Clinical Value of Applying Dual-energy CT Radio-mics Model to Evaluate Serosal Invasion of Advanced Gastric Cancer after Neoadjuvant Chemotherapy Treatment

Objective: We intend to evaluate the diagnostic efficacy of dual-energy CT radio-mics model based on Iodine Map (IM) in the application of preoperative re-staging of serosal invasion in locally advanced gastric cancer (LAGC) after neoadjuvant chemotherapy (NAC) treatment. Methods: A retrospective study was conducted on 155 patients with LAGC who were treated with standard NAC before operation (including 110 cases in training group and 45 cases in testing group). Two radiologists analyzed all the CT images and carried out the classification. After the semi-automatic drawing of region of interest volume (VOI), we extracted 1226 imaging features from each lesion based respectively on IM and 120kVp images. We adopted Spearman related analysis, Least Absolute Shrinkage and Selection Operator (LASSO) to punish Logistic regression in order to acquire important feature by getting rid of unstable and redundant features. Through multi-factor Logistic regression analysis, we established two prediction models (120kVp and IM-120kVp) based on the features selected respectively by 120kVp and 120kVp combined with IM. Results: Two radio-mics models both showed great prediction accuracy and efficiency in training and testing groups (IM-120kVp: AUC: training group, 0.953, testing group, 0.879; 120kVp: AUC: training group, 0.940, testing group, 0.831). The diagnostic accuracy of both models in the testing group (IM-120kVp: 84.4%, 120kVp: 80.0%) were higher than manual classification (68.9%). The diagnostic efficacy of IM-120kVp model was better than manual classification both in training (P<0.001) and testing groups (P=0.034). Conclusion: The radio-mics model based on dual-energy CT shows convincing diagnostic efficacy in differentiating serosal invasion in preoperative re-staging for LAGC patients after NAC treatment.