Development and Comparison of Prediction Models Based on Sonovue- and Sonazoid-Enhanced Ultrasound for Pathologic Grade and Microvascular Invasion in Hepatocellular Carcinoma

Abstract

ObjectiveThis study aimed to develop and compare the prediction model based on Sonovue and Sonazoid contrast enhanced ultrasound (CEUS) for pathologic grade and microvascular invasion (MVI) of hepatocellular carcinoma (HCC). And to investigate whether Kupffer phase images have additional predictive value for the above pathological features. Methods90 patients diagnosed as primary HCC and received curative hepatectomy were prospectively enrolled. All patients underwent conventional ultrasound (CUS), Sonovue and Sonazoid-CEUS examination preoperatively. Clinical, radiological and pathological features including pathologic grade, MVI and CD68 expression were collected. We developed prediction models composing of clinical, CUS and CEUS (Sonovue and Sonazoid, respectively) features for pathologic grade and MVI with both logistic regression and machine learning (ML) method. Results41 (45.6%) patients were poorly differentiated HCC (p-HCC) and 37(41.1%) were MVI positive. For pathologic grade, logistic model based on Sonazoid-CEUS had significant better performance than which based on Sonovue-CEUS (area under curve (AUC), 0.929 versus 0.848, P=0.035). While for MVI, these two models had similar accuracy (AUC, 0.810 versus 0.786, P=0.068). Meanwhile, we found well-differentiated HCC tends to have higher enhancement ratio in 6-12 minutes during Kupffer phase of Sonazoid-CEUS, as well as higher CD68 expression compared with p-HCC. Besides, all of these models can effectively predict the risk of recurrence (P < 0.05). ConclusionSonovue-CEUS and Sonazoid-CEUS showed comparable excellence in predicting MVI, while Sonazoid-CEUS was superior to Sonovue-CEUS in predicting pathologic grade due to the Kupffer phase. Enhancement ratio in Kupffer phase have additional predictive value for pathologic grade prediction.