Research On The Preoperative Identification Of Lymph Node Metastasis In CT Images Of Rectal Cancer Based On Deep Learning

In recent years,the incidence of rectal cancer has remained high.Rectal cancer is prone to infiltrate outside the intestine and cause lymph node and distant metastasis.In order to reduce the possibility of postoperative recurrence and metastasis,all rectal cancer patients with lymph node metastasis are completely removed during surgery.The lymph nodes are very necessary,but it is difficult to diagnose lymph node metastasis through the CT images of patients before surgery.Accurate judgment of lymph node metastasis before surgery is an important step in the treatment of rectal cancer.At present,using CT images of patients before surgery,there are no clinically recognized reports that can better and accurately judge lymph node metastasis before surgery.Based on imagingomics and deep learning methods,this paper studies the lymph node metastasis of rectal cancer.The data set comes from question B of the 7th Teddy Cup Data Mining Challenge.There are total of 107 patients with rectal cancer CT image data.Based on the imaging omics method,837 imaging omics features were extracted from the patient’s CT images,and then 20 features were selected through LASSO regression and the logistic regression,support vector machine and random forest machine learning models were established to identify lymph node metastasis The performance of the random forest model is better,with an F1 score of 83.33%.Aiming at the problem that doctors are susceptible to subjective factors when manually labeling ROI,this paper proposes an image segmentation algorithm that combines U-net network and residual structure to segment the rectal region of rectal cancer images.The improved Unet network in this paper deepens the network depth and solves the problems of gradient disappearance.The Dice coefficient reaches 85.54%,which realizes the accurate segmentation of the rectal area.In addition,this paper uses three commonly used deep learning networks to classify rectal cancer images and introduces the transfer learning method for training.The classification accuracy of the ResNet model reaches 81.63%.The results show that the ResNet network based on transfer learning has a good classification effect on rectal CT images.Compared with traditional imaging omics methods,this model improves the accuracy of preoperatively identifying lymph node metastasis by 5.55%.This article is based on imaging genomics and deep learning technology to study the lymph node metastasis of rectal cancer CT images,and evaluate the lymph node metastasis based on the extracted CT image features of rectal cancer,which improves the accuracy of imaging judgments of lymph node metastasis.The treatment of lymph node metastasis in patients with upper rectal cancer provides an auxiliary diagnostic method,which has certain guiding significance for the selection of early clinical treatment options for patients.