Objective To build macrotrabecular-massive hepatocellular carcinoma (MTM-HCC) differentiation model based on the quantitative pre-operative ultrasound features. Methods The hepatocellular carcinoma patients that underwent surgery in our hospital from Augest 1

2017 to Oct 1

2020 were retrospectively collected. After clinical and pathological sorting

the collected cases were separated into training set (70%) and validation set (30%). The quantitative pre-operative ultrasound features were extracted from training set data and selected by chi-square algorithm. Then

the trained random forest model performance was evaluated on validation set data. Results A total of 79 MTM-HCC and other HCCs were included in this study. The pre-operative AFP level

Edmondson-Steiner grade

satellite lesion and microvascular invasion status distribution were significantly different between MTM-HCC and other HCCs

while the age

sex

HBV infection status was not. The feature selection algorithm used high-dimension texture features in MTM-HCC prediction. The final random forest model achieved AUC=0.895

Accuracy=0.833

Precision=0.833

Sensitivity=60% and Specificity=89.5% on validation set. Conclusions The prognosis-benefit MTM-HCC differentiation model could be built based on the quantitative pre-operative ultrasound features

which has high specificity and was complementary with other differentiation model.