Research on Rectal Tumor Identification Method by Convolutional Neural Network Based on Multi-Feature Fusion

Abstract

Aiming at the obscure features of tumors in rectal CT images and their complexity, this paper proposes a multi-feature fusion-based convolutional neural network rectal tumor recognition method and uses it to model rectal tumors for classification experiments. This method extracts the convolutional features from rectal CT images using Alexnet, VGG16, ResNet, and DenseNet, respectively. At the same time, local features such as histogram of oriented gradient, local binary pattern, and HU moment invariants are extracted from this image. The above local features are fused linearly with the convolutional features. Then we put the new fused features into the fully connected layer for image classification. The experimental results finally reached the accuracy rates of 92.6 %, 93.1 %, 91.7 %, and 91.1 %, respectively. Comparative experiments show that this method improves the accuracy of rectal tumor recognition.

An effective study on the diagnosis of colon cancer with the developed local binary pattern method

“…Among traditional image enhancement methods, the LBP method was also used. In the study, only the accuracy rate was considered in the hybrid method developed by the researchers, and the highest accuracy rate obtained as a result of the study was reported as 93.1% [6]. In the [7] study, researchers conducted a very remarkable study on the detection of polyps, which have a vital place in colorectal cancer.…”