Deval Mehta

Deval Mehta, Yaniv Gal, Adrian Bowling et al. · ibm-research

Recent years have witnessed a rapid development of automated methods for skin lesion diagnosis and classification. Due to an increasing deployment of such systems in clinics, it has become important to develop a more robust system towards various Out-of-Distribution(OOD) samples (unknown skin lesions and conditions). However, the current deep learning models trained for skin lesion classification tend to classify these OOD samples incorrectly into one of their learned skin lesion categories. To address this issue, we propose a simple yet strategic approach that improves the OOD detection performance while maintaining the multi-class classification accuracy for the known categories of skin lesion. To specify, this approach is built upon a realistic scenario of a long-tailed and fine-grained OOD detection task for skin lesion images. Through this approach, 1) First, we target the mixup amongst middle and tail classes to address the long-tail problem. 2) Later, we combine the above mixup strategy with prototype learning to address the fine-grained nature of the dataset. The unique contribution of this paper is two-fold, justified by extensive experiments. First, we present a realistic problem setting of OOD task for skin lesion. Second, we propose an approach to target the long-tailed and fine-grained aspects of the problem setting simultaneously to increase the OOD performance.

Litao Yang, Deval Mehta, Dwarikanath Mahapatra et al. · ibm-research

Recently, a lot of automated white blood cells (WBC) or leukocyte classification techniques have been developed. However, all of these methods only utilize a single modality microscopic image i.e. either blood smear or fluorescence based, thus missing the potential of a better learning from multimodal images. In this work, we develop an efficient multimodal architecture based on a first of its kind multimodal WBC dataset for the task of WBC classification. Specifically, our proposed idea is developed in two steps - 1) First, we learn modality specific independent subnetworks inside a single network only; 2) We further enhance the learning capability of the independent subnetworks by distilling knowledge from high complexity independent teacher networks. With this, our proposed framework can achieve a high performance while maintaining low complexity for a multimodal dataset. Our unique contribution is two-fold - 1) We present a first of its kind multimodal WBC dataset for WBC classification; 2) We develop a high performing multimodal architecture which is also efficient and low in complexity at the same time.

Deval Mehta, Shobi Sivathamboo, Hugh Simpson et al.

In this work, we contribute towards the development of video-based epileptic seizure classification by introducing a novel framework (SETR-PKD), which could achieve privacy-preserved early detection of seizures in videos. Specifically, our framework has two significant components - (1) It is built upon optical flow features extracted from the video of a seizure, which encodes the seizure motion semiotics while preserving the privacy of the patient; (2) It utilizes a transformer based progressive knowledge distillation, where the knowledge is gradually distilled from networks trained on a longer portion of video samples to the ones which will operate on shorter portions. Thus, our proposed framework addresses the limitations of the current approaches which compromise the privacy of the patients by directly operating on the RGB video of a seizure as well as impede real-time detection of a seizure by utilizing the full video sample to make a prediction. Our SETR-PKD framework could detect tonic-clonic seizures (TCSs) in a privacy-preserving manner with an accuracy of 83.9% while they are only half-way into their progression. Our data and code is available at https://github.com/DevD1092/seizure-detection

Raja Kumar, Raghav Singhal, Pranamya Kulkarni et al.

Deep multimodal learning has shown remarkable success by leveraging contrastive learning to capture explicit one-to-one relations across modalities. However, real-world data often exhibits shared relations beyond simple pairwise associations. We propose M3CoL, a Multimodal Mixup Contrastive Learning approach to capture nuanced shared relations inherent in multimodal data. Our key contribution is a Mixup-based contrastive loss that learns robust representations by aligning mixed samples from one modality with their corresponding samples from other modalities thereby capturing shared relations between them. For multimodal classification tasks, we introduce a framework that integrates a fusion module with unimodal prediction modules for auxiliary supervision during training, complemented by our proposed Mixup-based contrastive loss. Through extensive experiments on diverse datasets (N24News, ROSMAP, BRCA, and Food-101), we demonstrate that M3CoL effectively captures shared multimodal relations and generalizes across domains. It outperforms state-of-the-art methods on N24News, ROSMAP, and BRCA, while achieving comparable performance on Food-101. Our work highlights the significance of learning shared relations for robust multimodal learning, opening up promising avenues for future research. Our code is publicly available at https://github.com/RaghavSinghal10/M3CoL.

