Classification of SD-OCT Volumes Using Local Binary Patterns: Experimental Validation for DME Detection

Table 6

Experiment —optimum number of words for each configuration as a result of LR classification, for high-level feature extraction of global and local-LBP, and local-LBP-TOP features with different preprocessing. The preprocessing includes NF, F, and F+A. The achieved performance is indicated in terms of ACC, F1, SE, and SP.

Features	Preprocessing
		ACC%	F1%	SE%	SP%	Number of words	ACC%	F1%	SE%	SP%	Number of words	ACC%	F1%	SE%	SP%	Number of words
Global-LBP	NF	81.2	78.5	68.7	93.7	500	62.5	58.0	56.2	62.5	80	62.5	62.5	62.5	62.5	80
	F	71.9	71.0	68.7	75.0	400	68.7	66.7	62.5	75.0	300	68.7	66.7	62.5	75.0	300
	F+A	71.9	71.0	68.7	75.0	500	71.9	71.0	68.7	75.0	200	75.0	68.7	68.7	68.7	500
Local-LBP	NF	75.0	75.0	75.0	75.0	70	65.6	64.5	62.5	68.7	90	62.5	60.0	56.2	68.7	30
	F	75.0	73.3	68.7	81.2	30	71.8	61.0	68.7	75.0	70	62.5	62.5	62.5	62.5	100
	F+A	75.0	69.0	62.5	81.2	40	71.9	71.0	68.7	75.0	200	68.7	66.7	68.7	62.5	10
Local-LBP-TOP	NF	68.7	68.7	68.7	68.7	400	75.0	75.0	75.0	75.0	500	71.9	71.0	68.7	75.0	60
	F	68.7	68.7	68.7	68.7	300	68.7	66.7	62.5	75.0	50	75.0	76.5	81.2	68.7	80
	F+A	75.0	73.3	68.7	81.2	100	75.0	73.3	68.7	81.2	90	75.0	69.0	62.5	81.2	70