Skip to main content
NCBI home page
Wiley Open Access Collection logoLink to Wiley Open Access Collection
. 2023 Oct 11;36(5):582–590. doi: 10.1111/den.14677

Artificial intelligence quantifying endoscopic severity of ulcerative colitis in gradation scale

Kaoru Takabayashi 1,✉, Taku Kobayashi 3, Katsuyoshi Matsuoka 5, Barrett G Levesque 8, Takuji Kawamura 6, Kiyohito Tanaka 6, Takeaki Kadota 7, Ryoma Bise 4,7, Seiichi Uchida 4,7, Takanori Kanai 2, Haruhiko Ogata 1
PMCID: PMC12136255  PMID: 37690125

Abstract

Objectives

Existing endoscopic scores for ulcerative colitis (UC) objectively categorize disease severity based on the presence or absence of endoscopic findings; therefore, it may not reflect the range of clinical severity within each category. However, inflammatory bowel disease (IBD) expert endoscopists categorize the severity and diagnose the overall impression of the degree of inflammation. This study aimed to develop an artificial intelligence (AI) system that can accurately represent the assessment of the endoscopic severity of UC by IBD expert endoscopists.

Methods

A ranking‐convolutional neural network (ranking‐CNN) was trained using comparative information on the UC severity of 13,826 pairs of endoscopic images created by IBD expert endoscopists. Using the trained ranking‐CNN, the UC Endoscopic Gradation Scale (UCEGS) was used to express severity. Correlation coefficients were calculated to ensure that there were no inconsistencies in assessments of severity made using UCEGS diagnosed by the AI and the Mayo Endoscopic Subscore, and the correlation coefficients of the mean for test images assessed using UCEGS by four IBD expert endoscopists and the AI.

Results

Spearman's correlation coefficient between the UCEGS diagnosed by AI and Mayo Endoscopic Subscore was approximately 0.89. The correlation coefficients between IBD expert endoscopists and the AI of the evaluation results were all higher than 0.95 (P < 0.01).

Conclusions

The AI developed here can diagnose UC severity endoscopically similar to IBD expert endoscopists.

Keywords: artificial intelligence, convolutional neural network, ranking, ulcerative colitis

INTRODUCTION

Ulcerative colitis (UC) is a chronic, persistent inflammatory bowel disease of unknown cause with recurrent flare‐ups and remissions, typically starting with mild inflammation of the rectum and progressing to extensive inflammation of the colon. 1 Evaluation of mucosal inflammation by colonoscopy is the gold standard for assessing and controlling inflammation to avoid colectomy. 2 , 3 , 4 , 5

Traditionally, endoscopic scoring systems, such as the Mayo Endoscopic Subscore (MES) 6 and the Ulcerative Colitis Endoscopic Index of Severity (UCEIS), 7 , 8 have been developed to assess inflammation in clinical trials. These scoring systems provide objectivity by incorporating the presence or absence of inflammatory findings in the most severely inflamed areas and assigning an endoscopic severity score categorized into four (MES) or nine (UCEIS) levels for each case. However, these scoring systems have several limitations. First, these systems typically evaluate inflammation by using categorical scores. Endoscopic inflammation with different degrees of clinical significance has been categorized using the same score. This can be a significant limitation compared with inflammatory bowel disease (IBD) expert endoscopist assessment when measuring the level of inflammation, subtle changes in response to therapy, or early signs of relapse. These assessments can be key to managing diseases with limited therapies, high rates of relapse, and surgical risks. However, it is difficult to measure experts' assessments accurately and reproducibly; therefore, it is generally expressed by categorized indices, such as MES and UCEIS. In addition, some reports indicate that inter‐ and intra‐observer differences in these indices are improving but are not yet sufficiently reproducible. 9

Artificial intelligence (AI) has recently been developed to increase the reproducibility of scores reported in recent years; however, it has only performed scoring in accordance with conventional scoring systems and has not represented the overall clinical endoscopic assessment of IBD expert endoscopists. 10 , 11 , 12 , 13 , 14 , 15 Therefore, this study aimed to develop an AI that can accurately represent the complex assessment of the endoscopic severity of UC by IBD expert endoscopists, similar to the visual analog scale.

