Nasir Hayat

Praveenkumar Kanithi, Clément Christophe, Marco AF Pimentel et al.

While Large Language Models (LLMs) achieve superhuman performance on standardized medical licensing exams, these static benchmarks have become saturated and increasingly disconnected from the functional requirements of clinical workflows. To bridge the gap between theoretical capability and verified utility, we introduce MEDIC, a comprehensive evaluation framework establishing leading indicators across various clinical dimensions. Beyond standard question-answering, we assess operational capabilities using deterministic execution protocols and a novel Cross-Examination Framework (CEF), which quantifies information fidelity and hallucination rates without reliance on reference texts. Our evaluation across a heterogeneous task suite exposes critical performance trade-offs: we identify a significant knowledge-execution gap, where proficiency in static retrieval does not predict success in operational tasks such as clinical calculation or SQL generation. Furthermore, we observe a divergence between passive safety (refusal) and active safety (error detection), revealing that models fine-tuned for high refusal rates often fail to reliably audit clinical documentation for factual accuracy. These findings demonstrate that no single architecture dominates across all dimensions, highlighting the necessity of a portfolio approach to clinical model deployment. As part of this investigation, we released a public leaderboard on Hugging Face.\footnote{https://huggingface.co/spaces/m42-health/MEDIC-Benchmark}

Nasir Hayat, Krzysztof J. Geras, Farah E. Shamout

Multi-modal fusion approaches aim to integrate information from different data sources. Unlike natural datasets, such as in audio-visual applications, where samples consist of "paired" modalities, data in healthcare is often collected asynchronously. Hence, requiring the presence of all modalities for a given sample is not realistic for clinical tasks and significantly limits the size of the dataset during training. In this paper, we propose MedFuse, a conceptually simple yet promising LSTM-based fusion module that can accommodate uni-modal as well as multi-modal input. We evaluate the fusion method and introduce new benchmark results for in-hospital mortality prediction and phenotype classification, using clinical time-series data in the MIMIC-IV dataset and corresponding chest X-ray images in MIMIC-CXR. Compared to more complex multi-modal fusion strategies, MedFuse provides a performance improvement by a large margin on the fully paired test set. It also remains robust across the partially paired test set containing samples with missing chest X-ray images. We release our code for reproducibility and to enable the evaluation of competing models in the future.

Nasir Hayat, Hazem Lashen, Farah E. Shamout

Despite the success of deep neural networks in chest X-ray (CXR) diagnosis, supervised learning only allows the prediction of disease classes that were seen during training. At inference, these networks cannot predict an unseen disease class. Incorporating a new class requires the collection of labeled data, which is not a trivial task, especially for less frequently-occurring diseases. As a result, it becomes inconceivable to build a model that can diagnose all possible disease classes. Here, we propose a multi-label generalized zero shot learning (CXR-ML-GZSL) network that can simultaneously predict multiple seen and unseen diseases in CXR images. Given an input image, CXR-ML-GZSL learns a visual representation guided by the input's corresponding semantics extracted from a rich medical text corpus. Towards this ambitious goal, we propose to map both visual and semantic modalities to a latent feature space using a novel learning objective. The objective ensures that (i) the most relevant labels for the query image are ranked higher than irrelevant labels, (ii) the network learns a visual representation that is aligned with its semantics in the latent feature space, and (iii) the mapped semantics preserve their original inter-class representation. The network is end-to-end trainable and requires no independent pre-training for the offline feature extractor. Experiments on the NIH Chest X-ray dataset show that our network outperforms two strong baselines in terms of recall, precision, f1 score, and area under the receiver operating characteristic curve. Our code is publicly available at: https://github.com/nyuad-cai/CXR-ML-GZSL.git

Nasir Hayat, Munawar Hayat, Shafin Rahman et al.

The existing zero-shot detection approaches project visual features to the semantic domain for seen objects, hoping to map unseen objects to their corresponding semantics during inference. However, since the unseen objects are never visualized during training, the detection model is skewed towards seen content, thereby labeling unseen as background or a seen class. In this work, we propose to synthesize visual features for unseen classes, so that the model learns both seen and unseen objects in the visual domain. Consequently, the major challenge becomes, how to accurately synthesize unseen objects merely using their class semantics? Towards this ambitious goal, we propose a novel generative model that uses class-semantics to not only generate the features but also to discriminatively separate them. Further, using a unified model, we ensure the synthesized features have high diversity that represents the intra-class differences and variable localization precision in the detected bounding boxes. We test our approach on three object detection benchmarks, PASCAL VOC, MSCOCO, and ILSVRC detection, under both conventional and generalized settings, showing impressive gains over the state-of-the-art methods. Our codes are available at https://github.com/nasir6/zero_shot_detection.

