Towards Case-based Interpretability for Medical Federated Learning.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Machine Learning

Background:

• Explaining AI decisions is crucial for clinical adoption.
• Federated learning is increasingly used in medical AI to protect data privacy.
• Access to past data is restricted in federated learning settings.

Purpose of the Study:

• To develop a method for generating case-based explanations in medical federated learning.
• To address the challenge of inaccessible past data in federated learning.
• To enhance trust and adoption of AI in clinical practice through interpretable models.

Main Methods:

• Utilizing deep generative models to create synthetic case-based explanations.
• Applying the approach to a proof-of-concept for pleural effusion diagnosis.
• Using publicly available Chest X-ray data for model training and validation.

Main Results:

• Demonstrated the feasibility of generating synthetic examples for case-based interpretability.
• Showcased a method to provide explanations without direct access to training data.
• Generated privacy-preserving explanations for AI decisions in a medical context.

Conclusions:

• Deep generative models can effectively generate case-based explanations in medical federated learning.
• This approach facilitates AI adoption in clinical settings by addressing privacy and interpretability concerns.
• The method offers a promising solution for explainable AI in regulated medical environments.