Radiologic error remains an enduring challenge in diagnostic medicine. Despite study of radiologic error since the mid-twentieth century, interpretive discrepancy rates have remained remarkably stable across modalities, institutions, and technologic eras. Moreover, despite compelling evidence that diagnostic performance is shaped by workload, feedback delays, information quality, and tradeoffs under pressure, radiology remains governed predominantly by discrepancy counting, individual remediation, and retrospective attribution. This persistence defies reductionist explanations narrowly centered on individual fallibility and highlights radiology's structural properties as a complex sociotechnical system. This Perspective reconceptualizes radiologic error through the lens of complex-adaptive systems theory whereby safety is understood as an emergent property of dynamic interactions rather than absence of individual failure. The article describes how radiology has not kept pace with epistemologic shifts in understanding error and proposes a reframing of radiologic safety grounded in adaptive capacity, resilience, and systems learning. The impact of artificial intelligence in reshaping system behavior and thereby introducing new challenges is considered. Drawing on the evolution of safety science from linear human-centric models to contemporary resilience-oriented frameworks, the analysis integrates empiric evidence on interpretive variability with theory from systems engineering, cognitive science, and organizational safety to identify conditions under which accurate diagnoses are routinely achieved.
Peter Karamoskos (2026) studied this question.