Yahong Li,1 Shaobo Duan,2 Shanshan Ren,3 Dujuan Li,4 Yujing Ma,5 Didi Bu,1 Yuanyuan Liu,1 Xiaoxiao Li,5 Xiguo Cai,6 Lianzhong Zhang1,3 

1Department of Ultrasound, Zhengzhou University People’s Hospital, Henan Provincial People’s Hospital, Henan, People’s Republic of China; 2Department of Health Management, Henan Provincial People’s Hospital, Henan, People’s Republic of China; 3Department of Ultrasound, Henan Provincial People’s Hospital, Henan, People’s Republic of China; 4Department of Pathology, Henan Provincial People’s Hospital, Henan, People’s Republic of China; 5Department of Ultrasound, Henan University People’s Hospital, Henan Provincial People’s Hospital, Henan, People’s Republic of China; 6Department of Rehabilitation, Henan Rehabilitation Clinical Medical Research Center, Henan Provincial People’s Hospital, Henan, People’s Republic of China

Correspondence: Lianzhong Zhang, Department of Ultrasound, Henan Provincial People’s Hospital, No. 7, Weiwu Road, Jinshui District, Zhengzhou City, Zhengzhou, 450003, People’s Republic of China, Tel +86-371-87160869, Email zlz8777@163.com

Purpose: Macrotrabecular-massive hepatocellular carcinoma (MTM-HCC) is a special pathological subtype of HCC, which is related to invasiveness and poor prognosis. We aimed to construct an ultrasomics model for preoperative noninvasive prediction of MTM-HCC.
Patients and Methods: Patients with pathologically confirmed HCC who underwent liver surgery between January 2021 and December 2023 were retrospectively enrolled. 211 eligible patients (169 males and 42 females) were divided 7:3 into the training set (n=147) and test set (n=64) by random stratified sampling. Ultrasomics models were constructed based on the ultrasound image features of the training set using five different ML algorithms, including random forest (RF), eXtreme gradient boosting (XGBoost), support vector machine (SVM), decision tree (DT), and logistic regression (LR). Additionally, a model based on clinical features and a combined model based on clinical and ultrasomics features were constructed to predict MTM-HCC. The performance of the models in the preoperative prediction of MTM-HCC was evaluated on the test set using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy.
Results: The ultrasomics models and the combined models of the five algorithms were effective in predicting MTM-HCC, and the combined models have improved AUC after adding clinical features compared with the ultrasomics model in the test set. The model constructed based on the RF algorithm in the test set has a high accuracy rate and specificity, and the overall performance of the models is better than that of the other four algorithm models, the AUC, accuracy, specificity, and sensitivity of its combined model and ultrasomics model are significantly higher than the clinical model.
Conclusion: ML-based ultrasomics model is an effective tool for predicting MTM-HCC before surgery. Integrating clinical and ultrasound image features enhances predictive performance, offering a novel approach for non-invasive preoperative diagnosis of MTM-HCC.

Keywords: prediction, aggressiveness, macrotrabecular-massive subtype, ultrasomics