Table 4.

Classification performance comparisons of ScSPM classifier with different preprocessing methods (300 AMD, 300 DME, and 300 NOR images as the training set).

Preprocessing	Classes	Accuracy (%)	Sensitivity (%)	Specificity (%)
MD method27	AMD	$96.85\pm 0.42$	$96.15\pm 0.92$	$97.20\pm 0.75$
	DME	$95.78\pm 0.47$	$91.17\pm 1.66$	$98.08\pm 0.64$
	NOR	$97.61\pm 0.37$	$98.04\pm 0.66$	$97.40\pm 0.43$
FCN method	AMD	$98.86\pm 0.16$	$98.67\pm 0.64$	$98.95\pm 0.31$
	DME	$98.43\pm 0.26$	$96.69\pm 1.15$	$99.29\pm 0.27$
	NOR	$98.77\pm 0.30$	$98.72\pm 0.34$	$98.79\pm 0.34$