SMKD is a generic knowledge selection formulation, which allows for either static or progressive selection thresholds. Additionally, SMKD covers two special cases: using no knowledge and using all knowledge, resulting in a unified MKD framework. We present extensive experimental results to demonstrate the effectiveness of SMKD and justify its design.

For medical image segmentation, contrastive learning is the dominant practice to improve the quality of visual representations by contrasting semantically similar and dissimilar pairs of samples. This is enabled by the observation that without accessing ground truth labels, negative examples with truly dissimilar anatomical features, if sampled, can significantly improve the performance. In reality, however, these samples may come from similar anatomical features and the models may struggle to distinguish the minority tail-class samples, making the tail classes more prone to misclassification, both of which typically lead to model collapse. In this paper, we propose $\texttt{ARCO}$, a semi-supervised contrastive learning (CL) framework with stratified group theory for medical image segmentation. In particular, we first propose building $\texttt{ARCO}$ through the concept of variance-reduced estimation, and show that certain variance-reduction techniques are particularly beneficial in pixel/voxel-level segmentation tasks with extremely limited labels. Furthermore, we theoretically prove these sampling techniques are universal in variance reduction. Finally, we experimentally validate our approaches on eight benchmarks, i.e., five 2D/3D medical and three semantic segmentation datasets, with different label settings, and our methods consistently outperform state-of-the-art semi-supervised methods. Additionally, we augment the CL frameworks with these sampling techniques and demonstrate significant gains over previous methods. We believe our work is an important step towards semi-supervised medical image segmentation by quantifying the limitation of current self-supervision objectives for accomplishing such challenging safety-critical tasks.

MOTIVATION: Language models pre-trained on biomedical corpora, such as BioBERT, have recently shown promising results on downstream biomedical tasks. Many existing pre-trained models, on the other hand, are resource-intensive and computationally heavy owing to factors such as embedding size, hidden dimension and number of layers. The natural language processing community has developed numerous strategies to compress these models utilizing techniques such as pruning, quantization and knowledge distillation, resulting in models that are considerably faster, smaller and subsequently easier to use in practice. By the same token, in this article, we introduce six lightweight models, namely, BioDistilBERT, BioTinyBERT, BioMobileBERT, DistilBioBERT, TinyBioBERT and CompactBioBERT which are obtained either by knowledge distillation from a biomedical teacher or continual learning on the Pubmed dataset. We evaluate all of our models on three biomedical tasks and compare them with BioBERT-v1.1 to create the best efficient lightweight models that perform on par with their larger counterparts. RESULTS: We trained six different models in total, with the largest model having 65 million in parameters and the smallest having 15 million; a far lower range of parameters compared with BioBERT's 110M. Based on our experiments on three different biomedical tasks, we found that models distilled from a biomedical teacher and models that have been additionally pre-trained on the PubMed dataset can retain up to 98.8% and 98.6% of the performance of the BioBERT-v1.1, respectively. Overall, our best model below 30 M parameters is BioMobileBERT, while our best models over 30 M parameters are DistilBioBERT and CompactBioBERT, which can keep up to 98.2% and 98.8% of the performance of the BioBERT-v1.1, respectively. AVAILABILITY AND IMPLEMENTATION: Codes are available at: https://github.com/nlpie-research/Compact-Biomedical-Transformers. Trained models can be accessed at: https://huggingface.co/nlpie.

Transformer is a promising neural network learner, and has achieved great success in various machine learning tasks. Thanks to the recent prevalence of multimodal applications and Big Data, Transformer-based multimodal learning has become a hot topic in AI research. This paper presents a comprehensive survey of Transformer techniques oriented at multimodal data. The main contents of this survey include: (1) a background of multimodal learning, Transformer ecosystem, and the multimodal Big Data era, (2) a systematic review of Vanilla Transformer, Vision Transformer, and multimodal Transformers, from a geometrically topological perspective, (3) a review of multimodal Transformer applications, via two important paradigms, i.e., for multimodal pretraining and for specific multimodal tasks, (4) a summary of the common challenges and designs shared by the multimodal Transformer models and applications, and (5) a discussion of open problems and potential research directions for the community.

Performance in an engineering system tends to degrade over time due to a variety of wearing or ageing processes. In supervisory controlled processes there are typically many signals being monitored that may help to characterize performance degradation. It is preferred to select the least amount of information to obtain high quality of predictive analysis from a large amount of collected data, in which labeling the data is not always feasible. To this end a novel unsupervised feature selection method, robust with respect to significant measurement disturbances, is proposed using the notion of "weak monotonicity" (WM). The robustness of this notion makes it very attractive to identify the common trend in the presence of measurement noises and population variation from the collected data. Based on WM, a novel suitability indicator is proposed to evaluate the performance of each feature. This new indicator is then used to select the key features that contribute to the WM of a family of processes when noises and variations among processes exist. In order to evaluate the performance of the proposed framework of the WM and suitability, a comparative study with other nine state-of-the-arts unsupervised feature evaluation and selection methods is carried out on well-known benchmark datasets. The results show a promising performance of the proposed framework on unsupervised feature evaluation in the presence of measurement noises and population variations.

Early detection of COVID-19 is an ongoing area of research that can help with triage, monitoring and general health assessment of potential patients and may reduce operational strain on hospitals that cope with the coronavirus pandemic. Different machine learning techniques have been used in the literature to detect potential cases of coronavirus using routine clinical data (blood tests, and vital signs measurements). Data breaches and information leakage when using these models can bring reputational damage and cause legal issues for hospitals. In spite of this, protecting healthcare models against leakage of potentially sensitive information is an understudied research area. In this study, two machine learning techniques that aim to predict a patient's COVID-19 status are examined. Using adversarial training, robust deep learning architectures are explored with the aim to protect attributes related to demographic information about the patients. The two models examined in this work are intended to preserve sensitive information against adversarial attacks and information leakage. In a series of experiments using datasets from the Oxford University Hospitals (OUH), Bedfordshire Hospitals NHS Foundation Trust (BH), University Hospitals Birmingham NHS Foundation Trust (UHB), and Portsmouth Hospitals University NHS Trust (PUH), two neural networks are trained and evaluated. These networks predict PCR test results using information from basic laboratory blood tests, and vital signs collected from a patient upon arrival to the hospital. The level of privacy each one of the models can provide is assessed and the efficacy and robustness of the proposed architectures are compared with a relevant baseline. One of the main contributions in this work is the particular focus on the development of effective COVID-19 detection models with built-in mechanisms in order to selectively protect sensitive attributes against adversarial attacks. The results on hold-out test set and external validation confirmed that there was no impact on the generalisibility of the model using adversarial learning.

