Accurate standard plane acquisition in fetal ultrasound (US) videos is crucial for fetal growth assessment, anomaly detection, and adherence to clinical guidelines. However, manually selecting standard frames is time-consuming and prone to intra- and inter-sonographer variability. Existing methods primarily rely on image-based approaches that capture standard frames and then classify the input frames across different anatomies. This ignores the dynamic nature of video acquisition and its interpretation. To address these challenges, we introduce Multi-Tier Class-Aware Token Transformer (MCAT); a visual query-based video clip localization (VQ-VCL) method to assist sonographers by enabling them to capture a quick US sweep. By then providing a visual query of the anatomy they wish to analyze, MCAT returns the video clip containing the standard frames for that anatomy, facilitating thorough screening for potential anomalies. We evaluate MCAT on two ultrasound video datasets and a natural image VQ-VCL dataset based on Ego4D. Our model outperforms state-of-the-art methods by 10% and 13% mtIoU on the ultrasound datasets and by 5.35% mtIoU on the Ego4D dataset, using 96% fewer tokens. MCAT’s efficiency and accuracy have significant potential implications for public health, especially in low- and middle-income countries (LMICs), where it may enhance prenatal care by streamlining standard plane acquisition, simplifying US based screening, diagnosis and allowing sonographers to examine more patients.

Large Vision-Language Models (VLMs), possessing millions or billions of parameters, typically require large text and image datasets for effective fine-tuning. However, collecting data from various sites, especially in healthcare, is challenging due to strict privacy regulations. An alternative is to fine-tune these foundation models on end-user devices, such as in medical clinics and hospitals, without sending data to a server. These local clients typically have limited computing power and small datasets, which are not enough for fully fine-tuning large VLMs on their own. A naive solution to these scenarios is to leverage parameter-efficient fine-tuning (PEFT) strategies such as adapters and apply federated learning (FL) algorithms to combine the learned adapter weights, thereby respecting the resource limitations and data privacy of the clients. However, this approach does not fully leverage the knowledge from multiple adapters trained on diverse data distributions and for diverse tasks. The adapters are adversely impacted by data heterogeneity and task heterogeneity across clients resulting in sub-optimal convergence. To this end, we propose a novel framework called FedPIA that improves upon the naive combinations of FL and PEFT by introducing Permutation and Integration of the local Adapters in the server and global Adapters in the clients exploiting Wasserstein barycenters for improved blending of client-specific and client-agnostic knowledge. This layerwise permutation helps to bridge the gap in the parameter space of local and global adapters before integration. We conduct over 2000 client-level experiments utilizing 48 medical image datasets across five different medical vision-language FL task settings encompassing visual question answering as well as image and report-based multi-label disease detection. Our experiments involving diverse client settings, ten different modalities, and two VLM backbones demonstrate that FedPIA consistently outperforms the state-of-the-art PEFT-FL baselines.

Federated Learning (FL) in healthcare ensures patient privacy by allowing hospitals to collaboratively train machine learning models while keeping sensitive medical data secure and localized. Most existing research in FL has concentrated on unimodal scenarios, where all healthcare institutes share the same type of data. However, in real-world healthcare situations, some clients may have access to multiple types of data pertaining to the same disease. Multimodal Federated Learning (MMFL) utilizes multiple modalities to build a more powerful FL model than its unimodal counterpart. However, the impact of missing modality in different clients, called modality incongruity, has been greatly overlooked. This paper, for the first time, analyses the impact of modality incongruity and reveals its connection with data heterogeneity across participating clients. We particularly inspect whether incongruent MMFL with unimodal and multimodal clients is more beneficial than unimodal FL. Furthermore, we examine three potential routes of addressing this issue. Firstly, we study the effectiveness of various self-attention mechanisms towards incongruity-agnostic information fusion in MMFL. Secondly, we introduce a modality imputation network (MIN) pre-trained in a multimodal client for modality translation in unimodal clients and investigate its potential towards mitigating the missing modality problem. Thirdly, we introduce several client-level and server-level regularization techniques including Modality-aware knowledge Distillation (MAD) and Leave-one-out teacher (LOOT) towards mitigating modality incongruity effects. Experiments are conducted with Chest X-Ray and radiology reports under several MMFL settings on two publicly available real-world datasets, MIMIC-CXR and Open-I.

