Artificial Intelligence Model for Analyzing Colonic Endoscopy Images to Detect Changes Associated with Irritable Bowel Syndrome

Background/AimsIBS is not an organic disease, and the patients typically show no abnormalities in lower gastrointestinal endoscopy. Recently, biofilm formation has been visualized by endoscopy, and the ability of endoscopy to detect microscopic changes due to dysbiosis and microinflammation has been reported. In this study, we investigated whether an Artificial Intelligence (AI) colon image model can detect IBS without biofilsm.MethodsStudy subjects were identified based on electronic medical records and categorized as IBS (group I; n=11), IBS with predominant constipation (IBS-C; group C; n=12), and IBS with predominant diarrhea (IBS-D; group D; n=12). Colonoscopy images from IBS patients and from asymptomatic healthy subjects (group N; n=88) were obtained. Google Cloud Platform AutoML Vision (single-label classification) was used to construct AI image models to calculate sensitivity, specificity, predictive value, and AUC. A total of 2479, 382, 538, and 484 images were randomly selected for groups N, IBS, IBC-C and IBS-D groups, respectively.ResultsThe AUC of the model discriminating between group N and group I was 0.95. Sensitivity, specificity, positive predictive value, and negative predictive value of group I detection were 30.8%, 97.6%, 66.7%, and 90.2%, respectively. The overall AUC of the model discriminating between groups N, C, and D was 0.83; sensitivity, specificity, and positive predictive value of group N were 87.5%, 46.2%, and 79.9%, respectively.ConclusionsUsing the image AI model, colonoscopy images of IBS could be discriminated from healthy subjects at AUC 0.95. Prospective studies are needed to further validate whether this externally validated model has similar diagnostic capabilities at other facilities and whether it can be used to determine treatment efficacy.


36
This work is supported by operating funds from the Mathematical, Data Science and AI 37 Education Program, University of Toyama.

Results
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 10, 2022. building of AI models without such programming expertise, and thus the application of 98 AI models in the medical field is likely to expand [6]. In fact, use of AI in the   The purpose of this study was to determine whether image AI models can differentiate 107 between different types of IBS and healthy colonoscopic images in real-world clinical 108 practice using Google cloud AutoML Vision.     respectively. We found that the confusing rate for group I and group N was 69% and . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 10, 2022.  In comparing Group C and Group D, the average precision (positive predictive value), 186 precision, and recall of the algorithm were 89.75%, 87.5%, and 87.5%, respectively, . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. distinguish between the two, the AUC was 0.90, which was greater than the difference 202 between the two groups and healthy subjects. To the best of our knowledge, this is the 203 first AI model that can detect IBS in endoscopic images. Further investigation is needed . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 10, 2022. ; https://doi.org/10.1101/2022.05.09.22274876 doi: medRxiv preprint 13 204 to determine whether AI can differentially detect histological abnormalities, the 205 presence of biofilm, and/or deformation of the colorectal lumen.

206
There are several limitations for this model. First, IBS diagnosis was defined not by 207 ROME criteria, but by disease names that were recorded for insurance purposes.

208
However, ROME IV criteria are not always used in clinical practice, and it is thought 209 that the model is rather accurate because it is designed for AI use in clinical practice.     Fig. 2. Images that scored relatively high (0 to 1) in the colon image AI model for . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 10, 2022.  Fig. 3. Images that scored relatively high (0 to 1) in the colon image AI model for 296 detecting healthy individuals are shown.   CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 10, 2022. ; https://doi.org/10.1101/2022.05.09.22274876 doi: medRxiv preprint Table 1 IBS model of the colon; AUC and contingency tables   312