Page 48 - Read Online

George et al. Mini-invasive Surg 2024;8:4 https://dx.doi.org/10.20517/2574-1225.2023.102 Page 13 of 23

tumours based on LCDH for polyp specificity of 96.20%
detection
[61]
Vieira et al. Polyps and 2020 Retrospective Portugal Construction of GMM and Database of 936 tumour images, 3,000 normal SVM Best result: sensitivity
tumours ensemble system for tumour images for training and testing MLP of 96.1%, specificity
detection of 98.3%
Yamada Polyps and 2021 Retrospective Japan Construction of CNN based on Training: 15,933 colorectal neoplasm images CNN Sensitivity of 79.0%,
[64]
et al. tumours SSMD for colorectal neoplasm Testing: 1,850 colorectal neoplasm images, 2,934 specificity of 87%
detection normal colon images
Saraiva Polyps and 2021 Retrospective Portugal Development of CNN for Database: 860 protruding lesions images, 2,780 CNN Sensitivity of 90.7%,
[65]
et al. tumours protruding lesion detection on normal mucosa images specificity of 92.6%
CCE imaging Training: 2,912 images of database
Testing: 728 images of database
[66] [110]
Jain et al. Polyps and 2021 Retrospective India Creation of deep CNN based Training and testing on KID database and CNN Sensitivity of 98%
[113]
tumours WCENet model for anomaly CVC-clinic database
detection in WCE images
[67]
Zhou et al. Polyps and 2022 Retrospective China Utilising neural network Training: 195 images CNN No sensitivity and
tumours ensembles to improve polyp Testing: 41 images specificity reported
segmentation
Mascarenhas Polyps and 2022 Retrospective Portugal Construction of CNN for Training: 1,928 protruding lesion images, 2,644 CNN Sensitivity of 90.0%,
[68]
et al. tumours protruding lesion detection on normal/other finding imagesTesting: 482 specificity of 99.1%
CCE protruding lesion images, 661 normal/other
finding images
Gilabert Polyps and 2022 Retrospective Spain Comparing AI tool to RAPID Testing: 18 videos CNN Sensitivity of 87.8%
[69]
et al. tumours Reader Software v9.0
(Medtronic)
Piccirelli Polyps and 2022 Retrospective Italy Testing the diagnostic Testing: 126 patients Express Sensitivity of 97%,
[75]
et al. tumours accuracy of Express View view specificity of 100%
(IntroMedic)
[70]
Liu et al. Polyps and 2022 Retrospective China Constructing DBMF fusion Training: 1,450 images DBMF No sensitivity and
tumours network with CNN and Testing: 636 images specificity given
transformer for polyp
segmentation
[71]
Souaidi et al. Polyps and 2023 Retrospective Morocco Modifying existing SSMD Training: 2,745 images SSMD No sensitivity and
tumours models for polyp detection Testing: 784 images specificity given
Mascarenhas Polyps and 2023 Retrospective Portugal Developing CNN for automatic Training: 14,900 images CNN Sensitivity of 96.8%,
[72]
Saraiva et al. tumours detection of small bowel Testing: 3,725 images specificity of 96.5%
protruding lesions
[114]
Lafraxo Polyps and 2023 Retrospective Morocco Proposing novel CNN-based MICCAI2017 : CNN No sensitivity or
[73]
et al. tumours architecture for GI image training: 2,796 images specificity given.
segmentation Testing: 652 images Accuracy of 99.16%
[115]
Kvasir-SEG dataset : on MICCAI2017
Training: 800 images reported,
[116]
Testing: 200 images CVC-ClinicDB dataset : 97.55% on Kvasir-
Training: 490 images SEG,