Xin Zhang, Deval Mehta, Yanan Hu et al.

Survival analysis holds a crucial role across diverse disciplines, such as economics, engineering and healthcare. It empowers researchers to analyze both time-invariant and time-varying data, encompassing phenomena like customer churn, material degradation and various medical outcomes. Given the complexity and heterogeneity of such data, recent endeavors have demonstrated successful integration of deep learning methodologies to address limitations in conventional statistical approaches. However, current methods typically involve cluttered probability distribution function (PDF), have lower sensitivity in censoring prediction, only model static datasets, or only rely on recurrent neural networks for dynamic modelling. In this paper, we propose a novel survival regression method capable of producing high-quality unimodal PDFs without any prior distribution assumption, by optimizing novel Margin-Mean-Variance loss and leveraging the flexibility of Transformer to handle both temporal and non-temporal data, coined UniSurv. Extensive experiments on several datasets demonstrate that UniSurv places a significantly higher emphasis on censoring compared to other methods.

Talha Ilyas, Deval Mehta, Zongyuan Ge

Skeleton-based human activity recognition has achieved strong empirical performance, yet most existing models remain black boxes and difficult to interpret. In this work, we introduce a neurosymbolic formulation of skeleton-based HAR that reframes action recognition as concept-driven first-order logical reasoning over motion primitives. Our framework bridges representation learning and symbolic inference by grounding first-order logic predicates in learnable spatial and temporal motion concepts. Specifically, we employ a standard spatio-temporal skeleton encoder to extract latent motion representations, which are then mapped to interpretable concept predicates via a spatio-temporal concept decoder that explicitly separates pose-centric and dynamics-centric abstractions. These concept predicates are composed through differentiable first-order logic layers, enabling the model to learn human-readable logical rules that govern action semantics. To impose semantic structure on the learned concepts, we align skeleton representations with LLM-derived descriptions of atomic motion primitives, establishing a shared conceptual space for perception and reasoning. Extensive experiments on NTU RGB+D 60/120 and NW-UCLA demonstrate that our approach achieves competitive recognition performance while providing explicit, interpretable explanations grounded in logical structure. Our results highlight neurosymbolic reasoning as an effective paradigm for interpretable spatio-temporal action understanding. Code: https://github.com/Mr-TalhaIlyas/REASON

Deval Mehta, Brigid Betz-Stablein, Toan D Nguyen et al.

The surge in developing deep learning models for diagnosing skin lesions through image analysis is notable, yet their clinical black faces challenges. Current dermatology AI models have limitations: limited number of possible diagnostic outputs, lack of real-world testing on uncommon skin lesions, inability to detect out-of-distribution images, and over-reliance on dermoscopic images. To address these, we present an All-In-One \textbf{H}ierarchical-\textbf{O}ut of Distribution-\textbf{C}linical Triage (HOT) model. For a clinical image, our model generates three outputs: a hierarchical prediction, an alert for out-of-distribution images, and a recommendation for dermoscopy if clinical image alone is insufficient for diagnosis. When the recommendation is pursued, it integrates both clinical and dermoscopic images to deliver final diagnosis. Extensive experiments on a representative cutaneous lesion dataset demonstrate the effectiveness and synergy of each component within our framework. Our versatile model provides valuable decision support for lesion diagnosis and sets a promising precedent for medical AI applications.

