Explainable Artificial Intelligence in Healthcare: Current Landscape, Challenges, and Future Directions.

Area of Science:

• Healthcare AI
• Explainable Artificial Intelligence (XAI)
• Machine Learning (ML)
• Deep Learning (DL)

Background:

• Artificial Intelligence (AI) transforms healthcare with improved diagnosis and personalized treatments.
• Opaque AI models ('black boxes') hinder interpretability, clinician trust, and adoption.
• Explainable Artificial Intelligence (XAI) offers transparent insights to address these limitations.

Purpose of the Study:

• To systematically review current evidence on XAI in healthcare.
• To map AI models to XAI techniques, healthcare domains, and clinical applications.
• To identify trends and challenges in XAI implementation.

Main Methods:

• Systematic search of six databases (2017-2025).
• PRISMA guidelines followed for study selection and data extraction.
• Data included AI model types, XAI techniques, healthcare domains, validation, and ethical reporting.

Main Results:

• Seventy studies included, focusing on oncology, cardiology, and infectious diseases.
• Deep learning models (76%) and tree-based models (24%) were prevalent.
• SHAP (54%) and LIME (30%) were common XAI techniques; ethical/regulatory aspects were underreported.

Conclusions:

• XAI improves transparency and trust in healthcare AI but faces challenges.
• Inconsistent validation, ethical frameworks, and interpretability metrics need development.
• Future work should prioritize clinically validated, ethical, and user-centered XAI models.