关键词: 超声检查, 影像组学, 肾小球肾炎, 机器学习, 预测模型

Abstract:

Objective To explore the application value of ultrasound radiomics technology based on grayscale ultrasound images in the differential diagnosis of histological classification of glomerulonephritis. Methods A total of 204 patients with renal biopsy were selected from our hospital, and according to pathological results, they were divided into the membranous nephropathy group (n=133) and the mesangial proliferative glomerulonephritis group (n=71). The ultrasound images were sketched and the image omics features were extracted by two physicians. The pathological results and ultrasound data of renal biopsy were collected from the two groups, and the ultrasound radiomics features were preliminarily screened by the maximum correlation and minimum redundancy algorithm (mRMR) algorithm for all the obtained omics feature data. Then the optimal effective features were selected from the screened features by minimum absolute shrinkage and selection operator (LASSO) algorithm, and random forest (RF), support vector machine (SVM), logistic regression (LR), four kinds of classifiers of K-nearest neighbor (KNN) method were used to establish a prediction model. All cases were randomly divided into the training set and the validation set according to the ratio of 7:3, and the four models were trained by the training set, and then validated in the validation set, and the best prediction model was selected by comparing the receiver operating characteristic (ROC) curve, Delong test and GiViTI calibration curve. Decision curve analysis (DCA) was used to evaluate the clinical utility of the model. Results The radiomics method was used to extract 837 radiomics features per image, and 16 meaningful features were finally screened out by using the mRMR + LASSO algorithm. Among the four prediction models of RF, SVM, LR and KNN, the best performing model was LR model, with the AUC of 0.944, the specificity of 0.867 and the sensitivity of 0.878. The GiViTI calibration curve showed that the model had good accuracy (P>0.05), and the decision curve showed that the prediction model had good clinical practical value. Conclusion Ultrasound radiomics has a good ability to distinguish the more common histological types of glomerulonephritis, and is a non-invasive method with good application prospects.

Key words: ultrasonography, radiomics, glomerulonephritis, machine learning, prediction model