Keyword: Radiology

ABSTRACT: Artificial intelligence (AI) has become increasingly integrated into radiology and nuclear medicine, particularly in oncology, where imaging plays a central role in diagnosis, staging, treatment planning, and response assessment. To date, evaluation of AI-enabled radiology has been dominated by diagnostic accuracy metrics derived from retrospective validation studies. While such measures are essential for technical assessment, they provide limited insight into real-world clinical value. High algorithmic performance does not necessarily translate into improved decision-making, workflow efficiency, patient outcomes, or health system performance. This narrative review critically examines AI-enabled radiology as a digital health intervention in oncology and nuclear medicine, emphasizing the need to move beyond accuracy-centric evaluation paradigms. We analyze the translational gap between controlled validation and routine clinical deployment, highlighting challenges related to dataset bias, generalizability, and human–AI interaction. Key domains of real-world impact are explored, including clinical decision-making, multidisciplinary integration, workflow and operational performance, patient-centered outcomes, and system-level implications. Methodological considerations for outcome-focused evaluation are discussed, alongside regulatory, ethical, and governance frameworks necessary for responsible implementation. We propose a clinical-impact–centered evaluation framework that links AI-assisted imaging to patient, clinician, and system-level outcomes within a continuous monitoring model. Reframing AI-enabled radiology as a clinical intervention rather than a standalone algorithm is essential for ensuring meaningful, equitable, and sustainable adoption in oncology and nuclear medicine practice.