MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202641092205 A) filed by Ravindra College Of Engineering For Women on July 29, 2026, for Federated Learning For Equitable Dermatological Care.
Inventors include T. Aditya Sai Srinivas; K. Dora Babu; S. Saritha; M. Jyothirmai; and Dr. Mohebbanaaz.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: Diagnosing skin diseases accurately remains a major challenge due to visual complexity, data privacy concerns, and biased performance on underrepresented skin tones. Traditional centralized AI models raise ethical issues, while federated learning (FL) approaches often falter with non-IID data and limited real-world validation. In this study, we introduce a privacyfocused, federated deep learning framework that integrates Enhanced CNNs (EfficientNet-B4 and ResNet-152) with FL algorithms (FedAvg and FedProx). Using a diverse dataset of 5,172 images across five common skin conditions 30% from darker skin tones we demonstrate that FedProx better handles data heterogeneity and achieves superior accuracy (95.7%) and faster convergence (27% improvement) compared to FedAvg. Edge deployment on Raspberry Pi ensures GDPR/HIPAA-compliant, real-time inference (23 ms latency). Notably, our model narrows diagnostic performance gaps across skin types and achieves 87% expert agreement and 92% patient satisfaction in clinical validation. This work advances AI-powered dermatology while upholding data privacy and inclusivity.
Disclaimer: Curated by HT Syndication.
