Assessing the Impact of MULLET AI System on Focal Liver Lesion Detection: A Multicenter, Multi-reader, Multi-case Clinical Trial

作者：Zhongquan Sun, Yu Chen, Hongfan Ding, Ying Chen, Haoze Cao, Xin Han, Wanyi Chen, Lei Wu, Feng Fang, Fan Wu, Lianyue Yang, Jiajun Bu, Yuan Ding, W Wang · 发表于：npj Precision Oncology · 年份：2026 · DOI：10.1038/s41698-026-01594-9 · 研究领域：Hepatocellular Carcinoma Treatment and Prognosis、Radiomics and Machine Learning in Medical Imaging、Retinal Imaging and Analysis

Early detection of focal liver lesions is critical for patient outcomes, but current imaging techniques have limitations. This study developed MULLET, an AI system using contrast-enhanced CT, and assessed its performance in assisting radiologists with FLL detection and characterization. A retrospective multicenter, multi-reader, multi-case trial was conducted, in which 10 radiologists independently interpreted 375 patients' clinical images with and without MULLET assistance. This trial was registered on ClinicalTrials.gov (ID: NCT06068413) with the registration name "A Retrospective, Multicenter, Multiple-viewer-multiple-case (MRMC) Clinical Trial to Evaluate the Safety and Efficacy of CT Image-assisted Detection Software for Focal Liver Lesions" in April 2023. Diagnostic performance was compared using area under the receiver operating characteristic curve (AUC) analysis. Without MULLET, the average AUC was 0.8188 (sensitivity: 74.45%, specificity: 87.77%). With MULLET, the average AUC significantly improved to 0.9268 (p < 0.0001), sensitivity increased to 89.95% (p = 0.0003), and specificity rose to 93.57% (p = 0.0063). MULLET also showed improved performance across different FLL sizes and subtypes. Moreover, the average reading time for radiologists was reduced by 21.8% (p < 0.0001). In conclusion, MULLET demonstrated promising performance in significantly improving radiologists' sensitivity, accuracy, and efficiency in FLL detection and diagnosis.