An image classification model based on transfer learning for ulcerative proctitis

Ulcerative colitis (UC) can be classified as proctitis, left-sided colitis or pancolitis, usually with rectal involvement at the beginning. Mucosal carcinogenesis is one of the most severe complications of UC. Persistent inflammation of the rectal mucosa may be an essential cause of mucosal cancer, thus the detection of rectal inflammation is of great significance. In this paper, we propose a transfer learning model to classify enteroscopy images to achieve adequate detection of ulcerative proctitis. First, with the support of senior doctors, a dataset of ulcerative proctitis is created with 1450 endoscopic images. Then, with trained in the dataset, a new multi-model fusion network is proposed to classify ulcerative proctitis images. The proposed model combines three pre-trained models which are Xception, ResNet and DenseNet, and these pre-trained models are used to extract features from the images, then the extracted features are fed into a fully connected layer to predict the label of the input image. Experimental results show that, compared with other models, the proposed model has better performance, achieving the classification accuracy of 97.93%, the sensitivity of 99% and the specificity of 99%.