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

Table 8

Experiment —classification results obtained from low-level representation of global LBP and LBP-TOP features with different preprocessing. Preprocessing steps include NF, F, and F+A. Different classifiers such as RF, GB, SVM, and -NN are used. The most relevant configurations are indicated with fonts. The configurations whose performance declines with additional preprocessing are indicated with italic font while those with the opposite behavior are indicated with underline font. The highest results which are specified in Table 5 are highlighted in bold.

(a)  

Features	Preprocessing	-NN						SVM
		SE%	SP%	SE%	SP%	SE%	SP%	SE%	SP%	SE%	SP%	SE%	SP%
		Global-LBP	NF	37.5	50.0	25.0	50.0	37.5	68.7	56.2	62.5	56.2	43.7	56.2	68.7
F	62.5		50.0	56.2	75.0	62.5	68.7	75.0	68.7	62.5	62.5	62.5	68.7
FA	56.2		50.0	56.2	75.0	62.5	68.7	75.0	68.7	62.5	62.5	62.5	68.7
Global-LBP-TOP	NF	31.2	93.7	37.5	100.0	37.5	81.2	62.5	75.0	62.5	93.7	56.2	87.5
	F	50.0	56.2	56.2	75.0	56.2	62.5	68.7	75.0	43.7	68.7	68.7	56.2
	F+A	75.0	43.7	56.2	43.7	68.7	50.0	68.7	62.5	62.5	56.2	56.2	68.7

(b)  

Features	Preprocessing	RF						GB
		8		16		24		8		16		24
		SE%	SP%	SE%	SP%	SE%	SP%	SE%	SP%	SE%	SP%	SE%	SP%
Global-LBP	NF	43.7	62.5	43.7	62.5	56.2	75	43.7	43.7	43.7	37.5	37.5	31.25
	F	56.2	56.2	68.7	62.5	62.5	68.7	25	56.2	50.0	43.7	25.0	43.7
	F+A	65.2	56.2	50.0	50.0	56.2	68.7	43.75	62.5	62.5	50.0	31.2	31.2
Global-LBP-TOP	NF	56.2	68.7	68.7	87.5	68.7	81.2	68.7	68.7	75.0	50.0	56.2	43.7
	F	56.2	62.5	81.2	68.7	81.2	81.2	56.2	62.5	62.5	68.7	68.7	81.2
	F+A	68.7	62.5	75.0	68.7	75.0	81.2	56.2	43.7	62.5	62.5	75.0	75.0