Radiomic Model Based on Multisequence MRI to Preoperative Predict Ki-67 Expression in Hepatocellular Carcinoma.
Authors
Affiliations (8)
Affiliations (8)
- Clinical Medicine College of Jining Medical University, Jining 272067, Shandong, China (S.Z.).
- Department of Radiology, Qilu Hospital of Shandong University, Jinan 250012, Shandong, China (L.Z.).
- Department of Ultrasound, The First Affiliated Hospital of Shandong First Medical University & Shandong Provincial Qianfoshan Hospital, Jinan 250014, Shandong, China (B.W.).
- Digital Imaging Center at People's Hospital of Laoling, Laoling 253600, Shandong, China (Q.L.).
- Department of General Surgery, The First Affiliated Hospital of Shandong First Medical University & Shandong Provincial Qianfoshan Hospital, Jinan 250014, Shandong, China (Z.X.).
- Department of Pathology, The First Affiliated Hospital of Shandong First Medical University & Shandong Provincial Qianfoshan Hospital, Shandong Medicine and Health Key Laboratory of Clinical Pathology, Shandong Lung Cancer Institute, Shandong Institute of Nephrology, Jinan 250014, Shandong, China (R.T.).
- Department of Radiology, The First Affiliated Hospital of Shandong First Medical University & Shandong Provincial Qianfoshan Hospital, Jinan 250014, Shandong, China (J.W., G.L.).
- Department of Radiology, The First Affiliated Hospital of Shandong First Medical University & Shandong Provincial Qianfoshan Hospital, Jinan 250014, Shandong, China (J.W., G.L.). Electronic address: [email protected].
Abstract
The expression of nuclear protein Ki-67 has demonstrated significant prognostic relevance in hepatocellular carcinoma (HCC). This study aimed to establish a prediction model for the expression of Ki-67 in HCC patients. Patients with pathologically confirmed solitary HCC were recruited from two institutions and stratified into the training (n = 197) and test (n = 58) cohorts. Based on the clinical and imaging features of patients in the original test cohort, we simulated the clinical and imaging features of an additional 40 patients, thus increasing the sample size of the test cohort to 98 cases. Clinical parameters and radiomic features from multi-sequence MR imaging were extracted. The feature selection process was performed to build models to predict Ki-67 expression using four machine learning algorithms. Receiver operating characteristic curve analysis, the area under the curve (AUC), and the DeLong test were used to assess the predictive performance of the models. Decision curve analysis (DCA) was used to evaluate the clinical utility. For single sequence radiomic models, the AUC values of In Phase-MLP, T2WI-LightGBM, DWI-LightGBM, AP-DecisionTree, and DP-DecisionTree models were 0.756, 0.749, 0.924, 0.804, and 0.821 in the training cohort, and 0.763, 0.738, 0.741, 0.697, and 0.753 in the test cohort, which were significantly higher than other models. For multi-sequence radiomic models, the LightGBM model demonstrated superior performance with AUCs of 0.965 in the training cohort and 0.766 in the test cohort, significantly exceeding single-sequence counterparts (Delong p < 0.05) in the training cohort. Notably, clinical models alone exhibited limited predictive value, and combined models including clinical and radiomic features failed to enhance the performance. The multi-sequence-LightGBM radiomic model could preoperatively predict Ki-67 expression in HCC, offering a non-invasive tool to stratify patients at high risk of recurrence and guiding personalized therapeutic strategies.