Healthcare is dynamic as demographics, diseases, and therapeutics constantly evolve. This dynamic nature induces inevitable distribution shifts in populations targeted by clinical AI models, often rendering them ineffective. Incremental learning provides an effective method of adapting deployed clinical models to accommodate these contemporary distribution shifts. However, since incremental learning involves modifying a deployed or in-use model, it can be considered unreliable as any adverse modification due to maliciously compromised or incorrectly labelled data can make the model unsuitable for the targeted application. This paper introduces self-aware stochastic gradient descent (SGD), an incremental deep learning algorithm that utilises a contextual bandit-like sanity check to only allow reliable modifications to a model. The contextual bandit analyses incremental gradient updates to isolate and filter unreliable gradients. This behaviour allows self-aware SGD to balance incremental training and integrity of a deployed model. Experimental evaluations on the Oxford University Hospital datasets highlight that self-aware SGD can provide reliable incremental updates for overcoming distribution shifts in challenging conditions induced by label noise.

Clinical settings are often characterized by abundant unlabelled data and limited labelled data. This is typically driven by the high burden placed on oracles (e.g., physicians) to provide annotations. One way to mitigate this burden is via active learning (AL) which involves the (a) acquisition and (b) annotation of informative unlabelled instances. Whereas previous work addresses either one of these elements independently, we propose an AL framework that addresses both. For acquisition, we propose Bayesian Active Learning by Consistency (BALC), a sub-framework which perturbs both instances and network parameters and quantifies changes in the network output probability distribution. For annotation, we propose SoQal, a sub-framework that dynamically determines whether, for each acquired unlabelled instance, to request a label from an oracle or to pseudo-label it instead. We show that BALC can outperform start-of-the-art acquisition functions such as BALD, and SoQal outperforms baseline methods even in the presence of a noisy oracle.

Most video-and-language representation learning approaches employ contrastive learning, e.g., CLIP, to project the video and text features into a common latent space according to the semantic similarities of text-video pairs. However, such learned shared latent spaces are not often optimal, and the modality gap between visual and textual representation can not be fully eliminated. In this paper, we propose Expectation-Maximization Contrastive Learning (EMCL) to learn compact video-and-language representations. Specifically, we use the Expectation-Maximization algorithm to find a compact set of bases for the latent space, where the features could be concisely represented as the linear combinations of these bases. Such feature decomposition of video-and-language representations reduces the rank of the latent space, resulting in increased representing power for the semantics. Extensive experiments on three benchmark text-video retrieval datasets prove that our EMCL can learn more discriminative video-and-language representations than previous methods, and significantly outperform previous state-of-the-art methods across all metrics. More encouragingly, the proposed method can be applied to boost the performance of existing approaches either as a jointly training layer or an out-of-the-box inference module with no extra training, making it easy to be incorporated into any existing methods.

The "Patient Instruction" (PI), which contains critical instructional information provided both to carers and to the patient at the time of discharge, is essential for the patient to manage their condition outside hospital. An accurate and easy-to-follow PI can improve the self-management of patients which can in turn reduce hospital readmission rates. However, writing an appropriate PI can be extremely time consuming for physicians, and is subject to being incomplete or error-prone for (potentially overworked) physicians. Therefore, we propose a new task that can provide an objective means of avoiding incompleteness, while reducing clinical workload: the automatic generation of the PI, which is imagined as being a document that the clinician can review, modify, and approve as necessary (rather than taking the human "out of the loop"). We build a benchmark clinical dataset and propose the Re$^3$Writer, which imitates the working patterns of physicians to first retrieve related working experience from historical PIs written by physicians, then reason related medical knowledge. Finally, it refines the retrieved working experience and reasoned medical knowledge to extract useful information, which is used to generate the PI for previously-unseen patient according to their health records during hospitalization. Our experiments show that, using our method, the performance of 6 different models can be substantially boosted across all metrics, with up to 20%, 11%, and 19% relative improvements in BLEU-4, ROUGE-L, and METEOR, respectively. Meanwhile, we show results from human evaluations to measure the effectiveness in terms of its usefulness for clinical practice. The code is available at https://github.com/AI-in-Health/Patient-Instructions.

Clinical time-series data retrieved from electronic medical records are widely used to build predictive models of adverse events to support resource management. Such data is often sparse and irregularly-sampled, which makes it challenging to use many common machine learning methods. Missing values may be interpolated by carrying the last value forward, or through linear regression. Gaussian process (GP) regression is also used for performing imputation, and often re-sampling of time-series at regular intervals. The use of GPs can require extensive, and likely adhoc, investigation to determine model structure, such as an appropriate covariance function. This can be challenging for multivariate real-world clinical data, in which time-series variables exhibit different dynamics to one another. In this work, we construct generative models to estimate missing values in clinical time-series data using a neural latent variable model, known as a Neural Process (NP). The NP model employs a conditional prior distribution in the latent space to learn global uncertainty in the data by modelling variations at a local level. In contrast to conventional generative modelling, this prior is not fixed and is itself learned during the training process. Thus, NP model provides the flexibility to adapt to the dynamics of the available clinical data. We propose a variant of the NP framework for efficient modelling of the mutual information between the latent and input spaces, ensuring meaningful learned priors. Experiments using the MIMIC III dataset demonstrate the effectiveness of the proposed approach as compared to conventional methods.