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/

Self-supervised learning (SSL) has achieved major advances in natural images and video understanding, but challenges remain in domains like echocardiography (heart ultrasound) due to subtle anatomical structures, complex temporal dynamics, and the current lack of domain-specific pre-trained models. Existing SSL approaches such as contrastive, masked modeling, and clustering-based methods struggle with high intersample similarity, sensitivity to low PSNR inputs common in ultrasound, or aggressive augmentations that distort clinically relevant features. We present DISCOVR (Distilled Image Supervision for Cross Modal Video Representation), a self-supervised dual-branch framework for cardiac ultrasound video representation learning. DISCOVR combines a clustering-based video encoder that models temporal dynamics with an online image encoder that extracts fine-grained spatial semantics. These branches are connected through a semantic cluster distillation loss that transfers anatomical knowledge from the evolving image encoder to the video encoder, enabling temporally coherent representations enriched with fine-grained semantic understanding. Evaluated on six echocardiography datasets spanning fetal, pediatric, and adult populations, DISCOVR outperforms both specialized video anomaly detection methods and state-of-the-art video-SSL baselines in zero-shot and linear probing setups, achieving superior segmentation transfer and strong downstream performance on clinically relevant tasks such as LVEF prediction. **Code available at:** [https://github.com/mdivyanshu97/DISCOVR](https://github.com/mdivyanshu97/DISCOVR)

Safe deployment of machine learning (ML) models in safety-critical domains such as medical imaging requires detecting inputs with characteristics not seen during training, known as out-of-distribution (OOD) detection, to prevent unreliable predictions. Effective OOD detection after deployment could benefit from access to the training data, enabling direct comparison between test samples and the training data distribution to identify differences. State-of-the-art OOD detection methods, however, either discard the training data after deployment or assume that test samples and training data are centrally stored together, an assumption that rarely holds in real-world settings. This is because shipping the training data with the deployed model is usually impossible due to the size of training databases, as well as proprietary or privacy constraints. We introduce the Isolation Network, an OOD detection framework that quantifies the difficulty of separating a target test sample from the training data by solving a binary classification task. We then propose Decentralized Isolation Networks (DIsoN), which enables the comparison of training and test data when data-sharing is impossible, by exchanging only model parameters between the remote computational nodes of training and deployment. We further extend DIsoN with class-conditioning, comparing a target sample solely with training data of its predicted class. We evaluate DIsoN on four medical imaging datasets (dermatology, chest X-ray, breast ultrasound, histopathology) across 12 OOD detection tasks. DIsoN performs favorably against existing methods while respecting data-privacy. This decentralized OOD detection framework opens the way for a new type of service that ML developers could provide along with their models: providing remote, secure utilization of their training data for OOD detection services. Code available at: https://github.com/FelixWag/DIsoN

Segmentation models for brain lesions in MRI are commonly developed for a specific disease and trained on data with a predefined set of MRI modalities. Such models cannot segment the disease using data with a different set of MRI modalities, nor can they segment other types of diseases. Moreover, this training paradigm prevents a model from using the advantages of learning from heterogeneous databases that may contain scans and segmentation labels for different brain pathologies and diverse sets of MRI modalities. Additionally, the confidentiality of patient data often prevents central data aggregation, necessitating a decentralized approach. Is it feasible to use Federated Learning (FL) to train a single model on client databases that contain scans and labels of different brain pathologies and diverse sets of MRI modalities? We demonstrate promising results by combining appropriate, simple, and practical modifications to the model and training strategy: Designing a model with input channels that cover the whole set of modalities available across clients, training with random modality drop, and exploring the effects of feature normalization methods. Evaluation on 7 brain MRI databases with 5 different diseases shows that this FL framework can train a single model that achieves very promising results in segmenting all disease types seen during training. Importantly, it can segment these diseases in new databases that contain sets of modalities different from those in training clients. These results demonstrate, for the first time, the feasibility and effectiveness of using FL to train a single 3D segmentation model on decentralised data with diverse brain diseases and MRI modalities, a necessary step towards leveraging heterogeneous real-world databases. Code: https://github.com/FelixWag/FedUniBrain

