Su1567 Adequate Treatment for Early Colorectal Cancer Diagnosed by Magnifying Endoscopy

Abstract

lesions. Interpretation of the magnifying NBI findings based on an objective standard is, therefore, demanded. Purpose: To evaluate the quantification of magnifying NBI findings of colorectal lesions and utility of computer-assisted automatic identification system for NBI classificationMethod: In this study, Sano’s classification which is the original colorectal NBI classification was further broadly divided into three classes A: CP type I, B: CP type II CP type IIIA and C: CP type IIIB. First, the following computer-assisted procedures were performed using front-view magnifying NBI images appropriate for each class (A: 5 images, B: 10 images, C: 6 images): determination of region of interest (ROI), conversion of images into grayscale images, visualization of rough vessels through binarization using the moving average method, deletion of all colors except the vessel color and removal of isolated points and holes. Subsequently, characteristics such as average visualized vessel width, average visualized vessel length, total visualized vessel length, average visualized vessel concentration and the fractal dimension were used to extract significant quantitative characteristics from these. Moreover, the images for the region of interest were added, and a total of 37 images (A: 11 images, B: 18 images, C: 8 images) were distinguished using an automatic identification system. In addition, identifications of the images of B and C alone were conducted. Results: As a result of the evaluation of each quantitative characteristic in the 21 images of ROI, significant differences were observed among several classes in three quantitative characteristics: average visualized vessel length, total visualized vessel length, and average visualized vessel concentration. As a result of the identification performed on a total of 37 images using the above quantitative characteristics, the identification rate of three classes was 81.1%. In addition, the identification rate between B and C was 96.2%. Conclusion: Admissible identification rates were obtained for the quantification and automatic identification of the magnifying colorectal NBI findings. Our future challenges are to improve the reliability of each quantitative characteristic and to seek for new quantitative characteristics. It can be expected that once automatic identification procedure with high accuracy is achieved, it will contribute to the education of less-experienced endoscopists.