Explainable Artificial Intelligence for Deep-Learning Based Classification of Cystic Fibrosis Lung Changes in MRI

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Radiology

Background:

• Deep learning models are increasingly used for medical image classification.
• Enhancing the transparency and explainability of these models is crucial for clinical adoption.
• Applying explainability algorithms to complex medical conditions like cystic fibrosis (CF) lung changes presents unique challenges.

Purpose of the Study:

• To apply and evaluate explainability algorithms for classifying lung perfusion changes and mucus plugging in CF patients using MRI.
• To assess the potential of explainable AI to support radiologists in diagnosing lung abnormalities.
• To investigate the feasibility of integrating explainability methods into existing deep learning pipelines for medical applications.

Main Methods:

• Implemented and tested several explainability algorithms, including integrated gradients and Grad-CAM, on a deep learning classification pipeline.
• Applied these methods to MRI data from cystic fibrosis patients to identify features relevant to lung changes.
• Evaluated the success and utility of four out of six implemented explainability algorithms.

Main Results:

• One explainability algorithm provided satisfactory results, successfully highlighting areas critical for classification.
• The applied methods demonstrated the potential of deep learning for classifying lung changes in CF patients.
• Explainable outputs showed promise in supporting radiological interpretation of CF-related lung abnormalities.

Conclusions:

• Explainable AI techniques can effectively highlight diagnostically relevant regions in medical images for CF lung conditions.
• The successful application of these algorithms can enhance clinician trust in deep learning diagnostic systems.
• Integrating explainable AI holds significant potential for developing advanced diagnostic decision support systems in radiology.