AI healthcare applications rely on sensitive electronic healthcare records (EHRs) that are scarcely labelled and are often distributed across a network of the symbiont institutions. It is challenging to train the effective machine learning models on such data. In this work, we propose dynamic neural graphs based federated learning framework to address these challenges. The proposed framework extends Reptile, a model agnostic meta-learning (MAML) algorithm, to a federated setting. However, unlike the existing MAML algorithms, this paper proposes a dynamic variant of neural graph learning (NGL) to incorporate unlabelled examples in the supervised training setup. Dynamic NGL computes a meta-learning update by performing supervised learning on a labelled training example while performing metric learning on its labelled or unlabelled neighbourhood. This neighbourhood of a labelled example is established dynamically using local graphs built over the batches of training examples. Each local graph is constructed by comparing the similarity between embedding generated by the current state of the model. The introduction of metric learning on the neighbourhood makes this framework semi-supervised in nature. The experimental results on the publicly available MIMIC-III dataset highlight the effectiveness of the proposed framework for both single and multi-task settings under data decentralisation constraints and limited supervision.

To train robust deep neural networks (DNNs), we systematically study several target modification approaches, which include output regularisation, self and non-self label correction (LC). Two key issues are discovered: (1) Self LC is the most appealing as it exploits its own knowledge and requires no extra models. However, how to automatically decide the trust degree of a learner as training goes is not well answered in the literature? (2) Some methods penalise while the others reward low-entropy predictions, prompting us to ask which one is better?To resolve the first issue, taking two well-accepted propositions–deep neural networks learn meaningful patterns before fitting noise [3] and minimum entropy regularisation principle [10]–we propose a novel end-to-end method named ProSelfLC, which is designed according to learning time and entropy. Specifically, given a data point, we progressively increase trust in its predicted label distribution versus its annotated one if a model has been trained for enough time and the prediction is of low entropy (high confidence). For the second issue, according to ProSelfLC, we empirically prove that it is better to redefine a meaningful low-entropy status and optimise the learner toward it. This serves as a defence of entropy minimisation.We demonstrate the effectiveness of ProSelfLC through extensive experiments in both clean and noisy settings. The source code is available at https://github.com/XinshaoAmosWang/ProSelfLC-CVPR2021.

The healthcare industry generates troves of unlabelled physiological data. This data can be exploited via contrastive learning, a self-supervised pre-training method that encourages representations of instances to be similar to one another. We propose a family of contrastive learning methods, CLOCS, that encourages representations across space, time, \textit{and} patients to be similar to one another. We show that CLOCS consistently outperforms the state-of-the-art methods, BYOL and SimCLR, when performing a linear evaluation of, and fine-tuning on, downstream tasks. We also show that CLOCS achieves strong generalization performance with only 25% of labelled training data. Furthermore, our training procedure naturally generates patient-specific representations that can be used to quantify patient-similarity.

The process of manually searching for relevant instances in, and extracting information from, clinical databases underpin a multitude of clinical tasks. Such tasks include disease diagnosis, clinical trial recruitment, and continuing medical education. This manual search-and-extract process, however, has been hampered by the growth of large-scale clinical databases and the increased prevalence of unlabelled instances. To address this challenge, we propose a supervised contrastive learning framework, CROCS, where representations of cardiac signals associated with a set of patient-specific attributes (e.g., disease class, sex, age) are attracted to learnable embeddings entitled clinical prototypes. We exploit such prototypes for both the clustering and retrieval of unlabelled cardiac signals based on multiple patient attributes. We show that CROCS outperforms the state-of-the-art method, DTC, when clustering and also retrieves relevant cardiac signals from a large database. We also show that clinical prototypes adopt a semantically meaningful arrangement based on patient attributes and thus confer a high degree of interpretability.

Antimicrobial resistance (AMR) poses a threat to global public health. To mitigate the impacts of AMR, it is important to identify the molecular mechanisms of AMR and thereby determine optimal therapy as early as possible. Conventional machine learning-based drug-resistance analyses assume genetic variations to be homogeneous, thus not distinguishing between coding and intergenic sequences. In this study, we represent genetic data from Mycobacterium tuberculosis as a graph, and then adopt a deep graph learning method-heterogeneous graph attention network ('HGAT-AMR')-to predict anti-tuberculosis (TB) drug resistance. The HGAT-AMR model is able to accommodate incomplete phenotypic profiles, as well as provide 'attention scores' of genes and single nucleotide polymorphisms (SNPs) both at a population level and for individual samples. These scores encode the inputs, which the model is 'paying attention to' in making its drug resistance predictions. The results show that the proposed model generated the best area under the receiver operating characteristic (AUROC) for isoniazid and rifampicin (98.53 and 99.10%), the best sensitivity for three first-line drugs (94.91% for isoniazid, 96.60% for ethambutol and 90.63% for pyrazinamide), and maintained performance when the data were associated with incomplete phenotypes (i.e. for those isolates for which phenotypic data for some drugs were missing). We also demonstrate that the model successfully identifies genes and SNPs associated with drug resistance, mitigating the impact of resistance profile while considering particular drug resistance, which is consistent with domain knowledge.

OBJECTIVE: To compare Cox models, machine learning (ML), and ensemble models combining both approaches, for prediction of stroke risk in a prospective study of Chinese adults. MATERIALS AND METHODS: We evaluated models for stroke risk at varying intervals of follow-up (<9 years, 0-3 years, 3-6 years, 6-9 years) in 503 842 adults without prior history of stroke recruited from 10 areas in China in 2004-2008. Inputs included sociodemographic factors, diet, medical history, physical activity, and physical measurements. We compared discrimination and calibration of Cox regression, logistic regression, support vector machines, random survival forests, gradient boosted trees (GBT), and multilayer perceptrons, benchmarking performance against the 2017 Framingham Stroke Risk Profile. We then developed an ensemble approach to identify individuals at high risk of stroke (>10% predicted 9-yr stroke risk) by selectively applying either a GBT or Cox model based on individual-level characteristics. RESULTS: For 9-yr stroke risk prediction, GBT provided the best discrimination (AUROC: 0.833 in men, 0.836 in women) and calibration, with consistent results in each interval of follow-up. The ensemble approach yielded incrementally higher accuracy (men: 76%, women: 80%), specificity (men: 76%, women: 81%), and positive predictive value (men: 26%, women: 24%) compared to any of the single-model approaches. DISCUSSION AND CONCLUSION: Among several approaches, an ensemble model combining both GBT and Cox models achieved the best performance for identifying individuals at high risk of stroke in a contemporary study of Chinese adults. The results highlight the potential value of expanding the use of ML in clinical practice.

