Most double or triple in size.Appl. Sci. 2021, 11,18 ofTable two. Classification tasks
Most double or triple in size.Appl. Sci. 2021, 11,18 ofTable two. Classification tasks summary. Process System Initial Author (Year) Database 2D/3D Field of View (FOV) 82 DR, 23 healthful 2D 6 6 mm2 20 DR, six AMD, four RVO, 26 healthy 2D three three mm2 30 DR, 30 NPDR, 40 healthy 2D 3 three mm2 33 no DR, 26 mild NPDR, 13 PDR, 22 healthy 2D 6 six mm2 80 DR, 90 healthy 2D 3 3 mm2 114 DR, 132 healthier 2D 3 three mm2 463 volumes 2D 3 three mm2 75 DR, 24 diabetes, 32 wholesome 2D six six mm2 303 photos 2D 3 three mm2 Description Attributes: blood vessel density, blood vessel caliber, distance map of FAZ area. Classifier: SVM classifier with RBF. Attributes: mean, regular Moveltipril Inhibitor deviation, skewness, and kurtosis of gray level histogram. No formal classifier. Functions: imply of the intercapillary areas, FAZ perimeter, circularity index, and vascular density. Classifier: neural network Attributes: Vessel tortuosity, fractal dimension ratio (FDR). Classifier: Ordinary least squares modeling approach. Options: multifractal parameter computation (maximum, shift, width, lacunarity, box counting dimension, facts dimension, correlation dimension). Classifier: SVM. Functions: wavelet transform on SVP, DVP, RVN. Classifiers: LR, LR-EN, SVM, XGBoost. VGG19, ResNet50, and DenseNet with superficial and deep plexus pictures, majority soft voting. ResultsSandhu 2018 [70]AUC = 95.22Aharony 2019 [21]Accuracy = 83.9 Total Accuracy = 97 Precision = 95.2 (healthy vs. diabetic) 96.7 (DR vs. NPDR) PDR Accuracy = 94 Mild NPDR vs. healthful Accuracy = 91Abdelsalam 2020 [32] Machine finding out Cano 2020 [65] Diabetic retinopathy classificationAbdelsalam 2021 [33]Accuracy = 98.5Liu 2021 [84]Sensitivity = 84 Specificity = 80 Ensemble network accuracy = 92 1.92 Accuracy = 87.27 AUC = 0.97 (wholesome) 0.98 (no DR) 0.97 (DR) Accuracy = 96.5 (two class) 80.0 (3 classes) 67.9 (four classes)Heisler 2020 [86]Deep learningLe 2020 [89]VGG16.Zang 2021 [90]A densely and constantly connected neural network with adaptive rate dropout (DcardNet).Appl. Sci. 2021, 11,19 ofTable two. Cont. Process Method Initial Author (Year) Database 2D/3D Field of View (FOV) Description Options: Haralick’s texture options, inverse difference normalized and inverse distinction moment normalized features, worldwide functions (like imply, normal deviation, skewness, kurtosis, and entropy), local structure options, thresholded cumulative count of microvasculature pixels). Classifier: SVM. Attributes: microvascular intensity median computed on 6 layers and 7 sectors. Classifiers: SVM, random forest, and gradient boosting. Options: rotation invariant uniform local binary pattern texture functions. Classifier: KNN classifier ResultsOng 2017 [29] Glaucoma classification Machine learning38 glaucoma, 120 healthy 2D 6 six mmSpecificity = 0.95 Sensitivity = 0.87 AUC = 0.Andrade De Jesus 2020 [24]82 glaucoma, 39 healthier 2D 3 three mmAUC = 0.760.06 (xGB) AUC = 0.670.06 (RNFL) Accuracy = 89 (all layers) 89 (superficial) 94 (deep) 98 (outer) one hundred (choriocapillaris) Accuracy = 93.4 (NV-AMD vs. healthful) 77.8 (NV-AMD vs. non-NV-AMD vs. healthful) All vessel Sensitivity = 0.9679 Specificity = 0.9572 Accuracy = 96.57 AUC = 98.05 Accuracy = 86.75Machine learning Age-Related Macular IEM-1460 MedChemExpress Degeneration Classification Deep learningAlfahaid 2018 [83]92 AMD, 92 wholesome 2D 160 non-NV-AMD, 80 NV-AMD, 97 healthy 2D 100 photos 2D 8 8 mm2 30 DR, 20 healthful 2D 6 six mm2 53 CSC, 47 healthful 2D 12 12 mmThakoor 2021 [91]Custom-made 3D CNN, consisting of four 3D convolutional layers, two dense layers, and final softma.