Skip to main content

Table 2 Bi-class accuracy of evaluated CNN training strategies, using the CWT-extracted vectors and either dropping strategy: CWT* with sensorimotor electrodes and CWT** with thresholding. In all compared cases, both sub-bands (\(\mu \) and \(\beta \)) are included and the CNN parameters are tuned individually

From: CNN-based framework using spatial dropping for enhanced interpretation of neural activity in motor imagery classification

Subjects

[41]

[42]

CWT

\(\kappa \)

CWT*

\(\kappa \)

CWT**

\(\kappa \)

A03T

88.2

91.7

95.0 ± 4.6

0.67

96.4 ± 4.8

0.92

95.0 ± 4.6

0.90

A09T

82.7

90.9

94.8 ± 4.2

0.68

93.1 ± 6.5

0.86

94.0 ± 6.3

0.88

A08T

91.8

92.3

\(\underline{94.0\pm 4.6}\)

0.90

97.0±3.6

0.94

94.7 ± 4.9

0.89

A06T

65.7

78.5

86.7 ± 7.2

0.71

84.9 ± 9.0

0.69

86.7 ± 7.2

0.73

A07T

51.7

86.5

86.4 ± 6.7

0.58

\(\underline{81.9\pm 6.2}\)

0.64

85.6 ± 9.3

0.71

A04T

53.9

80.4

85.4 ± 7.3

0.73

86.1 ± 7.5

0.72

87.6 ± 5.0

0.75

A02T

63.9

68.4

83.8 ± 6.5

0.73

80.3 ± 6.2

0.61

81.7 ± 4.8

0.63

A01T

79.4

87.8

83.4 ± 5.5

0.88

81.1 ± 5.0

0.62

83.2 ± 3.9

0.66

A05T

54.9

88.9

79.1 ± 4.8

0.90

78.3 ± 7.4

0.57

76.7 ± 6.9

0.53

Average

70.2

85.0 ± 7.4

87.6 ± 5.7

0.75

86.6 ± 6.2

0.73

87.4 ± 5.7

0.74