Xin Pei, Deval Mehta

Transcending the binary categorization of racist texts, our study takes cues from social science theories to develop a multi-dimensional model for racism detection, namely stigmatization, offensiveness, blame, and exclusion. With the aid of BERT and topic modeling, this categorical detection enables insights into the underlying subtlety of racist discussion on digital platforms during COVID-19. Our study contributes to enriching the scholarly discussion on deviant racist behaviours on social media. First, a stage-wise analysis is applied to capture the dynamics of the topic changes across the early stages of COVID-19 which transformed from a domestic epidemic to an international public health emergency and later to a global pandemic. Furthermore, mapping this trend enables a more accurate prediction of public opinion evolvement concerning racism in the offline world, and meanwhile, the enactment of specified intervention strategies to combat the upsurge of racism during the global public health crisis like COVID-19. In addition, this interdisciplinary research also points out a direction for future studies on social network analysis and mining. Integration of social science perspectives into the development of computational methods provides insights into more accurate data detection and analytics.

Litao Yang, Deval Mehta, Sidong Liu et al.

Digital pathology based on whole slide images (WSIs) plays a key role in cancer diagnosis and clinical practice. Due to the high resolution of the WSI and the unavailability of patch-level annotations, WSI classification is usually formulated as a weakly supervised problem, which relies on multiple instance learning (MIL) based on patches of a WSI. In this paper, we aim to learn an optimal patch-level feature space by integrating prototype learning with MIL. To this end, we develop a Trainable Prototype enhanced deep MIL (TPMIL) framework for weakly supervised WSI classification. In contrast to the conventional methods which rely on a certain number of selected patches for feature space refinement, we softly cluster all the instances by allocating them to their corresponding prototypes. Additionally, our method is able to reveal the correlations between different tumor subtypes through distances between corresponding trained prototypes. More importantly, TPMIL also enables to provide a more accurate interpretability based on the distance of the instances from the trained prototypes which serves as an alternative to the conventional attention score-based interpretability. We test our method on two WSI datasets and it achieves a new SOTA. GitHub repository: https://github.com/LitaoYang-Jet/TPMIL

Kunal Deo, Deval Mehta, Kshitij Jadhav

Long tail problems frequently arise in the medical field, particularly due to the scarcity of medical data for rare conditions. This scarcity often leads to models overfitting on such limited samples. Consequently, when training models on datasets with heavily skewed classes, where the number of samples varies significantly, a problem emerges. Training on such imbalanced datasets can result in selective detection, where a model accurately identifies images belonging to the majority classes but disregards those from minority classes. This causes the model to lack generalizability, preventing its use on newer data. This poses a significant challenge in developing image detection and diagnosis models for medical image datasets. To address this challenge, the One Shot GANs model was employed to augment the tail class of HAM10000 dataset by generating additional samples. Furthermore, to enhance accuracy, a novel metric tailored to suit One Shot GANs was utilized.

Yiwen Jiang, Deval Mehta, Wei Feng et al.

Concept Bottleneck Models (CBMs) decompose image classification into a process governed by interpretable, human-readable concepts. Recent advances in CBMs have used Large Language Models (LLMs) to generate candidate concepts. However, a critical question remains: What is the optimal number of concepts to use? Current concept banks suffer from redundancy or insufficient coverage. To address this issue, we introduce a dynamic, agent-based approach that adjusts the concept bank in response to environmental feedback, optimizing the number of concepts for sufficiency yet concise coverage. Moreover, we propose Conditional Concept Bottleneck Models (CoCoBMs) to overcome the limitations in traditional CBMs' concept scoring mechanisms. It enhances the accuracy of assessing each concept's contribution to classification tasks and feature an editable matrix that allows LLMs to correct concept scores that conflict with their internal knowledge. Our evaluations across 6 datasets show that our method not only improves classification accuracy by 6% but also enhances interpretability assessments by 30%.

Yiwen Jiang, Deval Mehta, Siyuan Yan et al.