Prateek Munjal, Nasir Hayat, Munawar Hayat et al.

Active learning (AL) is a promising ML paradigm that has the potential to parse through large unlabeled data and help reduce annotation cost in domains where labeling data can be prohibitive. Recently proposed neural network based AL methods use different heuristics to accomplish this goal. In this study, we demonstrate that under identical experimental settings, different types of AL algorithms (uncertainty based, diversity based, and committee based) produce an inconsistent gain over random sampling baseline. Through a variety of experiments, controlling for sources of stochasticity, we show that variance in performance metrics achieved by AL algorithms can lead to results that are not consistent with the previously reported results. We also found that under strong regularization, AL methods show marginal or no advantage over the random sampling baseline under a variety of experimental conditions. Finally, we conclude with a set of recommendations on how to assess the results using a new AL algorithm to ensure results are reproducible and robust under changes in experimental conditions. We share our codes to facilitate AL evaluations. We believe our findings and recommendations will help advance reproducible research in AL using neural networks. We open source our code at https://github.com/PrateekMunjal/TorchAL

Clément Christophe, Praveen K Kanithi, Prateek Munjal et al.

This study presents a comprehensive analysis and comparison of two predominant fine-tuning methodologies - full-parameter fine-tuning and parameter-efficient tuning - within the context of medical Large Language Models (LLMs). We developed and refined a series of LLMs, based on the Llama-2 architecture, specifically designed to enhance medical knowledge retrieval, reasoning, and question-answering capabilities. Our experiments systematically evaluate the effectiveness of these tuning strategies across various well-known medical benchmarks. Notably, our medical LLM Med42 showed an accuracy level of 72% on the US Medical Licensing Examination (USMLE) datasets, setting a new standard in performance for openly available medical LLMs. Through this comparative analysis, we aim to identify the most effective and efficient method for fine-tuning LLMs in the medical domain, thereby contributing significantly to the advancement of AI-driven healthcare applications.

Nasir Hayat, Krzysztof J. Geras, Farah E. Shamout

The healthcare domain is characterized by heterogeneous data modalities, such as imaging and physiological data. In practice, the variety of medical data assists clinicians in decision-making. However, most of the current state-of-the-art deep learning models solely rely upon carefully curated data of a single modality. In this paper, we propose a dynamic training approach to learn modality-specific data representations and to integrate auxiliary features, instead of solely relying on a single modality. Our preliminary experiments results for a patient phenotyping task using physiological data in MIMIC-IV & chest radiographs in the MIMIC- CXR dataset show that our proposed approach achieves the highest area under the receiver operating characteristic curve (AUROC) (0.764 AUROC) compared to the performance of the benchmark method in previous work, which only used physiological data (0.740 AUROC). For a set of five recurring or chronic diseases with periodic acute episodes, including cardiac dysrhythmia, conduction disorders, and congestive heart failure, the AUROC improves from 0.747 to 0.798. This illustrates the benefit of leveraging the chest imaging modality in the phenotyping task and highlights the potential of multi-modal learning in medical applications.

Ghadeer O. Ghosheh, Bana Alamad, Kai-Wen Yang et al.

Existing prognostic tools mainly focus on predicting the risk of mortality among patients with coronavirus disease 2019. However, clinical evidence suggests that COVID-19 can result in non-mortal complications that affect patient prognosis. To support patient risk stratification, we aimed to develop a prognostic system that predicts complications common to COVID-19. In this retrospective study, we used data collected from 3,352 COVID-19 patient encounters admitted to 18 facilities between April 1 and April 30, 2020, in Abu Dhabi (AD), UAE. The hospitals were split based on geographical proximity to assess for our proposed system's learning generalizability, AD Middle region and AD Western & Eastern regions, A and B, respectively. Using data collected during the first 24 hours of admission, the machine learning-based prognostic system predicts the risk of developing any of seven complications during the hospital stay. The complications include secondary bacterial infection, AKI, ARDS, and elevated biomarkers linked to increased patient severity, including d-dimer, interleukin-6, aminotransferases, and troponin. During training, the system applies an exclusion criteria, hyperparameter tuning, and model selection for each complication-specific model. The system achieves good accuracy across all complications and both regions. In test set A (587 patient encounters), the system achieves 0.91 AUROC for AKI and >0.80 AUROC for most of the other complications. In test set B (225 patient encounters), the respective system achieves 0.90 AUROC for AKI, elevated troponin, and elevated interleukin-6, and >0.80 AUROC for most of the other complications. The best performing models, as selected by our system, were mainly gradient boosting models and logistic regression. Our results show that a data-driven approach using machine learning can predict the risk of such complications with high accuracy.