Real-time endoscopic colonoscopy polyp detection system

Abstract

The present application discloses a real-time endoscopic colonoscopy polyp detection system, including: a video information acquisition unit, used to decompose the received video into video segments with several frames as a group; a video information extraction unit, used to analyze all Preprocessing the video clips described above, and extracting optical flow information; and, a video information analysis unit, used to obtain the video information processed by the video information extraction unit, and input it into a deep convolutional neural network for detection, thereby obtaining polyps Test results. Using the real-time endoscopic colonoscopy polyp detection system of the present application, spatio-temporal information can be extracted from the video stream generated in real time during the operation, and assist doctors to find intestinal polyps with low delay and high precision, improve the accuracy of doctor's judgment, and reduce the burden of surgery , and speed up the surgical process.

Description

technical field [0001] The application relates to an intestinal polyp detection system, in particular to a real-time endoscopic colonoscopy polyp detection system combining single-frame color images and multi-frame optical flow image spatio-temporal information, which belongs to the field of medical technology. Background technique [0002] Colorectal polyps are abnormal growths that start in the lining of the colon or rectum. While most polyps are safe, some carry a risk of becoming cancerous and leading to colorectal cancer. Colorectal cancer is currently the fourth most common cancer with the second highest mortality worldwide. At present, more than half of the people over the age of 60 have more than one colorectal polyp, so the early detection and removal of polyps is very important for the prevention of colorectal cancer. Endoscopic colonoscopy is the most common and effective preventive method to discover and remove polyps, but manual diagnosis is limited by the expe...

AI Technical Summary

Problems solved by technology

[0007] First, it is non-real-time. Due to its complexity, the speed of this method cannot exceed 30 frames per second on existing equipment, so it cannot be used for real-time purposes, and can only be used for postoperative analysis

[0008] Second, low accuracy and sensitivity. Most deep learning models require a large amount of accurately labeled data for training, but the polyp labeling task is limited by medical expertise. Most public data sets have the problem of small numbers and few types. Therefore, the training results are unsatisfactory

At the same time, because detection methods based on single-frame information are often trapped in the complex environment of the intestinal tract, there are a lot of noise such as foreign objects, liquids, and blurs in the field of view of the lens, it is difficult to make an accurate result evaluation based on a single image. There are reasons why the accuracy and sensitivity of the method are not satisfactory

[0009] In order to reduce noise interference, some researchers introduce time information to analyze consecutive multiple frames in the video. At present, there are three methods that have been published in the academic field, namely convolution long short-term memory neural network (ConvLSTM), conditional random field (CRF) Combined with deep learning-based target tracking methods to fuse multi-frame detection results, but these methods are limited by model complexity and cannot achieve real-time detection

Images