江苏大学学报（医学版）2025，Vol.35Issue(5)：421-428,8.DOI:10.13312/j.issn.1671-7783.y250004
基于低张水充盈胃CT图像的进展期胃癌智能T分期应用研究
Application of intelligent T-staging model for advanced gastric cancer based on low-tension water-filled stomach CT images
刘博文 1单秀红 1蒋鹏程 2王泽辉 3王霄霄 1王芷旋 1彭晨 1刘展鹏 1卢超 1潘冬刚1
作者信息
- 1. 江苏大学附属人民医院医学影像科,江苏 镇江 212002
- 2. 江苏大学附属人民医院普外科,江苏 镇江 212002
- 3. 镇江中澳人工智能研究院,江苏 镇江 212021
- 折叠
摘要
Abstract
Objective:To construct an end-to-end deep learning model for the multi-class classification task of T staging in advanced gastric cancer using CT images.Methods:This retrospective study included enhanced venous phase images from 423 gastric cancer patients,which were randomly divided into the training set and the test set at an 8∶2 ratio.We employed a deep learning automatic segmentation model based on the 3D-nnUNet for segmenting tumors.Simultaneously,a multi-class classification model based on the SmallFocusNet was developed for classification of T staging of advanced gastric cancer.Finally,these models were integrated to construct an end-to-end deep learning model for CT-T staging diagnosis of advanced gastric cancer.The performance of the segmentation model was evaluated using the Dice similarity coefficient(DSC),Intersection over Union(IoU),and 95%Hausdorff Distance(HD_95).The prediction efficacy of the deep learning model was assessed using area under the ROC curve(AUC)values,sensitivity,and specificity.Additionally,the diagnostic performance of the deep learning model for classification of T staging of advanced gastric cancer was compared with that of radiologists.Results:In the test set,the DSC and IoU of the automatic segmentation model were 0.869±0.095 and 0.779±0.137,respectively.The macro-average AUC value of the deep learning model was 0.882(95%CI:0.812-0.926).The AUC values for distinguishing T2,T3 and T4a stage tumors were 0.960(95%CI:0.915-0.990),0.739(95%CI:0.616-0.849)and 0.917(95%CI:0.812-0.926),respectively.The average sensitivity was 0.769(95%CI:0.676-0.853),with sensitivities for distinguishing T2,T3 and T4a stage tumors of 0.808(95%CI:0.654-0.923),0.750(95%CI:0.571-0.893)and 0.750(95%CI:0.594-0.906),respectively.Furthermore,the deep learning model outperformed radiologists in the diagnostic performance of T staging for advanced gastric cancer.Conclusion:The end-to-end deep learning model,which integrates multi-channel and attention mechanisms based on enhanced CT images,demonstrates high accuracy and consistency in preoperative T staging diagnosis of advanced gastric cancer.关键词
胃肿瘤/深度学习/进展期胃癌/自动分割/多分类/T分期Key words
gastric tumor/deep learning/advanced gastric cancer/automatic segmentation/multi-classification/T staging分类
医药卫生引用本文复制引用
刘博文,单秀红,蒋鹏程,王泽辉,王霄霄,王芷旋,彭晨,刘展鹏,卢超,潘冬刚..基于低张水充盈胃CT图像的进展期胃癌智能T分期应用研究[J].江苏大学学报（医学版）,2025,35(5):421-428,8.
