作者
Huiping Zhao,Jianbo Gao,Biaosheng Bai,Rui Wang,Juan Yu,Hao Lü,Ming Cheng,Pan Liang
摘要
Purpose Molecular testing for microsatellite instability (MSI) status plays a vital role in the clinical management of gastric cancer (GC). Nevertheless, challenges of routinely applied technology for MSI determination exist. This study aimed to develop and validate a non-invasive imaging biomarker for MSI assessment in GC and explore its prognostic value. Methods We retrospectively recruited 396 GC patients with pretreatment CT images from a single center and a public database and divided them into an original cohort (n = 356) and an external validation cohort (n = 40). The SMOTE algorithm was used to generate a balanced training cohort (n = 192) and the independent radiomics model, clinical model, and radiomics-clinic combined model were constructed for determining MSI status. The models’ discrimination, calibration, clinical usefulness, and prognosis significance were evaluated by AUC, calibration, decision curve analyses, and Kaplan-Meier curve analysis, respectively. Results The radiomics-clinic combined model derived from clinical and quantitative CT-based “Radscore” exhibited the best discriminatory abilities of MSI status in all cohorts, with AUCs of 0.836 (95% CI, 0.780–0.893) in the training cohort, 0.834 (95% CI, 0.688–0.981) in the external validation cohort, and 0.750 (95% CI, 0.682–0.819) in the original cohort, respectively. Meanwhile, the combined model demonstrated goodness of fitness, higher clinical net benefits, and significant positive integrated discrimination improvement compared with any independent model. While it showed no significant overall survival- or progression-free survival-based risk stratification ability (p > 0.05). Conclusions The radiomics-clinic combined model could be a potential non-invasive biomarker for MSI status in GC, which help clinical decision-making, nevertheless, provided limited prognostic ability.