SplitFed Learning for Dementia Diagnosis: A Privacy-Preserving Framework for Distributed Healthcare

Dementia affects millions of people worldwide, posing significant challenges for early diagnosis, monitoring, and treatment. With the growing adoption of artificial intelligence (AI) techniques in healthcare, it is crucial to address critical concerns related to data privacy and computational efficiency when developing AI-driven solutions for dementia care. AI-based methods require large amounts of patient data for training, this raises substantial privacy risks due to the sensitive nature of health information. The lack of robust privacy-preserving analysis techniques, combined with the increasing volume of health data, underscores the need for secure and efficient modeling approaches.In this study, we propose the application of SplitFed learning to safeguard patient data while optimizing computational resources in dementia diagnosis and prediction tasks. In proposed scheme, a AI model is divided between client and server for collaborative learning. Experiments were conducted with various client configurations, achieving an accuracy of 95% when the SplitFed model was trained with specific numbers of clients and communication rounds. The results demonstrate that SplitFed learning significantly enhances data privacy without compromising the accuracy of dementia predictions. Furthermore, it substantially reduces the processing burden on individual devices, highlighting its potential for real-time deployment in resource-constrained healthcare environments.