Artificial intelligence (AI) model for histological grading and prognostic identification of various epithelial dysplasia

OrganPurposeWSI scanning systemMagnificationPatchesAI modelModel performanceReference
Oral cavityHistological gradingAperio20×299 × 299 pixelsVGG16Testing AUC: 0.65Araújo et al. [16], Brazil (2023)
Prognostic identificationNanozoomer20×512 × 512 pixelsResNet50 and lightGBMTesting AUC: 0.81 (95% CI, 0.73–0.90)Cai et al. [3], China (2023)
Aperio, Nanozoomer20×, 40×512 × 512 pixelsIDaRSTesting AUC: 0.78Bashir et al. [12], UK (2023)
Aperio, Nanozoomer10×NANuClick (for cell detection)Testing AUC: 0.76 (95% CI, 0.68–0.85)Mahmood et al. [11], UK (2023)
LarynxHistological gradingNanozoomer20×224 × 224 pixelsDenseNet121Testing AUC: 0.89 (95% CI, 0.81–0.95)Lubrano et al. [2], France (2024)
EsophagusHistological gradingMirax Desk20×224 × 224 pixelsResNet50Testing AUC: 0.80Beuque et al. [9], Netherlands (2021)
Aperio40×1,280 × 1,280 pixelsYOLOv5 and ResNet1010.89 accuracy for 3-class, 0.96 accuracy for 2-classFaghani et al. [5], USA (2022)
StomachHistological gradingIscan Coreo20×320 × 320 pixelsResNet50 and domain adaptionTesting AUC: 0.82Shi et al. [4], China (2022)
ColorectumHistological gradingAT220×224 × 224 pixelsResNet18Testing AUC: 0.97 (95% CI, 0.95–0.99)Kim et al. [15], USA (2023)
CervixHistological gradingPloidyScanner20×224 × 224 pixelsVGG16Testing AUC: 0.76 (95% CI, 0.73–0.78)Bao et al. [8], China (2020)

WSI: whole slide image; lightGBM: light gradient boosting machine; IDaRS: iterative draw-and-rank sampling; AUC: area under the receiver operating characteristic curve; 95% CI: lower and upper values of the 95% confidence interval; NA: not applicable; VGG16: visual geometry group 16; ResNet50: residual neural network 50; DenseNet121: dense convolutional network 121; YOLOv5: You Only Look Once version 5