Radiomics-based automated machine learning for differentiating focal liver lesions on unenhanced computed tomography

• DOI: 10.1007/s00261-024-04685-y
• Publication Date: 2024-11-22T01:48:56Z
• Authors: Nan Yang; Zhuangxuan Ma; Ling Zhang; Wenbin Ji; Qian Xi; Ming Li; Liang Jin

Enhanced computed tomography (CT) is the primary method for focal liver lesion diagnosis. We aimed to use automated machine learning (AutoML) algorithms differentiate between benign and malignant lesions on basis of radiomics from unenhanced CT images. enrolled 260 patients 2 medical centers who underwent examinations January 2017 March 2023. This included 60 cases hepatic malignancies, 93 hemangiomas, 48 abscesses, 84 cysts. The Pyradiomics was used extract features By using mljar-supervised (MLJAR) AutoML framework, clinical, radiomics, fusion models combining clinical were established. In training validation sets, area under curve (AUC) values exceeded 0.900. external testing set, respective AUC as follows: 0.88, 1.00, 1.00 cysts; 0.81, 0.90, 0.97 hemangiomas; 0.89, 0.98, 0.92 abscesses; 0.23, 0.80, 0.93 malignancies. diagnostic accuracy rates cysts, abscesses by radiologists in cohort 0.96, 0.60, 0.79, 0.66, respectively. model based noninvasive images has high value distinguishing lesions. preoperative diagnosis essential choosing appropriate treatment. Thus, we MLJAR framework