Skin disease diagnostics through federated transfer learning on heterogeneous data.

Area of Science:

• Dermatology and Artificial Intelligence
• Medical Informatics
• Computational Health

Background:

• Skin diseases present significant physical and mental health challenges globally.
• Accurate and timely diagnosis is critical for effective dermatological treatment.
• Machine learning (ML) and deep learning (DL) show promise for enhancing diagnostic accuracy but require large datasets.

Purpose of the Study:

• To develop a privacy-preserving federated transfer learning framework for skin disease diagnosis.
• To address data shortages and privacy concerns in healthcare data sharing for ML models.
• To improve the efficiency and accuracy of dermatological diagnostic tools.

Main Methods:

• Utilized transfer learning with a dense neural network (DNN) for initial skin disease detection.
• Employed pre-trained architectures for feature extraction, followed by DNN classification.
• Integrated federated learning (FL) to train models across distributed nodes while preserving data privacy.
• Combined FL with transfer learning to create a secure and effective diagnostic ecosystem.

Main Results:

• The proposed feature extraction with federated learning model achieved high cross-validation accuracy: 99.528% on IID and 99.689% on non-IID databases.
• Demonstrated the capability to create efficient, lightweight models suitable for resource-constrained environments.
• Showcased ensemble learning's ability to enhance edge device performance.

Conclusions:

• Federated transfer learning offers a powerful, privacy-preserving solution for skin disease diagnosis.
• The developed model provides an efficient and accurate tool for modern healthcare settings.
• This approach effectively overcomes data limitations and privacy regulations in medical AI.