The articulatory geometric configurations of the vocal tract and the acoustic properties of the resultant speech sound are considered to have a strong causal relationship. This paper aims at finding a joint latent representation between the articulatory and acoustic domain for vowel sounds via invertible neural network models, while simultaneously preserving the respective domain-specific features. Our model utilizes a convolutional autoencoder architecture and normalizing flow-based models to allow both forward and inverse mappings in a semi-supervised manner, between the mid-sagittal vocal tract geometry of a two degrees-of-freedom articulatory synthesizer with 1D acoustic wave model and the Mel-spectrogram representation of the synthesized speech sounds. Our approach achieves satisfactory performance in achieving both articulatory-to-acoustic as well as acoustic-to-articulatory mapping, thereby demonstrating our success in achieving a joint encoding of both the domains.

We propose a mixed deep neural network strategy, incorporating parallel combination of Convolutional (CNN) and Recurrent Neural Networks (RNN), cascaded with deep autoencoders and fully connected layers towards automatic identification of imagined speech from EEG. Instead of utilizing raw EEG channel data, we compute the joint variability of the channels in the form of a covariance matrix that provide spatio-temporal representations of EEG. The networks are trained hierarchically and the extracted features are passed onto the next network hierarchy until the final classification. Using a publicly available EEG based speech imagery database we demonstrate around 23.45% improvement of accuracy over the baseline method. Our approach demonstrates the promise of a mixed DNN approach for complex spatialtemporal classification problems.

Speech-related Brain Computer Interfaces (BCI) aim primarily at finding an alternative vocal communication pathway for people with speaking disabilities. As a step towards full decoding of imagined speech from active thoughts, we present a BCI system for subject-independent classification of phonological categories exploiting a novel deep learning based hierarchical feature extraction scheme. To better capture the complex representation of high-dimensional electroencephalography (EEG) data, we compute the joint variability of EEG electrodes into a channel cross-covariance matrix. We then extract the spatio-temporal information encoded within the matrix using a mixed deep neural network strategy. Our model framework is composed of a convolutional neural network (CNN), a long-short term network (LSTM), and a deep autoencoder. We train the individual networks hierarchically, feeding their combined outputs in a final gradient boosting classification step. Our best models achieve an average accuracy of 77.9% across five different binary classification tasks, providing a significant 22.5% improvement over previous methods. As we also show visually, our work demonstrates that the speech imagery EEG possesses significant discriminative information about the intended articulatory movements responsible for natural speech synthesis.

Speech-related Brain Computer Interface (BCI) technologies provide effective vocal communication strategies for controlling devices through speech commands interpreted from brain signals. In order to infer imagined speech from active thoughts, we propose a novel hierarchical deep learning BCI system for subject-independent classification of 11 speech tokens including phonemes and words. Our novel approach exploits predicted articulatory information of six phonological categories (e.g., nasal, bilabial) as an intermediate step for classifying the phonemes and words, thereby finding discriminative signal responsible for natural speech synthesis. The proposed network is composed of hierarchical combination of spatial and temporal CNN cascaded with a deep autoencoder. Our best models on the KARA database achieve an average accuracy of 83.42% across the six different binary phonological classification tasks, and 53.36% for the individual token identification task, significantly outperforming our baselines. Ultimately, our work suggests the possible existence of a brain imagery footprint for the underlying articulatory movement related to different sounds that can be used to aid imagined speech decoding.

Vocal tract configurations play a vital role in generating distinguishable speech sounds, by modulating the airflow and creating different resonant cavities in speech production. They contain abundant information that can be utilized to better understand the underlying speech production mechanism. As a step towards automatic mapping of vocal tract shape geometry to acoustics, this paper employs effective video action recognition techniques, like Long-term Recurrent Convolutional Networks (LRCN) models, to identify different vowel-consonant-vowel (VCV) sequences from dynamic shaping of the vocal tract. Such a model typically combines a CNN based deep hierarchical visual feature extractor with Recurrent Networks, that ideally makes the network spatio-temporally deep enough to learn the sequential dynamics of a short video clip for video classification tasks. We use a database consisting of 2D real-time MRI of vocal tract shaping during VCV utterances by 17 speakers. The comparative performances of this class of algorithms under various parameter settings and for various classification tasks are discussed. Interestingly, the results show a marked difference in the model performance in the context of speech classification with respect to generic sequence or video classification tasks.