Advancing Healthcare Through Computer Vision

We are excited to share that our paper titled "AdaAdapting with an Open Mind: Leveraging Open-Vocabulary Detectors for Closed Set Source-Free Domain Adaptive Object Detection" has been accepted for publication in the Findings Track of the prestigious CVPR 2026! This work presents a novel method for source-free domain adaptive object detection that leverages open-vocabulary detectors to adapt to new domains without requiring access to source data. We are thrilled to contribute to the field of computer vision and look forward to sharing our research with the community at CVPR 2026!

We are thrilled to announce that two papers from our lab have been accepted for publication at the prestigious IEEE International Symposium on Biomedical Imaging (ISBI) 2026! The accepted papers are: 1) "Robust and Efficient 3D Gaussian Splatting for Diagnostic Imaging" by Mrinal Tyagi, Ashish Suri, and Chetan Arora, which presents a novel method for 3D reconstruction in diagnostic imaging using Gaussian splatting. 2) "Attend What Matters: Leveraging Vision Foundational Models for Breast Cancer Classification Using Mammograms" by Samyak Sanghvi, Piyush Miglani, Sarvesh Shashikumar, Kaustubh R Borgavi, and Chetan Arora, which explores the use of vision foundational models for improving breast cancer classification from mammograms. We are excited to share our research with the community and look forward to presenting our work at ISBI 2026!

Rohan Dhanakshirur successfully presented his PhD Synopsis, "Advancing Automation in Microscopic and Endoscopic Neurosurgical Skills Assessment." His thesis develops automated, data-driven methods for evaluating minimally invasive neurosurgical skills like micro-suturing, micro-drilling, and endoscopic navigation using novel machine learning techniques. The work introduces image-based and video-based models, improved surgical instrument segmentation, and contributes large annotated datasets. Real-world testing shows strong alignment with expert assessments, supporting scalable, objective neurosurgical training and evaluation.

Ankita Raj successfully presented her PhD Synopsis, "Novel Attack and Defense Techniques for Learning-based Computer Vision Systems," . Her research investigates key security and privacy threats in deep learning vision models used in critical domains like healthcare and autonomous driving. It introduces methods to detect and implement backdoor attacks using natural triggers in face recognition and prompt-tuning in open-vocabulary object detectors. The thesis also explores efficient model stealing attacks on black-box medical imaging models and shows that models fine-tuned from large pretrained vision foundations are more vulnerable to theft than traditional architectures.

Kshtiz Jain, a School of AI MSR student, presented his thesis defense on deep learning models for breast cancer detection from mammograms that mimic clinical practice by analyzing multiple views and incorporating patient history. The transformer-based models achieve state-of-the-art performance, support full-resolution images, and are optimized for deployment on edge devices with a clinician-friendly interface for practical use.

Dr. Krithika Rangarajan successfully presented her PhD Viva Voce on "Utility of Deep Learning in Breast Cancer Imaging in India." Her research focuses on applying deep learning to improve early detection and interpretation of mammograms, developing and testing AI tools tailored for Indian clinical needs, and enhancing radiology training and reporting.