Artificial Intelligence and Colonoscopy

Colonoscopy is an effective screening procedure for the diagnosis and prevention of colorectalcancer; however, colonoscopy can be challenging based on the lesion characteristics including itsdetection, classification, and removal. Endoscopists try to carefully examine the mucosal layer ofthe colon folds by flexing and torquing the shaft of the endoscope while withdrawing the instrument.To avoid missing any abnormalities, tips of the endoscopes are twisted in longitudinal oraxial views within the neutral, straight position. Simultaneously, lesions that are presumed to bepremalignant are removed. Moreover, recent studies showed that in ۱۴% of colonoscopy procedures,precursor lesions are not completely removed. All of the procedures from the detection ofthe lesion to its removal can be simple or complicated. That’s why colonoscopy is considered tobe an operator-dependent procedure. In an attempt to cover all of these shortcomings, researcheshave been made to integrate artificial intelligence into colonoscopy procedures. Artificial intelligence(AI) based systems have shown promise to increase adenoma detection rate (ADR) inrandomized clinical trials. By previously developed algorithms, AI can offer real-time supportto clinicians and automatically recognize polyps together with providing the probable histologyof the sample. Computer-assisted colonoscopy is recently implemented in two clinical areas ofComputer Assisted detection (CADe) and classification (CADx).Up to now, this procedure is restricted to non-advanced adenomas. The efficacy of AI-based systemsto detect flat lesions or advanced neoplasia is still in question because of the low prevalenceof subtle abnormalities or sessile serrated lesions (SSLs).We are currently gathering different datasets, consisting of colonoscopy images and videos, andhoping to reach ۲۰,۰۰۰ polyp-positive frames and ۵,۰۰۰ polyp-negative frames to develop a convolutionalneural network (CNN). All polyps are confirmed by histopathology. Polyp size, morphology,and location were also recorded. Full-length videos (white light only) were divided intoshorter polyp-positive and -negative sequences. We aim to detect subtle abnormalities and sessileserrated lesions (SSLs) in Iranian patients.

کلیدواژه ها:

artificial intelligence ، polyp ، colonoscopy ، adenoma detection rate ، convolutional neural network