Abstract
The aim of the study was to investigate the intelligent recognition of radiomics based on the convolutional neural network (CNN) in predicting endometrial cancer (EC). In this study, 158 patients with EC in hospital were selected as the research objects and divided into a training group and a test group. All the patients underwent magnetic resonance imaging (MRI) before surgery. Based on the CNN, the imaging model of EC prediction was constructed according to the characteristics. Besides, the comprehensive prediction model was established through the clinical information and imaging parameters. The results showed that the area under the working characteristic curve (AUC) of the radiomics model and comprehensive prediction model was 0.897 and 0.913 in the training group, respectively. In addition, the AUC of the radiomics model was 0.889 in the test group and that of the comprehensive prediction model was 0.897. The comprehensive prediction model was established through specific imaging parameters and clinical pathological information, and its prediction performance was good, indicating that radiomics parameters could be applied as noninvasive markers to predict EC.
Highlights
endometrial cancer (EC) is a group of epithelial malignant tumors that occur in the endometrium, the most frequent is adenocarcinoma originating from the endometrial glands [1]
Each patient underwent sagittal T1- and T2weighted images of pelvic enhanced magnetic resonance imaging (MRI) before surgery. e convolutional neural network (CNN) model was constructed, the imaging model for EC prediction was screened out based on features, and the comprehensive prediction model for EC was established based on clinical pathological information and imaging parameters. e patient’s region of interest (ROI) was drawn, and the area under the working characteristic curve (AUC), sensitivity, specificity, and accuracy were applied to evaluate the diagnostic effect of the constructed model, and its effect was verified in patients of the test group
Experimental Results of the Convolutional Neural Network in Dataset. e network structure of the CNN was similar to that of LeNet-5. e main difference was that the CNN did not adopt some of the previous parameters in LeNet-5 and applied a fully connected network in the final classifier part. e misclassification rate curve of the CNN in the training process is shown in Figure 8. e abscissa stood for the times of iterations, and the ordinate represented the misclassification rate. e test misclassification rate after the CNN convergence was higher than that of LeNet-5, and the CNN test misclassification rate and training misclassification rate were higher than those of LeNet-5 during the entire training process
Summary
EC is a group of epithelial malignant tumors that occur in the endometrium, the most frequent is adenocarcinoma originating from the endometrial glands [1]. EC often emerges in postmenopausal and perimenopausal women. It is one of the most common tumors of the female reproductive system, with nearly 200,000 new cases every year [2]. Among the gynecological malignant tumors that cause death, it ranks third after ovarian cancer and cervical cancer. Endometrial hyperplasia is the pathological change of the endometrium before canceration and has the potential to deteriorate. E morphology of atypical endometrial hyperplasia is similar to that of EC, and it is difficult to distinguish in clinical diagnosis [5]. Based on the danger of endometrial hyperplasia and the difficulty of clinical diagnosis, it is necessary to find a method that can intelligently identify normal endometrial, endometrial hyperplasia, and endometrial cancerous tissues
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call