Multimodal Large Language Models (MLLMs) have shown promise in visual-textual reasoning, with Multimodal Chain-of-Thought (MCoT) prompting significantly enhancing interpretability. However, existing MCoT methods rely on rationale-rich datasets and largely focus on inter-object reasoning, overlooking the intra-object understanding crucial for image classification. To address this gap, we propose WISE, a Weak-supervision-guided Step-by-step Explanation method that augments any image classification dataset with MCoTs by reformulating the concept-based representations from Concept Bottleneck Models (CBMs) into concise, interpretable reasoning chains under weak supervision. Experiments across ten datasets show that our generated MCoTs not only improve interpretability by 37% but also lead to gains in classification accuracy when used to fine-tune MLLMs. Our work bridges concept-based interpretability and generative MCoT reasoning, providing a generalizable framework for enhancing MLLMs in fine-grained visual understanding.

Deval Mehta, Yiwen Jiang, Catherine L Jan et al.

Recent advancements in deep learning have shown significant potential for classifying retinal diseases using color fundus images. However, existing works predominantly rely exclusively on image data, lack interpretability in their diagnostic decisions, and treat medical professionals primarily as annotators for ground truth labeling. To fill this gap, we implement two key strategies: extracting interpretable concepts of retinal diseases using the knowledge base of GPT models and incorporating these concepts as a language component in prompt-learning to train vision-language (VL) models with both fundus images and their associated concepts. Our method not only improves retinal disease classification but also enriches few-shot and zero-shot detection (novel disease detection), while offering the added benefit of concept-based model interpretability. Our extensive evaluation across two diverse retinal fundus image datasets illustrates substantial performance gains in VL-model based few-shot methodologies through our concept integration approach, demonstrating an average improvement of approximately 5.8\% and 2.7\% mean average precision for 16-shot learning and zero-shot (novel class) detection respectively. Our method marks a pivotal step towards interpretable and efficient retinal disease recognition for real-world clinical applications.

Talha Ilyas, Duong Nhu, Allison Thomas et al.

Accurate fetal movement (FM) detection is essential for assessing prenatal health, as abnormal movement patterns can indicate underlying complications such as placental dysfunction or fetal distress. Traditional methods, including maternal perception and cardiotocography (CTG), suffer from subjectivity and limited accuracy. To address these challenges, we propose Contrastive Ultrasound Video Representation Learning (CURL), a novel self-supervised learning framework for FM detection from extended fetal ultrasound video recordings. Our approach leverages a dual-contrastive loss, incorporating both spatial and temporal contrastive learning, to learn robust motion representations. Additionally, we introduce a task-specific sampling strategy, ensuring the effective separation of movement and non-movement segments during self-supervised training, while enabling flexible inference on arbitrarily long ultrasound recordings through a probabilistic fine-tuning approach. Evaluated on an in-house dataset of 92 subjects, each with 30-minute ultrasound sessions, CURL achieves a sensitivity of 78.01% and an AUROC of 81.60%, demonstrating its potential for reliable and objective FM analysis. These results highlight the potential of self-supervised contrastive learning for fetal movement analysis, paving the way for improved prenatal monitoring and clinical decision-making.

Anurag Shandilya, Swapnil Bhat, Akshat Gautam et al.

Generative models have proven to be very effective in generating synthetic medical images and find applications in downstream tasks such as enhancing rare disease datasets, long-tailed dataset augmentation, and scaling machine learning algorithms. For medical applications, the synthetically generated medical images by such models are still reasonable in quality when evaluated based on traditional metrics such as FID score, precision, and recall. However, these metrics fail to capture the medical/biological plausibility of the generated images. Human expert feedback has been used to get biological plausibility which demonstrates that these generated images have very low plausibility. Recently, the research community has further integrated this human feedback through Reinforcement Learning from Human Feedback(RLHF), which generates more medically plausible images. However, incorporating human feedback is a costly and slow process. In this work, we propose a novel approach to improve the medical plausibility of generated images without the need for human feedback. We introduce IMPROVE:Improving Medical Plausibility without Reliance on Human Validation - An Enhanced Prototype-Guided Diffusion Framework, a prototype-guided diffusion process for medical image generation and show that it substantially enhances the biological plausibility of the generated medical images without the need for any human feedback. We perform experiments on Bone Marrow and HAM10000 datasets and show that medical accuracy can be substantially increased without human feedback.

