Varsha Satish

Varsha Satish

Deep Learning | LLMs | Medical Image Analysis

Talks and presentations

Conference Proceedings: 2026 The Society for AI in Vision and Ophthalmology (SAIVO) Annual Meeting Program

May 01, 2026

Conference proceedings on our abstract: “Ultra-widefield Diabetic Retinopathy Prescreening Using a Patch-Based Attention Multiple Instance Learning Model”

Markdown icon

Abstract:
Ultra-widefield (UWF) fundus imaging enables comprehensive diabetic retinopathy (DR) assessment but poses challenges for automated analysis due to the trade-off between large field of view, image resolution, and computational efficiency, often causing loss of fine lesion detail during image down-sampling. To address this, we developed a weakly supervised patch-based attention Multiple Instance Learning (MIL) framework using EfficientNet_B0 features and attention-based aggregation for ETDRS ≥47 DR prescreening. The model achieved AUROC values of 0.74 in cross-validation and 0.76 on independent testing while preserving clinically relevant lesion information through attention-guided localization.

Slides Link

Conference Proceedings: KNIGHT Challenge 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI)

March 29, 2022

Conference proceedings on our paper: “Multi-modal Information Fusion for Classification of Kidney Abnormalities”

Markdown icon

Abstract:
Being able to predict the outcome of a treatment has obvious utility in treatment planning. Retrospective studies investigating the correlation of various tumor morphological characteristics to the treatment outcomes are becoming increasingly feasible due to data collection and advances in machine learning. For renal cancers, computed tomography (CT) imaging is a widely used diagnostic modality owing to its highly discernible visible features. However, manual inspection of several CT images are quite labour-intensive and often subjective. To automate this task, we propose an attention-based deep learning framework that automatically analyzes renal tumors by fusing both the clinical and imaging features. We demonstrate its effectiveness on the 2022 Knight challenge.

Paper Link

Conference Proceedings: 2022 IEEE 19th International Symposium on Biomedical Imaging (ISBI)

March 29, 2022

Conference proceedings on our paper: “Improved Histology Image Classification under Label Noise Via Feature Aggregating Memory Banks”

Markdown icon

Abstract:
Medical images are often subjectively labeled or weakly supervised, which introduces both closed-set and open-set label noise. Deep learning (DL) models underperform when the quality of supervisory labels degrades. We propose a unified framework for robust training of DL models that takes into account both closed-set and open-set noise, whereas the latter has been largely overlooked. We use a contrastive learning framework and feature aggregating memory banks to identify and increase the emphasis on clean training samples. Based on two publicly available datasets we demonstrate improved results for both synthetically introduced and naturally occurring (unknown) label noise.

Paper Link