Objective To establish radiomics models of preoperative prediction of microvascular invasion （MVI） in hepatocellular carcinoma （HCC） lesions based on Kupffer phase features of sonazoid contrast enhanced ultrasound （CEUS）and evaluate its clinical value. Methods Patients who were diagnosed with hepatic malignant tumors and who underwent sonazoid CEUS examinations were prospectively enrolled. Radiomics signatures were extracted from 6 regions of interest （ROIs） respectively，including tumor region（TR），peritumoral 5mm region （PR） and entire region （ER，TR+PR） on gray scale images and Kupffer phase images of sonazoid CEUS. Support vector machine method was used to establish predictive radiomics models and the predictive efficiency of 6 radiomics models were compared. Results A total of 50 patients with histologically confirmed single HCC lesion were prospectively enrolled （MVI-positive：31，MVI-negative：19）.Five radiomics features were extracted and screened from TR，PR and ER of gray scale images and Kupffer phase images

respectively. Among the 6 radiomics models，the performance of PR of Kupffer phase images performed the best （Accuracy 84%，Sensitivity 90.3% and Specificity 73.7%）.The decision curve analysis showed that PR of Kupffer phase provided a significantly higher clinical benefit than other models，when probability threshold value was set between 0.05 and 0.85. Conclusion The radiomics model based on Kupffer phase image of sonazoid CEUS is a potential method for preoperative prediction of MVI in HCC lesions.