Federated Learning-Based Multimodal Approach for Early Detection and Personalized Care in Cardiac Disease

Sultan Alasmari ¹ Rayed AlGhamdi ² Ghanshyam Tejani ^3,4 Sunil Kumar ¹ Seyed Jalaleddin Mousavirad ^5*

• ¹ Majmaah University, Al Majma'ah, Riyadh, Saudi Arabia
• ² King Abdulaziz University, Jeddah, Makkah, Saudi Arabia
• ³ Yuan Ze University, Zhongli District, Taiwan
• ⁴ Applied Science Private University, Amman, Amman, Jordan
• ⁵ Mid Sweden University, Sundsvall, Västernorrland, Sweden

This research presents a framework based on privacy for early detection of heart diseases using multimodal data analysis and federated learning. Any current technique remains ineffective due to the misintegration of health data, which contributes to the poor accuracy of such techniques. To achieve this result, we propose an attention-based feature fusion model for combining relevant features from cardiac images, ECG signals, patient records and nutrition data.Gradient Descent (SGD-DNN) to maintain data privacy and scalability. The fused feature vector serves as input to the SGD-DNN and provides a refined cardiac disease classification as output. Hence, it allows early diagnosis, personalized lifestyle recommendations, and remarkably improves patient care while ensuring data confidentiality. The high accuracies achieved (97.76% in Database 1, 98.43% in Database 2, and 99.12% in Database 3) demonstrate the potential application of the framework in real-world clinical practice, thus providing a solid, scalable, and privacy-centered solution for heart disease management.