A methodology for time-frequency image processing applied to the classification of non-stationary multichannel signals using instantaneous frequency descriptors with application to newborn EEG signals

EURASIP Journal on Advances in Signal Processing

Table 5 Confusion matrices for EEG classification using T-F features set {F_{S 1}, F_{S 2}, ..., F_{S 10}, F_{I 1}, ..., F_{I 5}} extracted from the TFD with SVM classifier and Neural Network-based classifier.

			Classifier outputs			Statistical parameters (%)
TFD	Class	Total	H	F	S	Sensitivity	Specificity	Total accuracy
Multi-class SVM classifier
	H	100	98 (98)	1 (2)	1 (0)	98 (98)	92 (90.5)
MBD	F	100	6 (5)	89 (92)	5 (3)	89 (92)	96.5 (93.5)	94 (93)
	S	100	1 (2)	4 (9)	95 (89)	95 (89)	93.5 (95)
	H	100	97 (99)	1 (1)	1 (0)	97 (99)	93 (91)
SPEC	F	100	5 (5)	89 (92)	6 (3)	89 (92)	97 (94.5)	94.33 (93.67)
	S	100	1 (2)	2 (7)	97 (90)	97 (90)	93 (95.5)
	H	100	99 (98)	0 (2)	1 (0)	99 (98)	93.5 (90)
SWVD	F	100	3 (5)	93 (91)	4 (4)	93 (91)	96.5 (93.5)	95.33 (92.67)
	S	100	0 (2)	4 (9)	94 (89)	94 (89)	96 (94.5)
	H	100	99 (98)	1 (2)	0 (0)	99 (98)	90.5 (89.5)
GKD	F	100	5 (5)	88 (91)	7 (4)	88 (91)	96 (93)	93.33 (92.33)
	S	100	1 (2)	5 (9)	93 (88)	93 (88)	93.5 (94.5)
	H	100	98 (98)	1 (2)	1 (0)	98 (98)	91.5 (90)
WVD	F	100	4 (8)	89 (87)	7 (5)	89 (87)	96.5 (95.5)	93.67 (92.67)
	S	100	2 (2)	3 (4)	94 (93)	94 (93)	93.5 (92.5)
Neural network-based classifier
	H	100	99 (89)	1 (11)	0 (0)	99 (89)	90.5 (93.5)
MBD	F	100	5 (8)	89 (88)	6 (0)	89 (88)	95.5 (94)	93.33 (92)
	S	100	0 (0)	8 (1)	92 (99)	92 (99)	94 (88.5)
	H	100	99 (91)	1 (7)	0 (0)	99 (91)	94 (95.5)
SPEC	F	100	2 (4)	93 (92)	5 (4)	93 (92)	97 (95)	95.67 (94)
	S	100	2 (0)	3 (1)	95 (99)	95 (99)	96 (91.5)
	H	100	98 (94)	0 (4)	0 (0)	98 (94)	93 (92)
SWVD	F	100	7 (7)	88 (89)	5 (4)	88 (89)	98 (94.5)	94.67 (92.67)
	S	100	0 (0)	2 (5)	98 (95)	98 (95)	93 (91.5)
	H	100	99 (98)	1 (0)	0 (0)	99 (98)	92 (91)
GKD	F	100	4 (8)	88 (85)	8 (7)	88 (85)	97.5 (97.5)	94.33 (93.33)
	S	100	1 (0)	3 (3)	96 (97)	96 (97)	93.5 (91.5)
	H	100	98 (91)	2 (7)	0 (0)	98 (91)	91.5 (92.5)
WVD	F	100	6 (7)	88 (88)	6 (5)	88 (88)	96.5 (94)	93.67 (92)
	S	100	0 (0)	4 (3)	95 (97)	95 (97)	93 (89.5)

The results, for each TFD, are given using two classifiers: multi-class SVM classifier and neural network-based classifier, with a real EEG database organized in three classes H, F and S. The T-F features are extracted from EEG segments of length 11.8 s (N = 2,048 samples). The results between parentheses are the classification results using the ten signal-related features {F_{S 1}, F_{S 2}, ..., F_{S 10}}. Sensitivity and specificity of each classifier and for each particular class as well as its total accuracy are also given.