Background: Gastrointestinal stromal tumors (GISTs) are common mesenchymal tumors with significant variations in prognosis and sensitivity to imatinib. Traditional detection of the DEL 557–558 mutation requires invasive procedures and is costly. Radiogenomics, by analyzing imaging, can predict tumor mutations and provide diagnostic and prognostic insight. Based on this, this study constructs and validates a radiomics-based prediction framework to preoperatively identify the KIT exon 11 codon 557–558 deletion mutation (DEL 557–558) in GISTs using contrast-enhanced CT (CE-CT). Methods: The CE-CT images and medical record data were examined for 126 GIST patients who underwent surgical resection and gene mutation testing between 2019 and 2021 at the medical center. Optimal radiomic features were extracted from the selected region of interest (ROI) on the CE-CT images. Two logistic regression (LR) models were established to forecast DEL 557–558: one utilized solely radiomic features, while the other incorporated both radiomic and clinicopathological parameters. The effectiveness of the LR models' predictions was assessed through receiver operating characteristic (ROC) curve analysis, with the mean area under the curve (AUC) value calculated via a five-fold cross-validation protocol. Results: Gastric location, higher mitotic count and higher Ki-67 expression were associated with GIST patients with DEL 557–558 mutation. The radiomic features model, incorporating 12 radiomic features, had AUCs of 0.90 ± 0.01 (95% CI: 0.87–0.94), 0.76 ± 0.04 (95% CI: 0.66–0.88), 0.892 (95% CI: 0.824–0.960) and 0.850 (95% CI: 0.720–0.980) in the prediction of DEL 557–558 in the training, validation, cross-validation and test datasets, respectively. The integrated model combining radiomic attributes with clinicopathological parameters displayed enhanced predictive performance, achieving AUC values of 0.93 ± 0.01 (95% CI: 0.90–0.96), 0.82 ± 0.04 (95% CI: 0.71–0.91), 0.916 (95% CI: 0.857–0.974) and 0.875 (95% CI: 0.745–1.000) in the training, validation, cross-validation and test datasets, respectively. Conclusion: Radiomics models may help predict DEL 557–558 mutations, thereby enabling more effective treatment selection and prognosis assessment.
Zhang et al. (Thu,) studied this question.