基于增强CT图像和Swin Transformer网络的食管癌T分期智能诊断模型的构建与评估

Construction and evaluation of an intelligent diagnostic model based on enhanced CT images and Swin Transformer network for T staging of esophageal cancer

目的基于增强CT图像和Swin Transformer网络,拟构建食管癌T分期智能诊断模型。方法收集2018年1月至2022年4月在陆军军医大学第一附属医院和山西省肿瘤医院胸外科经病理证实为食管癌的150例患者的45000张术前增强CT图像。经过UperNet Swin网络自动分割和肿瘤体积的计算,使用ResNet50、Swin Transformer和VIT 3个网络进行食管癌T分期智能诊断模型的构建。使用精准率、召回率、F1-score、特异度以及阴性预测值(negative predictive value,NPV)等指标在150例内部数据集上评价模型性能,描绘混淆矩阵和ROC曲线。结果在3个食管癌T分期诊断的模型中,Swin Transformer模型结合肿瘤体积、病理信息等特征的分期诊断效果最好,T1~T4期的精准率分别为1.00、0.67、0.83、1.00,AUC为0.861,优于ResNet50和VIT分期诊断模型,它们的精准率分别为0.13、0.27、0.59、0.81和0.03、0.14、0.56、0.75,AUC分别是0.611和0.542。结论与ResNet50和VIT网络比较,Swin Transformer网络能够更精准进行食管癌智能T分期诊断。

Objective To construct an intelligent diagnosis model for T stage of esophageal cancer based on the enhanced CT images and Swin Transformer net work.Methods A total of 45000 preoperative enhanced CT images of 150 patients with postoperative pathologiclly confirmed esophageal cancer in the Department of Thoracic Surgery of the First Affiliated Hospital of the Army Medical University and Shanxi Cancer Hospital from January 2018 to April 2022 were collected.Through automatic segmentation of the UperNet Swin network and calculation of tumor volume,ResNet50,Swin Transformer,and VIT were used to construct the intelligent diagnostic model for T staging of esophageal cancer.The model performance was evaluated using precision,recall,F l-score,speeificity and negative predictive value(NPV)in an in-house dataset of 150 cases,and confusion matrix and ROC curves were depicted.Results Among the three diagnostic models for T staging of esophageal cancer,the Swin Transformer model combined with the features of tumor volume and pathological information had the best effect in staging diagnosis,with precision rates of 1.00,0.67,0.83,and 1.00 in T1~T4 stages(AUC=0.861).The precision rate of ResNet50 was 0.13,0.27,0.59 and0.81(AUC=0.611),and that of VIT staging diagnosis model was 0.03,0.14,0.56 and 0.75(AUC=0.542).Conclusion Compared with ResNet50 and VIT networks,Swin Transformer network allow intelligent diagnosis for T Staging of esophageal cancer to be employed with greater preci sion.