Explore the Research Behind Swaasthyaa

India has a highly fragmented healthcare system with considerable variation in access and quality between urban and rural areas. We describe Swasth AI, an integrated platform that combines artificial intelligence (AI) diagnostics and model fusion to help fill gaps in care delivery. Based on prevalent diseases (diabetes, tuberculosis, cancer, cardiovascular diseases), Swasth AI applies deep learning models (CNNs, RNN/LSTMs, Transformers) and ensemble learning for early diagnosis, supporting the Ayushman Bharat Digital Mission (ABDM). Our system integrates patient data from different sources into a single AI-enabled framework that provides interoperability while enabling real-time analytics. Results demonstrate high performance (accuracy, precision, recall, F1-score, AUC) with significant improvements in earliest-stage detection and care coordination throughout India's healthcare ecosystem.

This comprehensive project report documents the complete development lifecycle of Swasth AI, from problem identification to implementation and validation. The report details the design methodology, system architecture, AI model development, and real-world testing of a unified healthcare platform for India. Key contributions include the development of a hybrid cloud-edge architecture, implementation of explainable AI techniques, and successful integration with India's Ayushman Bharat Digital Mission. The project demonstrates measurable improvements in early disease detection rates (18% for TB, 22% for diabetic retinopathy) and validates the effectiveness of model fusion approaches in healthcare diagnostics.

This paper presents a detailed analysis of model fusion techniques applied to healthcare diagnostics, using the Swasth AI platform as a comprehensive case study. We demonstrate how combining Convolutional Neural Networks (CNNs) for medical imaging, Recurrent Neural Networks (RNNs) for sequential health data, and Transformers for clinical text analysis can significantly improve diagnostic accuracy. Our ensemble approach achieved 95% overall accuracy, outperforming individual models by 3-7%. The paper includes detailed performance comparisons, ablation studies, and practical considerations for deploying model fusion in resource-constrained healthcare environments.

This study examines the challenges and solutions for deploying AI-powered healthcare diagnostics in resource-constrained rural settings. Through the implementation of Swasth AI in multiple Primary Health Centers (PHCs) across India, we demonstrate how edge computing and offline-capable AI models can provide specialist-level diagnostic support in areas with limited connectivity and infrastructure. Our hybrid architecture maintained 92-94% diagnostic accuracy in rural settings compared to 94-96% in urban hospitals, while providing essential healthcare AI capabilities during connectivity outages. The paper presents practical guidelines for adapting healthcare AI systems to diverse infrastructure environments.

Trust and transparency are critical factors for the successful adoption of AI in healthcare settings. This paper presents comprehensive explainable AI (XAI) techniques implemented in the Swasth AI platform, including Grad-CAM visualizations for medical imaging, SHAP values for feature attribution, and natural language explanations for clinical decisions. Through user studies with healthcare providers (n=25), we demonstrate that explainable AI features significantly improve clinician trust (4.2/5.0 rating) and adoption rates (85% active usage). The paper provides practical guidelines for implementing XAI in healthcare AI systems and addresses the unique challenges of explaining complex medical AI decisions to diverse stakeholders.

This paper examines the technical and policy challenges of integrating AI-powered healthcare platforms with national digital health infrastructure, using the integration of Swasth AI with India's Ayushman Bharat Digital Mission (ABDM) as a case study. We detail the implementation of Health ID integration, consent management frameworks, and interoperability standards (HL7 FHIR) that enable seamless data exchange across healthcare providers. Our experience demonstrates successful integration with 98% Health ID linkage rates and full compliance with ABDM data standards. The paper provides a blueprint for other healthcare AI systems seeking to align with national digital health initiatives.