Pragmatic RCT of AI-Assisted Reading for Malignant Hepatic Lesions on CE-CT

Multiphasic contrast-enhanced computed tomography (CE-CT) is widely used to evaluate focal liver lesions, but clinically significant malignancies may be missed or incompletely characterized. This prospective, multicenter, pragmatic, cluster-randomized, center-day crossover trial will compare AI-assisted reporting with standard radiology reporting. Center-days will be assigned in an approximately 1:1 ratio using a centrally generated randomized sequence, and each participant will receive only one workflow. In the AI-assisted arm, the radiologist will save an unaided initial assessment before the locked AI output is revealed and may revise or retain the assessment before authorizing the final report. In the control arm, examinations will be interpreted without access to AI outputs. The primary outcome will be the proportion of participants with a reference-standard-confirmed clinically significant malignant focal liver lesion that was missed or inadequately characterized in the final report. The reference standard will incorporate histopathology, follow-up imaging, multidisciplinary consensus, and relevant clinical records. Non-inferiority will be assessed using a margin prespecified in the statistical analysis plan before enrollment, followed by superiority testing if non-inferiority is demonstrated. Secondary outcomes include senior-review workload, final report turnaround time, detection rate of clinically significant malignant liver lesions, clinically appropriate management-change rate, and time to initiation of treatment for liver malignancy.