Minireviews
Copyright ©The Author(s) 2021.
World J Gastroenterol. Apr 14, 2021; 27(14): 1392-1405
Published online Apr 14, 2021. doi: 10.3748/wjg.v27.i14.1392
Table 1 Artificial intelligence application for esophageal squamous cell cancer
AI ApplicationStudy designData categoryType of ImagesAI architectureTraining datasetValidation Method or datasetAUCSENSPEACCPPVNPVCompared with expertsRef.
DiagnosisRetrospectiveStill imageHRM2-class LDA104 sites167 sites0.9384%95%NANANANAShin et al[18], 2016
DiagnosisProspectiveStill imageHRMFully automated algorithm104 sites167 sites0.93795%91%NANANANAQuang et al[19], 2016
DetectionRetrospectiveStill imageWCEJDPCA + CCV400 images10-fold-CV0.947193.33%89.20%90.75%NANANALiu et al[20], 2016
DiagnosisRetrospectiveStill imageWLI/NBICNN-SSD8428 imagesCaffe DL framework/1118 imagesNA81% (WLI)/89% (NBI) (per-patient) 72% (WLI)/86% (NBI) (per-image)79%99%39%95%NAHorie et al[21], 2019
DetectionRetrospectiveStill imageWLIDNN-CAD2428 images187 images0.963797.8%85.4%91.4%86.4%97.6%SuperiorCai et al[22], 2019
DiagnosisRetrospectiveStill imageWLI/NBI/BLICNN-SSD22562 imagesCaffe DL framework/727 imagesNA100% (Non-ME + NBI/BLI) 90% (Non-ME + WLI) 98% (ME)63% (Non-ME + NBI/BLI) 76% (Non-ME + WLI) 56% (ME)77% (Non-ME + NBI/BLI) 81% (Non-ME + WLI) 77% (ME)NANAEquivalentOhmori et al[23], 2020
DiagnosisRetrospectiveStill imageME-NBIFCN-CAD1383 lesions3-fold-CVNA87% (lesion level)84.1 (lesion level)89.2 (lesion level) 93.0 (pixel level)NANAEquivalentZhao et al[24], 2019
DiagnosisRetrospectiveStill imageME-NBICNN7046 images5-fold-CVNA89.3%98%93.7%NANANAEverson et al[25], 2019
DiagnosisRetrospectiveStill imageECSCNN-GoogLeNet4715 imagesCaffe DL framework/1520 images0.8592.6%89.3%90.9%NANANAKumagai et al[27], 2019
Invasion depth measurementRetrospectiveStill imageWLI/NBI/BLICNN-SSD14338 imagesCaffe DL framework/914 imagesNA90.1%95.8%91.0%99.2%63.9%EquivalentNakagawa et al[28], 2019
Invasion depth measurementRetrospectiveStill imageWLI/NBICNN-SDD-GoogLeNet1751 imagesCaffe DL framework/291 imagesNA84.1%73.3%80.9%NANASuperiorTokai et al[29], 2020
DiagnosisRetrospectiveStill image/Real-time videoNBICAD-SegNet6473 images6671 images/80 videos0.98998.04% (per-image) 91.5%(per-frame)95.03% (per-image) 99.9%(per-frame)NANANANAGuo et al[30], 2020
DiagnosisRetrospective/ ProspectiveStill image/Real-time imageWLIGRAIDS/DeepLab V.3+4091 images3323 imagesNANANANANANAEquivalentLuo et al[31], 2020
Detection/Invasion depth measurementRetrospectiveStill image/Real-time videoNBI/BLICNN-SSD17274 images5277 images/144 videosNA91.1%51.5%63.9%46.1%92.7%SuperiorFukuda et al[32], 2020
Invasion depth measurementRetrospectiveStill image/video imagesWLI/NBI/BLICNN-SSD23977 imagesPyTorch DL framewor/102 video imagesNA50% (non-ME) 70.8%(ME)98.7% (non-ME) 94.9%(ME)87.3% (non-ME) 89.2%(ME)92.3% (non-ME) 81.0%(ME)86.5% (non-ME) 91.4%(ME)SuperiorShimamoto et al[33], 2020