Abstract: Objective To develop a machine learning （ML） model based on shear wave elastography （SWE） radiomics for predicting the risk of hepatic fibrosis progression in patients with type 2 diabetes mellitus （T2DM） and non-alcoholic fatty liver disease （NAFLD）, aiming to reduce the necessity for surgery through early non-surgical interventions. Methods A retrospective analysis was performed on SWE image data of 218 patients with T2DM and NAFLD. Liver region image segmentation and radiomics feature extraction were conducted using the 3D Slicer software. After feature selection, random forest, support vector machine, and extreme gradient boosting （XGBoost） were employed to build ML predictive models. The performance of the models was evaluated using receiver operating characteristic （ROC） curves, calibration curves, and decision curve analyses （DCA） to determine the optimal ML predictive model. Results Among the 218 patients, 40 （18.3%） had an initial liver stiffness ≥ 6.5 kPa, indicating a risk of hepatic fibrosis. A total of 849 radiomics features were extracted from each segmented SWE image. After data standardization and feature selection, five radiomics features closely associated with hepatic fibrosis were identified. Three ML models were constructed based on these radiomic features, and the XGBoost model demonstrated optimal discriminatory ability, achieving an area under the ROC curve of 0.864. The calibration curve and DCA curve indicated that the model accurately predicted the risk of hepatic fibrosis when the predicted probability was within the range of 0.4-0.7. Conclusion This study has successfully trained and validated an XGBoost model based on hepatic SWE radiomics features, demonstrating its potential applicability in predicting the risk of hepatic fibrosis progression in T2DM patients with NAFLD.

Key words: Type 2 diabetes mellitus, Non-alcoholic fatty liver disease, Hepatic fibrosis, Shear wave elastography, Radiomics, Machine learning