The m-out-of-n bootstrap—proposed by \cite{bickel1992resampling}—approximates the distribution of a statistic by repeatedly drawing $m$ subsamples ($m \ll n$) without replacement from an original sample of size n; it is now routinely used for robust inference with heavy-tailed data, bandwidth selection, and other large-sample applications. Despite this broad applicability across econometrics, biostatistics, and machine-learning workflows, rigorous parameter-free guarantees for the soundness of the m-out-of-n bootstrap when estimating sample quantiles have remained elusive. This paper establishes such guarantees by analysing the estimator of sample quantiles obtained from m-out-of-n resampling of a dataset of length n. We first prove a central limit theorem for a fully data-driven version of the estimator that holds under a mild moment condition and involves no unknown nuisance parameters. We then show that the moment assumption is essentially tight by constructing a counter-example in which the CLT fails. Strengthening the assumptions slightly, we derive an Edgeworth expansion that delivers exact convergence rates and, as a corollary, a Berry–Esséen bound on the bootstrap approximation error. Finally, we illustrate the scope of our results by obtaining parameter-free asymptotic distributions for practical statistics, including the quantiles for random walk MH, and rewards of ergodic MDP's, thereby demonstrating the usefulness of our theory in modern estimation and learning tasks.

Education-level or socioeconomic background of patients may dictate their ability to understand medical jargon. Inability to understand primary findings from a radiology report may lead to unnecessary anxiety among patients or missed follow up. We aim to meet this challenge by developing a patient-sensitive summarization model for radiology reports. We selected computed tomography (CT) exams of chest as a use-case and collected 7,000 studies from Mayo Clinic. Summarization model was built on top of the T5 large language model (LLM) as our experiments indicated that its text-to-text transfer architecture was suited for abstractive text summarization, resulting in a model with 0.77B trainable parameters. Noisy ground truth for model training was collected by prompting LLaMA-13B model. We recruited experts (board-certified radiologists) and laymen to manually evaluate model-generated summaries generated by model. Our model rarely missed information as marked by majority opinion of radiologists. Laymen indicated 63% improvement in their understanding by reading model-generated layman summaries. Comparison with zero-shot performance of ChatGPT indicated that the proposed model reduced the rate of hallucination by half and rate of missing important information by fivefold. The proposed model can generate reliable summaries for radiology reports understandable by patients with vastly different levels of medical knowledge.

The integration of Large Language Models (LLMs) into the healthcare domain has the potential to significantly enhance patient care and support through the development of empathetic, patient-facing chatbots. This study investigates an intriguing question Can ChatGPT respond with a greater degree of empathy than those typically offered by physicians? To answer this question, we collect a de-identified dataset of patient messages and physician responses from a hospital and generate alternative replies using ChatGPT. We then introduce a set of empathy ranking evaluation (EMRank) metrics to automatically judge the empathy degree. We further conduct human study to gauge the empathy level of responses. Our findings indicate that LLM-powered chatbots have the potential to surpass human physicians in delivering empathetic communication, suggesting a promising avenue for enhancing patient care and reducing professional burnout. To summary, this study not only highlights the importance of clinical empathy in patient interactions but also proposes a set of automatic empathy ranking metrics, paving the way for the broader adoption of LLMs in healthcare.

