<p>Clinical decision support systems are increasingly integrated with artificial intelligence, which improves diagnostic accuracy and operational efficiency. The transparency of many deep learning models remains a limiting factor to clinical trust and accountability. Explainable artificial intelligence (XAI) has thus become an important field of research to make AI-based diagnostic systems more interpretable and clinically relevant. Recently, agentic AI has garnered attention as a promising paradigm for improving existing diagnostic pipelines through multistep reasoning, tool use, and workflow-based decision support. This review provides systematic and scientometric insights into the development, approaches, and clinical relevance of XAI in computer-aided diagnosis (CADx), particularly in cancer diagnosis based on biomedical imaging and physiological signals and discusses the future role of agentic AI in clinical processes. The analysis covers 1,472 publications from 2015 to 2025, outlines the knowledge field, identifies key sources, and describes the prevailing trend in interpretability across MRI, CT, ultrasound, EEG, and ECG applications. The findings show a rapid growth in XAI-enabled CADx research in the post-2023 period, driven by increased demand for trustworthy, clinically usable AI. Simultaneously, ongoing issues persist, including dataset heterogeneity, limited generalizability, inconsistent validation and reporting procedures, unequal evaluation of explanation quality, and unstable clinician trust. The paper concludes that agentic AI can benefit explainable CADx through adaptive, workflow-aware support, while current systems are research prototypes rather than clinically validated decision-support tools. This work provides a highly critical and methodologically sound perspective on XAI-CADx and establishes strategic directions for developing transparent, robust, and clinically credible diagnostic AI systems.</p>