Machine learning applied to retinal image processing for glaucoma detection: review and perspective

Table 2 Main studies using features extraction

Methods	Year	Data	Preprocessing	Features Extract	No. of features	Best classifier^a	Results^a (%)
Methods	Year	Data	Preprocessing	Features Extract	No. of features	Best classifier^a	Acc	Sp	Sn
Noronha et al. [38]	2014	272	Image resize with interpolation method	Higher order cumulant features	35	NB	92.65	100.00	92.00
Acharya et al. [39]	2015	510	Image resizing with histogram equalization	Gabor transform	32	SVM	90.98	91.63	91.32
Issac et al. [40]	2015	67	Image resizing with statistical features	Cropped input image after segmentation	3	SVM	94.11	90	100
Raja et al. [45]	2015	158	Grayscale conversion and histogram equalization	Hyper-analytic wavelet transformation	16	SVM	90.14	85.66	94.30
Singh et al. [47]	2016	63	N/A	Wavelet feature extraction	18	k-NN	94.75	100	90.91
Maheshwari et al. [30]	2017	488	Grayscale conversion	Variational mode decomposition	4	LS-SVM	94.79	95.88	93.62
Soltani et al. [48]	2018	104	Histogram equalization and noise filtering	Randomized Hough transform	4	Fuzzy logic	90.15	94.80	97.80
Koh et al. [49]	2018	2220	NA	Pyramid histogram of visual words and Fisher vector	4 x 4 (grid)	RF	96.05	95.32	96.29
Mohamed et al. [50]	2019	166	Color channel selection and illumination correction	Superpixel feature extraction module	256	SVM	98.63	97.60	92.30
Rehman et al. [51]	2019	110	Bilateral filtering	Intensity-based statistical features and texton-map histogram	2	SVM	99.30	99.40	96.90

^aOnly the best results obtained in each method were left
k-NN classifier, least-squares support vector machine LS-SVM, random forest RF, naive Bayes NB, support vector machine SVM

ISSN: 1475-925X