Korean J Helicobacter Up Gastrointest Res > Volume 21(4); 2021 > Article |
|
Study | Nationality of datasets | Type of artificial intelligence | Type of endoscopic image | Type of controls | Total number of images | Number of cases in test dataset | Number of controls in test dataset | |
---|---|---|---|---|---|---|---|---|
Image-based analysis | ||||||||
de Groof et al. [5] (2020) | Europe | CNN | WLI | Non-dysplastic BE | 144 | 33 Barrett’s neoplasias | 111 non-dysplastic BEs | |
Guo et al. [6] (2020) | Multi-national data | CNN | NBI | Non-cancer | 6,671 | 1,480 precancerous and ESCCs (ESCN) | 5,191 non-cancers | |
García-Peraza-Herrera et al. [7] (2020) | Asia | CNN | ME-NBI | Normal IPCL | 67,740 | 39,662 abnormal IPCLs (ESCN) | 28,078 normal IPCLs | |
Hashimoto et al. [8] (2020) | US | CNN | WLI | Non-dysplastic BE | 448 | 225 Barrett’s neoplasias | 223 non-dysplastic BEs | |
de Groof et al. [9] (2020) | Multi-national data (Europe) | CNN | WLI | Non-dysplastic BE | 297 | 129 Barrett’s neoplasias | 168 non-dysplastic BEs | |
Everson et al. [10] (2019) | Asia | CNN | ME-NBI | Normal IPCL | 1,437 | 791 abnormal IPCLs (ESCN) | 646 normal IPCLs | |
Cai et al. [11] (2019) | Asia | CNN | WLI | Normal image | 187 | 91 ESCN | 96 normal images | |
Horie et al. [12] (2019) | Asia | CNN | WLI with NBI | Non-cancer | 97 | 47 esophageal cancers | 50 non-cancers | |
Liu et al. [13] (2016) | Asia | SVM | WLI | Normal image | 400 | 150 early esophageal cancers | 250 normal images | |
van der Sommen et al. [14] (2016) | Europe | SVM | WLI | Non-dysplastic BE | 100 | 60 Barrett’s neoplasias | 40 non-dysplastic BEs | |
Patient-based analysis | ||||||||
de Groof et al. [5] (2020) | Europe | CNN | WLI | Non-dysplastic BE | 20 | 10 Barrett’s neoplasias | 10 non-dysplastic BEs | |
Ohmori et al. [15] (2020) | Asia | CNN | Non-ME detection, ME diagnosis, NBI, BLI | Non-cancer or normal | 102 | 52 superficial ESCC | 50 | |
Ebigbo et al. [16] (2020) | Europe | CNN | WLI | Non-dysplastic BE | 62 | 36 early EACs | 26 non-dysplastic BE | |
de Groof et al. [9] (2020) | Multi-national data (Europe) | CNN | WLI | Non-dysplastic BE | 297 | 129 Barrett’s neoplasias | 168 non-dysplastic BEs | |
Zhao et al. [17] (2019) | Asia | CNN | ME-NBI | Non-cancerous IPCL | 1,383 | 1,176 IPCLs (early ESCC) | 207 non-cancerous IPCLs | |
Ebigbo et al. [18] (2019) | Europe | CNN | WLI with NBI | Non-dysplastic BE | 74 | 33 early EACs | 41 non-dysplastic BE | |
de Groof et al. [19] (2019) | Europe | SVM | WLI | Non-dysplastic BE | 60 | 40 Barrett’s neoplasias | 20 non-dysplastic BE | |
Sehgal et al. [20] (2018) | Europe | Decision tree algorithm | WLI | Non-dysplastic BE image | 40 | 17 Barrett’s neoplasias | 23 non-dysplastic BE | |
van der Sommen et al. [21] (2014) | Europe | SVM | WLI | Non-dysplastic BE | 64 | 32 early EAC | 32 non-dysplastic BEs |
CNN, convolutional neural network; WLI, white light imaging; BE, Barrett’s esophagus; NBI, narrow-band imaging; BLI, blue-light imaging; ESCC, esophageal squamous cell carcinoma; ESCN, early squamous cell neoplasia; ME, magnification endoscopy; IPCL, intrapapillary capillary loop classification; SVM, support vector machine; EAC, early adenocarcinoma.
