A Tri-Factor Adaptive Federated Learning Framework for Parkinson's Disease Diagnosis via Multi-Source Facial...

Area of Science:

• Artificial Intelligence in Healthcare
• Medical Data Privacy
• Neurological Disorder Diagnostics

Background:

• Early Parkinson's disease (PD) diagnosis is critical for effective management.
• Facial expression analysis, particularly 'masked face' symptoms, shows promise for PD detection.
• Data silos due to privacy concerns limit the development of robust diagnostic models.

Purpose of the Study:

• To propose the Tri-Factor Adaptive Federated Learning (TriAFL) framework for privacy-preserving collaborative analysis of facial expression data.
• To address challenges of data heterogeneity and Non-IID data in federated learning for PD diagnosis.
• To develop a scalable and accurate AI-driven tool for non-invasive Parkinson's disease screening.

Main Methods:

• Development of the TriAFL framework incorporating a three-dimensional client contribution evaluation (gradient, data, learning efficiency).
• Creation of the largest known facial expression dataset for Parkinson's disease patients through hospital collaboration.
• Integration of local data augmentation strategies to enhance diagnostic model performance.

Main Results:

• TriAFL demonstrated superior classification performance compared to conventional federated learning methods.
• The framework effectively handled Non-IID data, improving model generalizability.
• Experimental validation confirmed TriAFL's efficacy in accurately diagnosing Parkinson's disease.

Conclusions:

• TriAFL offers a robust solution for collaborative, privacy-preserving analysis of sensitive medical data.
• The framework significantly enhances the accuracy and reliability of AI-based diagnostic tools for neurological disorders.
• This approach facilitates the development of rapid, non-invasive screening methods, advancing AI in healthcare.