Establishment And Applicability Of A Diagnostic System For Advanced Gastric Cancer T Staging Based On A Faster Region-Based Convolutional Neural Network

Objective Based on Faster R-CNN,an automatic tumor diagnosis system with enhanced CT images of the upper abdomen of advanced gastric cancer is constructed to automatically delineate the tumor lesion area of advanced gastric cancer and analyze and identify the T staging of advanced gastric cancer.Method The research team retrospectively collected 293 patients with advanced gastric cancer who were in the Affiliated Hospital of Qingdao University from January 2017 to June2018.After double screening of inclusion and exclusion criteria,a total of 225 study subjects were included(64 of them were in T2 stage.72 cases of T3 stage,89 cases of T4stage),the upper abdomen enhanced CT images of the above subjects were collected.Two senior imaging specialists used label Img software to identify the cancerous area consistent with the postoperative pathological T stage on the enhanced CT image of the upper abdomen.The research team used a random number method to divide the study subjects into a training group(51 cases in T2 stage,58 cases in T3 stage,71 cases in T4stage)and a test group(13 cases in T2 stage,T3 stage)according to a 4:1 ratio.There are14 cases and 18 cases in T4 stage).According to statistics,the training group contains2800 positive images and the test group contains 700 positive images.In order to reduce the over-fitting problem when training the diagnostic system,the research team carried out data enhancement processing on the positive images in the training set.After data enhancement processing,3855 feature images were obtained.Subsequently,the research team used the training group to train the automatic diagnosis system with the full sequence of 3855 feature images.From the result of the learning curve loss function of the training diagnosis system,it can be concluded that after 600 learning cycles,the diagnosis system obtains the best optimized parameters.In order to verify the performance of the diagnosis system to automatically segment the tumor and identify the T stage of the tumor,the automatic diagnosis system was tested using the full sequence of images of the test group containing 700 positive images.In order to quantify the diagnostic accuracy of the diagnostic system in T staging of gastric cancer,the research team used accuracy,specificity,sensitivity,positive predictive value,negative predictive value,receiver operating characteristic curve(ROC curve)and area under the curve.(AUC value)for comprehensive evaluation.Result The research team counted the number of true positives and false positives of the gastric cancer tumor location and T stage automatically identified by the diagnostic system in the images of the test group,and further calculated the true positive and false positive rates of the diagnostic system correctly identifying gastric cancer images under different probability thresholds.The ROC curve is drawn based on the true and false positive rate,and the area under the curve is 0.93(AUC value is 0.93).From this,it can be concluded that the overall accuracy of the T staging diagnosis system for advanced gastric cancer is93%,and the sensitivity and specificity are further calculated They are 95% and 95%respectively.In the same way,it can be concluded that the accuracy of the diagnosis system for gastric cancer T2,T3 and T4 stages is 90%,93% and 95%,respectively.The time it takes for this diagnostic system to automatically recognize and analyze a single image is 0.2s,while the average time it takes for an imaging expert to interpret a single image is 10 s.ConclusionThe T staging diagnosis system for advanced gastric cancer constructed based on Faster RCNN has high accuracy and practical operability.The diagnosis system can autonomously complete the automatic segmentation of advanced gastric cancer tumors in the enhanced CT image of the upper abdominal venous phase and the T stage of gastric cancer.Recognition can assist radiologists to quickly and accurately interpret the enhanced CT images of the upper abdomen of advanced gastric cancer.