U-FDL-PPE: A Unified Federated Deep Learning Framework with Privacy-Preserving Explainability for Early and Accurate Viral Disease Prediction

Anupam Agrawal¹ASADI SRINIVASULU¹Anant Mohan²Ramchand Vedaiyan^3*Kalavagunta Varshita⁴K Vijaya Bhaskar⁵

• ¹Indian Institute of Information Technology Allahabad, Allahabad, India
• ²All India Institute of Medical Sciences New Delhi Department of Pedodontics and Preventive Dentistry, New Delhi, India
• ³Villa college, Malé, Maldives
• ⁴BITS Pilani - Dubai Campus, Dubai, United Arab Emirates
• ⁵Chadalawada Ramanamma Engineering College Department of Management Studies, Tirupati, India

Accurate and early detection of viral diseases plays a vital role in improving public health outcomes and enabling prompt clinical decisions. However, traditional centralized deep learning systems struggle with serious issues like patient data privacy, lack of interoperability across institutions, and limited transparency especially when multiple healthcare centers are involved. To overcome these challenges, this research introduces U-FDL-PPE, a unified federated deep learning framework designed with built-in privacy-preserving and explainable AI capabilities. Its goal is to support accurate and trustworthy disease prediction within a decentralized setup. Our approach uses federated learning (FL), which allows deep neural networks to be trained collaboratively across different institutions without ever sharing raw patient data. The experiment was conducted using the publicly available COVID-19 Radiography Database, simulating a three-client federated environment to classify cases into COVID-19, Normal, and Viral Pneumonia. The framework employs MobileNetV2 as the backbone for image classification and integrates Grad-CAM to provide visual explanations for its predictions. After five rounds of federated training, the model achieved strong results 88% accuracy, an F1 score of 89.66%, and a multi-class AUC of 0.5192. The confusion matrix showed consistent class-wise accuracy, and the Grad-CAM outputs highlighted clinically relevant features in the images. Overall, U-FDL-PPE delivers a practical, scalable, and privacy-2 compliant AI solution suitable for real-world healthcare deployment. It not only meets the critical need for early diagnosis of viral infections but also builds clinician confidence through interpretable outputs and strict data confidentiality. This work sets a foundation for advancing trustworthy federated AI in medical diagnostics.