Real-time AI-assisted Endocyroscopy for the Diagnosis of Colorectal Lesions

Colonoscopy is currently the gold standard of screening for CRC. Endocytoscopy is a kind of ultra-high magnification endoscopy. Combined with chemical staining and narrow band imaging technology, endoscopists can observe the cell nucleus morphology, gland tube morphology and microvascular morphology with the naked eye, so as to avoid pathological examination and realize the purpose of real-time biopsy in the body. However, the judgment of endocytoscopic images needs a lot of experience to improve the judgment accuracy. Moreover, endoscopists have certain subjective judgments and errors in the process of judging the results. Therefore, artificial intelligence (AI) is proposed for computer-assisted diagnosis in clinic to solve this problem. At present, the available artificial intelligence systems for assisting endoscopists in diagnosing using endocytoscopy are still relatively scarce, and they are based on traditional machine learning methods, which still have certain limitations in terms of accuracy. With the continuous development of computer vision technology, deep learning has been widely applied in the development of endoscopic assistance diagnosis systems. Therefore, the investigators developed a CADx system trained using deep learning to assist endoscopists in making diagnoses when using endocytoscopy. This CADx system can predict the captured endocytoscopy images in real time and display the prediction results, which can assist endoscopists in providing diagnostic references.

However, currently this CADx technology has not yet undergone prospective clinical validation in the clinical setting. The investigators plan to conduct a prospective clinical trial to validate the accuracy of CADx for prediction of colorectal lesions histology in real-time endocytoscopy. This study will prospectively collect the lesions that meet the inclusion and exclusion criteria. After the endoscopists make the diagnosis through endoscopic images and CADx and then undergo endoscopic resection or surgical resection followed by pathological diagnosis, they will compare the artificial intelligence diagnosis results with the gold standard pathological results, and summarize the diagnostic accuracy of this artificial intelligence-assisted diagnostic system for the colorectal lesions.

The investigators plan to conduct a prospective, multi-centre clinical trial to validate the accuracy of CADx support for prediction of colorectal lesions histology in real-time endocytoscopy.

Inclusion criteria

• Patients who have at least one colorectal lesion detected during endocytoscopy
• Consent obtained for the study

Exclusion criteria

• lesions lacking high-quality images;
• Inflammatory bowel disease, familial adenomatous polyposis and other special diseases;
• Submucosal tumors;
• Pathological diagnosis of inflammatory polyps, Peutz-Jeghers polyps, juvenile polyps, lymphoma and other special pathological types.