RadJEPA Research Achievement

At the Indian Institute of Technology Bombay, researchers are advancing clinically relevant AI for medical imaging with RadJEPA, a lightweight yet high-performance vision encoder for chest X-ray analysis. Developed at the Artificial Intelligence in Digital Health (AIDE) Lab, Koita Centre for Digital Health, the work is led by Prof. Dr. Kshitij Jadhav (Assistant Professor, KCDH | Chief Project Coordinator, AI Centres of Excellence, MoE, GoI), along with MS student Anas Khan. 

RadJEPA was pre-trained on over 8 lakh unlabelled chest X-ray images, learning robust visual representations without relying on diagnostic annotations. Across multiple radiology tasks including disease classification, anatomical segmentation, and report generation, the model consistently outperformed existing radiology-specific and general-purpose vision models. 

Notably, RadJEPA achieves these results with just 86 million parameters, outperforming models up to 10× larger, including a model developed by Microsoft trained under comparable data and computational settings. This demonstrates how carefully designed architectures from academic labs can rival and exceed large-scale industry models while using significantly fewer computational resources. 

RadJEPA also shows strong robustness across datasets from different hospitals and scanners, a key requirement for real-world clinical deployment. Now released as an open-source model, RadJEPA is being explored globally, reinforcing IIT Bombay’s leadership in efficient, clinically grounded healthcare AI. 🔗 Project page: https://huggingface.co/AIDElab-IITBombay/RadJEPA