A Pipeline for the Implementation and Visualization of Explainable Machine Learning for Medical Imaging Using...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Medicine
• Radiomics

Background:

• Machine learning (ML) models accurately predict clinical factors from medical images but lack interpretability, hindering clinical trust.
• Current explainable ML (XML) methods like saliency maps offer limited insight into image feature importance.
• Trustworthy AI in medicine requires transparent prediction models for critical clinical decisions.

Purpose of the Study:

• To introduce a novel pipeline for explainable machine learning (XML) in medical imaging analysis.
• To enhance the interpretability of complex ML models used in medical imaging.
• To increase physician trust and aid decision-making in AI-driven medical diagnostics.

Main Methods:

• Developed a novel XML pipeline integrating radiomics data and Shapley values.
• Utilized well-defined predictors from medical imaging for model building.
• Created a clinician-focused dashboard for visualizing XML imaging results.

Main Results:

• Successfully explained outcome predictions from complex ML models built with medical imaging.
• Demonstrated a workflow for developing and explaining prediction models using MRI data.
• Provided interpretable insights into genetic mutation prediction in glioma patients.

Conclusions:

• The proposed XML pipeline enhances trust and interpretability in medical imaging AI.
• Radiomics and Shapley values offer a robust approach for explaining ML predictions in clinical settings.
• The clinician-focused dashboard facilitates generalized application and adoption of explainable AI in medical imaging.