Radiological Image-Based Condition Detection and Report Correlation
NHA Hackathon - Problem Statement 2
Team Kantaka Sodhana - Felicitation Ceremony - IISc Bengaluru
The Event
The AB PM-JAY Auto-Adjudication Hackathon was organized by the National Health Authority in collaboration with IndiaAI Mission (under MeitY) and the Indian Institute of Science, Bengaluru. Over 3,500 participants registered nationwide, with solutions evaluated by an expert jury from government, academia, healthcare, and technology institutions. The hackathon addressed three critical problem statements aimed at strengthening speed, transparency, accuracy, and programme integrity in health claims adjudication under Ayushman Bharat PM-JAY, the world's largest publicly funded health insurance scheme.
The Problem
Healthcare insurance fraud in India often involves mismatches between radiological images (X-rays, CT scans, MRI) and the textual diagnostic reports accompanying insurance claims. Fraudulent actors submit radiology reports that do not correspond to the actual images, inflating conditions or fabricating findings to claim higher reimbursements under AB PM-JAY.
The National Health Authority posed a challenge: build an AI system capable of reading radiological images, detecting medical conditions present in them, and cross-correlating those findings with the accompanying textual reports. The system needed to flag discrepancies where a report claims a condition that the image does not support, or where image findings are omitted from the report to hide certain patterns.
Our Approach
Team Kantaka Sodhana built an end-to-end AI pipeline combining computer vision models for radiological image analysis with Natural Language Processing for report parsing. The system extracts condition indicators from medical images using deep learning models trained on radiology datasets, then parses the associated discharge summaries and diagnostic reports to verify consistency. An anomaly scoring engine compares both outputs and flags claims where the image evidence and text narrative diverge beyond a confidence threshold. The pipeline was designed for scale, capable of processing thousands of claims daily with minimal manual intervention.
Results and Impact
The solution demonstrated the ability to surface fraud patterns that manual review would miss, particularly in high-volume claims processing environments. With AB PM-JAY processing approximately 50,000 claims daily across 1,900+ treatment packages, even a small percentage improvement in fraud detection translates to crores saved in public healthcare funds. The jury recognized the solution for its innovation, scalability, and direct applicability to real-world healthcare claims workflows.
Technology Stack
Gallery
All winning teams with cheques at IISc Bengaluru
Andhra Prabha newspaper coverage - Telugu engineers shine at NHA Hackathon
Press and Media Mentions
National Health Authority concludes AB PM-JAY Auto-Adjudication Hackathon Showcase 2026 at IISc Bengaluru
Press Information Bureau (Government of India) - 9 May 2026
NHA recognises innovators for developing AI systems for healthcare claims processing and fraud detection
The Hindu - May 2026
IndiaAI Mission and NHA Felicitate Winners of AB PM-JAY Auto-Adjudication Hackathon Showcase 2026
Ease My Prep - 13 May 2026
See this solution in action
Watch DemoKantaka Sodhana - Recognition Log - PS-02