Randomised controlled trial of WISENSE, a real-time quality improving system for monitoring blind spots during esophagogastroduodenoscopy

Abstract

Objective: Esophagogastroduodenoscopy (EGD) is the pivotal procedure in the diagnosis of upper gastrointestinal lesions. However, there are significant variations in EGD performance among endoscopists, impairing the discovery rate of gastric cancers and precursor lesions. The aim of this study was to construct a real-time quality improving system, WISENSE, to monitor blind spots, time the procedure and automatically generate photodocumentation during EGD and thus raise the quality of everyday endoscopy.

Design: WISENSE system was developed using the methods of deep convolutional neural networks and deep reinforcement learning. Patients referred because of health examination, symptoms, surveillance were recruited from Renmin hospital of Wuhan University. Enrolled patients were randomly assigned to groups that underwent EGD with or without the assistance of WISENSE. The primary end point was to ascertain if there was a difference in the rate of blind spots between WISENSE-assisted group and the control group.

Results: WISENSE monitored blind spots with an accuracy of 90.40% in real EGD videos. A total of 324 patients were recruited and randomised. 153 and 150 patients were analysed in the WISENSE and control group, respectively. Blind spot rate was lower in WISENSE group compared with the control (5.86% vs 22.46%, p<0.001), and the mean difference was -15.39% (95% CI -19.23 to -11.54). There was no significant adverse event.

Conclusions: WISENSE significantly reduced blind spot rate of EGD procedure and could be used to improve the quality of everyday endoscopy.

Trial registration number: ChiCTR1800014809; Results.

Keywords: endoscopy.

Conflict of interest statement

Competing interests: None declared.

© Author(s) (or their employer(s)) 2019. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.

Figures

Figure 1

Representative images predicted by the WISENSE in classifying gastric images into 26 sites or NA. The displays showed the gastric sites determined by the WISENSE and the prediction confidence. Class 0, NA, images that could not be classified in any site due to the absence of anatomical landmarks. (1) oesophagus; (2) squamocolumnar junction; (3–6) antrum (G, P, A, L); (7) duodenal bulb; (8) duodenal descending; (9–12) lower body (G, P, A, L); (13–16) middle-upper body in forward view (G, P, A, L); (17–20) fundus (G, P, A, L); (21–23) middle-upper body in retroflex view (P, A, L); (24–26) angulus (P, A, L). A, anterior wall; G, greater curvature; L, lesser curvature; P, posterior wall.

Figure 2

A diagram of the DRL model. DRL makes an action (at) based on the state (st) in environment, lighting a site of 1–26 or do nothing (action is 0) and get a reward (positive score) for a correct action. Labels and confidences of images are projected into a 10×26 grid into a state that can be input to the DRL. Numbers in the abscissa of the matrix represents 26 gastric sites or NA, and the ordinate represents when frames appear. Small cubes in the nine rows from top to bottom represent EGD frames appeared in different times, with their respective positions in abscissa showing their sites predicted by DCNN. The colour shade of cubes represents the confidence of the DCNN’s prediction (the whiter, the higher). The cube representing the first frame appears at the top of the matrix when a video is played, and the previous cube moves down and the next cube appears at the top when the second frame comes. Cubes keep falling down from top to bottom, and for a while, we could see nine cubes dynamically displayed in the matrix until the end of the video, showing predictions and confidences of DCNN on nine consecutive frames. Grey cubes in the bottom row of the matrix show sites that identified to be observed by DRL. DCNN, deep convolutional neural networks; DRL, deep reinforcement learning; EGD, esophagogastroduodenoscopy.

Figure 3

A schematic illustration of how the WISENSE obtains photodocumentation during EGD procedure. (A) For obtaining accurate photodocumentation, WISENSE first filtered unqualified images and then extracted the most representative frame in each site during the process of EGD. (B) A representative photodocumentation generated by WISENSE. A, anterior wall; G, greater curvature; F, forward view; L, lesser curvature; P, posterior wall; R, retroflex view. EGD, esophagogastroduodenoscopy.

Figure 4

Real-time use of WISENSE with an endocytoscope during esophagogastroduodenoscopy. The computer on which WISENSE is installed was directly connected to an endoscopy unit (Evis Lucera Elite CV 290, Olympus) and placed side by side with the original screen, achieving real-time monitoring blind spots during esophagogastroduodenoscopy.

Figure 5

The accuracy, sensitivity and specificity of WISENSE for monitoring blind spots in real EGD videos. In 107 real EGD videos, WISENSE monitored blind spots with an average accuracy of 90.02%, and a separate accuracy for each site ranging from 70.21% to 100%. The average sensitivity and specificity were of 87.57% and 95.02%, ranging from 63.4% to 100% and 75% to 100%, respectively. All EGD videos contain the oesophagus and duodenum; therefore, the negative value of oesophagus and duodenum was zero and specificity of the two sites was unavailable. True positive, WISENSE lights up site A in the stomach model when endoscopists also label site A; true negative, WISENSE leaves site B in transparent in the stomach model and site B is also not labelled by endoscopists. The number of videos containing site C is the ‘positive’ value of site C, and the number of videos missing site D is the ‘negative’ value of site D. Acccuracy=true predictions/(positive+negative), sensitivity=true positive/positive, specificity=true negative/negative. EGD, esophagogastroduodenoscopy.

Figure 6

Trial flow diagram.