In this article, we propose a novel neuromuscular password-based user authentication method. The method consists of two parts: surface electromyogram (sEMG) based finger muscle isometric contraction password (FMICP) and neuromuscular biometrics. FMICP can be entered through isometric contraction of different finger muscles in a prescribed order without actual finger movement, which makes it difficult for observers to obtain the password. In our study, the isometric contraction patterns of different finger muscles were recognized through high-density sEMG signals acquired from the right dorsal hand. Moreover, both time-frequency-space domain features at macroscopic level (interference-pattern EMG) and motor neuron firing rate features at microscopic level (via decomposition) were extracted to represent neuromuscular biometrics, serving as a second defense. The FMICP and macro-micro neuromuscular biometrics together form a neuromuscular password. The proposed neuromuscular password achieved an equal error rate (EER) of 0.0128 when impostors entered a wrong FMICP. Even when impostors entered the correct FMICP, the neuromuscular biometrics, as the second defense, inhibited impostors with an EER of 0.1496. To the best of our knowledge, this is the first study to use individually unique neuromuscular information during unobservable muscle isometric contractions for user authentication, with training and testing data acquired on different days.

Because of the rapid and serious nature of acute cardiovascular disease (CVD) especially ST segment elevation myocardial infarction (STEMI), a leading cause of death worldwide, prompt diagnosis and treatment is of crucial importance to reduce both mortality and morbidity. During a pandemic such as coronavirus disease-2019 (COVID-19), it is critical to balance cardiovascular emergencies with infectious risk. In this work, we recommend using wearable device based mobile health (mHealth) as an early screening and real-time monitoring tool to address this balance and facilitate remote monitoring to tackle this unprecedented challenge. This recommendation may help to improve the efficiency and effectiveness of acute CVD patient management while reducing infection risk.

With the soaring development of body sensor network (BSN)-based health informatics, information security in such medical devices has attracted increasing attention in recent years. Employing the biosignals acquired directly by the BSN as biometrics for personal identification is an effective approach. Noncancelability and cross-application invariance are two natural flaws of most traditional biometric modalities. Once the biometric template is exposed, it is compromised forever. Even worse, because the same biometrics may be employed as tokens for different accounts in multiple applications, the exposed template can be used to compromise other accounts. In this work, we propose a cancelable and cross-application discrepant biometric approach based on high-density surface electromyogram (HD-sEMG) for personal identification. We enrolled two accounts for each user. HD-sEMG signals from the right dorsal hand under isometric contractions of different finger muscles were employed as biometric tokens. Since isometric contraction, in contrast to dynamic contraction, requires no actual movement, the users' choice to login to different accounts is greatly protected against impostors. We realized a promising identification accuracy of 85.8% for 44 identities (22 subjects × 2 accounts) with training and testing data acquired 9 days apart. The high identification accuracy of different accounts for the same user demonstrates the promising cancelability and cross-application discrepancy of the proposed HD-sEMG-based biometrics. To the best of our knowledge, this is the first study to employ HD-sEMG in personal identification applications, with signal variation across days considered.

Different from the traditional methods of assessing the cardiac activities through heart rhythm statistics or P-QRS-T complexes separately, this study demonstrates their interactive effects on the power density spectrum (PDS) of ECG signal with applications for the diagnosis of ST-segment elevation myocardial infarction (STEMI) diseases. Firstly, a mathematical model of the PDS of ECG signal with a random pacing pulse train (PPT) mimicking S-A node firings was derived. Secondly, an experimental PDS analysis was performed on clinical ECG signals from 49 STEMI patients and 42 healthy subjects in PTB Diagnostic Database. It was found that besides the interactive effects which are consistent between theoretical and experimental results, the ECG PDSs of STEMI patients exhibited consistently significant power shift towards lower frequency range in ST-elevated leads in comparison with those of reference leads and leads of health subjects with the highest median frequency shift ratios at 51.39 ± 12.94% found in anterior MI. Thirdly, the results of ECG simulation with systematic changes in PPT firing statistics over various lengths of ECG data ranging from 10 s to 60 mins revealed that the mean and median frequency parameters were less affected by the heart rhythm statistics and the data length but more depended on the alterations of P-QRS-T complexes, which were further confirmed on 33 more STEMI patients in European ST-T Database, demonstrating that the frequency indexes could be potentially used as alternative indicators for STEMI diagnosis even with ultra-short-term ECG recordings suitable for wearable and mobile health applications in living-free environments.

OBJECTIVE: This paper presents a deep learning method of predicting where in a hospital emergency patients will be admitted after being triaged in the Emergency Department (ED). Such a prediction will allow for the preparation of bed space in the hospital for timely care and admission of the patient as well as allocation of resource to the relevant departments, including during periods of increased demand arising from seasonal peaks in infections. METHODS: The problem is posed as a multi-class classification into seven separate ward types. A novel deep learning training strategy was created that combines learning via curriculum and a multi-armed bandit to exploit this curriculum post-initial training. RESULTS: We successfully predict the initial hospital admission location with area-under-receiver-operating-curve (AUROC) ranging between 0.60 to 0.78 for the individual wards and an overall maximum accuracy of 52% where chance corresponds to 14% for this seven-class setting. Our proposed network was able to interpret which features drove the predictions using a 'network saliency' term added to the network loss function. CONCLUSION: We have proven that prediction of location of admission in hospital for emergency patients is possible using information from triage in ED. We have also shown that there are certain tell-tale tests which indicate what space of the hospital a patient will use. SIGNIFICANCE: It is hoped that this predictor will be of value to healthcare institutions by allowing for the planning of resource and bed space ahead of the need for it. This in turn should speed up the provision of care for the patient and allow flow of patients out of the ED thereby improving patient flow and the quality of care for the remaining patients within the ED.

