Anal Center Detection with Superpixel Segmentation

Abstract

Anal center detection is of great significance for the diagnosis of anorectal diseases, for accurate anal center detection can help gastrointestinal (GI) robot enter the human body automatically and check the patient’s anal and lower intestinal diseases, which is the first step to realize autonomous diagnosis. However, there is no available result on the anal center detection. In this work, the superpixel method is employed to find the anal center. In the first step, the collected image dataset is expanded through the data augmentation method. In the second step, we use the superpixel segmentation method, a machine learning algorithm, to segment the image by pixels with similar features in the image. Then we determine the region of interest (ROI) based on the threshold and the size of the connected region. After that, the gray barycenter method is used to determine the center of gravity of the ROI i.e., the anal center. The ground-truth anal center is obtained by the average of the anal center coordinates determined by ten anorectal surgeons. By the proposed algorithm, it is found that the ROI detected in 70.59% of the images in the dataset includes the anal center, and the positioning accuracy of the anal center is 88.87% averagely. Thus, the method can provide the anal center for GI robot.