About 60%of people who died of colon cancer after colonic examination were missed lesions during the screening. To this end, two Japanese researchers spent 8 years to develop artificial intelligence (AI) technology to help endoscopy doctors.
The Director of the Gastrointestinal Division Center of the Showa University and the joint responsibility of the project of the project believe that about 25%of the colorectal tumors are ignored by the experts looking for their experts through micro cameras. However, if you enter the correct data, deep learning algorithms can improve human detection capabilities. Since 2013, Kudo’s team has been working with the team leader of the Director of the Information Technology Center of Nagoya University to develop colonoscopy artificial intelligence technology.
The group designed a computer-aided detection (CADE) system called Endobrain-Eye to reduce the number of negative colorectal tumors, and a computer auxiliary identification (CADX) tool called Endobrain. It provides more Accurate type evaluation.
The basic artificial intelligence technology of these two systems are the same. Cade applies neural networks to the features of the video box colon polyps (abnormal lumps). CADX uses the same feature extraction process to help classify lesions so that more detailed pathological diagnosis in the laboratory.
Both technologies have brought significant improvements to testing. When a large -scale open colonoscopy database containing the five academic centers of Japan, the CADE system is 98%sensitivity and 93.7%specific recognition of colorectal lesions. The adenoma detection rate (ADR) is an established quality indicator of colonoscopy. A META analysis recently published by researchers and their collaborators shows that compared with standard human body screening, ADR increases 50%; every 1%increase in ADR, and colorectal cancer -related deaths decreased by 3%.
Subsidity but important cancer
It is easy to leak flat or depressed tumors (referred to as non -polypylid lesions) in colonoscopy. The signs are very small, which may be just the slight changes in the color of the mucosa and the offset of the subcutaneous capillaries. Kudo explained that no matter what they are, the depression lesions are more likely to be invasive in essence.
The depression and horizontal diffusion tumors are usually the most easily neglected colorectal lesions in colonoscopy. It has malignant potential. In 2017, the accuracy of the early version of CADX was as high as 94%in diagnosis of invasive colorectal cancer. If it is used in conjunction with super -existence observation, CADX is also proved to have a particularly strong ability in the body’s invasive cancer recognition. Surgery colonoscopy can easily describe the characteristics of tumor through the nucleus on the surface of the tumor on the visual body. Narrow -band imaging, which is used in specific blue and green wavelengths to enhance the details of microvascular and sub -blue dyeing, which helps the nucleus of the visual tumor, and also improves the diagnostic accuracy in the superproidered large endoscopic examination. Surgery and dyeing mirror technology also helps artificial intelligence to distinguish tumor and non -tumor polyps, as well as large -scale invasion and other tumor lesions. Therefore, artificial intelligence, combined with amplification and dyeing, can make the endoscopic doctor predict the pathological pathology on the spot.
The extra speed and accuracy provided by artificial intelligence will reduce the cost, time and risk of biopsy and repeated colonoscopy. Accurate diagnosis will also reduce excess treatment.
Research by Kudo and others shows that using CADX in real time in the endoscope can meet the diagnosis of small, non -tumor rectal polyps. A study conducted in 2020 found that CADX may reduce colonoscopy -related costs by 7-20%.
When a suspected colon lesion occurs on the screen, the real -time artificial intelligence system should be immediately notified to the endoscopic doctor immediately, which will provide more time for observation, resection or biopsy. In addition, the research team also found that through the use of real -time three -dimensional reconstruction technology, the blind spots in colonoscopy can be reduced, thereby better monitoring areas that are difficult to visually visual.
Source of the original text