Accurate and reliable prediction of hospital admission location is important due to resource-constraints and space availability in a clinical setting, particularly when dealing with patients who come from the emergency department. In this work we propose a student-teacher network via reinforcement learning to deal with this specific problem. A representation of the weights of the student network is treated as the state and is fed as an input to the teacher network. The teacher network’s action is to select the most appropriate batch of data to train the student network on from a training set sorted according to entropy. By validating on three datasets, not only do we show that our approach outperforms state-of-the-art methods on tabular data and performs competitively on image recognition, but also that novel curricula are learned by the teacher network. We demonstrate experimentally that the teacher network can actively learn about the student network and guide it to achieve better performance than if trained alone.

The papers in this special section were presented at the 2019 IEEE-EMBS International Conferences on Biomedical and Health Informatics (BHI’19) and Wearable and Implantable Body Sensor Networks (BSN’19). Topics of integrative sensor networks, informatics and modeling bring together the tightly coupled and rapidly developing fields of biomedical and health informatics and body sensor networks. Biomedical and health informatics encompasses methods to extract and communicate information from data in order to impact health, healthcare, life sciences and biomedicine. Body sensor networks provide one means to measure the needed data, through continuous monitoring in both clinical and free-living environments.

The paucity of physiological time-series data collected from low-resource clinical settings limits the capabilities of modern machine learning algorithms in achieving high performance. Such performance is further hindered by class imbalance; datasets where a diagnosis is much more common than others. To overcome these two issues at low-cost while preserving privacy, data augmentation methods can be employed. In the time domain, the traditional method of time-warping could alter the underlying data distribution with detrimental consequences. This is prominent when dealing with physiological conditions that influence the frequency components of data. In this paper, we propose PlethAugment; three different conditional generative adversarial networks (CGANs) with an adapted diversity term for the generation of pathological photoplethysmogram (PPG) signals in order to boost medical classification performance. To evaluate and compare the GANs, we introduce a novel metric-agnostic method; the synthetic generalization curve. We validate this approach on two proprietary and two public datasets representing a diverse set of medical conditions. Compared to training on non-augmented class-balanced datasets, training on augmented datasets leads to an improvement of the AUROC by up to 29% when using cross validation. This illustrates the potential of the proposed CGANs to significantly improve classification performance.

Assessment of physiological instability preceding adverse events on hospital wards has been previously investigated through clinical early warning score systems. Early warning scores are simple to use yet they consider data as independent and identically distributed random variables. Deep learning applications are able to learn from sequential data, however they lack interpretability and are thus difficult to deploy in clinical settings. We propose the 'Deep Early Warning System' (DEWS), an interpretable end-to-end deep learning model that interpolates temporal data and predicts the probability of an adverse event, defined as the composite outcome of cardiac arrest, mortality or unplanned ICU admission. The model was developed and validated using routinely collected vital signs of patients admitted to the the Oxford University Hospitals between 21st March 2014 and 31st March 2018. We extracted 45 314 vital-sign measurements as a balanced training set and 359 481 vital-sign measurements as an imbalanced testing set to mimic a real-life setting of emergency admissions. DEWS achieved superior accuracy than the state-of-the-art that is currently implemented in clinical settings, the National Early Warning Score, in terms of the overall area under the receiver operating characteristic curve (AUROC) (0.880 vs. 0.866) and when evaluated independently for each of the three outcomes. Our attention-based architecture was able to recognize 'historical' trends in the data that are most correlated with the predicted probability. With high sensitivity, improved clinical utility and increased interpretability, our model can be easily deployed in clinical settings to supplement existing EWS systems.

In low and middle income countries, infectious diseases continue to have a significant impact, particularly amongst the poorest in society. Tetanus and hand foot and mouth disease (HFMD) are two such diseases and, in both, death is associated with autonomic nervous system dysfunction (ANSD). Currently, photoplethysmogram or electrocardiogram monitoring is used to detect deterioration in these patients, however expensive clinical monitors are often required. In this study, we employ low-cost and mobile wearable devices to collect patient vital signs unobtrusively; and we develop machine learning algorithms for automatic and rapid triage of patients that provide efficient use of clinical resources. Existing methods are mainly dependent on the prior detection of clinical features with limited exploitation of multi-modal physiological data. Moreover, the latest developments in deep learning (e.g. cross-domain transfer learning) have not been sufficiently applied for infectious disease diagnosis. In this paper, we present a fusion of multi-modal physiological data to predict the severity of ANSD with a hierarchy of resource-aware decision making. First, an on-site triage process is performed using a simple classifier. Second, personalised longitudinal modelling is employed that takes the previous states of the patient into consideration. We have also employed a spectrogram representation of the physiological waveforms to exploit existing networks for cross-domain transfer learning, which avoids the laborious and data intensive process of training a network from scratch. Results show that the proposed framework has promising potential in supporting severity grading of infectious diseases in low-resources settings, such as in the developing world.

