上海交通大学学报（医学版）

上海交通大学学报（医学版） >

2022 , Vol. 42 >Issue 2: 205 - 210

DOI: https://doi.org/10.3969/j.issn.1674-8115.2022.02.011

论著 · 技术与方法

深度卷积神经网络模型辅助下结直肠息肉检测系统对初级医师结直肠小息肉检出率的影响

  • 王晓峰 ,
  • 周璐 ,
  • 姚乐宇 ,
  • 何凡 ,
  • 彭海霞 ,
  • 杨大明 ,
  • 黄晓霖
展开
  • 1.上海交通大学医学院附属同仁医院内镜中心，上海 200336
    2.上海交通大学电子信息与电气工程学院，上海 200240
    3.上海交通大学医疗机器人研究院，上海 200240
王晓峰（1990—），男，住院医师，硕士；电子信箱：wxf3908@shtrhospital.com
王晓峰（1990—），男，住院医师，硕士；电子信箱：wxf3908@shtrhospital.com
杨大明，电子信箱：ydm1100@shtrhospital.com。
黄晓霖，电子信箱：xiaolinhuang@sjtu.edu.cn

收稿日期: 2021-11-01

  网络出版日期: 2022-03-17

基金资助

国家自然科学基金(61977046);上海市科技计划项目(2021SHZDZX0102);上海市科学技术委员会基金(21ZR1458600);上海市长宁区卫生健康委员会项目(20214T001);上海交通大学“交大之星”计划医工交叉研究基金(YG2022ZD031)

收起

Effect of DCNN model-assisted colorectal polyp detection system on the detection of colorectal polyps by junior physicians

  • Xiaofeng WANG ,
  • Lu ZHOU ,
  • Leyu YAO ,
  • Fan HE ,
  • Haixia PENG ,
  • Daming YANG ,
  • Xiaolin HUANG
Expand
  • 1.Digestive Endoscopy Center, Tongren Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai 200336, China
    2.School of Electronics Information and Electrical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China
    3.Institute of Medical Robotics, Shanghai Jiao Tong University, Shanghai 200240, China
YANG Daming, E-mail: ydm1100@shtrhospital.com.
HUANG Xiaolin, E-mail: xiaolinhuang@sjtu.edu.cn

Received date: 2021-11-01

  Online published: 2022-03-17

Supported by

National Natural Science Foundation of China(61977046);Shanghai Science and Technology Planning Project(2021SHZDZX0102);Fund of Science and Technology Commission of Shanghai Municipality(21ZR1458600);Project of Health Commission of Changning District, Shanghai(20214T001);Interdisciplinary Program of Shanghai Jiao Tong University(YG2022ZD031)

Fold

摘要

目的·探究基于人工智能（artificial intelligence，AI）技术中深度卷积神经网络（deep convolutional neural network，DCNN）模型构建的计算机辅助检测（computer-aided detection，CADe）系统对缺乏电子结肠镜操作经验的初级医师结直肠息肉检出率的影响。方法·选取上海交通大学医学院附属同仁医院内镜中心数据库2019年1月—2020年12月的结肠镜图像及2021年1月—3月的结肠镜视频。将筛选出的图像和视频分为数据集1（5 908张图像）和数据集6（360条短视频），数据集1分为数据集1a（4 906张图像）、数据集1b（300张图像）和数据集1c（702张图像）；其中，数据集1c为从视频数据集6中截取的图像。数据集2~5分别为公共数据集CVC-ClinicDB、CVC-ColonDB、ETIS-Larib Polyp DB和KVASIR，共包含2 188张图片。数据集1a和数据集2~5为模型训练集，数据集1b和数据集1c为模型测试集。将10名经过培训且无结肠镜操作经验的初级医师随机分为AI辅助组（A组，n=5）及无AI辅助组（B组，n=5）。2组医师分别对360条结肠镜视频进行判读。前180条视频两组均无AI辅助。后180条视频中，AI辅助组辅以息肉检测系统，将视频数据集经检测系统处理后标记息肉，供研究者判读；无AI辅助组则观看原始视频，判读是否存在息肉。所有视频先由2位高年资内镜医师确认是否存在息肉；若存在争议，则剔除该视频，并以这2位医师共同确认的诊断结果为金标准。视频中存在息肉，受试者未能检出，视为漏诊；视频中无息肉，受试者判断存在息肉，视为误诊。结果·前180条视频均无AI辅助时，A组与B组结直肠息肉漏诊例数比较，差异无统计学意义；后180条视频中，A组息肉检出漏诊例数明显小于B组（P=0.031）；在A组内比较，后180条视频中的息肉漏诊例数小于前180条视频，差异具有统计学意义（P=0.007）。2组间及各自组内的息肉误诊例数比较，差异均无统计学意义。结论·该研究所构建的DCNN模型辅助下的结直肠息肉检测模型可以明显改善缺乏结肠镜操作经验医师的结直肠息肉检出率，同时不会增加结直肠息肉的误诊。

关键词： 人工智能; 深度卷积神经网络; 计算机辅助检测; 结直肠息肉

本文引用格式

王晓峰 , 周璐 , 姚乐宇 , 何凡 , 彭海霞 , 杨大明 , 黄晓霖 . 深度卷积神经网络模型辅助下结直肠息肉检测系统对初级医师结直肠小息肉检出率的影响[J]. 上海交通大学学报（医学版）, 2022 , 42(2) : 205 -210 . DOI: 10.3969/j.issn.1674-8115.2022.02.011

Abstract

Objective

·To explore the effect of a computer-aided detection (CADe) system constructed by a deep convolutional neural network (DCNN) model in artificial intelligence (AI) technology on the detection rate of colorectal polyps among the junior physicians lacking of colonoscopy operation experience.

