Article Sidebar

PDF
Date Published: 12/01/2022
Abstract Views: 415
Views PDF: 298
DOI: https://doi.org/10.51298/vmj.v509i1.1763
Issue
Vol. 509 No. 1 (2021)
Section
Các bài báo
How to Cite
Việt Hằng, Đào ., Thanh Tùng, N., Ngọc Hoa, L. ., Phúc Bình, N. ., Văn Long, Đào ., Thị Thủy, N., & Viết Sang, Đinh . (2022). EVALUATING THEACCURACY OF EFFICIENTNET IN COLON POLYP DETECTION AND DELINEATION. Vietnam Medical Journal, 509(1). https://doi.org/10.51298/vmj.v509i1.1763

EVALUATING THEACCURACY OF EFFICIENTNET IN COLON POLYP DETECTION AND DELINEATION

Đào Việt Hằng1,2,3,, Nguyễn Thanh Tùng1, Lâm Ngọc Hoa1, Nguyễn Phúc Bình1, Đào Văn Long1,2,3, Nguyễn Thị Thủy4, Đinh Viết Sang5
1 The Institute of Gastroenterology and Hepatology
2 Department of Internal Medicine, Hanoi Medical University
3 Endoscopy center, Hanoi Medical University Hospital
4 Vietnam National University of Agriculture
5 Hanoi University of Science and Technology

Main Article Content

Abstract

Objective: to evaluate the accuracy of EfficientNet algorithm in detecting colon polyps and to determine factors associated with the rate of missed polyp and false detection. Methods: Cross-sectional study. EfficientNet algorithm was validated on a set of 4000 still image (2000 images containing 2111 polyps, 2000 images with no polyp) by comparing with the ground-truth delineated by experts. Accuracy was assessed by sensitivity (Se), specificity (Sp), positivepredictive value (PPV) and negative predictive value (NPV). Regression models were used to determine factors related to the rate ofmissed polyp and false detection. Results: Se, PPV, Sp, NPV and accuracy were 97.60%, 94.44%, 94.25%, 97.52% and 95.93%, respectively. Multivariate regression analysis showed that cleanliness, polyp’s sizeand numberof polyps on image were significantly associated with the missedrate; cleanlinessand diagnosis were related to the false detection rate. Conclusion: EfficientNet algorithm had high accuracy, can be further developed using big data to support endoscopists during endoscopy or improve endoscopy and medicaltraining.

Article Details

Keywords

Artificial intelligent, EfficientNet, colonoscopy, polyp detection, polyp localization

References

1. Đào Việt Hằng và cộng sự. (2020).Kết quả bước đầu của ứng dụng trí tuệ nhân tạo trong phát hiện polyp đại tràng tại Việt Nam.Tạp chí Nghiên cứu Y học,130 (6), 101-110.
2. International Agency for Research on Cancer (2020). The Globocan Cancer Obseratory. Country profile: Vietnam.
3. Lai E.J., Calderwood A.H., Doros G. và cộng sự. (2009). The Boston Bowel Preparation Scale: A valid and reliable instrument for colonoscopy-oriented research. Gastrointest Endosc, 69(3 Pt 2), 620–625.
4. Lui T.K.L., Hui C.K.Y. và cộng sự. (2021). New insights on missed colonic lesions during colonoscopy through artificial intelligence-assisted real-time detection (with video). Gastrointest Endosc, 93(1), 193-200.
5. Ozawa T., Ishihara S. và cộng sự. (2020). Automated endoscopic detection and classification of colorectal polyps using convolutional neural networks. Therap Adv Gastroenterol,13, 1756284820910659.
6. Seo J.H., Lee B.-I. và cộng sự. (2020). Adenoma miss rate of polypectomy-referring hospitals is high in Korea. Korean J Intern Med, 35(4), 881–888.
7. Valarini S.B.M.và cộng sự. (2011). Correlation between location, size and histologic type of colorectal polyps at the presence of dysplasia and adenocarcinoma. Journal of Coloproctology (Rio de Janeiro), 31(3), 241–247.
8. Wang P., Xiao X., Glissen Brown J.R. và cộng sự. (2018). Development and validation of a deep-learning algorithm for the detection of polyps during colonoscopy. Nat Biomed Eng,2(10), 741–748.