Page 38 - Read Online

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

5,000 patients
[33]
Majid et al. Erosions and 2020 Retrospective Pakistan Using multi-type features Training: 6,922 images of bleeding, CNN Sensitivity of 96.5%
ulcers extraction, fusion, and features oesophagitis, polyp, and ulcerative colitis
selection to detect ulcer, polyp, Testing: 2,967 images of bleeding,
esophagitis, and bleeding oesophagitis, polyp, and ulcerative colitis
[29]
Kundu et al. Erosions and 2020 Retrospective Bangladesh Employing LDA for ROI separation Training: 65 bleeding, 31 ulcers, SVM Sensitivity of 85.96%,
ulcers and 30 tumour images specificity of 92.24%
Testing: 15 continuous video clips
[34]
Otani et al. Erosions and 2020 Retrospective Japan Multiple lesion detection using Database of 398 images of erosions and Deep No sensitivity and
ulcers RetinaNet ulcers, 538 images of angiodysplasias, neural specificity reported
4,590 images of tumours, and 34,437 network
normal images for training and testing
[35]
Xia et al. Erosions and 2021 Retrospective China Novel CNN and RCNN system to Training: 822,590 images CNN, Sensitivity of 96.2%, specificity
ulcers detect 7 types of lesions in MCE Testing: 201,365 images RCNN of 76.2%
imaging
[36]
Afonso et al. Erosions and 2021 Retrospective Portugal Identify but also differentiate ulcers Training: 18,976 images CNN Sensitivity of 86.6%, specificity
ulcers and erosions based on Testing: 4,744 images of 95.9%
haemorrhagic potential
Mascarenhas Erosions and 2021 Retrospective Portugal Identify various lesions on CE Training: 42,844 images CNN Sensitivity of 88%, specificity
[37]
Saraiva et al. ulcers images and differentiate using Testing: 10,711 images of 99%
Saurin’s classification
[38]
Afonso et al. Erosions and 2022 Retrospective Portugal Identify but also differentiate ulcers Training: 4,904 images CNN Sensitivity of 90.8%, specificity
ulcers and erosions based on Testing: 379 normal images, 266 erosion, of 97.1%
haemorrhagic potential 286 P1 Ulcer images, 295 P2 Ulcer
images
Mascarenhas Erosions and 2022 Retrospective Portugal Develop CNN-based method to Training: 7,204 images CNN Sensitivity of 96.3%, specificity
[39]
et al. ulcers detect and distinguish colonic Testing: 1,801 of 98.2%
mucosal lesions and luminal blood
in CCE imaging
[40]
Xiao et al. Erosions and 2022 Retrospective China Classify capsule gastroscope Training: 228 images CNN No sensitivity and specificity,
ulcers sensitivity of 96.9% images into normal, chronic erosive Testing: 912 images accuracy of 94.81%
and a specificity of gastritis, and gastric ulcer
99.9% specific categories
[41]
Ribeiro et al. Erosions and 2022 Retrospective Portugal Accurately detect ulcers and Training: 26,869 images CNN Sensitivity of 96.9%, specificity
ulcers erosions in CCE images Testing: 3,375 normal images, 357 of 99.9%
images with ulcers or colonic erosions
[43]
Nakada et al. Erosions and 2023 Retrospective Japan Utilise RetinaNet to diagnose Training: 6,476 erosion and ulcer images, Deep Erosions and ulcers: sensitivity
ulcers erosions and ulcers, vascular 1,916 angiodysplasias images, 7,127 neural of 91.9%, specificity of 93.6%
lesions, and tumours in WCE tumour images, 14,014,149 normal network Vascular lesions:
imaging images sensitivity of 87.8%,
Testing: images from 217 patients specificity of 96.9%
Tumours: sensitivity of 87.6%,
specificity of 93.7%
[42] [110]
Raut et al. Erosions and 2023 Retrospective India Use various feature extraction Training and testing on KID dataset Deep Sensitivity of 97.23%,