Table 3.

Average classification accuracy of the automatic classification system after random resampling.

Feature	Method	Average performance on the testing set
		Average sensitivity	P-value	Average specificity	P-value
Global geometrical features	PCA + SVM	23.8 ± 15.2%	0.56	80.8 ± 8.1%	0.63
Global texture features	PCA + SVM	44.0 ± 15.7%	0.01*	87.5 ± 5.7%	1.00
Fusion of local features	AdaBoost	67.6 ± 14.5%	0.57	87.9 ± 4.5%	0.47

SVM, support vector machine.

PCA, principal component analysis.

The P-value was used for comparing the results of the specific sampling of participants in Table 2 with the average results of 50 times of resampling, using the t-test. *P < 0.05.