Artificial intelligence CT radiomics to predict early recurrence of intrahepatic cholangiocarcinoma: a multicenter study

Objectives In this multicenter study, we sought to develop and validate a preoperative model for predicting early recurrence (ER) risk after curative resection of intrahepatic cholangiocarcinoma (ICC) through artificial intelligence (AI)-based CT radiomics approach. Materials and methods A total of 311 patients (Derivation: 160; Internal and two external validations: 36, 74 and 61) from 8 medical centers who underwent curative resection were collected retrospectively. In derivation cohort, radiomics and clinical–radiomics models for ER prediction were constructed by LightGBM (a machine learning algorithm). A clinical model was also developed for comparison. Model performance was validated in internal and two external cohorts by ROC. In addition, we investigated the interpretability of the LightGBM model. Results The combined clinical–radiomics model that included 15 radiomic features and 3 clinical features (CA19-9 > 1000 U/ml, vascular invasion and tumor margin), resulting in the area under the curves (AUCs) of 0.974 (95% CI 0.946–1.000) in the derivation cohort, and 0.871–0.882 (95% CI 0.672–0.962) in the internal and external validation cohorts, respectively, which are higher than the AJCC 8th TNM staging system (AUCs: 0.686–0.717, p all < 0.05). Especially, the sensitivity of this machine learning model could reach 94.6% on average for all the cohorts. Conclusions This AI-driven combined radiomics model may provide as a useful tool to preoperatively predict ER and improve therapeutic management of ICC patients.