Abstract: Objective To investigate the utility of an MRI imaging omics-based machine learning model for the preoperative prediction of microvascular invasion in the peripheral zone of hepatic alveolar echinococcosis (HAE). Methods A retrospective analysis was conducted on 200 patients diagnosed with HAE, confirmed via postoperative pathology. MR imaging and histopathological data were collected, with histological characteristics of the lesion being extracted from the MR images. The cohort was divided into a training set of 88 patients and test set of 22 patients , following a 8∶2 ratio. Machine learning models, including XGBoost, Random forest(RF), Logistic Regression, Support Vector Machine, and Classical Decision Tree, were developed to predict microvascular invasion in the peripheral zone of HAE. Model validation was performed, and Receiver operating characteristic(ROC) curves were generated to evaluate the predictive performance of the various models on microvascular invasion in the marginal zone of HAE. Results Pathological analysis revealed that out of 200 patients, 75 exhibited microvascular invasion in the marginal zone, whereas 125 did not. Statistical analysis indicated no significant differences in gender, age, lesion location, and lesion size between patients with and without marginal microvascular invasion (P>0.05). The training set included 60 cases of marginal zone microvascular invasion, with the remaining 15 cases in the test set. From the collected data, 1380 imageomics features were intitially extracted, out of which 406 features were retained using the variance threshold method. Subsequently, six optimal imageomics features were identified through univariate selection, comprising 1 first-order statistical feature and 5 higher-order statistical features. The process of feature selection was refined to ultimately select 6 optimal imaging omics features employing bothe the square difference threshold and univariate selection methods. In the evaluation of diagnostic performance, the XGBoost and RF models demonstrated the highest AUC values in the training set for the diagnosing marginal microvascular invasion, registering AUC values of 0.95 and 0.96, respectively. In the test set, the AUC values for XGBoost and RF were also superior to those of the other model groups, at 0.88 and 0.84 respectively, indicating a higher diagnostic accuracy. Statistical comparison between the AUC values of the XGBoost and RF models in both training and testing sets revealed no significant difference(P>0.05). Conclusion In the preoperative forecasting of microvascular invasion within the marginal zone of HAE, models based on MRI imaging omics, specifically XGBoost and RF, demonstrated superior predictive capabilities.

Key words: Hepatic alveolar hydatid disease, Magnetic resonance imaging, Marginal zone microvascular invasion, Imaging omics, Machine learning model