A multi-class scoring system based on CT features for preoperative prediction in gastric gastrointestinal stromal tumors

Our study aimed to establish and validate a multi-class scoring system for preoperative gastric gastrointestinal stromal tumors (GISTs) risk stratifications based on CT features. 150 gastric GIST patients who underwent contrast-enhanced CT examination and surgical resection from hospital 1 were retrospectively analyzed as the training cohort, and 61 patients from hospitals 2 and 3 were included as the validation cohort. A model was established by logistic regression analysis and weighted to be a scoring model. A calibration test, area under the receiver operating characteristic (ROC) curve (AUC), and cutoff points were determined for the score model. The model was also divided into three score ranges for convenient clinical evaluation. Five CT features were included in the score model, including tumor size (4 points), ill-defined margin (6 points), intratumoral enlarged vessels (5 points), heterogeneous enhancement pattern (4 points), and exophytic or mixed growth pattern (2 points). Then, based on the calibration results, performance was merely assessed as very low and high* risk. The AUCs of the score model for very low risk and high* risk were 0.973 and 0.977, and the cutoff points were 3 points (97.30%, 93.81%) and 7 points (92.19%, 94.19%), respectively. In the validation cohort, the AUCs were 0.912 and 0.972, and the cutoff values were 3 points (92.31%, 85.42%) and 5 points (100%, 87.88%), respectively. The model was stratified into 3 ranges: 0-3 points for very low risk, 4-8 points for low risk, and 9-21 points for high* risk. A concise and practical score system for gastric GISTs risk stratification was proposed.