Enhancing Alzheimer's disease classification through split federated learning and GANs for imbalanced datasets.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Computational Neuroscience

Background:

• Healthcare data imbalance and privacy concerns hinder accurate medical classification.
• Existing models struggle with decentralized and imbalanced datasets for diseases like Alzheimer's.
• Advanced technologies are crucial for improving diagnostic accuracy and patient care.

Purpose of the Study:

• To develop a privacy-preserving medical classification model for Alzheimer's disease detection.
• To address challenges of imbalanced datasets and data decentralization in medical AI.
• To enhance the generalization capabilities of AI models in clinical settings.

Main Methods:

• Utilized split federated learning (SFL) for decentralized model training without data sharing.
• Integrated conditional generative adversarial networks (cGANs) to synthesize realistic data for minority classes.
• Developed a hybrid approach combining SFL and cGANs for robust Alzheimer's disease classification.

Main Results:

• Achieved approximately 83.54% accuracy in Alzheimer's disease classification.
• Demonstrated effective learning from decentralized and imbalanced medical datasets.
• Successfully preserved patient data privacy through the federated learning framework.

Conclusions:

• The proposed SFL and cGANs model effectively addresses data privacy and imbalance in medical classification.
• This approach enhances diagnostic capabilities for Alzheimer's disease, improving potential patient outcomes.
• The methodology offers a scalable solution for developing AI in healthcare while respecting data protection regulations.