Imaging phenotypes extracted via radiomics of magnetic resonance imaging have shown great potential in predicting the treatment response in breast cancer patients after administering neoadjuvant systemic therapy (NST). Understanding the causal relationships between the treatment response and Imaging phenotypes, Clinical information, and Molecular (ICM) features are critical in guiding treatment strategies and management plans. Counterfactual explanations provide an interpretable approach to generating causal inference. However, existing approaches are either computationally prohibitive for high dimensional problems, generate unrealistic counterfactuals, or confound the effects of causal features by changing multiple features simultaneously. This paper proposes a new method called Sparse CounteRGAN (SCGAN) for generating counterfactual instances to reveal causal relationships between ICM features and the treatment response after NST. The generative approach learns the distribution of the original instances and, therefore, ensures that the new instances are realistic. We propose dropout training of the discriminator to promote sparsity and introduce a diversity term in the loss function to maximize the distances among generated counterfactuals. We evaluate the proposed method on two publicly available datasets, followed by the breast cancer dataset, and compare their performance with existing methods in the literature. Results show that SCGAN generates sparse and diverse counterfactual instances that also achieve plausibility and feasibility, making it a valuable tool for understanding the causal relationships between ICM features and treatment response.Note to Practitioners— Determining the suitability of NST for a breast cancer patient before surgery is complex and depends on factors such as patient demographics, tumor characteristics, clinical history, and molecular subtypes. Understanding the causal relationships between different features and pathologic responses to NST may lead to opportunities for targeted therapies and help oncologists make informed decisions about continuing or limiting systemic therapy after the initial consultation. The lack of causal explanations in traditional machine learning models for predicting NST has limited their applicability in clinical decision-making. SCGAN proposed in this paper overcomes the limitations of existing methods in generating causal inference via counterfactual explanations. This approach helps identify causal relationships between imaging phenotypes, clinical history, molecular features, and pathologic response to NST. The resulting information can aid in developing personalized treatment plans for patients and ultimately improve patient outcomes. The proposed approach may be extended to broader applications beyond healthcare, such as manufacturing, where it could improve quality control processes and enhance production efficiency.

OBJECTIVE: Hospital acquired infections (HAIs) are one of the top 10 leading causes of death within the United States. While current standard of HAI risk prediction utilizes only a narrow set of predefined clinical variables, we propose a graph convolutional neural network (GNN)-based model which incorporates a wide variety of clinical features. MATERIALS AND METHODS: Our GNN-based model defines patients' similarity based on comprehensive clinical history and demographics and predicts all types of HAI rather than focusing on a single subtype. An HAI model was trained on 38 327 unique hospitalizations while a distinct model for surgical site infection (SSI) prediction was trained on 18 609 hospitalization. Both models were tested internally and externally on a geographically disparate site with varying infection rates. RESULTS: The proposed approach outperformed all baselines (single-modality models and length-of-stay [LoS]) with achieved area under the receiver operating characteristics of 0.86 [0.84-0.88] and 0.79 [0.75-0.83] (HAI), and 0.79 [0.75-0.83] and 0.76 [0.71-0.76] (SSI) for internal and external testing. Cost-effective analysis shows that the GNN modeling dominated the standard LoS model strategy on the basis of lower mean costs ($1651 vs $1915). DISCUSSION: The proposed HAI risk prediction model can estimate individualized risk of infection for patient by taking into account not only the patient's clinical features, but also clinical features of similar patients as indicated by edges of the patients' graph. CONCLUSIONS: The proposed model could allow prevention or earlier detection of HAI, which in turn could decrease hospital LoS and associated mortality, and ultimately reduce the healthcare cost.

Automated curation of noisy external data in the medical domain has long been in high demand, as AI technologies need to be validated using various sources with clean, annotated data. Identifying the variance between internal and external sources is a fundamental step in curating a high-quality dataset, as the data distributions from different sources can vary significantly and subsequently affect the performance of AI models. The primary challenges for detecting data shifts are - (1) accessing private data across healthcare institutions for manual detection and (2) the lack of automated approaches to learn efficient shift-data representation without training samples. To overcome these problems, we propose an automated pipeline called MedShift to detect top-level shift samples and evaluate the significance of shift data without sharing data between internal and external organizations. MedShift employs unsupervised anomaly detectors to learn the internal distribution and identify samples showing significant shiftness for external datasets, and then compares their performance. To quantify the effects of detected shift data, we train a multi-class classifier that learns internal domain knowledge and evaluates the classification performance for each class in external domains after dropping the shift data. We also propose a data quality metric to quantify the dissimilarity between internal and external datasets. We verify the efficacy of MedShift using musculoskeletal radiographs (MURA) and chest X-ray datasets from multiple external sources. Our experiments show that our proposed shift data detection pipeline can be beneficial for medical centers to curate high-quality datasets more efficiently.

Reduction in 30-day readmission rate is an important quality factor for hospitals as it can reduce the overall cost of care and improve patient post-discharge outcomes. While deep-learning-based studies have shown promising empirical results, several limitations exist in prior models for hospital readmission prediction, such as: (a) only patients with certain conditions are considered, (b) do not leverage data temporality, (c) individual admissions are assumed independent of each other, which ignores patient similarity, (d) limited to single modality or single center data. In this study, we propose a multimodal, spatiotemporal graph neural network (MM-STGNN) for prediction of 30-day all-cause hospital readmission, which fuses in-patient multimodal, longitudinal data and models patient similarity using a graph. Using longitudinal chest radiographs and electronic health records from two independent centers, we show that MM-STGNN achieved an area under the receiver operating characteristic curve (AUROC) of 0.79 on both datasets. Furthermore, MM-STGNN significantly outperformed the current clinical reference standard, LACE+ (AUROC = 0.61), on the internal dataset. For subset populations of patients with heart disease, our model significantly outperformed baselines, such as gradient-boosting and Long Short-Term Memory models (e.g., AUROC improved by 3.7 points in patients with heart disease). Qualitative interpretability analysis indicated that while patients' primary diagnoses were not explicitly used to train the model, features crucial for model prediction may reflect patients' diagnoses. Our model could be utilized as an additional clinical decision aid during discharge disposition and triaging high-risk patients for closer post-discharge follow-up for potential preventive measures.

Language modality within the vision language pre-training framework is innately discretized, endowing each word in the language vocabulary a semantic meaning. In contrast, visual modality is inherently continuous and high-dimensional, which potentially prohibits the alignment as well as fusion between vision and language modalities. We therefore propose to "discretize" the visual representation by joint learning a codebook that imbues each visual token a semantic. We then utilize these discretized visual semantics as self-supervised ground-truths for building our Masked Image Modeling objective, a counterpart of Masked Language Modeling which proves successful for language models. To optimize the codebook, we extend the formulation of VQ-VAE which gives a theoretic guarantee. Experiments validate the effectiveness of our approach across common vision-language benchmarks.

This work presents a fusion artificial intelligence (AI) framework that combines patient electronic medical record (EMR) and physiological sensor data to accurately predict early risk of sepsis 4 hours before onset. The fusion AI model has two components - an on-chip AI model that continuously analyzes patient electrocardiogram (ECG) data and a cloud AI model that combines EMR and prediction scores from on-chip AI model to predict fusion sepsis onset score. The on-chip AI model is designed using analog circuits for high energy efficiency that allows integration with resource constrained wearable device. The on-chip AI reduces by 4.5× compared to digital baseline, and by 4× compared to state-of-the-art bio-medical AI ICs. Combination of EMR and sensor physiological data improves prediction performance compared to EMR or physiological data alone, and the late fusion model has an accuracy of 92.2% in predicting sepsis 4 hours before onset. The key differentiation of this work over existing sepsis prediction literature is the use of single modality patient vital (ECG) and simple demographic information, instead of comprehensive laboratory test results and multiple vital signs.

Improving the retrieval relevance on noisy datasets is an emerging need for the curation of a large-scale clean dataset in the medical domain. While existing methods can be applied for class-wise retrieval (aka. inter-class), they cannot distinguish the granularity of likeness within the same class (aka. intra-class). The problem is exacerbated on medical external datasets, where noisy samples of the same class are treated equally during training. Our goal is to identify both intra/inter-class similarities for fine-grained retrieval. To achieve this, we propose an Outlier-Sensitive Content-based rAdiologhy Retrieval System (OSCARS), consisting of two steps. First, we train an outlier detector on a clean internal dataset in an unsupervised manner. Then we use the trained detector to generate the anomaly scores on the external dataset, whose distribution will be used to bin intra-class variations. Second, we propose a quadruplet (a, p, nintra, ninter) sampling strategy, where intra-class negatives nintra are sampled from bins of the same class other than the bin anchor a belongs to, while n_inter are randomly sampled from inter-classes. We suggest a weighted metric learning objective to balance the intra and inter-class feature learning. We experimented on two representative public radiography datasets. Experiments show the effectiveness of our approach. The training and evaluation code can be found in https://github.com/XiaoyuanGuo/oscars.

