CNN-Based Segmentation Network Applied to Image Recognition of Angiodysplasias Lesion Under Capsule Endoscopy

Background & Objective: Small intestinal vascular malformations (angiodysplasias) are the frequent causes of small intestinal bleeding [1-2]. Capsule endoscopy has become the primary diagnostic method of angiodysplasias. However, manual reading of the entire gastrointestinal tract is a heavy workload, time-consuming, and affects the accuracy of diagnosis. Artificial intelligence assists the diagnosis and can increase the detection rate of angiodysplasias in the small intestine, achieve automatic disease detection, and shorten the CE reading time.Methods: This study proposed an improved convolutional neural network (CNN) semantic segmentation network model that automatically recognizes the angiodysplasias category under CE and draws the outline of the lesion, improving the manual diagnosis of angiodysplasias lesions. The Skeleton network was extracted using Resnet-50, improved and optimized by fusion of shallow features and deep features, and the image was classified at the pixel level to achieve the segmentation of angiodysplasias lesions. The training and test sets were constructed, and the comparative experiments were performed with PSPNet, DeeplabV3+ and UperNet.Results: The test set constructed in the study achieved satisfactory results: pixel accuracy was 99%, mean IOU was 0.69, negative predictive value was 98.74%, and positive predictive value was 94.27%. The model parameter was 46.38 M, the float calculation was 467.2 G, and the duration was 0.6 s to segment and recognized a picture.Conclusions: Thus, constructing a segmentation network based on deep learning to segment and recognize angiodysplasias lesions is an effective and feasible method in the diagnosis of angiodysplasias lesions.