Preoperative Evaluation of Hepatocellular Carcinoma Differentiation Using Contrast-Enhanced Ultrasound-Based Deep-Learning Radiomics Model

Abstract

Xiachuan Qin,1,2 Xiaomin Hu,2 Weihan Xiao,2 Chao Zhu,3 Qianqin Ma,1 Chaoxue Zhang1 1Department of Ultrasound, The First Affiliated Hospital of Anhui Medical University, Hefei, People’s Republic of China; 2Department of Ultrasound, Nanchong Central Hospital, The Second Clinical Medical College, North Sichuan Medical College (University), Nan Chong, People’s Republic of China; 3Department of Radiology, The First Affiliated Hospital of Anhui Medical University, Hefei, People’s Republic of ChinaCorrespondence: Chaoxue Zhang, Department of Ultrasound, The First Affiliated Hospital of Anhui Medical University, Hefei, Anhui, 230022, People’s Republic of China, Tel [phone], Email [email protected]: Distinguishing the degree of differentiation, hepatocellular carcinoma (HCC) has important clinical significance in the therapeutic decision-making and patient prognosis evaluation.Methods: We developed a deep-learning radiomics (DLR) model based on contrast-enhanced ultrasound (CEUS) to evaluate the differentiation of HCC noninvasive. We retrospectively analyzed HCC patients who had undergone resection and CEUS one week preoperatively between November 2015 and August 2022. Enrolled patients were randomly divided into training (n=190) and testing (n=82) cohorts in a 7:3 ratio. The depth of learning and radiological characteristics reflecting the differentiation degree of HCC were extracted, and the least absolute shrinkage and selection operator(LASSO) was used for feature selection to obtain the most valuable features and then build a DLR model based on the useful features.Results: The deep-learning Radiomics model could accurately predict the degree of differentiation of HCC; the area under the curve of the DLR model in the training and testing cohorts was 0.969 and 0.932, respectively. The accuracy, sensitivity, and specificity of the CEUS-based DLR model for predicting the differentiation of HCC were 0.915, 0.938, and 0.900, respectively, in the testing cohort. The decision curve analysis confirmed that the combined model predicted good overall net income for differentiation.Conclusion: The CEUS-based DLR model provides an easy-to-use, visual, and personalized tool for predicting the differentiation of HCC and can help doctors formulate more favorable treatment plans for patients.Keywords: hepatocellular carcinoma, differentiated, contrast-enhanced ultrasound, deep-learning, radiomics

Keywords

• hepatocellular carcinoma，differentiated，contrast-enhanced ultrasound，deep-learning，radiomics