Abstract: Background and purpose: Radiomics refers to the comprehensive quantification of human tissues through assessing a large number of quantitative image features. Radiomics approach is used to decode tumor phenotypes and predict treatment outcomes. Here we present a study investigating radiomic analysis to assess normal liver features and predict chemotherapy-associated liver injury. Methods: Gastric cancer patients treated with surgery and adjuvant chemotherapy were enrolled in this study retrospectively. CT images were obtained before chemotherapy. The whole liver organ was delineated by radiation oncologists. Images were extracted and filtered by radiomic approach to extract radiomic features. Clustering was performed to reveal clusters of patients with similar radiomic expression patterns. Chi-squared tests were used to assess the association of radiomic data with clinical data and chemotherapy-related liver injury. Results: Radiomic features of 73 patients were clustered into two clusters. A significant association with gender (P=0.004, chi-squared test) was observed, where in male showed a higher presence in cluster Ⅰ. Incidence of abnormal liver function after chemotherapy was 48.7% in cluster I and 67.6% in cluster Ⅱ, respectively (Δ=18.9%). Clinical data including age, gender, chemotherapy modality, number of chemotherapy cycles, HBV infection history, HBs-antigen presence were not associated with liver function abnormalities after chemotherapy. Accuracy of radiomic analysis to predict liver injury is 0.59. Conclusion: Radiomic approach revealed different imaging features of liver between men and women. It could help to predict chemotherapy-associated liver injury. It is feasible to use radiomics approach to decode normal liver features and predict treatment-associated liver injury.

Key words: Radiomics, Liver, Cluster, Gender, Chemotherapy-associated liver injury