Deep Learning for Relevant Findings in Colonoscopy

Colorectal cancer is a leading cause of death in the world, thus, early prevention and treatment became a major priority. Improved detection, on-site evaluation and the accuracy of the medical exploration are essential for this attempt. Automatic detection of polyps, diverticula and membrane bleedings or lesions, might enhance the objectivity, precision and efficiency of this examination, aiding the endoscopy expert. Video colonoscopy files can be split in frames and these can be classified into informative or non-informative images. Applying the technology advancements on the chosen frames, the system can learn to detect and point on the relevant findings. It can assist on-line object detection and off-line classification and segmentation of images. Using diverse structures of deep learning, it is able to help the physicians or to assess the quality of the video colonoscopy procedures. We will further describe some affordable, efficient, solutions that might be used to tackle these problems worldwide.