Small-scale data centers at the edge are becoming prominent in offering various services to the end-users following the cloud model while avoiding the high latency inherent to the classic cloud environments when accessed from remote Internet regions. However, we should address several challenges to facilitate the end-users finding and consuming the relevant services from the edge at the Internet scale. First, the scale and diversity of the edge hinder seamless access. Second, a framework where researchers openly share their services and data in a secured manner among themselves and with external consumers over the Internet does not exist. Third, the lack of a unified interface and trust across the service providers hinder their interchangeability in composing workflows by chaining the services. Thus, creating a workflow from the services deployed on the various edge nodes is presently impractical. This paper designs Viseu, a latency-aware blockchain framework to provide Virtual Internet Services at the Edge. Viseu aims to solve the puzzle of network service discovery at the edge, considering the peers' reputation and latency when choosing the service instances. Viseu enables peers to share their computational resources, services, and data among each other in an untrusted environment, rather than relying on a set of trusted service providers. By composing workflows from the peers' services, rather than confining them to the pre-established service provider and consumer roles, Viseu aims to facilitate scientific collaboration across the peers natively. Furthermore, by offering services from multiple peers close to the end-users, Viseu also minimizes end-to-end latency and data loss in the service execution at the Internet scale.

Free-form radiology reports associated with coronary computed tomography angiography (CCTA) include nuanced and complicated linguistics to report cardiovascular disease. Standardization and interpretation of such reports is crucial for clinical use of CCTA. Coronary Artery Disease Reporting and Data System (CAD-RADS) has been proposed to achieve such standardization by implementing a strict template-based report writing and assignment of a score between 0 and 5 indicating the severity of coronary artery lesions. Even after its introduction, free-form unstructured report writing remains popular among radiologists. In this work, we present our attempts at bridging the gap between structured and unstructured reporting by natural language processing. We present machine learning models that while being trained only on structured reports, can predict CAD-RADS scores by analysis of free-text of unstructured radiology reports. The best model achieves 98% accuracy on structured reports and 92% 1-margin accuracy (difference of \le 1 in the predicted and the actual scores) for free-form unstructured reports. Our model also performs well under very difficult circumstances including nuanced and widely varying terminology used for reporting cardiovascular functions and diseases, scarcity of labeled data for training our model, and uneven class label distribution.

OBJECTIVE: We sought to ascertain perceived factors affecting women's career development efforts in the American Medical Informatics Association (AMIA) and to provide recommendations for improvements. MATERIALS AND METHODS: Data were collected using a 27-item survey administered via the AMIA newsletter and other social channels. Survey questions comprised 3 demographics, 15 Likert-scale, and 9 open-ended items. Likert-scale responses were summarized across respondent ages, career stages, and career domains, and open-ended responses were thematically analyzed. RESULTS: We received survey responses from 109 AMIA women members. Our findings demonstrate that AMIA had made strides in promoting career development, and the most effective AMIA efforts included social events (83%), panel discussions (80%), and scientific sessions (79%). However, despite these efforts, women members perceived that gender-specific challenges persisted within AMIA, and recognized the need for increased networking opportunities (96%), raising awareness of gender-specific challenges (95%), and encouraging gender proportional representation in leadership (92%). DISCUSSION: International and national biomedical informatics professional communities have put forth efforts to address gender-specific issues in career development. Yet, our study identified that some of these, including the deep-rooted gender power hierarchy and bias, are still perceived as profound in AMIA. CONCLUSION: Even though existing career development efforts for women are highly effective, important perceived gender-specific career development issues require further attention and investigation to improve existing AMIA activities.

A recurrent neural network (RNN) is presented in this work for automatic detection of atrial fibrillation from raw ECG signals without any hand-crafted feature extraction. We designed a stacked long-short term memory (LSTM) network - a special RNN with capability of learning long-term temporal dependencies in the ECG signal. The RNN is digitally synthesized in 65nm CMOS process, and consumes 21.8nJ/inference at 1kHz operating frequency, while achieving state-of-the-art classification accuracy of 85.7% and f1-score of 0.82. The energy consumption of the proposed RNN is 8 χ lower than state-of-the-art integrated circuits for arrhythmia detection.

This paper presents fully-integrated analog neural network classifier architecture for low resolution image classification that eliminates memory access. We design custom activation functions using single-stage common-source amplifiers, and apply a hardware-software co-design methodology to incorporate knowledge of the custom activation functions into the training phase to achieve high accuracy. Performing all computations entirely in the analog domain eliminates energy cost associated with memory access and data movement. We demonstrate our classifier on multinomial classification task of recognizing downsampled handwritten digits from MNIST dataset. Fabricated in 65nm CMOS process, the measured energy consumption for down-sampled MNIST dataset is 173pJ/classification, which is 3× better than state-of-the-art. The prototype IC achieves mean classification accuracy of 81.3% even after down-sampling the original MNIST images by 96% from 28 × 28 pixels to 5 × 5 pixels.

