Survey of explainable artificial intelligence techniques for biomedical imaging with deep neural networks

Area of Science:

• Biomedical Imaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Deep learning (DL) excels in medical image classification but faces slow clinical adoption.
• The 'black box' nature of DL models hinders trust due to lack of interpretability.
• Ensuring patient safety requires understanding AI diagnostic predictions, akin to autonomous vehicle safety.

Purpose of the Study:

• To provide a comprehensive review of Explainable AI (XAI) for biomedical imaging diagnostics.
• To categorize existing XAI techniques applicable to medical imaging.
• To discuss challenges and future directions in XAI for healthcare.

Main Methods:

• Literature review of XAI techniques in biomedical imaging.
• Categorization of XAI methods based on their approach and application.
• Analysis of current challenges and future research avenues.

Main Results:

• XAI techniques are essential for building trust in AI-driven medical diagnoses.
• Categorization of XAI methods aids in understanding their applicability.
• Identified open challenges and future directions for XAI research in this domain.

Conclusions:

• XAI is vital for the safe and effective integration of DL into clinical practice.
• Further research in XAI will accelerate disease diagnosis and regulatory compliance.
• This survey offers valuable insights for clinicians, regulators, and AI developers.