Abstract:

Recently, the success of deep learning, especially deep convolution neural network (CNN), has promoted the development of polyp recognition and segmentation based on image and video. At present, most diagnostic colonoscopy rooms use artificial intelligence methods. Through extensive research and testing, the feasibility of classifying and monitoring colorectal lesions in colonoscopy has been verified. This paper reviews the application of CNN in the characteristics of tumors or precancerous lesions, automatic anatomical classification of colonoscopy images, variable intestinal preparation / colonoscopy automatic quality control system, as well as the shortcomings and future prospects of CNN in colonoscopy diagnosis.

Key words:

Artificial intelligence, Deep convolution neural network, Colonoscopy, Polyp, Adenoma, Detection rate of adenoma

摘要：

最近，深度学习特别是深度卷积神经网络（convolution neural network，CNN）的成功推动了基于图像和视频的息肉识别和分割的发展。目前，大多数诊断性结肠镜检查室使用人工智能方法。现通过广泛研究及测试，验证了其在结肠镜检查中分类并监测结直肠病变的可行性。本文对关于国内外CNN在肿瘤或癌前病变特征、对结肠镜图像进行自动解剖学分类、可变肠道准备/结肠镜检查自动质量控制系统等方面的应用以及CNN在结肠镜诊断中存在的不足及未来发展前景研究进展进行综述。

关键词:

人工智能, 深度卷积神经网络, 结肠镜检查, 息肉, 腺瘤, 腺瘤检出率