Bipasha Kundu

Bipasha Kundu

Research Assistant at Chester F. Carlson Center for Imaging Science, Rochester Institute of Technology

About me

I am a 4th year Ph.D. student in Imaging Science at Rochester Institute of Technology (RIT), working as a Research Assistant at the BiMVisIGN Lab. My research sits at the intersection of medical image analysis, deep learning, and computer vision, with a primary focus on cardiac MRI, specifically the left atrium (LA). My work spans automatic segmentation, deformable image registration, cardiac motion estimation, and vision-language model evaluation for clinically grounded image quality assessment. I am passionate about building robust, data-efficient deep learning pipelines that can directly support clinical decision-making — particularly for patients undergoing ablation therapy for atrial fibrillation (AF).

Prior to RIT, I earned my M.S. in Electrical Engineering from the University of Minnesota Duluth and my B.S. in Electronics & Communication Engineering from Khulna University of Engineering & Technology, Bangladesh. I have also gained industry experience as a Pre-Sales Engineer and Power & Infrastructure Specialist at multinational telecommunications companies in Bangladesh.

Research

The heart is one of the most complex and dynamic organs to image — and yet, accurate analysis of cardiac MRI is critical for millions of patients living with atrial fibrillation. My doctoral work is driven by a simple but ambitious question: how can we teach machines to see and understand the heart the way a skilled cardiologist does?

I work at the intersection of deep learning, computer vision, and cardiac imaging, building computational tools that help clinicians plan and evaluate ablation therapy, a procedure used to treat atrial fibrillation by deliberately creating small scars in heart tissue to disrupt abnormal electrical signals. Getting this right matters enormously: too little ablation leaves the arrhythmia untreated; too much damages healthy tissue. My research aims to give clinicians better, faster, and more reliable information at every stage of this process.

A central challenge I tackle is the left atrium segmentation problem — precisely delineating the left atrium from LGE-MRI scans, which are notoriously difficult to interpret due to low contrast, imaging artifacts, and high variability across patients. I have explored how large-scale vision foundation models, including self-supervised Vision Transformers like DINOv2, can be adapted to this task without requiring massive amounts of labeled medical data.

A second thread of my research asks: can we infer what a heart is doing from a single static image? Through cardiac motion estimation, I developed a framework that extracts motion patterns of the left atrium across the cardiac cycle, using those patterns as a signal to detect and localize scar tissue — even from images that capture just one frozen moment in time. This work was recognized as an oral presentation at SPIE Medical Imaging 2026

I have also worked on automated image quality assessment for LGE-MRI, recognizing that even the best segmentation model will fail if fed a poor-quality scan. By evaluating vision-language models for this task, I aim to create an early filter in the clinical pipeline that flags problematic acquisitions before they influence downstream analysis. I am currently exploring the integration of vision-language frameworks into the segmentation pipeline, with the goal of leveraging semantic and textual supervision to produce anatomically consistent, clinically interpretable delineations of the left atrium.

News

  • Our paper “A Two Stage Pipeline for Left Atrial Wall Constrained Scar Segmentation and Localization from LGE-MR Images” was accepted in EMBC 2026 (April 2026)

  • Our paper “Motion-Guided Scar Detection from Static Left Atrial MRI via Deformable Registration to a Healthy Atlas” was accepted in SPIE Medical Imaging 2026 for Oral Presentation (October 2025)

  • Our paper “Multi-Scale Feature Fusion with Image-Driven Spatial Integration for Left Atrium Segmentation from Cardiac MR Images” was accepted in EMBC 2025 (April 2025)

  • Our paper “Investigating the Domain Adaptability of General-Purpose Foundation Models for Left Atrium Segmentation from MR Images” was accepted in Functional Imaging and Modeling of the Heart (FIMH) 2025 (April 2025)

  • Attended SPIE Medical Imaging, 2025 to present our paper on “Assessing the Performance of the DINOv2 Self-supervised Learning Vision Transformer Model for the Segmentation of the Left Atrium from MRI Images”.

  • Received Travel Award from RIT Women in science (WISe) program (December 2024)

  • Our paper, “Assessing the Performance of the DINOv2 Self-supervised Learning Vision Transformer Model for the Segmentation of the Left Atrium from MRI Images” was accepted in SPIE Medical Imaging 2025 for Oral Presentation (October 2024)

  • Our paper, “Comparative analysis of non-rigid registration techniques for liver surface registration,” was accepted in SPIE Medical Imaging, 2024 (October 2023)

  • Passed Qualifying Exam (July 2023)

  • Started Ph.D. in Imaging Science, Rochester Institute of Technology (August 2022)

  • Started Ph.D. in Electrical Engineering, Texas Tech University (August 2021)

  • Started MS in Electrical Engineering, University of Minnesota Duluth (August 2018)

  • Joined Banglalink Digital Communications Ltd. as Power & Infrastructure Specialist (January 2018)

  • Joined Express Systems Ltd. as Pre-sales Engineer (January 2016)

  • Graduated (BSc in ECE) from Khulna University of Engineering & Technology (July 2015)