Xin Pei, Deval Mehta

Transcending the binary categorization of racist and xenophobic texts, this research takes cues from social science theories to develop a four dimensional category for racism and xenophobia detection, namely stigmatization, offensiveness, blame, and exclusion. With the aid of deep learning techniques, this categorical detection enables insights into the nuances of emergent topics reflected in racist and xenophobic expression on Twitter. Moreover, a stage wise analysis is applied to capture the dynamic changes of the topics across the stages of early development of Covid-19 from a domestic epidemic to an international public health emergency, and later to a global pandemic. The main contributions of this research include, first the methodological advancement. By bridging the state-of-the-art computational methods with social science perspective, this research provides a meaningful approach for future research to gain insight into the underlying subtlety of racist and xenophobic discussion on digital platforms. Second, by enabling a more accurate comprehension and even prediction of public opinions and actions, this research paves the way for the enactment of effective intervention policies to combat racist crimes and social exclusion under Covid-19.

Deval Mehta, Umar Asif, Tian Hao et al.

This paper presents a novel deep learning enabled, video based analysis framework for assessing the Unified Parkinsons Disease Rating Scale (UPDRS) that can be used in the clinic or at home. We report results from comparing the performance of the framework to that of trained clinicians on a population of 32 Parkinsons disease (PD) patients. In-person clinical assessments by trained neurologists are used as the ground truth for training our framework and for comparing the performance. We find that the standard sit-to-stand activity can be used to evaluate the UPDRS sub-scores of bradykinesia (BRADY) and posture instability and gait disorders (PIGD). For BRADY we find F1-scores of 0.75 using our framework compared to 0.50 for the video based rater clinicians, while for PIGD we find 0.78 for the framework and 0.45 for the video based rater clinicians. We believe our proposed framework has potential to provide clinically acceptable end points of PD in greater granularity without imposing burdens on patients and clinicians, which empowers a variety of use cases such as passive tracking of PD progression in spaces such as nursing homes, in-home self-assessment, and enhanced tele-medicine.

Umar Asif, Deval Mehta, Stefan von Cavallar et al.

Existing action recognition methods mainly focus on joint and bone information in human body skeleton data due to its robustness to complex backgrounds and dynamic characteristics of the environments. In this paper, we combine body skeleton data with spatial and motion features from face and two hands, and present "Deep Action Stamps (DeepActs)", a novel data representation to encode actions from video sequences. We also present "DeepActsNet", a deep learning based ensemble model which learns convolutional and structural features from Deep Action Stamps for highly accurate action recognition. Experiments on three challenging action recognition datasets (NTU60, NTU120, and SYSU) show that the proposed model trained using Deep Action Stamps produce considerable improvements in the action recognition accuracy with less computational cost compared to the state-of-the-art methods.

Xin Pei, Deval Mehta

Situated in the global outbreak of COVID-19, our study enriches the discussion concerning the emergent racism and xenophobia on social media. With big data extracted from Twitter, we focus on the analysis of negative sentiment reflected in tweets marked with racist hashtags, as racism and xenophobia are more likely to be delivered via the negative sentiment. Especially, we propose a stage-based approach to capture how the negative sentiment changes along with the three development stages of COVID-19, under which it transformed from a domestic epidemic into an international public health emergency and later, into the global pandemic. At each stage, sentiment analysis enables us to recognize the negative sentiment from tweets with racist hashtags, and keyword extraction allows for the discovery of themes in the expression of negative sentiment by these tweets. Under this public health crisis of human beings, this stage-based approach enables us to provide policy suggestions for the enactment of stage-specific intervention strategies to combat racism and xenophobia on social media in a more effective way.