AI-Powered Osteoarthritis Classification & Reporting

What it does

Doctors capture a knee X-ray on a React web app or an Android app. The system grades osteoarthritis automatically and produces a web-based report.

How it works

• Flow: capture on React/Android → encrypted upload to S3 → Lambda trigger → SageMaker/EC2 inference → prediction → Lambda → report written to the EMR.
• A closed feedback loop retrains the model on doctor-validated labels, so it keeps improving with real clinical use.
• Earlier, as an intern, Rishabh ran EDA on DICOM images and built baseline classifiers (logistic regression, random forest) before moving to deep learning.

Results

• 96% classification accuracy.
• More than 90% concordance with radiologists in trials at SRM Trichy.
• Outperformed competitor RBKnee.
• Rishabh led the clinicians and engineers who delivered it.
• Follow-on research, "OA Classification and Reporting using Agentic AI and Deep Neural Networks", is ongoing with OsteosConnect.