Table 5.

Different methods for detection of exudates.

Algorithm	Image processing techniques	Database	Color space	Sensitivity	Specificity	Accuracy
Ram and Sivaswamy [41]	Clustering-based method and color space features	DIARETDB1	RGB, CIE L ∗ u ∗ v ∗, HSV, HIS	71.96%	—	89.7%
Soares et al. [42]	Morphological operators and adaptive thresholding	DIARETDB1	Green channel	97.49%	99.95%	99.91%
Jayakumari and Santhanam [43]	Energy minimization method using echo state neural network	Private Hospital	—	90%	—	—
Karegowda et al. [44]	KNNFP and WKNNFP classifiers	DIARETDB1	HIS	—	—	97.50% WKNNFP 96.67% KNNFP
Amel et al. [45]	Combine the K-means  clustering algorithm and mathematical morphology	Ophthalmologic Images	CIELab	95.92%	99.78%	99.70%
Rokade and Manza [46]	Haar wavelets transformation, KNN classifier	MISP DIRETDB0, DIRETDB1, STARE	Green channel	37.14%, 21.87%, 12.50%, 25.47%	—	—
Kayal and Banerjee [2]	Median filtering, image thresholding	DIARETDB0 DIARETDB1	Gray scale	97.25%	96.85%	—
Jaya et al. [47]	Morphological operations, Circular Hough transform, Fuzzy support vector machine	Private Hospital	—	94.1%	90.0%	—
Rozlan et al. [48]	Morphology operation, columnwise neighborhoods operation	Sungai Buloh Hospital	Green channel	—	—	60%
Soman and Ravi [49]	Circular Hough transform and bit plane slicing, morphological operations	Standard Diabetic Retinopathy	Green channel	0.9362	—	88%
Annunziata et al. [50]	Multiple scale Hessian approach	STARE HRF	Green channel	—	—	95.62% 95.81%
Van Grinsven et al. [51]	Bag of Words approach	Messidor EUGENDA	HSV, YCbCr	—	—	0.90 AUC
Kaur and Mittal [52]	Dynamic region growing method	SGHS hospital	Gray scale	—	—	98.65%