计算机科学与探索 ›› 2024, Vol. 18 ›› Issue (3): 627-645.DOI: 10.3778/j.issn.1673-9418.2310062

• 前沿·综述 • 上一篇    下一篇

卷积神经网络在结直肠息肉辅助诊断中的应用综述

考文涛,李明,马金刚   

  1. 山东中医药大学 智能与信息工程学院,济南 250355
  • 出版日期:2024-03-01 发布日期:2024-03-01

Review of Application of Convolutional Neural Network in Auxiliary Diagnosis of Colorectal Polyps

KAO Wentao, LI Ming, MA Jingang   

  1. College of Intelligence and Information Engineering, Shandong University of Traditional Chinese Medicine, Jinan 250355, China
  • Online:2024-03-01 Published:2024-03-01

摘要: 结直肠癌是一种恶性肿瘤,主要发生在结肠和直肠的组织中,其早期发现和治疗具有重要意义。结直肠癌的早期检测和预防主要是对病人的肠道进行视觉检查,从而筛查结直肠息肉,但人工检查存在漏诊率高等弊端。基于卷积神经网络(CNN)的辅助诊断系统在结直肠息肉的诊断方面表现出最先进的性能,是目前计算机辅助诊断领域的研究热点。根据近几年发表的相关重要文献,对卷积神经网络在结直肠息肉辅助诊断中的应用进行系统综述。首先介绍了结直肠息肉诊断领域的常用数据集,其中包括图片和视频数据集;其次分别对CNN在结直肠息肉检测、分割以及分类中的应用进行系统阐述,对各算法的主要改进思路、优缺点以及性能进行深入分析,旨在为研究人员提供更系统的参考,并对深度学习模型的可解释性进行总结;最后对基于CNN的结直肠息肉辅助诊断的各类算法进行总结,并对未来的研究方向进行展望。

关键词: 结直肠息肉, 卷积神经网络(CNN), 计算机辅助诊断, 可解释性

Abstract: Colorectal cancer is a malignant tumor that mainly occurs in the tissues of the colon and rectum, and its early detection and treatment are of great significance. The early detection and prevention of colorectal cancer mainly involves visual examination of the patient??s intestines to screen for colorectal polyps, but manual examination has the disadvantage of high misdiagnosis rate. The auxiliary diagnostic system based on convolutional neural networks (CNN) has shown the most advanced performance in the diagnosis of colorectal polyps, and is currently a research hotspot in the field of computer-aided diagnosis. Based on important literature published in recent years, a systematic review of the application of convolutional neural networks in the auxiliary diagnosis of colorectal polyps is conducted. Firstly, the commonly used datasets in the field of colorectal polyp diagnosis are introduced, including image and video datasets. Secondly, the application of CNN in colorectal polyp detection, segmentation, and classification is systematically elaborated. The main improvement ideas, advantages and disadvantages, and performance of each algorithm are analyzed in depth, aiming to provide researchers with a more systematic reference, and summarize the interpretability of deep learning models. Finally, a summary of various algorithms for assisting the diagnosis of colorectal polyps based on CNN is provided, and future research directions are prospected.

Key words: colorectal polyps, convolutional neural networks (CNN), computer aided diagnosis, interpretability