A convolutional neural network model for T-stage prediction of rectal cancer using CT images

• Published in: 2023 IEEE International Symposium on Medical Measurements and Applications (MeMeA) pp. 1 - 5
• Main Authors: Han, Mingye, Jia, Qingzhu, Xiong, Tingwei, Gao, Yixing, Liu, Peng, Yan, Jia
• Format: Conference Proceeding
• Language: English
• Published: IEEE 14.06.2023
• Subjects: Computed tomography; CT slices; Deep learning; Degradation; Neural networks; Predictive models; rectal cancer; Surgery; T-staging; Training
• DOI: 10.1109/MeMeA57477.2023.10171892

Rectal cancer is a common malignant disease that accounts for a high proportion of tumors of the gastrointestinal system and poses a high risk of death. Therefore, it is important for patients to be preoperatively staged accurately, which helps define an effective surgical treatment plan. The aim of this paper is to combine deep learning methods with CT for preoperative T-staging of rectal cancer. In this paper, we improved AlexNet and proposed a fast and effective classification network called the attention residual convolution neural network (ARCNN). On the one hand, residual structures are introduced to prevent the degradation of neural networks, and on the other hand, the convolutional block attention module (CBAM) is added to improve model performance from both spatial and channel dimensions. The combination of residual structure and attention mechanism can improve the ability of the model to extract features, effectively reduce the interference of invalid features, and thus enhance the model's ability to classify CT images. We used all 3,090 CT images from 318 patients with rectal cancer for training and testing. The model efficiently learns the characteristics of rectal cancer in different stages during training. The classification accuracy on the test set can reach 99.78%. Compared with other comparison deep learning models, our proposed classification model is an efficient and accurate T-staging prediction method for rectal cancer.