Volume 155, Issue 4, Pages 1069-1078.e8 (October 2018) Deep Learning Localizes and Identifies Polyps in Real Time With 96% Accuracy in Screening Colonoscopy Gregor Urban, Priyam Tripathi, Talal Alkayali, Mohit Mittal, Farid Jalali, William Karnes, Pierre Baldi Gastroenterology Volume 155, Issue 4, Pages 1069-1078.e8 (October 2018) DOI: 10.1053/j.gastro.2018.06.037 Copyright © 2018 AGA Institute Terms and Conditions
Gastroenterology 2018 155, 1069-1078. e8DOI: (10. 1053/j. gastro. 2018 Copyright © 2018 AGA Institute Terms and Conditions
Figure 1 Examples of dataset. (Top row) Images containing a polyp with a superimposed bounding box. (Bottom row) Non-polyp images. Three pictures on the left were taken using NBI and 3 pictures on the right include tools (eg, biopsy forceps, cuff devices, etc) that are commonly used in screening colonoscopy procedures. Gastroenterology 2018 155, 1069-1078.e8DOI: (10.1053/j.gastro.2018.06.037) Copyright © 2018 AGA Institute Terms and Conditions
Figure 2 Representative frame shots of CNN-overlaid colonoscopy videos. Presence of a green box indicates that a polyp is detected with greater than 95% confidence by our CNN polyp localization model; the location and size of the box are predictions of the CNN model. Expert confidence that the box contained a true polyp is shown in the upper left of the images (video collages of CNN localization predictions available at: http://www.igb.uci.edu/colonoscopy/AI_for_GI.html). Gastroenterology 2018 155, 1069-1078.e8DOI: (10.1053/j.gastro.2018.06.037) Copyright © 2018 AGA Institute Terms and Conditions
Supplementary Figure 1 Effect of the size of the filtering window on the sensitivity and specificity of CNN predictions using a fixed threshold of 0.4. Gastroenterology 2018 155, 1069-1078.e8DOI: (10.1053/j.gastro.2018.06.037) Copyright © 2018 AGA Institute Terms and Conditions
Supplementary Figure 2 Receiver operator characteristic curve for all 5 CNN architectures trained on subsets of the 8641 colonoscopy images. Results obtained on the test splits of the 8641 colonoscopy images. Gastroenterology 2018 155, 1069-1078.e8DOI: (10.1053/j.gastro.2018.06.037) Copyright © 2018 AGA Institute Terms and Conditions