Toward explainable AI in radiology: Ensemble-CAM for effective thoracic disease localization in chest X-ray images...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Chest X-ray (CXR) is crucial for diagnosing thoracic diseases.
• Deep learning (DL) offers computer-aided diagnosis (CAD) but faces challenges with data annotation and explainability.
• Existing DL models require large annotated datasets and lack interpretable outputs.

Purpose of the Study:

• To develop a Class Activation Mapping (CAM)-based ensemble model, Ensemble-CAM, for improved CXR analysis.
• To address the limitations of data dependency and lack of explainability in DL-based CAD tools.
• To leverage weakly supervised learning and explainable AI (XAI) for disease localization and outcome justification.

Main Methods:

• Developed Ensemble-CAM, a novel model integrating ensemble learning, transfer learning, and CAM functions.
• Employed weakly supervised learning to minimize reliance on heavily annotated data.
• Utilized explainable AI (XAI) for interpretable feature visualization.

Main Results:

• Ensemble-CAM effectively predicts disease locations using interpretable features.
• The model demonstrates reduced dependency on strongly annotated data.
• Visualizations enhance confidence in diagnostic predictions.
• Fusion functions optimized cumulative performance.

Conclusions:

• Ensemble-CAM offers enhanced performance and reliability in thoracic disease detection from CXR.
• The model successfully integrates explainability with diagnostic prediction.
• This approach advances the application of DL in medical imaging by addressing key practical challenges.