Minireviews
Copyright ©The Author(s) 2020.
Artif Intell Gastrointest Endosc. Oct 28, 2020; 1(2): 33-43
Published online Oct 28, 2020. doi: 10.37126/aige.v1.i2.33
Table 1 Deep learning applications in wireless capsule endoscopy for classifying gastrointestinal disorders
Ref.
Class/outcome variable
Deep network architecture
Device/image resolution
Training and internal validation dataset
Testing/external validation dataset
Accuracy (%)/AUC
Sensitivity (%)/specificity (%)
Majid et al[19], 2020, NAMultiple lesions (bleeding, esophagitis, ulcer, polyp)CNN with classical features fusion and selectionNA/224 × 224 pixels70% of 12889 images from multiple databases30% of 12889 images from multiple databases96.5/NA96.5/NA
Ding et al[20], 2019, ChinaMultiple SB lesions1CNN (ResNet 152)SB-CE by Ankon Technologies/480 × 480 pixels158235 images from 1970 patients113268334 images from 5000 patientsNA/NA99.88/100 (per patient); 99.90/100 (per lesion)
Iakovidis et al[21], 2018, NAMultiple SB lesions2CNN and iterative cluster unification(1) NA/489 × 409 pixels; and (2) MiroCam CE/320 × 320 pixels(1) 465 images from 1063 volunteers; and (2) 852 images(1) 233 images from 1063 volunteers; and (2) 344 images(1) 89.9/0.963; and (2) 77.5/0.814(1) 90.7/88.2; and (2) 36.2/91.3
Aoki et al[22], 2020, JapanBleeding (blood content)CNN (ResNet50)Pillcam SB2 or SB3 CE / 224 × 224 pixels27847 images from 41 patients10208 images from 25 patients99.89/0.999896.63/99.96
Tsuboi et al[23], 2019, JapanBleeding (SB angioectasia)CNN (SSD)Pillcam SB2 or SB3 CE/300 × 300 pixels2237 images from 141 patients10488 images from 28 patientsNA/0.99898.8/98.4
Leenhardt et al[24], 2019, FranceBleeding (SB angioectasia)CNN-based semantic segmentationPillcam SB3 CE / NA600 images600 imagesNA/NA96/100
Li et al[25], 2017, ChinaBleeding (intestinal hemorrhage)CNNs: (1) LeNet; (2) AlexNet; (3) GoogLeNet; and (4) VGG-NetNA/NA9672 images2418 imagesNA/NA(1) 99.91/96.2; (2) 99.96/98.72; (3) 100/98.73; and (4) 99.96/98.72
Jia et al[26], 2017, Hong Kong, ChinaBleeding (both active and inactive)CNNNA/240 × 240 pixels1000 images500 imagesNA/NA91.0/NA
Jia et al[27], 2016, Hong Kong, ChinaBleeding (both active and inactive)CNN (Inspired by AlexNet)NA/240 × 240 pixels8200 images1800 imagesNA/NA99.2/NA
Aoki et al[28], 2019, JapanUlcer (erosion or ulceration)CNN (SSD)Pillcam SB2 or SB3 CE/300 × 300 pixels5360 images from 115 patients10440 images from 65 patients90.8/0.95888.2/90.9
Wang et al[29], 2019, ChinaUlcerCNN (RetinaNet)Magnetic-guided CE by Ankon Technologies/480 × 480 pixels37278 images from 1204 patient cases9924 images from 300 patient cases90.10/0.946989.71/90.48
Wang et al[30], 2019, ChinaUlcerCNN (based on ResNet 34)Magnetic-guided CE by Ankon Technologies/480 × 480 pixels80% of dataset from 1416 patients20% of dataset from 1416 patients92.05/0.972691.64/92.42
Alaskar et al[31], 2019, NAUlcerCNN: (1) GoogLeNet; and (2) AlexNetNA /(1) 224 × 224 pixels; and (2) 227 × 227 pixels336 images105 images(1) 100/1; and (2) 100/1(1) 100/100; and (2) 100/100
Fan et al[32], 2018, China(1) Ulcer; and (2) ErosionCNN (AlexNet)NA/511 × 511 pixels(1) 5500 images; and (2) 7410 images(1) 2750 images; and (2) 5500 images(1) 95.16/0.9891; and (2) 95.34/0.9863(1) 96.80/94.79; and (2) 93.67/95.98
Zhou et al[6], 2017, USACeliac diseaseCNN (GoogLeNet)Pillcam SB2 CE/512 × 512 pixels8800 images from 11 patients8000 images from 10 patientsNA/NA100/100
Klang et al[33], 2020, IsraelCrohn’s diseaseCNN (Xception)Pillcam SB2 CE/299 × 299 pixelsExperiment 1: 80% of 17640 images from 49 patients; Experiment 2: Images from 48 patientsExperiment 1: 20% of 17,640 images from 49 patients; Experiment 2: Images from 1 individual patientExperiment 1: 95.4-96.7/0.989-0.994; Experiment 2: 73.7–98.2/0.940-0.999Experiment 1: 92.5-97.1/96.0-98.1; Experiment 2: 69.5-100/56.8-100
Saito et al[34], 2020, JapanPolyp (protruding lesion)CNN (SSD)Pillcam SB2 or SB3 CE/300 × 300 pixels30584 images from 292 patients17507 images from 93 patients84.5/0.91190.7/79.8
Yuan et al[7], 2017, Hong Kong, ChinaPolypDeep neural networkPillcam SB CE/64 × 64 pixelsUnknown proportion of 4000 images from 35 patientsUnknown proportion of 4000 images from 35 patients98/NA98/99
He et al[35], 2018, IsraelHookwormCNNPillcam SB CE/227 × 227 pixels10 out of 11 patients (436796 images from 11 patients)1 individual patient (11-fold cross-validation)88.5/NA84.6/88.6
Table 2 Deep learning applications in wireless capsule endoscopy for classifying non-disease objects
Ref.
Class/outcome variable
Deep network architecture
Device/image resolution
Training and internal validation dataset
Testing/external validation dataset
Accuracy (%)/AUC
Sensitivity (%)/specificity (%)
Seguí et al[38], 2016, SpainScenes (turbid, bubbles, clear blob, wrinkles, wall)CNNPillcam SB2 CE/100 × 100 pixels100000 images from 50 videos20000 images from 50 videos96/NANA/NA
Zou et al[5], 2015, NAOrgan locations (stomach, small intestine, and colon)CNN (AlexNet)NA/480 × 480 pixels60000 images15000 images95.52/NANA/NA
Table 3 Deep learning applications in wireless capsule endoscopy for improving the reading efficiency of wireless capsule endoscopy
Ref.
Experiment type
Scope of WCE reading /device
Conventional reading
Deep learning assisted reading
P value
Aoki et al[39], 2019, JapanRetrospective study using anonymized dataSB section only/Pillcam SB3mean reading time (min): Trainee: 20.7; Expert: 12.2mean reading time (min): Trainee: 5.2; Expert: 3.1 < 0.001
Overall lesion detection rate: Trainee: 47%; Expert: 84%Overall lesion detection rate: Trainee: 55%; Expert: 87%NS
Ding et al[20], 2019, ChinaRetrospective study by randomly selected videosSmall bowel abnormalities/SB-CE by Ankon Technologiesmean reading time ± standard deviation (min): 96.6 ± 22.53mean reading time ± standard deviation (min): 5.9 ± 2.23< 0.001
Overall lesion detection rate: 41.43%Overall lesion detection rate: 47.00%NA1