Work | Image representation | Preprocessing | Features | Classifier | Volumes | Images are publicly available |
---|---|---|---|---|---|---|
Liu et al. [4] | 2D | Image warping | Multi-scale spatial pyramid, LBP histogram + PCA | Non-linear Support Vector Machine | 457 | No |
Serrano et al. [7] | 2D | Normalization | Haar-Like features and Haralick texture features (curtosis and skewness) | Decision Trees | 200 | No |
Albarrak et al. [8] | 3D | Split Bregman Isotropic Total Variation algorithm and a second order polynomial least-square curve fitting for image flattening | Oriented gradient local binary pattern histograms | Bayes network | 140 | No |
Zhang et al. [10] | 3D | Bregman Isotropic Total Variation algorithm with a least squares approach | Local binary patterns of three orthogonal planes (LBP-TOP), local phase quantization (LPQ) and multi-scale spatial pyramid (MSSP) | Ensemble of one-class kernel principal component analysis (KPCA) models | 140 | No |
Farsiu et al. [5] | 3D | Segmentation of tree retinal layers | Abnormal RPEDC thickness and thinness scores | Generalized linear model regression | 384 | Yes |
Srinivasan et al. [9] | 3D | Denoise with BM3D | HOG descriptors | Three linear one-class Support Vector Machines | 45 | Yes |
Venhuizen et al. [11] | 2D | First order vertical Gaussian gradient filter | Unsupervised feature learning approach based in patches of images | Random forest classifier | 384 | Yes |
Wang et al. [12] | 2D | – | Multi-scale linear configuration patterns (LCP) | Sequential minimal optimization (SMO) | 45 | Yes |
Sun et al. [13] | 2D | Retina aligning and crop SIFT descriptors | Three two-class Support Vector Machines (SVM) | 45/678 scans | Yes/no | |
Ravenscroft et al. [15] | 2D | Manual segmentation and labelling of choroid | Learnable features by Convolutional Neural Network (CNN) | Neural Network | 75 | No |
Fang et al. [16] | 3D | Patch mean removal | PCA features | Extreme learning machine (ELM) classifier | 45/54 | Yes/no |
Karri et al. [17] | 2D | RPE estimation based in intensity and BM3D filter is used for noise reduction | Learnable features | Convolutional Neural Network (Transfer learning/GoogLeNet) | 45 | Yes |
Lee et al. [18] | 2D | – | Learnable features | Convolutional Neural Network | 100,000 B-scans | No |
Kermany et al. [19] | 2D | – | Learnable features | Convolutional Neural Network (Transfer Learning) | 207,130 B-scans | Yes |