Methods

·The colonoscopy images from January 2019 to December 2020 and colonoscopy videos from January 2021 to March 2021 from the Endoscopy Center Database of Digestive Endoscopy Center, Tongren Hospital, Shanghai Jiao Tong University School of Medicine were collected. The collected images and videos were divided into dataset 1 (5 908 images) and dataset 6 (360 short videos). Dataset 1 was divided into dataset 1a (4 906 images), dataset 1b (300 images) and dataset 1c (702 images), and dataset 1c was the intercepted images from video dataset 6. Datasets 2 to 5 were the public datasets CVC-ClinicDB, CVC-ColonDB, ETIS-Larib Polyp DB and KVASIR, containing a total of 2 188 images. Dataset 1a and datasets 2?5 were the model training sets, and dataset 1b and dataset 1c were the model testing sets. Ten trained junior physicians with no experience in colonoscopy were randomized into AI-assisted group (Group A, n=5) and non-AI-assisted group (Group B, n=5) to interpret 360 colonoscopy videos and determine the presence of polyps in the videos, respectively. For the first 180 videos, both groups had no AI assistance. For the second 180 videos, the AI-assisted group was supplemented with a DCNN model-assisted colorectal polyp detection system, which processed the video data set through the detection system to mark polyps for the investigator's interpretation, while the non-AI-assisted group watched the original video to interpret the presence of polyps. All the videos were first confirmed by 2 senior endoscopists for the presence of polyps. If there was a dispute, the video was excluded and the diagnosis confirmed by these 2 physicians together was the gold standard. If a polyp was present in the video and the investigator failed to detect it, the diagnosis was considered missed; if no polyp was present in the video and the investigator judged that a polyp was present, the diagnosis was considered misdiagnosed.

Results

·In the first 180 videos without AI assistance, there was no significant difference in the number of missed colon polyps between group A and group B. In the second 180 videos, the number of missed polyps in group A with AI assistance was significantly less than that in group B without AI assistance (P=0.031). Meanwhile, the number of missed polyps in the second 180 videos in group A was less than that in the first 180 videos, and the difference was also statistically significant (P=0.007). In addition, there was no significant difference in the number of misdiagnosed polyps detected between and within the groups.

Conclusion

·The DCNN model-assisted colorectal polyp detection system can significantly improve the polyps detection rate by physicians lacking of colonoscopy operation experience without increasing misdiagnosis of colorectal polyps.

Key words： artificial intelligence (AI); deep convolutional neural network (DCNN); computer-aided detection (CADe); colorectal polyp