Study | Nationality of datasets | Type of artificial intelligence | Type of endoscopic image | Aim of study | Design of study | Number of cases | Outcomes |
---|---|---|---|---|---|---|---|
Cho et al. [25] (2019) | Asia | CNN | WLI | Diagnosis of gastric neoplasms | Retrospective model establishment and prospective validation | Training and testing: 5,017 images, validation: 200 images | AUCs of classifying gastric cancer: 0.877; gastric neoplasm: 0.927 |
Cho et al. [26] (2020) | Asia | CNN | WLI | Diagnosis of depth of invasion in gastric neoplasms | Retrospective model establishment and prospective validation | Training and testing: 2,899 white-light endoscopic images, validation: 206 images | External test accuracy 77.3% |
Yoon et al. [27] (2019) | Asia | CNN | WLI | Classification of endoscopic images as early gastric cancer (T1a or T1b) or non-cancer | Retrospective | 11,539 endoscopic images (896 T1a-, 809 T1b-, and 9,834 non-early gastric cancer) | AUC of early gastric cancer detection: 0.981, depth prediction: 0.851 |
Zhu et al. [28] (2019) | Asia | CNN | WLI | Diagnosis of depth of invasion in gastric cancer (mucosa/SM1/deeper than SM1) | Retrospective | Training: 790 images, testing: 203 images | Accuracy: 89.2%, AUC: 0.94 |
Kubota et al. [29] (2012) | Asia | ANN | WLI | Diagnosis of depth of invasion in gastric cancer | Retrospective | 902 images | Accuracy: 77.2%, 49.1%, 51.0%, and 55.3% for T1-4 staging, respectively |
Hirasawa et al. [30] (2018) | Asia | CNN | WLI, chromoendoscopy, narrow-band imaging | Detection of gastric cancers | Retrospective | Training: 13,584 images, testing: 2,296 images | Accurate detection rate with a diameter of 6 mm or more: 98.6% |
Kanesaka et al. [31] (2018) | Asia | SVM | Magnifying narrow-band imaging | Diagnosis and delineation of early gastric cancer using magnifying narrow-band imaging images | Retrospective | Training: 126 images, testing: 81 images | Accuracy: 96.3% |
Lee et al. [32] (2019) | Asia | CNN | WLI | Classification of normal, benign ulcer, and gastric cancer | Retrospective | 200 normal, 367 cancer, and 220 ulcer cases | Accuracy: normal vs. ulcer/normal vs. cancer: above 90%; ulcer vs. cancer: 77.1% |
Modified from tables in the study of Bang et al. [1]
CNN, convolutional neural network; WLI, white light imaging; AUC, area under the curve; SM, submucosa; ANN, artificial neural network; SVM, support vector machine.
Study | Nationality of datasets | Type of artificial intelligence | Type of endoscopic image | Diagnostic method of H. pylori infection | Number of cases in test dataset | Number of controls in test dataset | Unit of analysis |
---|---|---|---|---|---|---|---|
Yasuda et al. [48] (2020) | Japan | SVM | Linked color imaging | More than 2 different tests in each case (histology, serum antibody, stool antigen, urea breath test) | 42 H. pylori patients | 63 controls (46 post-eradication patients and 17 uninfected patients) | Patient-based |
210 H. pylori positive images | 315 control images (230 post-eradication and 85 uninfected images) | Image-based | |||||
210 H. pylori positive images | 85 uninfected images (H. pylori naïve) | Image-based (infected vs. uninfected) | |||||
210 H. pylori positive images | 230 after eradication images | Image-based (infected vs. after-eradication) | |||||
85 uninfected images | 230 after eradication images | Image-based (uninfected vs. after-eradication) | |||||
Zheng et al. [49] (2019) | China | CNN | WLI | Histology with immunohistochemistry (if negative, urea breath test was done) | 2,575 H. pylori positive images | 1,180 control images (whether post-eradication or uninfected images are unknown) | Image-based |
Shichijo et al. [50] (2019) | Japan | CNN | WLI | Serum or urine antibody, stool antigen, urea breath test | 70 H. pylori positive patients | 777 controls (284 post-eradication and 493 uninfected images) | Patient-based |
59 H. pylori positive images | 477 uninfected images (H. pylori naïve) | Image-based (infected vs. uninfected) | |||||
55 H. pylori positive images | 182 after eradication images | Image-based (infected vs. after-eradication) | |||||
481 uninfected images | 249 after eradication images | Image-based (uninfected vs. after-eradication) | |||||
Japan | CNN | WLI | Serum antibody (H. pylori IgG ≥10 U/mL was considered positive) | 30 H. pylori patients | 30 controls (uninfected patients) (H. pylori naïve) | Patient-based | |
Linked color imaging | Patient-based | ||||||
Blue laser imaging-bright | Patient-based | ||||||
Nakashima et al. [51] (2018) | Japan | CNN | WLI | Serum antibody (H. pylori IgG ≥10 U/mL was considered positive) | 15 H. pylori positive images | 15 control images (uninfected patients) (H. pylori naïve) | Image-based |
Shichijo et al. [53] (2017) | Japan | CNN | WLI | Serum or urine anti-body, stool antigen, urea breath test | 72 H. pylori patients | 325 controls (uninfected patients) (H. pylori naïve) | Patient-based |
Huang et al. [54] (2008) | Taiwan | Sequential forward floating selection with SVM | WLI | Histology (3 pairs of samples from the topographic sites, including antrum, body, and cardia were obtained in a uniform way) | 130 H. pylori patients | 106 controls (whether post-eradication or uninfected patients are unknown) | Patient-based |
Huang et al. [55] (2004) | Taiwan | Refined feature selection with neural network | WLI | Histology (3 pairs of samples from the topographic sites, including antrum, body, and cardia were obtained in a uniform way) | 41 H. pylori patients | 33 controls (whether post-eradication or uninfected patients are unknown) | Patient-based |
Modified from tables in the study of Bang et al. [39]
SVM, support vector machine; CNN, convolutional neural network; WLI, white light imaging.
Role of the Small Intestinal Microbiota in Gastrointestinal Disorders2024 December;24(4)
Fungal Infection in Upper Gastrointestinal Tract2024 June;24(2)
Parasitic Diseases of Upper Gastrointestinal Tract2024 June;24(2)
Viral Infection in Upper Gastrointestinal Tract2024 June;24(2)
Endoscopic Treatments in Perforation or Fistula in Upper Gastrointestinal Tract2024 March;24(1)