Explainable AI in Healthcare: Methods, Concepts, and Challenges

AI has reached performance levels comparable to human capabilities, particularly in medicine and healthcare. This includes managing health services, predictive medicine, and making data-driven decisions based on patient information and diagnostics. However, the integration of AI in healthcare must be done thoughtfully due to inherent limitations. In our paper we have presented how the main limitation arises from the opaque nature of AI, leading to reduced transparency and less dependable decision-making. To tackle this issue, explainable AI (XAI) has been introduced, emphasizing the need for AI algorithms to reveal their internal processes through transparency and granular approaches. XAI aims to instill confidence in the accuracy of model predictions, thereby enhancing trust in their derivation. This review focuses on methods and concepts of XAI specifically relevant to healthcare. Despite being a relatively new concept, integrating XAI with AI poses significant challenges. Our review discusses these challenges and presents real-world healthcare references. Furthermore, the review explores how the Shapley additive explanations (SHAP) tool contributes to improving transparency and interpretability in this context. In a time where AI is transforming healthcare, this review serves as a valuable resource for researchers, clinicians, and policymakers navigating the complexities of implementing XAI in healthcare.