Page 56 - Read Online

George et al. Mini-invasive Surg 2024;8:4 https://dx.doi.org/10.20517/2574-1225.2023.102 Page 17 of 23

Table 6. Table of AI applications in capsule endoscopy for coeliac disease
Year of Study Study Training/Validation
Ref. Application Aim and goals AI type Results
publication design location dataset
Zhou Coeliac 2017 Retrospective China Develop CNN- Training: 6 coeliac disease CNN Sensitivity of
et al. [93] disease based methodology patient CE videos, 5 control 100%,
for coeliac disease patient CE videos specificity of
identification Testing: 5 coeliac disease 100%
patient CE videos, 5 control
patient CE videos
Wang Coeliac 2020 Retrospective China Construct novel Database of 1,100 normal CNN Sensitivity of
et al. [94] disease deep learning mucosa images, 1,040 CD SVM 97.20%,
recalibration mucosa images used for KNN specificity of
module for the training and testing LDA 95.63%
diagnosis of coeliac
disease on VCE
images
Li Coeliac 2021 Retrospective China Utilise novel SPCA Training: 184 images KNN No
et al. [95] disease method for image Testing: 276 images SVMCNN Sensitivity or
processing to specificity
detect coeliac given,
disease accuracy of
93.9%
Chetcuti Coeliac 2023 Retrospective United Evaluate and Training: 444,659 images MLA No
Zammit disease Kingdom/ compare coeliac Testing: 63 VCE videos Sensitivity or
et al. [96] United disease severity specificity
States of assessment of AI given
America tool and human
readers

AI: Artificial intelligence; CNN: convolutional neural network; CE: capsule endoscopy; VCE: video capsule endoscopy; SVM: support vector
machine; KNN: K nearest neighbour; LDA: linear discriminant analysis; SPCA: strip principal component analysis; MLA: machine learning
algorithm.

Table 7. Table of AI applications in capsule endoscopy for hookworm detection
Year of Study Study Training/Validation AI
Ref. Application Aim and goals Results
publication design location dataset type
Wu Hookworm 2016 Retrospective China Automatically detect 440,000 images from 11 MLA Sensitivity of
[97]
et al. detection hookworm on WCE patients used for training 77.3%,
images and testing specificity of
77.9%
He Hookworm 2018 Retrospective China Utilise deep learning for 440,000 images from 11 CNN Sensitivity of
et al. [98] detection automatic hookworm patients used for training 84.6%,
detection and testing specificity of
88.6%
Gan Hookworm 2021 Retrospective China Construct CNN for the Training: 11,236 images of CNN Sensitivity of
et al. [99] detection automatic detection of hookworm 92.2%,
hookworm on CE Testing: 531 hookworm specificity of
images images, 9,998 normal 91.1%
images

AI: Artificial intelligence; WCE: wireless capsule endoscopy; MLA: machine learning algorithm; CNN: convolutional neural network; CE: capsule
endoscopy.

positives such as in those with haemorrhoids or who do not wish to partake in FOBT-based screening
programs. Furthermore, the non-invasiveness and cost-effectiveness of AI Capsule colonoscopy offer
advantages over traditional procedures, making it a promising option for mass screening in the near future.
It is expected that AI tools will replace parts of the endoscopy procedure after undergoing further clinical
evaluation, especially with examples such as AnX Robotica’s ProScan receiving FDA approval in 2024.