Establishment And Clinical Application Of An Automatic Diagnosis Platform For Rectal Cancer T-staging Based On A Deep Neural Network

Background:As a common gastrointestinal malignancy,Colorectal cancer seriously threatens the life of patients.With the continuous understanding of the pathogenesis of rectal cancer,the treatment of rectal cancer has formed a comprehensive treatment of surgical treatment,combined with radiotherapy and chemotherapy treatment.Different patients should have different treatment plans to have a long life.Scientific treatment plan will bring better prognosis and quality life to patients.The development of patients’ treatment plans depends on accurate preoperative staging of cancer.Magnetic resonance imaging(MRI),as the preferred method of rectal cancer examination,can accurately determine the tumor staging in patients with rectal cancer.Artificial intelligence technology,mainly represented by machine learning,has made remarkable achievements in image recognition in recent years.Introducing artificial intelligence into image recognition can establish a stable and reliable platform for image recognition and judgment in a short time.Objective:The objective of this study was to establish an automatic diagnostic platform for preoperative T staging of rectal cancer patients by learning high resolution T2 sequence images of rectal cancer patients through deep neural network.Methods:The imaging data and clinical data of 183 patients with rectal cancer in our hospital were retrospectively collected.And the patients were randomly divided into two groups,training group(146 cases)and the validation group(37 cases).Tumor infiltrating the deepest areas in the images of training group were annotated.Faster region-based convolutional neural networks(Faster R-CNN)was used to build the platform.Images of training group were input into the platform to train the platform and completed the preliminary platform.Then using the validation group validated the performance of the platform,and evaluate platform by receiver operating characteristic(ROC)curve.Results:An automatic diagnostic platform for T staging of rectal cancer was established by learning magnetic resonance images of the included patients.The platform can accurately judge the images in different imaging directions(AUC=1.00)and identify the location of lesions.The areas under the ROC curve in different imaging directions were as follows:AUC in the horizontal plane =0.99,AUC in the sagittal plane =0.97,and AUC in the coronal plane =0.98.In each layer,the identification results of each stage are as follows: in the horizontal plane,AUC(T1)=1.00,AUC(T2)=1.00,AUC(T3)=1.00,AUC(T4)=1.00;In coronal plane,AUC(T1)=0.96,AUC(T2)=0.97,AUC(T3)=0.97,AUC(T4)=0.97;In sagittal plane,AUC(T1)=0.95,AUC(T2)=0.99,AUC(T3)=0.96,AUC(T4)=1.00Conclusion:Faster R-CNN AI might be an effective and objective method to build the platform for predicting rectal cancer T-staging.