Survey of XAI in digital pathology

Explainability is essential for clinical adoption: The paper emphasizes that AI algorithms in medical imaging need to go beyond just high accuracy—they must be transparent and understandable to be trusted and used in actual clinical practice, particularly in digital pathology where diagnostic decisions directly impact patient care.

Uncertainty matters as much as explanations: The survey identifies uncertainty estimation as an integral part of explainable AI, not a separate concern. This means AI systems should not only explain their reasoning but also communicate how confident they are in their predictions, which is critical for clinicians deciding whether to rely on an AI recommendation.

One-size-fits-all explanations don't work: The paper categorizes XAI techniques from multiple perspectives, suggesting that different explanation methods serve different purposes. Medical professionals and researchers need different types of explanations depending on their needs—what helps a pathologist validate a diagnosis may differ from what regulators need to approve the system.

Bridging the gap between AI experts and doctors is crucial: By explicitly targeting both technical researchers and medical professionals, the survey highlights that responsible AI in pathology requires genuine cross-disciplinary collaboration, not just technologists building systems that doctors must accept.

Digital pathology has unique requirements: The paper notes that pathology imaging has "particular characteristics and needs" that distinguish it from other medical imaging domains, implying that generic AI solutions must be adapted thoughtfully rather than applied wholesale to clinical settings.