Abstract
Objectives: Radiomics has enhanced quantitative ultrasound (QUS) imaging based on envelope statistics for liver fibrosis evaluation. However, early detection of liver fibrosis in patients with hepatic steatosis remains challenging. This study is to develop ultrasound scatteromics prediction models, utilizing simplified feature sets from multimodal QUS envelope statistics imaging, for detecting early-stage liver fibrosis (stage ≥ F1) and significant fibrosis (≥F2) in the presence of hepatic steatosis. Methods: The dataset in this prospective study included 252 subjects (n = 125 for training and validation; n = 127 subjects for independent testing), which underwent blood tests, liver biopsy, and ultrasound radiofrequency data acquisition. In scatteromics analysis, multimodal QUS envelope statistics imaging (Nakagami, homodyned K, and information entropy statistics) was employed. For each imaging, a predefined simplified feature set was calculated, followed by feature selection for machine learning using support vector machine (SVM), random forest (RF), and linear discriminant analysis (LDA). The scatteromics model was obtained using a repeated five-fold stratified cross-validation and then independently tested. The performance was evaluated by the area under the receiver operating characteristic curve (AUROC); scatteromics features were also compared with aspartate aminotransferase (AST) and alanine aminotransferase (ALT). Results: Scatteromics features showed no significant correlation with AST and ALT, with correlation coefficients ranging from 0.003 to 0.28. In patients with coexisting hepatic steatosis, scatteromics significantly outperformed QUS envelope statistics imaging in identifying early-stage liver fibrosis, achieving AUROC values of 0.85 to 0.87 for the training and validation datasets, and 0.78 to 0.81 for the testing dataset. In comparison, scatteromics demonstrated modest performance in detecting significant liver fibrosis (≥F2), with AUROC ranging from 0.66 to 0.71 in the training cohort and 0.64 to 0.76 in the testing cohort. Conclusions: The proposed scatteromics model streamlines the data analysis workflow of conventional QUS radiomics, enabling early detection of liver fibrosis with reduced dependence on inflammation and hepatic steatosis.