Radiomics-Based Prediction of Microvascular Invasion Grade in Nodular Hepatocellular Carcinoma Using Contrast-Enhanced Magnetic Resonance Imaging.

Objective: The aim of this study is to develop and verify a magnetic resonance imaging (MRI)-based radiomics model for predicting the microvascular invasion grade (MVI) before surgery in individuals diagnosed with nodular hepatocellular carcinoma (HCC).

Methods: A total of 198 patients were included in the study and were randomly stratified into two groups: a training group consisting of 139 patients and a test group comprising 59 patients. The tumor lesion was manually segmented on the largest cross-sectional slice using ITK SNAP, with agreement reached between two radiologists. The selection of radiomics features was carried out using the LASSO (Least Absolute Shrinkage and Selection Operator) algorithm. Radiomics models were then developed through maximum correlation, minimum redundancy, and logistic regression analyses. The performance of the models in predicting MVI grade was assessed using the area under the receiver operating characteristic curve (AUC) and metrics derived from the confusion matrix.

Results: There were no notable statistical differences in sex, age, BMI (body mass index), tumor size, and location between the training and test groups. The AP and PP radiomic model constructed for predicting MVI grade demonstrated an AUC of 0.83 (0.75-0.88) and 0.73 (0.64-0.80) in the training group and an AUC of 0.74 (0.61-0.85) and 0.62 (0.48-0.74) in test group, respectively. The combined model consists of imaging data and clinical data (age and AFP), achieved an AUC of 0.85 (0.78-0.91) and 0.77 (0.64-0.87) in the training and test groups, respectively.

Conclusion: A radiomics model utilizing-contrast-enhanced MRI demonstrates strong predictive capability for differentiating MVI grades in individuals with nodular HCC. This model could potentially function as a dependable and resilient tool to support hepatologists and radiologists in their preoperative decision-making processes.