Development of a magnetic resonance imaging-derived radiomics model to predict microvascular invasion in patients with hepatocellular carcinoma

Abstract

Background: Hepatocellular carcinoma (HCC) with microvascular invasion (MVI) has a poor prognosis, is prone to recurrence and metastasis, and requires more complex surgical techniques. Radiomics is expected to enhance the discriminative performance for identifying HCC, but the current radiomics models are becoming increasingly complex, tedious, and difficult to integrate into clinical practice. The purpose of this study was to investigate whether a simple prediction model using noncontrast-enhanced T2-weighted magnetic resonance imaging (MRI) could preoperatively predict MVI in HCC. Methods: A total of 104 patients with pathologically confirmed HCC (training cohort, n=72; test cohort, n=32; ratio, about 7:3) who underwent liver MRI within 2 months prior to surgery were retrospectively included. A total of 851 tumor-specific radiomic features were extracted on T2-weighted imaging (T2WI) for each patient using AK software (Artificial Intelligence Kit Version; V. 3.2.0R, GE Healthcare). Univariate logistic regression and least absolute shrinkage and selection operator (LASSO) regression were used in the training cohort for feature selection. The selected features were incorporated into a multivariate logistic regression model to predict MVI, which was validated in the test cohort. The model’s effectiveness was evaluated using the receiver operating characteristic and calibration curves in the test cohort. Results: Eight radiomic features were identified to establish a prediction model. In the training cohort, the area under the curve, accuracy, specificity, sensitivity, and positive and negative predictive values of the model for predicting MVI were 0.867, 72.7%, 84.2%, 64.7%, 72.7%, and 78.6%, respectively; while in the test cohort, they were 0.820, 75%, 70.6%, 73.3%, 75%, and 68.8%, respectively. The calibration curves displayed good consistency between the prediction of MVI by the model and actual pathological results in both the training and validation cohorts. Conclusions: A prediction model using radiomic features from single T2WI can predict MVI in HCC. This model has the potential to be a simple and fast method to provide objective information for decision-making during clinical treatment.