Cielab Automatic Colonoscopy Post-Evaluation

Aims Creating an automatic software tool to objectively review and evaluate colonoscopy images. By using CIELAB color space we may detect feces residues, artefacts, and clear intestinal mucosa at pixel level in each frame. By computing pixel percentages, we propose an objective quantification of bowel preparation degree and colonoscopy uninterpretable images.

Methods We took samples of each kind of pixels as small image rectangles inside representative regions of frames. Then, some colorprints defining each type of pixels are build automatically. Based on these colorprints, colonoscopy images are automatically processed and pixel representing feces, artefacts and clear intestinal mucosa are labeled with different colors and counted accordingly. Using several empirical thresholds, frames can be classified as usable or not for diagnosis.

Results We ran our method on a 12 minutes and 50 seconds colonoscopy, consisting of 19,245 frames. The color samples were taken out of the first 800 frames. The computer software found 138 frames (0.72 %) with feces residues over 15 %, 794 frames (4.12 %) with artefacts over 30 % and only 1,665 very good frames (8,65 %) with feces residues below 5 %, artefacts below 10 % and over 50 % of pixels representing clear intestinal mucosa. The rest of 16,648 frames (86.5 %) had less than 50 % intestinal mucosa pixels and therefore were not classified as diagnosis interpretable.

Conclusions We proposed a method to automatically analyze colonoscopy images to classify them as suitable or not to reach a diagnosis. It can be used to post-evaluate colonoscopies. Using a deep-learning algorithm this method may be successful for automatic live or recorded evaluation of a colonoscopy quality.

Citation: Ciobanu A, Luca M, Oltean A eP293 CIELAB AUTOMATIC COLONOSCOPY POST-EVALUATION. Endoscopy 2021; 53: S194.

Publication History

Article published online:
19 March 2021