Demystifying Artificial Intelligence: A Systematic Review of Explainable Artificial Intelligence in Medical Imaging

Area of Science:

• Artificial Intelligence in Medicine
• Medical Imaging Analysis
• Explainable AI (XAI)

Background:

• Machine learning (ML) and deep learning (DL) show promise in medical imaging (MI) for diagnosis and segmentation.
• The
• black-box
• nature of many AI models hinders clinical adoption due to lack of transparency and trust.
• Explainable AI (XAI) aims to address these transparency issues in healthcare algorithms.

Purpose of the Study:

• To conduct a comprehensive literature review of XAI techniques in medical imaging.
• To evaluate the potential of XAI in enhancing transparency and trust in clinical AI applications.
• To identify challenges and gaps in current XAI research for healthcare.

Main Methods:

• Systematic review of existing literature on XAI techniques (intrinsic and post-hoc).
• Comparative analysis of over 18 XAI methods, detailing their mathematical foundations and applicability.
• Development of standardized evaluation metrics for XAI performance in medical imaging tasks.

Main Results:

• XAI methods can significantly improve the interpretability of AI models in medical imaging.
• Key challenges include limited interpretability, computational complexity, and lack of standardized evaluation.
• A comprehensive comparative analysis of 18+ XAI techniques is presented, highlighting strengths and weaknesses.

Conclusions:

• XAI is crucial for bridging the gap between AI advancements and clinical integration in healthcare.
• Standardized evaluation metrics and actionable recommendations are proposed for effective XAI implementation.
• Encouraging adoption of XAI practices will foster transparent, interpretable, and reliable AI systems in medical settings.