Explainable Artificial Intelligence in Mammography: A Systematic Review of Methods, Evaluation Practices, and...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Explainable artificial intelligence (XAI) is proposed to enhance trust in AI mammography systems.
• The clinical validity and readiness of current XAI explanations are not well-established.
• This review systematically assesses XAI methods in mammography and their evaluation.

Purpose of the Study:

• To systematically review XAI methods applied to mammography.
• To synthesize how explanations are evaluated for validity, robustness, and clinical usefulness.
• To identify gaps in current XAI research for mammography.

Main Methods:

• Systematic review following PRISMA 2020 guidelines.
• Searched major databases (MEDLINE, Embase, Scopus, Web of Science, Cochrane) from 2015-2026.
• Included studies using mammography with explicit explanation mechanisms; narrative synthesis and adapted XAI appraisal were used.

Main Results:

• Fourteen studies were included, focusing on detection, classification, or prediction.
• XAI methods included interpretable-by-design, saliency/attribution, and feature-level approaches.
• Most studies showed only qualitative plausibility; few had external validation or human-factor assessment; saliency benchmarks indicated modest lesion-pointing reliability.

Conclusions:

• Mammography XAI predominantly features visually plausible but inconsistently validated explanations.
• While intrinsically interpretable designs are emerging, external validation and clinician-centered evaluations are infrequent.
• Future research should prioritize pre-specified claims, quantitative metrics, robustness testing, and human decision-making studies.