|
Author | Method | Data | Results/description |
|
Hoyt et al. (1973), [11] | The first subjective attempt to utilize fundus cameras for glaucoma detection by the evaluation of RNFL visual appearance. Comparison with perimetric findings. | A few number of black-and-white photographs | Funduscopic signs of the RNFL pattern provide the earliest objective evidence of nerve fiber layer atrophy in the retina. |
|
Lundstrom and Eklundh (1980), [12] | Subjective visual evaluation of the changes in RNFL pattern intensity using fundus photographs. | A few number of black-and-white photographs | Findings that consecutive changes in RNFL pattern intensity are connected to progression of glaucoma disease. |
|
Airaksinen et al. (1984), [3] | Subjective scoring of visual RNFL appearance in fundus photographs. | Black-and-white photographs (84 normals, 58 glaucomatous) | Confirmation of the dependence between changes in RNFL pattern and glaucoma progression in fundus photographs. |
|
Peli (1988), [13] | Semiautomatic analysis of RNFL texture based on intensity information. | Digitized black-and-white photographs (5 normal, 5 glaucomatous, and 5 suspected of glaucoma) | Additional confirmation of the changes in RNFL intensity caused by glaucoma atrophy. |
|
Yogesan et al. (1998), [5] | Automatic method for texture analysis of RNFL based on gray level run length matrices. | Digitized fundus photographs of size 648 × 560 pixels (5 normals, 5 glaucomatous) | Promising results for large focal wedge-shaped RNFL losses well outlined by surrounding healthy nerve fiber bundles. Diffuse RNFL loses could not be detected. |
|
Tuulonen et al. (2000), [6] | Semiautomatic method using microtexture analysis of the RNFL pattern. | Digitized fundus photographs 1280 × 1024 pixels (7 normals, 9 glaucomatous, and 8 suspected of glaucoma | Showing that changes in a microtexture of RNFL pattern are related to glaucoma damage. There is a lack of small sample size. |
|
Oliva et al. (2007), [14] | Semiautomatic method to texture analysis based on RNFL pattern intensity. Comparison with OCT measurement. | DCFI with size of 2256 × 2032 pixels (9 normals, 9 glaucomatous) | Correlation was only 0.424 between the intensity related parameters extracted from fundus images and RNFL thickness was measured by OCT. |
|
Kolář and Jan (2008), [7] | Automatic method to texture analysis of RNFL based on fractal dimensions. | DCFI with size of 3504 × 2336 pixels (14 normal, 16 glaucomatous) | Local fractal coefficient was used as a feature for glaucomatous eye detection. There were problems with robust estimation of this coefficient. |
|
Muramatsu, et al. (2010), [10] | Automatic approach with Gabor filters to enhance certain regions with RNFL pattern and clustering of these regions aimed to glaucoma detection. | DCFI with size of 768 × 768 pixels (81 normals, 81 glaucomatous) | The method is suitable only for detection of focal and wider RNFL losses expressed by significant changes in intensity. |
|
Odstrcilik et al. (2010), [8] | Automatic method to texture analysis of RNFL based on Markov random fields. | DCFI with size of 3504 × 2336 pixels (18 normals, 10 glaucomatous) | The features ability to differentiate between healthy and glaucomatous cases is validated using OCT RNFL thickness measurement. |
|
Prageeth et al. (2011), [15] | Automatic method to texture analysis using only intensity information about RNFL presence. | DCFI with size of 768 × 576 pixels (300 normals, 529 glaucomatous) | Intensity criteria were used. Detection of the substantial RNFL atrophy. |
|
Acharya et al. (2011), [16] | Automatic analysis of RNFL texture using higher order spectra, run length, and cooccurrence matrices. | DCFI with size of 560 × 720 pixels (30 normals, 30 glaucomatous) | Specificity to detect glaucomatous eye is over 91%. The article does not explain thoroughly how the features were extracted and in which area of the image were computed. |
|
Jan et al. (2012), [9] | Automatic method to RNFL texture analysis based on combination of intensity, edge representation, and Fourier spectral analysis. | DCFI with size of 3504 × 2336 pixels (8 normals, 4 glaucomatous) | The ability of proposed features to classify RNFL defects has been proven via comparison with OCT. The comparison was done only in a heuristic manner. |
|