基于深度学习算法的胰腺癌CT自动分期系统的构建与应用

Construction and application of automatic CT staging system for pancreatic cancer based on deep learning algorithm

目的:构建基于深度学习算法的胰腺癌计算机断层扫描(CT)自动分期系统,并探讨其应用价值。方法:回顾性分析我院2014年01月至2021年12月收治的286例胰腺癌患者的临床资料,均经CT检查且明确TNM分期,利用CT检查信息基于深度学习算法的胰腺癌CT自动分期系统。另选取2022年01月至2023年02月胰腺癌患者92例,均经CT检查,并利用上述系统进行TNM分期,分析该系统的准确性。结果:基于深度学习算法的胰腺癌CT自动分期系统共包括7个模块,可以实现胰腺癌TNM自动分期;92例患者中共有Ⅰ期12例、Ⅱ期31例、Ⅲ期36例、Ⅳ期13例,经基于深度学习算法的胰腺癌CT自动分期系统诊断共有Ⅰ期10例、Ⅱ期31例、Ⅲ期38例、Ⅳ期13例;该系统诊断胰腺癌TNM分期的灵敏度、特异度和准确度高,且与金标准高度一致(Kappa值=0.912,P<0.001)。结论:本研究构建了基于深度学习算法的胰腺癌CT自动分期系统,诊断价值高。

Objective:To construct an automatic staging system of pancreatic cancer computed tomography(CT)based on deep learning algorithm and explore its application value.Methods:Retrospective analysis was made on the clinical data of 286 patients with pancreatic cancer admitted to the hospital from January 2014 to December 2021.All of them had undergone CT examination and identified TNM staging.The automatic CT staging system for pancreatic cancer was based on deep learning algorithm using CT examination information.In addition,92 patients with pancreatic cancer from January 2022 to February 2023 were selected,all of whom were examined by CT,and the above system was used for TNM staging to analyze the accuracy of the system.Results:The automatic CT staging system for pancreatic cancer based on deep learning algorithm included 7 modules,which could realize automatic TNM staging of pancreatic cancer.There were 12 cases of stageⅠ,31 cases of stageⅡ,36 cases of stageⅢand 13 cases of stageⅣin 92 patients,and 10 cases of stageⅠ,31 cases of stageⅡ,38 cases of stageⅢand 13 cases of stageⅣwere diagnosed by the automatic CT staging system based on deep learning algorithm.The system had high sensitivity,specificity and accuracy in diagnosing TNM staging of pancreatic cancer,and which was highly consistent with the gold standard(Kappa value=0.912,P<0.001).Conclusion:In this study,an automatic CT staging system for pancreatic cancer based on deep learning algorithm is constructed,which has high diagnostic value.