双能CT影像组学模型评估进展期胃癌新辅助化疗后浆膜侵犯的临床价值研究

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

目的:探讨基于碘图(IM)的双能CT影像组学模型在新辅助化疗(NAC)后局部进展期胃癌(LAGC)浆膜浸润的术前再分期中的诊断效能。方法:对155例(训练组110例,测试组45例)术前经过标准NAC治疗的LAGC患者进行回顾性研究。所有CT图像由两名放射科医生分析,并进行人工分类。半自动勾画感兴趣区体积(VOI),在IM和120 k Vp图像上分别从每个病变中提取了1226个影像组学特征。采用Spearman相关分析和最小绝对收缩选择算子(LASSO)惩罚Logistic回归过滤不稳定及冗余特征,从而筛选出重要特征。通过多因素Logistic回归分析,分别得到了基于120 kVp选择的特征和120 kVp结合IM选择的特征建立的两个预测模型(120 k Vp和IM-120 kVp)。结果:两种影像组学模型(IM-120 k Vp AUC:训练组,0.953,测试组,0.879;120 kVp AUC:训练组,0.940,测试组,0.831)在训练和测试组中均显示出较高的预测准确度和效能。所有模型在测试组的诊断准确率(IM-120 k Vp:84.4%,120 k Vp:80.0%)均高于人工分类(68.9%)。IM-120 kVp模型在训练(P<0.001)和测试组中的诊断效能(P=0.034)均优于人工分类。结论:基于双能CT的影像组学模型在NAC治疗后LAGC术前再分期鉴别浆膜侵犯方面表现出令人信服的诊断效能。

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 120 k Vp 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(120 k Vp and IM-120 k Vp)based on the features selected respectively by 120 k Vp and 120 k Vp combined with IM.Results:Two radio-mics models both showed great prediction accuracy and efficiency in training and testing groups(IM-120 k Vp:AUC:training group,0.953,testing group,0.879;120 k Vp:AUC:training group,0.940,testing group,0.831).The diagnostic accuracy of both models in the testing group(IM-120 k Vp:84.4%,120 k Vp:80.0%)were higher than manual classification(68.9%).The diagnostic efficacy of IM-120 k Vp 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.