2024 Volume 53 Issue 4 Pages 255-262
This paper presents a method for removing the tube region in endoscopic images used for the diagnosis of internal hemorrhoids. This method can work for a preprocessing in the existing computer- aided diagnosis system of the hemorrhoids. Since the appearance of the tube region depends on the lighting and the condition of the subject, the region had to be removed by hand in the existing system. Utilizing characteristics of pixel values in the region and the location of the region, the proposed method segments the image into plural candidates of the tube region, and discriminates whether each candidate could be a part of the tube region. In the experiments of the proposed method for 192 images (96 normal and 96 abnormal cases), we confirmed the performance of the tube removal. Besides, after applying the removal to the images, they were tried in the existing CAD system, which discriminates the images between normal and abnormal cases. In the experiments, we confirmed the tube region was correctly removed in 172 images. And, in the trials by the existing CAD system, the accuracy averaged 79.1%, i.e., we confirmed that automatic diagnosis of internal hemorrhoids can conducted with approximately 80% by using the proposed method.