MOTIVATION: Timely identification of Mycobacterium tuberculosis (MTB) resistance to existing drugs is vital to decrease mortality and prevent the amplification of existing antibiotic resistance. Machine learning methods have been widely applied for timely predicting resistance of MTB given a specific drug and identifying resistance markers. However, they have been not validated on a large cohort of MTB samples from multi-centers across the world in terms of resistance prediction and resistance marker identification. Several machine learning classifiers and linear dimension reduction techniques were developed and compared for a cohort of 13 402 isolates collected from 16 countries across 6 continents and tested 11 drugs. RESULTS: Compared to conventional molecular diagnostic test, area under curve of the best machine learning classifier increased for all drugs especially by 23.11%, 15.22% and 10.14% for pyrazinamide, ciprofloxacin and ofloxacin, respectively (P < 0.01). Logistic regression and gradient tree boosting found to perform better than other techniques. Moreover, logistic regression/gradient tree boosting with a sparse principal component analysis/non-negative matrix factorization step compared with the classifier alone enhanced the best performance in terms of F1-score by 12.54%, 4.61%, 7.45% and 9.58% for amikacin, moxifloxacin, ofloxacin and capreomycin, respectively, as well increasing area under curve for amikacin and capreomycin. Results provided a comprehensive comparison of various techniques and confirmed the application of machine learning for better prediction of the large diverse tuberculosis data. Furthermore, mutation ranking showed the possibility of finding new resistance/susceptible markers. AVAILABILITY AND IMPLEMENTATION: The source code can be found at http://www.robots.ox.ac.uk/ davidc/code.php. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Resistance co-occurrence within first-line anti-tuberculosis (TB) drugs is a common phenomenon. Existing methods based on genetic data analysis of Mycobacterium tuberculosis (MTB) have been able to predict resistance of MTB to individual drugs, but have not considered the resistance co-occurrence and cannot capture latent structure of genomic data that corresponds to lineages. RESULTS: We used a large cohort of TB patients from 16 countries across six continents where whole-genome sequences for each isolate and associated phenotype to anti-TB drugs were obtained using drug susceptibility testing recommended by the World Health Organization. We then proposed an end-to-end multi-task model with deep denoising auto-encoder (DeepAMR) for multiple drug classification and developed DeepAMR_cluster, a clustering variant based on DeepAMR, for learning clusters in latent space of the data. The results showed that DeepAMR outperformed baseline model and four machine learning models with mean AUROC from 94.4% to 98.7% for predicting resistance to four first-line drugs [i.e. isoniazid (INH), ethambutol (EMB), rifampicin (RIF), pyrazinamide (PZA)], multi-drug resistant TB (MDR-TB) and pan-susceptible TB (PANS-TB: MTB that is susceptible to all four first-line anti-TB drugs). In the case of INH, EMB, PZA and MDR-TB, DeepAMR achieved its best mean sensitivity of 94.3%, 91.5%, 87.3% and 96.3%, respectively. While in the case of RIF and PANS-TB, it generated 94.2% and 92.2% sensitivity, which were lower than baseline model by 0.7% and 1.9%, respectively. t-SNE visualization shows that DeepAMR_cluster captures lineage-related clusters in the latent space. AVAILABILITY AND IMPLEMENTATION: The details of source code are provided at http://www.robots.ox.ac.uk/∼davidc/code.php. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Patients in a hospital step-down unit require a level of care that is between that of the intensive care unit (ICU) and that of the general ward. While many patients remain physiologically stabilized, others will suffer clinical emergencies and be readmitted to the ICU, with a subsequent high risk of mortality. Had the associated physiological deterioration been detected early, the emergency may have been less severe or avoided entirely. Current clinical monitoring is largely heuristic, requiring manual calculation of risk scores and the use of heuristic decision criteria. Technical drawbacks include ignoring the time-series dynamics of physiological measurements, and lacking patient-specificity (i.e., personalization of models to the individual patient). In this paper, we demonstrate how Gaussian process regression models can supplement current monitoring practice by providing interpretable and intuitive illustrations of erratic vital-sign volatility. These personalized volatility metrics may provide significantly advanced warning of deterioration, while minimizing the false alarms that induce so-called alarm fatigue. While many AI-based approaches to healthcare are criticized for being uninterpretable "black-box" methods, the cause of alarms generated from the proposed methods are explicitly interpretable and intuitive. We conclude that intelligent computational inference using methods such as those proposed can enhance current clinical decision making and potentially save lives.

The role of sensing technologies, such as wearables, in delivering precision care is becoming widely acceptable. Given the very large quantities of sensor data that rapidly accumulate, there is a need to employ automated algorithms to label biosignal sensor data. In many real-life clinical applications, no such expert labels are available, and algorithms for processing sensor data must be relied upon, without access to the "ground truth." It is therefore extremely difficult to choose which algorithms to trust or discard at any point in time, where different algorithms may be optimal for different patients, or even for different points in time for the same patient. We propose two fully Bayesian approaches for fusing labels from independent and potentially correlated annotators (i.e., algorithms or, where available, experts). These are generative models to aggregate labels (i.e., the outputs of the algorithms, such as identified ECG morphology) in an unsupervised manner, to estimate jointly the assumed bias and precision of each algorithm without access to the ground truth. The latter fused estimate may then be used to infer the underlying ground truth. For the first time in the biomedical context, we show that modeling correlations between annotators, and fusing information concerning task difficulty (such as the estimated quality of the sensor data), improve these estimates with respect to commonly employed strategies in the literature. Also, we adopt a strongly Bayesian approach to inference using Gibbs sampling to improve estimates over the existing state of the art. We present results from applying the proposed pair of models to simulated and two publicly available biomedical datasets, to demonstrate proof-of-principle. We show that our proposed models outperform all existing approaches recreated from the literature. We also show that the proposed methods are robust when dealing with missing values (as often occurs in real-life biomedical applications), and that they are suitably efficient for use in real-time applications, thereby providing the basis for the reliable use of sensors for personalizing the care of the individual.

A visualisation method for automatically clustering heart beats in single channel electrocardiography (ECG) signals was developed and applied to a dataset of post-intensive care patients who received long-term continuous monitoring. We first segmented the ECG signal into individual beats using an R-peak detection algorithm. A matrix was constructed by storing the segmented ECG beats in row-wise format. Singular value decomposition (SVD) was applied to remove sparse invalid detected R peaks, thus smoothing the matrix. Treating the matrix of ECG beat values as an image, an edge detection algorithm was applied, resulting in a binary matrix containing traces of heart beats with the contiguous and discontiguous components extracted. We considered each component to be a cluster of heart beats. This method was robust to signal noise by exploiting detected R peaks and ECG raw cycles represented in a matrix format for estimation of heart beats. The algorithm also eliminated the effect of underestimated R peaks in the estimation of heart beats, and minimised the effects of overestimated R peaks using the SVD algorithm. This method allows clusters of beats to be visualized, which may assist clinicians in estimating the components of long-term ECG signals.

