基于支持向量机多模态超声模型诊断肾疾病

Objective: To compare the effectiveness of multi-modal ultrasound, including conventional ultrasound, color Doppler ultrasound and shear wave elastic imaging in diagnosis of renal diseases based on support vector machine support vector machine (SVM) and traditional Logistic regression. Methods: Totally 94 patients with pathologically proved renal diseases (RD group) and 109 patients without renal diseases (control group) were collected and examined with conventional ultrasound, color Doppler ultrasound and shear wave elastic imaging, respectively. SVM and Logistic regression were used for modeling. Then all 203 patients were divided into 2 groups according to 3:1, then 153 cases were used as SVM's training samples for single factor variable judgment and model establishment, the other 50 cases were used as validation samples to evaluate the prediction effect of SVM model. Results: The elastic hardness of left renal cortex and the width of right kidney entered the regression equation in Logistic regression. The accuracy of Logistic regression model for diagnosis of renal diseases was 83.74%, of SVM model was 85.10% (χ²=0.17, P=0.68). Conclusion: Multimodal ultrasound has high effectiveness for diagnosis of renal diseases. SVM and Logistic models have similar diagnostic effectiveness.