Evaluation Of Rectal Cancer Circumferential Resection Margin Using Faster Region-based Convolutional Neural Network(R-CNN) In High-resolution Magnetic Resonance Images

Background: High-resolution MRI is regarded as the best method to evaluate whether there is an involved circumferential resection margin(CRM)in rectal cancer.Objective: We explored the application of the Faster Region-based Convolutional Neural Network(Faster R-CNN)to identify positive circumferential resection margins in high-resolution MRI images.Main Outcome Measures: This was a retrospective study conducted at a single surgical unit of a public university hospital.We studied 240 patients with rectal cancer in the Affiliated Hospital of Qingdao University from 07/2016 to 08/2018,who were determined to have a positive circumferential resection margin and who had received a high-resolution MRI.All posttreatment cases were excluded from this study.The Faster Region-based Convolutional Neural Network was trained by 12,258 transverse relaxation weighted imaging(T2WI)images of pelvic high-resolution MRI to build an artificial intelligence platform and complete clinical tests.In this network,the proportion of positive and negative circumferential resection margin images was 1:2.In accordance with the test results of the validation group,the metrics of the receiver operating characteristic curves(ROC)and the area under the curve(AUC)were applied to compare the diagnosis results of the artificial intelligence platform with those of senior radiology experts.Results: In this artificial intelligence platform,the accuracy,sensitivity and specificity of the circumferential resection margin status as determined were 0.932,0.838 and 0.956,respectively.The area under the receiver operating characteristic curves was0.953.The time required to automatically recognize an image was 0.2 seconds.Limitations: This is a single-center retrospective study with limited data volume and a highly selected patient cohort.Conclusions: In high-resolution MRI images of rectal cancer before treatment,the application of Faster Region-based Convolutional Neural Network to segment the positive circumferential resection margin has high accuracy and feasibility.