Department of Medical Education, King Saud University, College of Medicine, Riyadh 11461, Saudi Arabia.
Medicina (Kaunas). 2019 Aug 12;55(8):473. doi: 10.3390/medicina55080473.
Colorectal cancer (CRC) is one of the most common causes of cancer mortality in the world. The incidence is related to increases with age and western dietary habits. Early detection through screening by colonoscopy has been proven to effectively reduce disease-related mortality. Currently, it is generally accepted that most colorectal cancers originate from adenomas. This is known as the "adenoma-carcinoma sequence", and several studies have shown that early detection and removal of adenomas can effectively prevent the development of colorectal cancer. The other two pathways for CRC development are the Lynch syndrome pathway and the sessile serrated pathway. The adenoma detection rate is an established indicator of a colonoscopy's quality. A 1% increase in the adenoma detection rate has been associated with a 3% decrease in interval CRC incidence. However, several factors may affect the adenoma detection rate during a colonoscopy, and techniques to address these factors have been thoroughly discussed in the literature. Interestingly, despite the use of these techniques in colonoscopy training programs and the introduction of quality measures in colonoscopy, the adenoma detection rate varies widely. Considering these limitations, initiatives that use deep learning, particularly convolutional neural networks (CNNs), to detect cancerous lesions and colonic polyps have been introduced. The CNN architecture seems to offer several advantages in this field, including polyp classification, detection, and segmentation, polyp tracking, and an increase in the rate of accurate diagnosis. Given the challenges in the detection of colon cancer affecting the ascending (proximal) colon, which is more common in women aged over 65 years old and is responsible for the higher mortality of these patients, one of the questions that remains to be answered is whether CNNs can help to maximize the CRC detection rate in proximal versus distal colon in relation to a gender distribution. This review discusses the current challenges facing CRC screening and training programs, quality measures in colonoscopy, and the role of CNNs in increasing the detection rate of colonic polyps and early cancerous lesions.
结直肠癌（CRC）是全球癌症死亡率最高的原因之一。发病率与年龄增长和西方饮食习惯有关。通过结肠镜检查进行早期筛查已被证明能有效降低与疾病相关的死亡率。目前，人们普遍认为大多数结直肠癌起源于腺瘤。这被称为“腺瘤-癌序列”，几项研究表明，早期发现和切除腺瘤可以有效预防结直肠癌的发展。CRC 发展的另外两个途径是林奇综合征途径和无蒂锯齿状途径。腺瘤检出率是结肠镜检查质量的既定指标。腺瘤检出率提高 1%，结直肠癌间隔期的发生率就会降低 3%。然而，在结肠镜检查过程中，有几个因素可能会影响腺瘤的检出率，文献中也对解决这些因素的技术进行了深入讨论。有趣的是，尽管在结肠镜检查培训计划中使用了这些技术，并引入了结肠镜检查质量措施，但腺瘤的检出率差异很大。考虑到这些局限性，使用深度学习，特别是卷积神经网络（CNN）来检测癌性病变和结肠息肉的举措已经出现。CNN 架构在这个领域似乎有几个优势，包括息肉分类、检测和分割、息肉跟踪，以及提高准确诊断率。鉴于影响升结肠（近端）的结肠癌检测的挑战，升结肠癌在 65 岁以上的女性中更为常见，也是这些患者死亡率较高的原因之一，因此仍有一个问题有待回答，即 CNN 是否有助于最大限度地提高与性别分布相关的近端与远端结肠的 CRC 检出率。本文综述了 CRC 筛查和培训计划、结肠镜检查中的质量措施以及 CNN 提高结肠息肉和早期癌性病变检出率的作用所面临的当前挑战。
