Federated learning framework integrating REFINED CNN and Deep Regression Forests.

Area of Science:

• Medical Informatics
• Machine Learning
• Data Privacy

Background:

• Predictive learning from medical data faces challenges due to privacy and security concerns.
• Federated learning offers a privacy-preserving approach for cross-silo predictions in healthcare.
• The impact of population heterogeneity on federated learners remains underexplored.

Purpose of the Study:

• To propose a methodology for detecting population heterogeneity in federated settings.
• To develop a federated version of Deep Regression Forests to handle heterogeneity.
• To evaluate the performance of the proposed federated approach combined with a CNN framework.

Main Methods:

• Development of a novel methodology to detect population heterogeneity in federated learning environments.
• Implementation of a federated Deep Regression Forests model.
• Integration of the REpresentation of Features as Images with NEighborhood Dependencies (REIFIN) CNN framework with federated learning.

Main Results:

• The proposed methodology effectively detects population heterogeneity in federated settings.
• The federated Deep Regression Forests demonstrate improved performance in handling heterogeneity.
• Combining federated Deep Regression Forests with REIFIN-CNN yields superior results compared to existing methods.

Conclusions:

• Federated learning is a viable approach for privacy-preserving medical data analysis.
• Addressing population heterogeneity is crucial for robust federated learning in healthcare.
• The proposed federated Deep Regression Forests offer a promising solution for heterogeneous medical data.