METHODS

Patients and endoscopic images

We conducted a multicenter, retrospective study using 39,553 endoscopic images collected from 424 patients with UC who underwent colonoscopy at Keio University Hospital between April 2012 and October 2019 and 10,256 endoscopic images collected from 388 patients with UC who underwent colonoscopy at the Japanese Red Cross Kyoto Daini Hospital during the same period, resulting in 49,809 images. Endoscopic imaging was performed using PCF‐H290ZI, PCF‐H290I, CF‐H290I, CF‐H260AI, or CF‐H260AZI devices, which are standard endoscopes manufactured by Olympus Medical Systems (Tokyo, Japan), or with EC‐L600ZP7, EC‐L600ZP7/L, or EC‐L600XP7/L devices, which are standard endoscopes manufactured by Fujifilm Corporation (Tokyo, Japan). The study was approved by the ethics committee of each medical institution (Keio Ethics Committee approval no. 20180315), and each participant provided informed consent before participation.

Dataset for endoscopic images of UC and definition of IBD expert endoscopists

To enable the AI to perform continuous evaluations of inflammation in line with the strategy used by IBD expert endoscopists, we did not directly use the scores determined from images using MES or UCEIS as classification tasks when training the AI. Instead, we trained the AI as a ranking task using comparative information on the UC severity of image pairs annotated by IBD expert endoscopists. In the MES and UCEIS systems, findings that reflect the degree of inflammation are converted to a numeric form, and assessment is performed on that basis. These findings included the visible vascular pattern, degree of adhesion of the mucopurulent matter, depth, number of erosional and ulcerous surfaces, degree of mucosal edema, degree and range of redness, mucosal fragility, and extent of regenerated mucosa (in this study, hemorrhage was excluded from evaluation because assessments were performed using a stationary screen). However, because assessments based on these criteria cannot reliably reproduce composite assessments of UC severity made by IBD expert endoscopists, we relied on each diagnostic assessment without restrictions related to detailed severity criteria.

Here, an IBD expert endoscopist with >15 years of endoscopic experience, performed endoscopy on >2000 patients with UC, and had published at least one report on the endoscopic diagnosis of UC. Seven IBD expert endoscopists participated in the study: three involved in the AI development process only and four in the AI validation process. To prepare image datasets for AI, an IBD expert endoscopist assessed which of the two randomly arranged images (total training pool: 14,208 images) indicated greater severity (or whether they were of approximately equal severity). The MES data were used for pretraining and to evaluate the performance of the AI when preparing the data. A total of 14,208 images were used, consisting of 7897, 2847, 2684, and 780 images with MES values of 0, 1, 2, and 3, respectively. All still images were preprocessed to exclude blurred images. Narrow‐band imaging, blurred, or out‐of‐focus images were excluded.

AI algorithm

This study uses a ranking‐convolutional neural network (ranking‐CNN) as a machine learning model to provide more detailed severity assessment than conventional scoring. First, an IBD expert endoscopist labeled pairs of endoscopic images with relative severity labels to provide training data. Appendix S1 discusses the AI algorithm. Engineering researchers developed the AI algorithm without annotating training data or evaluating performance. Furthermore, the algorithm was designed to correct the output from the ranking‐CNN to a severity score ranging from 0 to 10 after training the model. Details of the correction method are provided in Appendix S1. We named this scale, which represents the endoscopic severity of UC as a continuous value from 0 to 10, the UC Endoscopic Gradation Scale (UCEGS) (Fig. 1).

Figure 1.

Figure 1

Open in a new tab

Overview of the process of developing a novel artificial intelligence (AI) system capable of diagnosing the Ulcerative Colitis Endoscopic Gradation Scale (UCEGS). CNN, convolutional neural network; IBD, inflammatory bowel disease.

