Federated Machine Learning for Skin Lesion Diagnosis: An Asynchronous and Weighted Approach.

Area of Science:

• Artificial Intelligence in Medicine
• Computational Dermatology
• Machine Learning for Healthcare

Background:

• Accurate and timely skin cancer diagnosis is critical for patient outcomes.
• Traditional machine learning for healthcare faces significant data privacy challenges.
• Existing methods may not adequately balance diagnostic accuracy with privacy preservation.

Purpose of the Study:

• To propose a privacy-aware machine learning approach for enhanced skin cancer detection.
• To address data privacy concerns in the application of machine learning in healthcare settings.
• To improve the efficiency and accuracy of federated learning models for medical diagnosis.

Main Methods:

• Utilized asynchronous federated learning combined with convolutional neural networks (CNNs).
• Optimized communication rounds by stratifying CNN layers into shallow and deep components.
• Implemented a temporally weighted aggregation strategy for improved central model convergence.

Main Results:

• The proposed privacy-aware approach demonstrated superior accuracy compared to existing methods.
• Achieved higher diagnostic accuracy with a significant reduction in communication rounds.
• Outperformed current techniques in terms of overall communication cost efficiency.

Conclusions:

• The developed method offers a promising solution for improving skin cancer diagnosis.
• Effectively addresses critical data privacy concerns in machine learning-based healthcare applications.
• Federated learning with temporal weighting presents a viable strategy for privacy-preserving medical AI.