We describe an evaluation of photoplethysmography (PPG) signals with two wavelengths channels (infrared and red) using a wrist-worn sensor for the estimation of heart rate variability (HRV) and oxygen saturation (SpO2). Five healthy subjects were equipped with a commercial wrist-worn pulse oximeter (Wavelet Health, USA) on the right hand, and both a commercial smart watch for fitness use (Huawei Watch, Series 2) and a clinically-validated transmission-mode pulse oximeter (Creative Medical PC-68B) on their left hand as a reference. Synchronised PPG signals from the Wavelet Health, the Huawei watch, and the PC-68B were recorded for approximately 10 minutes. Subjects were asked to leave the left hand in a resting state, while moving the right hand with two types of movement (periodic and random). A method is proposed to incorporate coupling information between the two wavelengths of PPGs based on the bivariate empirical mode decomposition algorithm. Our method is shown to improve the quality of red PPG allowing improved estimation of SpO2. A comparison of average heart rate (HR), HRV, and SpO2 from all devices is provided.

Motivation: Correct and rapid determination of Mycobacterium tuberculosis (MTB) resistance against available tuberculosis (TB) drugs is essential for the control and management of TB. Conventional molecular diagnostic test assumes that the presence of any well-studied single nucleotide polymorphisms is sufficient to cause resistance, which yields low sensitivity for resistance classification. Summary: Given the availability of DNA sequencing data from MTB, we developed machine learning models for a cohort of 1839 UK bacterial isolates to classify MTB resistance against eight anti-TB drugs (isoniazid, rifampicin, ethambutol, pyrazinamide, ciprofloxacin, moxifloxacin, ofloxacin, streptomycin) and to classify multi-drug resistance. Results: Compared to previous rules-based approach, the sensitivities from the best-performing models increased by 2-4% for isoniazid, rifampicin and ethambutol to 97% (P < 0.01), respectively; for ciprofloxacin and multi-drug resistant TB, they increased to 96%. For moxifloxacin and ofloxacin, sensitivities increased by 12 and 15% from 83 and 81% based on existing known resistance alleles to 95% and 96% (P < 0.01), respectively. Particularly, our models improved sensitivities compared to the previous rules-based approach by 15 and 24% to 84 and 87% for pyrazinamide and streptomycin (P < 0.01), respectively. The best-performing models increase the area-under-the-ROC curve by 10% for pyrazinamide and streptomycin (P < 0.01), and 4-8% for other drugs (P < 0.01). Availability and implementation: The details of source code are provided at http://www.robots.ox.ac.uk/~davidc/code.php. Contact: [email protected]. Supplementary information: Supplementary data are available at Bioinformatics online.

Gaussian process regression (GPR) provides a means to generate flexible personalized models of time series of patient vital signs. These models can perform useful clinical inference in ways that population-based models cannot. A challenge for the use of personalized models is that they must be amenable to a wide range of parameterizations, to accommodate the plausible physiology of any patient in the population. Additionally, optimal performance is typically achieved when models are regularized in light of the knowledge of the physiology of the individual patient. In this paper, we describe a method to build GP models with varying complexity (via covariance kernels) and regularization (via fixed priors over hyperparameters) on a patient-specific level, for the purpose of robust vital-sign forecasting. To this end, our results present evidence in support of two main hypotheses: 1) the use of patient-specific models can outperform population-based models for useful clinical tasks, such as vital-sign forecasting; and 2) the optimal values of (hyper)parameters of these models are best determined by sophisticated methods of optimization, due to high correlation between dimensions of the search space. The resulting models are sufficiently robust to inform clinicians of a patient's vital-sign trajectory and warn of imminent deterioration.

Novelty detection or one-class classification starts from a model describing some type of `normal behaviour' and aims to classify deviations from this model as being either novelties or anomalies. In this paper the problem of novelty detection for point patterns $S=\{\mathbf{x}_1,\ldots ,\mathbf{x}_k\}\subset \mathbb{R}^d$ is treated where examples of anomalies are very sparse, or even absent. The latter complicates the tuning of hyperparameters in models commonly used for novelty detection, such as one-class support vector machines and hidden Markov models. To this end, the use of extreme value statistics is introduced to estimate explicitly a model for the abnormal class by means of extrapolation from a statistical model $X$ for the normal class. We show how multiple types of information obtained from any available extreme instances of $S$ can be combined to reduce the high false-alarm rate that is typically encountered when classes are strongly imbalanced, as often occurs in the one-class setting (whereby `abnormal' data are often scarce). The approach is illustrated using simulated data and then a real-life application is used as an exemplar, whereby accelerometry data from epileptic seizures are analysed - these are known to be extreme and rare with respect to normal accelerometer data.

Clinical data management systems typically provide caregiver teams with useful information, derived from large, sometimes highly heterogeneous, data sources that are often changing dynamically. Over the last decade there has been a significant surge in interest in using these data sources, from simply re-using the standard clinical databases for event prediction or decision support, to including dynamic and patient-specific information into clinical monitoring and prediction problems. However, in most cases, commercial clinical databases have been designed to document clinical activity for reporting, liability and billing reasons, rather than for developing new algorithms. With increasing excitement surrounding "secondary use of medical records" and "Big Data" analytics, it is important to understand the limitations of current databases and what needs to change in order to enter an era of "precision medicine." This review article covers many of the issues involved in the collection and preprocessing of critical care data. The three challenges in critical care are considered: compartmentalization, corruption, and complexity. A range of applications addressing these issues are covered, including the modernization of static acuity scoring; on-line patient tracking; personalized prediction and risk assessment; artifact detection; state estimation; and incorporation of multimodal data sources such as genomic and free text data.

The ability to determine patient acuity (or severity of illness) has immediate practical use for clinicians. We evaluate the use of multivariate timeseries modeling with the multi-task Gaussian process (GP) models using noisy, incomplete, sparse, heterogeneous and unevenly-sampled clinical data, including both physiological signals and clinical notes. The learned multi-task GP (MTGP) hyperparameters are then used to assess and forecast patient acuity. Experiments were conducted with two real clinical data sets acquired from ICU patients: firstly, estimating cerebrovascular pressure reactivity, an important indicator of secondary damage for traumatic brain injury patients, by learning the interactions between intracranial pressure and mean arterial blood pressure signals, and secondly, mortality prediction using clinical progress notes. In both cases, MTGPs provided improved results: an MTGP model provided better results than single-task GP models for signal interpolation and forecasting (0.91 vs 0.69 RMSE), and the use of MTGP hyperparameters obtained improved results when used as additional classification features (0.812 vs 0.788 AUC).

