A radiomics-based model for prediction of lymph node metastasis in gastric cancer
Xujie Gao1, Tingting Ma1, Jingli Cui2
1Department of Radiology, Tianjin Medical University Cancer Institute and Hospital, Huanhuxi Road, Hexi District, Tianjin 300060, China; National Clinical Research Center for Cancer, Huanhuxi Road, Hexi District, Tianjin 300060, China; Tianjin's Clinical Research Center for Cancer, Huanhuxi Road, Hexi District, Tianjin 300060, China; The Key Laboratory of Cancer Prevention and Therapy, Huanhuxi Road, Hexi District, Tianjin 300060, China.
Purpose:
To develop and validate a radiomics-based model for preoperative prediction of lymph node metastasis (LNM) in gastric cancer (GC).
Method:
A total of 768 GC patients were enrolled in this retrospective study. Radiomics features were extracted from portal venous phase computed tomography (CT) scans. A radiomics signature was built with highly reproducible features using the least absolute shrinkage and selection operator (LASSO) method in the training cohort (n = 486). The signature was further validated in internal validation (n = 240) and external testing cohorts (n = 42). Multivariate logistic regression analysis was conducted to build a model that combined radiomics signature, serum biomarkers, and lymph node status according to CT. Performance of the model was determined by its discrimination, calibration, and clinical usefulness. The predictive value of the model was also evaluated in early stage GC (EGC) subgroup.
Results:
The radiomics signature comprised 7 robust features showed favorable prediction efficacy in all cohorts. A radiomics-based model that incorporated radiomics signature, serum CA72-4, and CT-reported lymph node status had good calibration and discrimination in training cohort [AUC, 0.92; 95% confidence interval (CI), 0.89-0.95] and validation cohort (AUC 0.86; 95% CI, 0.81-0.91). The model also showed a favorable predictive performance for EGC patients with an AUC of 0.85 (95% CI, 0.76-0.94). Decision curve analysis confirmed the clinical utility of this model.
Conclusions:
The radiomics-based model showed favorable accuracy for prediction of LNM in GC. The model may also serve as a noninvasive tool for preoperative evaluation of LNM in EGC.