We built a natural language processing (NLP) language model that can be used to extract cancer treatment information using structured and unstructured electronic medical records (EMR). Our work appears to be the first that combines EMR and NLP for treatment identification.

An analog artificial neural network (ANN) classifier using a common-source amplifier based nonlinear activation function is presented in this work. A shallow ANN is designed using transistor level circuits and a multinomial (10 classes) classification accuracy of 0.82 is achieved on the MNIST dataset which consists of handwritten images of digits from 0-9. Use of common-source amplifier structure simplifies the ANN and results in 5X lower energy consumption than existing analog classifiers. The classifier performance is validated using Spectre and Matlab simulations.

The fuzzy$k$-nearest neighbor classifier (F$k$NN) improves upon the flexibility of the$k$-nearest neighbor classifier by considering each class as a fuzzy set and estimating the membership of an unlabeled data instance for each of the classes. However, the question of validating the quality of the class memberships estimated by F$k$NN for a regular multiclass classification problem still remains mostly unanswered. In this paper, we attempt to address this issue by first proposing a novel direction of evaluating a fuzzy classifier by highlighting the importance of focusing on the class memberships estimated by F$k$NN instead of its misclassification error. This leads us to finding novel theoretical upper bounds, respectively, on the bias and the mean squared error of the class memberships estimated by F$k$NN. Additionally the proposed upper bounds are shown to converge toward zero with increasing availability of the labeled data points, under some elementary assumptions on the class distribution and membership function. The major advantages of this analysis are its simplicity, capability of a direct extension for multiclass problems, parameter independence, and practical implication in explaining the behavior of F$k$NN in diverse situations (such as in presence of class imbalance). Furthermore, we provide a detailed simulation study on artificial and real data sets to empirically support our claims.

Obtaining enough labeled data to robustly train complex discriminative models is a major bottleneck in the machine learning pipeline. A popular solution is combining multiple sources of weak supervision using generative models. The structure of these models affects the quality of the training labels, but is difficult to learn without any ground truth labels. We instead rely on weak supervision sources having some structure by virtue of being encoded programmatically. We present Coral, a paradigm that infers generative model structure by statically analyzing the code for these heuristics, thus significantly reducing the amount of data required to learn structure. We prove that Coral's sample complexity scales quasilinearly with the number of heuristics and number of relations identified, improving over the standard sample complexity, which is exponential in n for learning n-th degree relations. Empirically, Coral matches or outperforms traditional structure learning approaches by up to 3.81 F1 points. Using Coral to model dependencies instead of assuming independence results in better performance than a fully supervised model by 3.07 accuracy points when heuristics are used to label radiology data without ground truth labels.

Nowadays, a wide range of advanced techniques provides accurate and detailed 3D data about patients' anatomy, as captured by medical scans (MRI, CT, Micro CT, etc.). While medical imaging assists daily clinical practice, 3D patient-specific models (3D-PSMs) of anatomy have still a quite limited use. We consider part-based semantic annotation beneficial to bring 3D-PSMs into clinical practice. To this end, tools are needed to extract clinically relevant information from 3D models, to associate such knowledge with their corresponding parts, and to support the storage, sharing and searching of annotated 3D-PSMs in a structured manner. In this context, we present the Sem Anatomy3D framework, which demonstrates the idea of ontology-driven annotation and indexing of 3D-PSMs and their Parts-of-Relevance, characterized by anatomical landmarks and pathological markers (e.g. Articular and non-articular facets, ligament insertion sites, erosions). The key functionaity is to offer services for part-base annotation of 3D-PSMs which enables search or browse the 3D-PSM according to the annotation attached to its Parts-of-Relevance. The paper describes the results in terms of methods to support the part-based annotation of 3D-PSM, and the formalization of the data model to store and manage global and part-based annotation to improve search and analysis of 3D patient-specific anatomical models and subparts. Finally, we specialized our framework to support the diagnosis of rheumatoid arthritis in the carpal bones, but, in principle, it can support similar tasks in other clinical applications.