The identification of invalid data in recordings obtained using wearable sensors is of particular importance since data obtained from mobile patients is, in general, noisier than data obtained from nonmobile patients. In this paper, we present a signal quality index (SQI), which is intended to assess whether reliable heart rates (HRs) can be obtained from electrocardiogram (ECG) and photoplethysmogram (PPG) signals collected using wearable sensors. The algorithms were validated on manually labeled data. Sensitivities and specificities of 94% and 97% were achieved for the ECG and 91% and 95% for the PPG. Additionally, we propose two applications of the SQI. First, we demonstrate that, by using the SQI as a trigger for a power-saving strategy, it is possible to reduce the recording time by up to 94% for the ECG and 93% for the PPG with only minimal loss of valid vital-sign data. Second, we demonstrate how an SQI can be used to reduce the error in the estimation of respiratory rate (RR) from the PPG. The performance of the two applications was assessed on data collected from a clinical study on hospital patients who were able to walk unassisted.

The majority of patients in the hospital are ambulatory and would benefit significantly from predictive and personalized monitoring systems. Such patients are well suited to having their physiological condition monitored using low-power, minimally intrusive wearable sensors. Despite data-collection systems now being manufactured commercially, allowing physiological data to be acquired from mobile patients, little work has been undertaken on the use of the resultant data in a principled manner for robust patient care, including predictive monitoring. Most current devices generate so many false-positive alerts that devices cannot be used for routine clinical practice. This paper explores principled machine learning approaches to interpreting large quantities of continuously acquired, multivariate physiological data, using wearable patient monitors, where the goal is to provide early warning of serious physiological determination, such that a degree of predictive care may be provided. We adopt a one-class support vector machine formulation, proposing a formulation for determining the free parameters of the model using partial area under the ROC curve, a method arising from the unique requirements of performing online analysis with data from patient-worn sensors. There are few clinical evaluations of machine learning techniques in the literature, so we present results from a study at the Oxford University Hospitals NHS Trust devised to investigate the large-scale clinical use of patient-worn sensors for predictive monitoring in a ward with a high incidence of patient mortality. We show that our system can combine routine manual observations made by clinical staff with the continuous data acquired from wearable sensors. Practical considerations and recommendations based on our experiences of this clinical study are discussed, in the context of a framework for personalized monitoring.

Novelty detection, or one-class classification, is of particular use in the analysis of high-integrity systems, in which examples of failure are rare in comparison with the number of examples of stable behaviour, such that a conventional multi-class classification approach cannot be taken. Support Vector Machines (SVMs) are a popular means of performing novelty detection, and it is conventional practice to use a train-validate-test approach, often involving cross-validation, to train the one-class SVM, and then select appropriate values for its parameters. An alternative method, used with multi-class SVMs, is to calibrate the SVM output into conditional class probabilities. A probabilistic approach offers many advantages over the conventional method, including the facility to select automatically a probabilistic novelty threshold. The contributions of this paper are (i) the development of a probabilistic calibration technique for one-class SVMs, such that on-line novelty detection may be performed in a probabilistic manner; and (ii) the demonstration of the advantages of the proposed method (in comparison to the conventional one-class SVM methodology) using case studies, in which one-class probabilistic SVMs are used to perform condition monitoring of a high-integrity industrial combustion plant, and in detecting deterioration in patient physiological condition during patient vital-sign monitoring.

We consider an integrated patient monitoring system, combining electronic patient records with high-rate acquisition of patient physiological data. There remain many challenges in increasing the robustness of "e-health" applications to a level at which they are clinically useful, particularly in the use of automated algorithms used to detect and cope with artifact in data contained within the electronic patient record, and in analyzing and communicating the resultant data for reporting to clinicians. There is a consequential "plague of pilots," in which engineering prototype systems do not enter into clinical use. This paper describes an approach in which, for the first time, the Emergency Department (ED) of a major research hospital has adopted such systems for use during a large clinical trial. We describe the disadvantages of existing evaluation metrics when applied to such large trials, and propose a solution suitable for large-scale validation. We demonstrate that machine learning technologies embedded within healthcare information systems can provide clinical benefit, with the potential to improve patient outcomes in the busy environment of a major ED and other high-dependence areas of patient care.

Vital-sign monitoring of patients within a hospital setting is a big component in the recognition and treatment of early signs of deterioration. Current vital-sign monitoring systems, including both manual early warning systems, and more sophisticated data fusion systems, typically make use of the most recently recorded data, and are unable to deal with missing data in a principled manner. The latter is particularly pertinent in the field of ambulatory monitoring, in which patient movement can result in sensor disconnections and other artefact. This paper presents a Gaussian process regression technique for estimating missing data and how it can be incorporated within an automated data fusion monitoring system. The technique is then demonstrated using vital-sign data from a recent clinical study conducted at the John Radcliffe Hospital, Oxford, showing an improvement over an existing data fusion algorithm by providing both an estimate of missing vital sign data and the uncertainty in the estimated value.

Hospital patient outcomes can be improved by the early identification of physiological deterioration. Automatic methods of detecting patient deterioration in vital-sign data typically attempt to identify deviations from assumed normal physiological conditions, which is a one-class approach to classification. This paper investigates the use of a two-class approach, in which abnormal physiology is modelled explicitly. The success of such a method relies on the accuracy of data labels provided by clinical experts, which may be incomplete (due to large dataset size) or imprecise (due to clinical labels covering intervals, rather than each data point within those intervals). We propose a novel method of refining clinical labels such that the two-class classification approach may be adopted for identifying patient deterioration. We demonstrate the effectiveness of the proposed methods using a large dataset acquired in a 24-bed hospital step-down unit.