Construction of a user interface to visualize the output of novel AI

To accurately determine the severity and area of colorectal inflammation associated with UC, we developed a user interface in which the UCEGS results diagnosed using this novel AI were incorporated into the schema for the colon. Specifically, a physician performing the endoscopy stipulated an area in the colon in the endoscopic images, calculated the mean image severity for each locus as determined by the AI, and generated schematic images of the colon with numeric values and color gradations proportional to severity. Figure 2 shows the output images of the novel AI system (i.e., AI with user interface [UI]). The UI comprises five screens and the operational flow includes five steps. Step 1 launches the application, Step 2 registers patient information, Step 3 analyzes the endoscopic images and determines the severity of each image, and Step 4 displays the results of the severity assessment of the entire bowel. The overall bowel severity screen (overall severity screen 1) can be switched to a screen that compares bowel severity with previous data (overall severity screen 2). Step 5 displays the resulting image.

Figure 2.

Figure 2

Open in a new tab

User interface (UI) structure and operation flow. The figure shows the structure of the UI and the operation flow. The UI comprises five screens and the operational flow consists of five steps. Step 1 starts the application, Step 2 registers the patient information, Step 3 reads the endoscopic images and determines the severity of each image, and Step 4 displays the results of the severity assessment of the entire bowel. The overall bowel severity screen (overall severity screen 1) can be switched to a screen that compares bowel severity with previous data (overall severity screen 2). Step 5 displays the resulting image. The UI was designed based on the opinions of inflammatory bowel disease expert endoscopists and has a structure that is easy to use in clinical practice.

Outcome

Using 1479 images for which MES had been assessed beforehand by an IBD expert endoscopist, correlation coefficients were calculated to ensure no inconsistencies in the assessments of severity based on the results of this novel AI‐diagnosed UCEGS and the current MES score. The correlation coefficients and standard errors of the means of the UCEGS results for the 50 test images evaluated by four IBD expert endoscopists who were not involved in the AI development process, and the novel AI were calculated. The standard deviation was also calculated to assess the diagnostic variability between each IBD expert endoscopist. Finally, to confirm whether a diagnosis of inflammation within the correct range was feasible, the system was evaluated using endoscopic images obtained from patients with UC. The 1479 test set images were selected from different patients and on different dates than the training set images, and 50 test images were randomly selected from the test set.

Statistical analysis

All statistical analyses were performed using Python version 3.6.12 (Python Software Foundation, Fredericksburg, VA, USA). Correlations between continuous variables were evaluated using the Spearman's rank correlation coefficient and linear regression. Statistical significance was defined as P < 0.05.

RESULTS

Workflow of the AI system

Figure 3 shows the flow of inflammation evaluation of endoscopic images using the novel AI system. Once the endoscopic images are input into the AI system, the AI can calculate the scale value of inflammation along the UCEGS for each image.

Figure 3.

Figure 3

Open in a new tab

Workflow of the artificial intelligence (AI) system. When endoscopic images are input into this AI system, the AI can calculate the scale value of inflammation along the Ulcerative Colitis (UC) Endoscopic Gradation Scale for each image.

Relationship between MES and AI‐diagnosed UCEGS

Figure 4 shows box plots of the AI‐diagnosed UCEGS and MES results for the test data. The MES assessed by IBD expert endoscopists is shown on the x‐axis, whereas the AI‐diagnosed UCEGS is shown on the y‐axis. The distribution of AI‐diagnosed UCEGS exhibited no bias for any MES value. At MES values of 0 through 2, there was no overlap in the interquartile range for AI‐diagnosed UCEGS, and the Spearman's correlation coefficient between MES and AI‐diagnosed UCEGS was approximately 0.89, indicating a strong positive correlation for the order of severity between AI‐diagnosed UCEGS and MES. However, at MES values of 2 and 3, there was an overlap in the interquartile range, suggesting that the AI‐diagnosed UCEGS and MES were not necessarily consistent in the high‐severity range. This inconsistency may be related to the exclusion of hemorrhage during relational annotation or a lack of scoring for UC‐specific individual findings. However, it may also be related to the comprehensive assessments performed by IBD expert endoscopists.

Figure 4.

Figure 4

Open in a new tab

Relationship between Mayo Endoscopic Subscore (MES) and artificial intelligence (AI)‐diagnosed Ulcerative Colitis (UC) Endoscopic Gradation Scale (UCEGS). The Spearman's correlation coefficient between MES and AI‐diagnosed UCEGS was approximately 0.89, indicating a strong positive correlation for the order of severity between the AI‐diagnosed UCEGS and MES.

Correlation between severity assessments made by IBD expert endoscopists and AI

Fifty endoscopic images were evaluated to validate UCEGS. Four IBD expert endoscopists graded the endoscopic severity as a continuous value from 0 to 10, and the correlation with UCEGS was calculated. The correlation coefficients were extremely high: 0.96, 0.98, 0.97, and 0.96 for each endoscopist (P < 0.01; Fig. 5), and the standard errors of the means of the severity scores for the four IBD expert endoscopists and the novel AI were 0.7 and ≤1, respectively. Figure 6 shows the mean score estimated by the IBD expert endoscopists and the error range by standard deviation. The IBD expert endoscopist's estimates tended to be more variable for mild and moderate disease images than for severe images.

Figure 5.

Figure 5

Open in a new tab

Correlation between severity assessments made by inflammatory bowel disease (IBD) expert endoscopists and artificial intelligence (AI). Fifty endoscopic images were evaluated by four IBD expert endoscopists and using AI according to the Ulcerative Colitis Endoscopic Gradation Scale, and the correlation coefficients of the evaluation results were found to be extremely high: 0.96, 0.98, 0.97, and 0.96, respectively (P < 0.01).

Figure 6.

Figure 6

Open in a new tab

The mean scale estimated by inflammatory bowel disease (IBD) expert endoscopists, and the error range by standard deviation. This figure shows the mean score estimated by IBD expert endoscopists and the error range by standard deviation. The horizontal axis represents the ordinal number of the 50 images used for scoring, arranged in Ulcerative Colitis Endoscopic Gradation Scale (UCEGS) order, and the vertical axis represents the severity score. The error bars indicate the standard deviation (±1 SD). The IBD expert endoscopist's estimates tended to be more variable for mild and moderate disease images than for severe images. AI, artificial intelligence.

DISCUSSION

In this study, we developed a novel AI that does not rely on conventional scoring methods but instead aims to leverage the intelligence of IBD expert endoscopists when evaluating the disease status of UC. This AI quantifies inflammation as a gradient, allowing for automated visualization of the endoscopist's assessment of mucosal inflammation.

Traditionally, the endoscopic severity of UC has been evaluated using scoring indices such as MES and UCEIS, and previously developed AIs have been aimed at scoring these indices alternatively to human endoscopists. 11 , 16 , 17 However, expert endoscopists incorporate important information into endoscopic images beyond the findings that define the conventional score and judge the severity of each case. Specifically, the amount of mucus, edema, and erythema at the ulcer margins, regenerative epithelium at the ulcer base, and other complex and sometimes difficult‐to‐verbalize findings are all taken into account. 7 , 10 , 13 , 18 To reproduce the assessments of IBD expert endoscopists more accurately, it is necessary to incorporate unlimited aspects of these findings; this has been impossible. However, we abandoned the concept of expressing inflammation as a categorical score. We developed an AI that expressed inflammation on a continuous scale, making it possible to computerize and reproduce the endoscopist's assessment. The correlation between continuous values from 0 to 10 graded by IBD expert endoscopists and the UCEGS using this novel AI was high, whereas the standard error was low. These results indicate that this novel AI can reliably reproduce the evaluation of IBD experts based on all the inflammation‐related findings described here and suggest that this system can be used as a computerized substitute for scoring by an IBD expert endoscopist.

To the best of our knowledge, this is the first study to apply a novel deep learning method, ranking‐CNN, for the development of a clinical diagnostic tool. The details of ranking‐CNN are described in the section “AI algorithm,” ranking‐CNN can accurately reorder objects that are difficult to evaluate quantitatively according to their size or strength. In other words, it can be expressed in the same manner as the visual analog scale, which we named UCEGS.

One of the strengths of our AI is that it can represent inflammation on a UCEGS scale, which enables a more detailed assessment of inflammation than conventional scoring systems. This novel AI was found to subdivide images into different levels of severity, even when evaluated according to the MES criteria (Fig. 4), which is believed to be more effective than conventional scoring systems at accurately determining treatment efficacy and evaluating mucosal healing.

Finally, to apply this AI in clinical practice, we developed a UI that depicted the severity and distribution of inflammation (Fig. 5). This is intended to visualize the entire status of the colon at a glance by superimposing the UCEGS on the image of the colon so that it can reflect what the endoscopists envisioned when performing the endoscopy. Bresci et al. 19 reported that the degree of lesion extension is important in determining the prognosis of UC. In view of this, this system is expected to enable immediate comprehension of the present severity and extent of inflammation and to more accurately determine improvement in response to treatment, which will aid in selecting subsequent treatment options.

The AI developed in this study had some limitations. First, a selection bias may have affected the collection of endoscopic images for AI learning. Among the images used for learning, the proportion of images showing mild inflammation was the highest, whereas the proportion showing severe inflammation was the lowest. However, because these data were collected from multiple medical institutions, we considered them to represent a dataset from real‐world clinical practice. This supports the notion that the novel AI system exhibits sufficient performance even in this environment and is thus applicable to clinical practice. Second, no external validation was performed to evaluate the performance of the AI. We believe that the possibility of the validation process of this AI being overestimated cannot be ruled out because it was evaluated at four relevant facilities but not at facilities that were not involved in the development of this AI. Third, the usefulness of this novel AI has not yet been validated in clinical practice. Future prospective studies are needed to verify whether UCEGS correlates with disease prognosis and whether AI can detect treatment‐induced mucosal changes accurately. Fourth, although IBD expert endoscopists can recognize the site of the colon on endoscopic images, AI is not yet able to automatically diagnose the site of the colon. Therefore, the development of an automatic site‐identification function is desirable.

In conclusion, the innovative AI developed in this study can act as a proxy for IBD expert endoscopists by reliably presenting comprehensive assessments of UC severity, which was not previously possible. Future studies should focus on the responsiveness of this novel scale to effective therapies.

CONFLICT OF INTEREST

Author K.Takabayashi has received personal fees from Olympus and Janssen Co Ltd. T.Kobayashi has received personal fees from AbbVie GK, Ajinomoto Pharma, Alfresa Pharma, Asahi Kase Medical, Astellas Covidien, Celltrion, EA Pharma Co. Ltd, Eisai Co. Ltd, Eli Lilly, Ferring Pharmaceuticals, Galapagos, Gilead Sciences, Google, Janssen, JIMRO Co. Ltd, Kyorin Pharmaceutical Co. Ltd, Mitsubishi Tanabe Pharma, Mochida Pharmaceutical, Nippon Kyaku, Pfizer, Takeda Pharmaceutical, Thermo Fisher Scientific, and ZERIA; and research grants from AbbVie GK, Alfresa Pharma, EA Pharma Co., Ltd., Kyorin Pharmaceutical Co. Ltd, Mitsubishi Tanabe Pharma, Mochida Pharmaceutical, Nippon Kyaku, Otsuka Holdings Co., Ltd., Sekisui Medical, Thermo Fisher Scientific, and ZERIA. K.M. has received personal fees from Mitsubishi Tanabe Pharma, Takeda Pharmaceutical Co., Ltd, Janssen Pharmaceutical K.K., AbbVie Inc., EA Pharma Co., Ltd, Pfizer Inc., Mochida Pharmaceutical Co., Ltd, Kyorin Pharmaceutical Co., Ltd, ZERIA Pharmaceutical Co., Ltd, Kissei Pharmaceutical Co., Ltd, Nippon Kayaku Co., Ltd, JIMRO Co., Ltd, and Gilead Sciences; research grants from Mitsubishi Tanabe Pharma, AbbVie Inc., EA Pharma Co., Ltd., Mochida Pharmaceutical Co., Ltd., Kyorin Pharmaceutical Co., Ltd., ZERIA Pharmaceutical Co., Ltd., Kissei Pharmaceutical Co., Ltd., Nippon Kayaku Co., Ltd., and JIMRO Co., Ltd.; and has a leadership advisory role for EA Pharma Co., Ltd., Eli Lilly, Bristol‐Myers Squibb, and Janssen Pharmaceutical K.K. B.L. is a shareholder and employee of Gossamer Bio, Inc. T.Kawamura is an Associate Editor of Digestive Endoscopy and has received personal fees from Olympus Corporation, Boston Scientific Co. Ltd, MC Medical Inc., Kyorin Pharmaceutical Co. Ltd, Takeda Pharmaceutical Co. Ltd, Bayer Yakuhin Ltd, EA Pharma Co. Ltd, Daiichi Sankyo Co. Ltd, and Mitsubishi Tanabe Pharma Corporation. S.U. has received personal fees from Nippon Telegraph and Telephone and CyberAgent, Inc. T.Kanai has received personal fees from Mitsubishi Tanabe Pharma, Takeda Pharmaceutical Co., Ltd, AbbVie Inc., EA Pharma Co., Ltd, and Miyarisan Pharmaceutical Co., Ltd; research grants from Miyarisan Pharmaceutical Co., Ltd, Glico Co., Ltd, and Nissin Foods Co., Ltd; and has a leadership advisory role for Chugai Pharmaceutical Co., Ltd. H.O. has received personal fees from Fujifilm Co., Ltd, Olympus Co., Ltd, and Covidien; and research grants from Boston Scientific Co., Ltd.

FUNDING INFORMATION

This study was supported by AMED under Grant Numbers JP19lk1010026h0002 and 20lk1010026h0003.

Supporting information

Appendix S1 Detailed artificial intelligence algorithm.

DEN-36-582-s002.docx (14.3KB, docx)

Figure S1 Flowchart for novel artificial intelligence system construction. By learning the results of severity comparisons diagnosed by an inflammatory bowel disease expert endoscopist, ranking‐convolutional neural network can create an array of inflammation severities with rank values. Four inflammatory bowel disease expert endoscopists then rescaled these rank values to a continuous 10‐point scale to create the Ulcerative Colitis Endoscopic Gradation Scale. Finally, a user interface for visualizing the scale is added to the ranking‐convolutional neural network to construct a novel artificial intelligence system.

DEN-36-582-s001.tif (90.2KB, tif)

ACKNOWLEDGMENTS

We thank Dr Liana Adam (MD, PhD) and Editage for editing the manuscript draft.

Kaoru Takabayashi and Taku Kobayashi contributed equally to this work.

REFERENCES

  • 1. Kobayashi T, Siegmund B, Le Berre C et al. Ulcerative colitis. Nat Rev Dis Primers 2020; 6: 74. [DOI] [PubMed] [Google Scholar]
  • 2. Eaden JA, Abrams KR, Mayberry JF. The risk of colorectal cancer in ulcerative colitis: A meta‐analysis. Gut 2001; 48: 526–535. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Frøslie KF, Jahnsen J, Moum BA, Vatn MH, IBSEN Group . Mucosal healing in inflammatory bowel disease: Results from a Norwegian population‐based cohort. Gastroenterology 2007; 133: 412–422. [DOI] [PubMed] [Google Scholar]
  • 4. Colombel JF, Rutgeerts P, Reinisch W et al. Early mucosal healing with infliximab is associated with improved long‐term clinical outcomes in ulcerative colitis. Gastroenterology 2011; 141: 1194–1201. [DOI] [PubMed] [Google Scholar]
  • 5. Ardizzone S, Cassinotti A, Duca P et al. Mucosal healing predicts late outcomes after the first course of corticosteroids for newly diagnosed ulcerative colitis. Clin Gastroenterol Hepatol 2011; 9: 483–489.e3. [DOI] [PubMed] [Google Scholar]
  • 6. Schroeder KW, Tremaine WJ, Ilstrup DM. Coated oral 5‐aminosalicylic acid therapy for mildly to moderately active ulcerative colitis. A randomized study. N Engl J Med 1987; 317: 1625–1629. [DOI] [PubMed] [Google Scholar]
  • 7. Travis SPL, Schnell D, Krzeski P et al. Developing an instrument to assess the endoscopic severity of ulcerative colitis: The Ulcerative Colitis Endoscopic Index of Severity (UCEIS). Gut 2012; 61: 535–542. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8. Travis SPL, Schnell D, Krzeski P et al. Reliability and initial validation of the Ulcerative Colitis Endoscopic Index of Severity. Gastroenterology 2013; 145: 987–995. [DOI] [PubMed] [Google Scholar]
  • 9. Feagan BG, Khanna R, Sandborn WJ et al. Agreement between local and central reading of endoscopic disease activity in ulcerative colitis: Results from the tofacitinib OCTAVE trials. Aliment Pharmacol Ther 2021; 54: 1442–1453. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Takenaka K, Ohtsuka K, Fujii T et al. Development and validation of a deep neural network for accurate evaluation of endoscopic images from patients with ulcerative colitis. Gastroenterology 2020; 158: 2150–2157. [DOI] [PubMed] [Google Scholar]
