Computed tomography texture-based models for predicting KIT exon 11 mutation of gastrointestinal stromal tumors

Abstract

BackgroundKIT exon 11 mutation in gastrointestinal stromal tumors (GISTs) is associated with treatment strategies. However, few studies have shown the role of imaging-based texture analysis in KIT exon 11 mutation in GISTs. In this study, we aimed to show the association between computed tomography (CT)-based texture features and KIT exon 11 mutation. MethodsNinety-five GISTs confirmed by surgery and identified with mutational genotype of KIT were included in this study. By amplifying the samples using over-sampling technique, a total of 183 region of interest (ROI) segments were extracted from 63 patients as training cohort. The 63 new ROI segments were extracted from the 63 patients as internal validation cohort. Thirty-two patients who underwent KIT exon 11 mutation test during 2021–2023 was selected as external validation cohort. The textural parameters were evaluated both in training cohort and validation cohort. Least absolute shrinkage and selection operator (LASSO) algorithms and logistic regression analysis were used to select the discriminant features. ResultsThree of textural features were obtained using LASSO analysis. Logistic regression analysis showed that patients’ age, tumor location and radiomics features were significantly associated with KIT exon 11 mutation (p < 0.05). A nomogram was developed based on the associated factors. The area under the curve (AUC) of clinical features, radiomics features and their combination in training cohort was 0.687 (95 % CI: 0.604–0.771), 0.829 (95 % CI: 0.768–0.890) and 0.874 (95 % CI: 0.822–0.926), respectively. The AUC of radiomics features in internal validation cohort and external cohort was 0.880 (95 % CI: 0.796–0.964) and 0.827 (95%CI: 0.667–0.987), respectively. ConclusionThe CT texture-based model can be used to predict KIT exon 11 mutation in GISTs.