Poster: Cardiology Focus
AI-Deployable ECG Risk Stratification for Indian Cardiovascular Public Health
Authors & Affiliations
- Umathurappan Chandrasekar, BE (Mech)
Principal IT Architect, Infraspace, Apple Valley, MN, USA - Kannaiayan Prakash, BE (CS)
Senior Engineering Manager, Huntington National Bank, Minneapolis, MN, USA - Radhika Gupta, BDS, MPH (Epidemiology)
Public Health Specialist, Infraspace, Apple Valley, MN, USA
Correspondence: Umathurappan Chandrasekar, 15224 Florist Circle, Apple Valley, MN 55124; uma@infraspace.net
Conflict of Interest & Funding: No conflicts of interest. Funding: This work received no external funding.
Abstract
Background: Cardiovascular diseases cause 31% of Indian deaths, affecting younger demographics across diverse ethnic groups. Resource-limited PHCs need scalable AI screening.
Objective: Develop fair, robust ECG-based AI for acute MI/arrhythmia prediction with clinician oversight for rural deployment.
Methods: Multimodal CNN-LSTM models trained on 25,000 CREATE registry ECGs across 53 centers. Fairness via demographic debiasing; robustness tested with field noise. Bihar PHC pilots (n=4,500) integrated with ABDM. Federated learning preserved privacy.
Results: 93% AUC MI prediction, 91% arrhythmias. Fairness: <4% disparity (North vs South India, M/F). Robustness: 87% accuracy post-noise. Pilots reduced referrals 35%, 92% clinician acceptance.
Conclusion: Deployable AI enables equitable CVD screening, demonstrating transportability across India's diversity with auditable governance.
Table 1: Model Performance Across Indian Cohorts
| Cohort | MI AUC | Arrhythmia AUC | Fairness Gap | Rural Accuracy |
|---|---|---|---|---|
| North India | 0.94 | 0.92 | 3.2% | 88% |
| South India | 0.93 | 0.91 | - | 87% |
| Overall | 0.93 | 0.91 | <4% | 87% |
View Posters
