A quantitative model using multi-parameters in dual-energy CT to preoperatively predict serosal invasion in locally advanced gastric cancer

Abstract

ObjectivesTo develop and validate a quantitative model for predicting serosal invasion based on multi-parameters in preoperative dual-energy CT (DECT).Materials and methodsA total of 342 LAGC patients who underwent gastrectomy and DECT from six centers were divided into one training cohort (TC), and two validation cohorts (VCs). Dual-phase enhanced DECT-derived iodine concentration (IC), water concentration, and monochromatic attenuation of lesions, along with clinical information, were measured and collected. The independent predictors among these characteristics for serosal invasion were screened with Spearman correlation analysis and logistic regression (LR) analysis. A quantitative model was developed based on LR classifier with fivefold cross-validation for predicting the serosal invasion in LAGC. We comprehensively tested the model and investigated its value in survival analysis.ResultsA quantitative model was established using IC, 70 keV, 100 keV monochromatic attenuations in the venous phase, and CT-reported T4a, which were independent predictors of serosal invasion. The proposed model had the area-under-the-curve (AUC) values of 0.889 for TC and 0.860 and 0.837 for VCs. Subgroup analysis showed that the model could well discriminate T3 from T4a groups, and T2 from T4a groups in all cohorts (all p < 0.001). Besides, disease-free survival (DFS) (TC, p = 0.015; and VC1, p = 0.043) could be stratified using this quantitative model.ConclusionThe proposed quantitative model using multi-parameters in DECT accurately predicts serosal invasion for LAGC and showed a significant correlation with the DFS of patients.Critical relevance statementThis quantitative model from dual-energy CT is a useful tool for predicting the serosal invasion of locally advanced gastric cancer.Key PointsSerosal invasion is a poor prognostic factor in locally advanced gastric cancer that may be predicted by DECT.DECT quantitative model for predicting serosal invasion was significantly and positively correlated with pathologic T stages.This quantitative model was associated with patient postoperative disease-free survival.Graphical