参考文献

1 SUNG H, FERLAY J, SIEGEL R L, et al. Global cancer statistics 2020: globocan estimates of incidence and mortality worldwide for 36 cancers in 185 countries[J]. CA Cancer J Clin, 2021, 71(3): 209-249.
2 CORLEY D A, JENSEN C D, MARKS A R, et al. Adenoma detection rate and risk of colorectal cancer and death[J]. N Engl J Med, 2014, 370(14): 1298-1306.
3 KAMINSKI M F, REGULA J, KRASZEWSKA E, et al. Quality indicators for colonoscopy and the risk of interval cancer[J]. N Engl J Med, 2010, 362(19): 1795-1803.
4 AHN S B, HAN D S, BAE J H, et al. The miss rate for colorectal adenoma determined by quality-adjusted, back-to-back colonoscopies[J]. Gut Liver, 2012, 6(1): 64-70.
5 LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436-444.
6 KARKANIS S A, IAKOVIDIS D K, MAROULIS D E, et al. Computer-aided tumor detection in endoscopic video using color wavelet features[J]. IEEE Trans Inf Technol Biomed, 2003, 7(3): 141-152.
7 GONG D X, WU L L, ZHANG J, et al. Detection of colorectal adenomas with a real-time computer-aided system (ENDOANGEL): a randomised controlled study[J]. Lancet Gastroenterol Hepatol, 2020, 5(4): 352-361.
8 WANG P, BERZIN T M, GLISSEN BROWN J R, et al. Real-time automatic detection system increases colonoscopic polyp and adenoma detection rates: a prospective randomised controlled study[J]. Gut, 2019, 68(10): 1813-1819.
9 SU J R, LI Z, SHAO X J, et al. Impact of a real-time automatic quality control system on colorectal polyp and adenoma detection: a prospective randomized controlled study (with videos)[J]. Gastrointest Endosc, 2020, 91(2): 415-424.e4.
10 BERNAL J, SáNCHEZ F J, FERNáNDEZ-ESPARRACH G, et al. WM-DOVA maps for accurate polyp highlighting in colonoscopy: validation vs. saliency maps from physicians[J]. Comput Med Imaging Graph, 2015, 43: 99-111.
11 FERNáNDEZ-ESPARRACH G, BERNAL J, LóPEZ-CERóN M, et al. Exploring the clinical potential of an automatic colonic polyp detection method based on the creation of energy maps[J]. Endoscopy, 2016, 48(9): 837-842.
12 SILVA J, HISTACE A, ROMAIN O, et al. Toward embedded detection of polyps in WCE images for early diagnosis of colorectal cancer[J]. Int J Comput Assist Radiol Surg, 2014, 9(2): 283-293.
13 JHA D, SMEDSRUD P H, RIEGLER M A, et al. Kvasir-seg: a segmented polyp dataset[C]//26th International Conference, MMM 2020, Daejeon, South Korea, January 5-8, 2020, Proceedings, Part Ⅱ. Cham: Springer, 2020: 451-462.
14 REDMON J, FARHADI A. YOLOv3: an incremental improvement[Z/OL]. (2018-04-08)[2021-09-10]. https://arxiv.org/pdf/1804.02767v1.pdf
15 DENG J, DONG W, SOCHER R, et al. ImageNet: a large-scale hierarchical image database[C]//Proceedings of IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Miami: IEEE, 2009: 248-255.
16 KINGMA D P, BA J. Adam: a method for stochastic optimization[Z/OL]. (2017-01-30)[2021-09-10]. https//arxiv.org/pdf/1412.6298.pdf.
17 REX D K, KAHI C, O'BRIEN M, et al. The American Society for Gastrointestinal Endoscopy PIVI (Preservation and Incorporation of Valuable Endoscopic Innovations) on real-time endoscopic assessment of the histology of diminutive colorectal polyps[J]. Gastrointest Endosc, 2011, 73(3): 419-422.
18 FUKUTOMI H. Endoscopic diagnosis of protruding lesions of the stomach[J]. Iryo, 1967, 21(8):940-946.
19 KAMINSKI M F, ANDERSON J, VALORI R, et al. Leadership training to improve adenoma detection rate in screening colonoscopy: a randomised trial[J]. Gut, 2016, 65(4): 616-624.
20 WANG W H, XU L, BAO Z F, et al. Differences with experienced nurse assistance during colonoscopy in detecting polyp and adenoma: a randomized clinical trial[J]. Int J Colorectal Dis, 2018, 33(5): 561-566.
21 LIU A H, WANG H S, LIN Y J, et al. Gastrointestinal endoscopy nurse assistance during colonoscopy and polyp detection: a PRISMA-compliant meta-analysis of randomized control trials[J]. Medicine (Baltimore), 2020, 99(34): e21278.
22 REPICI A, HASSAN C. Artificial intelligence for colonoscopy: the new Silk Road[J]. Endoscopy, 2021, 53(3): 285-287.
23 ALI S, DMITRIEVA M, GHATWARY N, et al. Deep learning for detection and segmentation of artefact and disease instances in gastrointestinal endoscopy[J]. Med Image Anal, 2021, 70: 102002.
24 BERZIN T M, TOPOL E J. Adding artificial intelligence to gastrointestinal endoscopy[J]. Lancet, 2020, 395(10223): 485.
25 MISAWA M, KUDO S E, MORI Y, et al. Current status and future perspective on artificial intelligence for lower endoscopy[J]. Dig Endosc, 2021, 33(2): 273-284.
26 LI J W, CHIA T, FOCK K M, et al. Artificial intelligence and polyp detection in colonoscopy: use of a single neural network to achieve rapid polyp localization for clinical use[J]. J Gastroenterol Hepatol, 2021, 36(12): 3298-3307.
27 DURAK S, BAYRAM B, BAKIRMAN T, et al. Deep neural network approaches for detecting gastric polyps in endoscopic images[J]. Med Biol Eng Comput, 2021, 59(7/8): 1563-1574.
28 KOGA S, IKEDA A, DICKSON D W. Deep learning-based model for diagnosing Alzheimer's disease and tauopathies[J]. Neuropathol Appl Neurobiol, 2022, 48(1): e12759.
29 KALAGER M, WIESZCZY P, LANSDORP-VOGELAAR I, et al. Overdiagnosis in colorectal cancer screening: time to acknowledge a blind spot[J]. Gastroenterology, 2018, 155(3): 592-595.
30 PAGGI S, RADAELLI F, REPICI A, et al. Advances in the removal of diminutive colorectal polyps[J]. Expert Rev Gastroenterol Hepatol, 2015, 9(2): 237-244.
31 SCHOEFL R, ZIACHEHABI A, WEWALKA F. Small colorectal polyps[J]. Dig Dis, 2014, 33: 38-41.
32 VON RENTELN D, BARKUN A N. Increasing detection rates for diminutive adenomas: are we on the right track? [J]. Gut, 2016, 65(6): 1056-1057.
Options
文章导航

/

〈 〉