Answers are grounded in this paper and cite their sources.
The integration of artificial intelligence (AI) into medical tasks has introduced powerful tools for diagnosis, prognosis, and treatment planning, yet the opacity of many deep learning systems has limited their acceptance in clinical environments where trust, accountability, and interpretability are essential. Explainable artificial intelligence (XAI) addresses these concerns by revealing the reasoning behind AI-driven decisions. This systematic review examines XAI methods applied to medical tasks, focusing on how explanations are generated and delivered through visual, textual, counterfactual, example-based, and multimodal formats. Drawing on a structured search of six databases covering literature published between 2020 and March 2026, we analyzed 309 studies, categorizing state-of-the-art techniques by their underlying algorithms, explanation formats, and clinical applicability. Visual explanations such as saliency maps and Grad-CAM predominate in medical imaging, whereas textual, counterfactual, example-based, and multimodal methods are increasingly applied to clinical prediction, decision support, and automated report generation. We further examine benchmark datasets, evaluation practices, and empirical studies assessing explanation quality, interpretability, and user trust, and highlight ethical considerations including fairness, bias, and transparency in deployment. By synthesizing recent advances and identifying gaps, this review aims to support the development of more interpretable, trustworthy, and human-centered AI systems for healthcare.
No takes yet. Share an insight, caveat, or question.
Sakib et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: