Page 53 - Read Online

George et al. Mini-invasive Surg 2024;8:4 https://dx.doi.org/10.20517/2574-1225.2023.102 Page 15 of 23

Table 5. Table of AI applications in capsule endoscopy for inflammatory bowel disease

Year of Study Study AI
Ref. Application Aim and goals Training/Validation dataset Results
publication design location type
Haji-Maghsoudi Inflammatory 2012 Retrospective Iran Develop method for the detection of Stenosis: 45 images CED Crohn’s: sensitivity of
[79]
et al. bowel disease lymphangiodysplasia, xanthoma, CD, and CD: 74 images 89.32%,
stenosis in WCE image Lymphangiectasia: 32 images specificity of 65.37%
Lymphoid hyperplasia: 27 images Stenosis: sensitivity of
Xanthoma: 28 images 91.27%,
specificity of 87.27%
Lymphangiectasia:
sensitivity of 95.45%,
specificity of 94.1%
Lymphoid: sensitivity of
87.01%,
specificity of 79.71%
Xanthoma: sensitivity of
97%,
specificity of 97.13%
[78]
Kumar et al. Inflammatory 2012 Retrospective United States Constructing classifier cascade for classifying Training: 355 images SVM Sensitivity over 90% was
bowel disease of America CD lesions into normal, mild, and severe Testing: 212 normal images, 213 mild, found
108 severe images

Charisis and Inflammatory 2016 Retrospective Greece Utilise novel feature extraction method for Database of 466 normal images and SVM Sensitivity of 95.2%,
Hadjileontiadis [80] bowel disease detecting CD lesions 436 CD images specificity of 92.4%
de Maissin Inflammatory 2018 Retrospective France Develop CNN for automatic detection of SB CD Training: 589 images CNN Sensitivity of 62.18%,
[82]
et al. bowel disease lesions Testing: 73 images specificity of 66.81%
[83]
Klang et al. Inflammatory 2019 Retrospective Israel Utilise CNN for CD monitoring and diagnosis by Training: 1,090 images CNN Sensitivity of 96.9%,
bowel disease SB ulcer detection Testing: 273 images specificity of 96.6%
[81]
Barash et al. Inflammatory 2020 Retrospective Israel Automatic severity grading of CD ulcers into Training: 1,242 images CNN Sensitivity of 71%,
bowel disease grades 1 to 3 Testing: 248 images specificity of 34%
[84]
Klang et al. Inflammatory 2020 Retrospective Israel Construction of CNN to differentiate normal and Training: 14,112 images CNN Sensitivity of 97.1%,
bowel disease ulcerated mucosa Testing: 3,528 images specificity of 96%
de Maissin Inflammatory 2021 Retrospective France Assessing importance of annotation quality on Database of 3,498 images was RANN Sensitivity of 93%,
[85]
et al. bowel disease CNN annotated by different readers for specificity of 95%
different trials
Inflammatory 2021 Retrospective Israel Identify intestinal strictures on CE images from Database of 1,942 stricture images, CNN Sensitivity of 92%,
[86]
Klang et al. bowel disease CD patients 14,266 normal mucosa images, specificity of 89%
7,075 mild ulcer images,
2,386 moderate ulcer images,
2,223 severe ulcer images used for
training and testing
[87]
Klang et al. Inflammatory 2021 Retrospective Israel Identify NSAID ulcers, which are common Training: 7,391 CD mucosal ulcer CNN Sensitivity of 92%,
bowel disease differentials for CD ulcers on CE images images, specificity of 95%
10,249 normal mucosa
Testing: 980 NSAIDs ulcer images,