Table 1.

Summary of the state-of-the-art methods for DME detection

References	Diseases			Data size	Pre-processing				Features	Representation	Classifier	Evaluation	Results
	AMD	DME	Normal		De-noise	Flatten	Aligning	Cropping
Srinivansan et al. [10]	$✓$	$✓$	$✓$	45	$✓$	$✓$		$✓$	HoG		Linear-SVM	ACC	86.7%,100%,100%
Venhuizen et al. [11]	$✓$		$✓$	384					Texton	BoW, PCA	RF	AUC	0.984
Liu et al. [12]	$✓$	$✓$	$✓$	326		$✓$	$✓$		Edge, LBP	PCA	SVM-RBF	AUC	0.93
Lemaître et al. [13]		$✓$	$✓$	32	$✓$				LBP, LBP-TOP	PCA, BoW, Histogram	RF	SE, SP	87.5%, 75%
Sankar et al. [15]		$✓$	$✓$	32	$✓$	$✓$		$✓$	Pixel-intensities	PCA	Mahalanobis-distance to GMM	SE, SP	80%, 93%
Albarrak et al. [16]	$✓$		$✓$	140	$✓$	$✓$			LBP-TOP, HoG	PCA	Bayesian network	SE, SP	92.4%, 90.5%