  • 11. Ozawa T, Ishihara S, Fujishiro M et al. Novel computer‐assisted diagnosis system for endoscopic disease activity in patients with ulcerative colitis. Gastrointest Endosc 2019; 89: 416–421.e1. [DOI] [PubMed] [Google Scholar]
  • 12. Nakase H, Hirano T, Wagatsuma K et al. Artificial intelligence‐assisted endoscopy changes the definition of mucosal healing in ulcerative colitis. Dig Endosc 2021; 33: 903–911. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Stidham RW, Liu W, Bishu S et al. Performance of a deep learning model vs human reviewers in grading endoscopic disease severity of patients with ulcerative colitis. JAMA Netw Open 2019; 2: e193963. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Le Berre C, Sandborn WJ, Aridhi S et al. Application of artificial intelligence to gastroenterology and hepatology. Gastroenterology 2020; 158: 76–94.e2. [DOI] [PubMed] [Google Scholar]
  • 15. Holmer AK, Dulai PS. Using artificial intelligence to identify patients with ulcerative colitis in endoscopic and histologic remission. Gastroenterology 2020; 158: 2045–2047. [DOI] [PubMed] [Google Scholar]
  • 16. Maeda Y, Kudo SE, Ogata N et al. Evaluation in real‐time use of artificial intelligence during colonoscopy to predict relapse of ulcerative colitis: A prospective study. Gastrointest Endosc 2022; 95: 747–756.e2. [DOI] [PubMed] [Google Scholar]
  • 17. Takenaka K, Ohtsuka K, Fujii T, Oshima S, Okamoto R, Watanabe M. Deep neural network accurately predicts prognosis of ulcerative colitis using endoscopic images. Gastroenterology 2021; 160: 2175–2177.e3. [DOI] [PubMed] [Google Scholar]
  • 18. Hirai F, Matsui T. A critical review of endoscopic indices in ulcerative colitis: Inter‐observer variation of the endoscopic index. Clin J Gastroenterol 2008; 1: 40–45. [DOI] [PubMed] [Google Scholar]
  • 19. Bresci G, Parisi G, Capria A. Duration of remission and long‐term prognosis according to the extent of disease in patients with ulcerative colitis on continuous mesalamine treatment. Colorectal Dis 2008; 10: 814–817. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Appendix S1 Detailed artificial intelligence algorithm.

DEN-36-582-s002.docx (14.3KB, docx)

Figure S1 Flowchart for novel artificial intelligence system construction. By learning the results of severity comparisons diagnosed by an inflammatory bowel disease expert endoscopist, ranking‐convolutional neural network can create an array of inflammation severities with rank values. Four inflammatory bowel disease expert endoscopists then rescaled these rank values to a continuous 10‐point scale to create the Ulcerative Colitis Endoscopic Gradation Scale. Finally, a user interface for visualizing the scale is added to the ranking‐convolutional neural network to construct a novel artificial intelligence system.

DEN-36-582-s001.tif (90.2KB, tif)

Articles from Digestive Endoscopy are provided here courtesy of Wiley

RESOURCES