Enhanced Approach for Classification of Ulcerative Colitis Severity in Colonoscopy Videos Using CNN

Ulcerative Colitis (UC) is an inflammatory bowel disease that causes inflammation, ulcers and bleeding of the colon affecting more than 500,000 people in the United States. To achieve the therapeutic objectives for UC, which are to first induce and then maintain disease remission, physicians need to evaluate the severity of UC. However, objective assessment of US is difficult because of the non-uniform nature of symptoms and large variations in disease presentation. To address this, in our previous work, we developed two different approaches in which one uses the image textures, and the other uses CNN (Convolutional Neural Network) to measure and objectively classify the severity of UC as seen on optical colonoscopy video frames. However, we found that the image texture based approach could not handle large numbers of variations in patterns, and the CNN based approach could not achieve very high accuracy. In this paper, we improve our CNN based approach in two ways to provide better accuracy for UC severity classification. We add more thorough and essential preprocessing, subdivide each class of UC severity and generate more classes for the classification to accommodate large variations in patterns. The experimental results show that the proposed preprocessing and generation of more classes can improve the overall accuracy of automated classification of the severity of UC.