1. a. Key Laboratory of Laser Life Science, Ministry of Education; b. Guangdong Key Laboratory of Laser Life Science, College of
Biophotonics, South China Normal University, Guangzhou Guangdong 510631, China; 2. Department of Gastrointestinal Surgery,
The Third Affiliated Hospital of Sun Yat-sen University, Guangzhou Guangdong 510630, China; 3. South China Normal University
(Qingyuan) Science and Technology Innovation Research Institute Co., Ltd., Qingyuan Guangdong 511517, China
Abstract：Objective To accurately locate the tumor location of rectal cancer and conduct preoperative T staging of rectal cancer.
Methods First, CT images of rectal cancer were segmented by FCN-8s, U-Net and SegNet models to locate the tumor location. On
the basis of segmentation, simple CNN, AlexNet and InceptionV3 models were used to identify segmented CT images to determine T
stage of rectal cancer. CT images of rectal cancer from 240 patients were selected for training and testing. Dice coefficient, precision rate
and recall rate were used to evaluate segmentation results, and accuracy, specificity and sensitivity were used to evaluate classification
results. Results U-Net model had the best segmentation effect, and the average Dice coefficient, accuracy rate and recall rate were
84.6%, 84.1% and 85.2%, respectively. After segmentation, the InceptionV3 model had the best classification effect, and the accuracy,
specificity and sensitivity of its classification of T2 and T3 rectal cancer images were 95.4%, 94.3% and 96.1%. The experimental
results showed that U-Net model and InceptionV3 model could obtain better segmentation results and higher classification accuracy.
Conclusion The neural network model can achieve good results in tumor segmentation and preoperative T staging on CT images
of rectal cancer, and can be used as a clinical auxiliary diagnostic tool to help doctors make clinical treatment plans for rectal cancer.
肖波1a,1b，朱旭东2，魏华江1a,1b，魏波2，陈同生1a,1b,3. 基于神经网络的直肠癌CT图像自动分割和分类[J]. 中国医疗设备, 2022, 37(2): 60-64.
XIAO Bo1a,1b, ZHU Xudong2, WEI Huajiang1a,1b, WEI Bo2, CHEN Tongsheng1a,1b,3. Automatic CT Image Segmentation and Classification of Rectal Cancer Based on Neural Network. China Medical Devices, 2022, 37(2): 60-64.
