基于Vision Transformer的肠镜图像识别模型在结肠疾病中的诊断作用研究OA

Objective To investigate the diagnostic role of a Vision Transformer(ViT)-based artificial intelligence(AI)system in colonic diseases by analyzing clinical endoscopic imaging data.Methods A total of 3000 standard white-light colonoscopy images from 1082 patients with histologically confirmed colonic diseases(including colon polyps,colitis,and colon cancer)were retrospectively col-lected from the Endoscopy Center database of Hebei Provincial Hospital of Traditional Chinese Medicine.The processed datasets of these three disease types were divided in a ratio of 7:2:1.For each disease type,70%of the images were randomly selected as the train-ing set,20%as the test set,and 10%as the validation set.Finally,the ViT model was used to identify and classify the images.Results In the test set,the classification accuracy for colonoscopy images was 99.61%for colon polyps,99.67%for colitis,and 100.00%for colon cancer.Conclusion ViT demonstrated high diagnostic accuracy in detecting colonic diseases.This model can assist primary hospi-tals in enhancing the diagnostic accuracy of colonic diseases and help primary endoscopists enhance their ability to identify colonic dis-eases,thus demonstrating relatively reliable clinical application value.