Skip to main content

Table 1 Detailed comparison of the two classifiers

From: Detection of motor execution using a hybrid fNIRS-biosignal BCI: a feasibility study

 

Subject

1

2

3

4

5

62

7

Average

NIRS only

Acc. ± SD [%]

(75.0 ± 4.6 †)

74.8 ± 5.2 †

85.5 ± 3.1 †

91.7 ± 1.3 †

71.7 ± 5.4 †

90.3 ± 4.5 †

62.4 ± 6.9

79.4 ± 11.7

       

(89.7 ± 3.5 †)

 

(78.7 ± 10.6)

 

Sens. ± SD [%]

(73.3 ± 5.4)

72.3 ± 5.7

85.7 ± 2.9

88.7 ± 1.5

72.8 ± 6.7

87.2 ± 7.6

62.8 ± 11.6

78.3 ± 10.5

       

(89.9 ± 6.3)

 

(77.9 ± 10.2)

 

Spec. ± SD [%]

(76.7 ± 4.9)

77.2 ± 7.6

85.2 ± 5.6

94.8 ± 2.7

70.6 ± 5.1

93.5 ± 4.2

61.9 ± 6.8

80.5 ± 13.0

       

(89.6 ± 5.8)

 

(79.4 ± 11.3)

 

N s

(1)

4

1

1

1

3 (3)

2

 
 

N m

(1)

1

1

2

1

1 (2)

1

 

Combined

Acc. ± SD [%]

 

81.7 ± 4.1 †

87.7 ± 3.0 †

98.8 ± 0.8 †

89.2 ± 4.5 †

94.2 ± 1.2 †

79.6 ± 3.2 †

88.5 ± 7.3

 

Sens. ± SD [%]

 

82.1 ± 4.8

85.6 ± 3.8

99.6 ± 1.2

84.7 ± 8.3

93.8 ± 0.0

77.3 ± 5.1

87.2 ± 8.1

 

Spec. ± SD [%]

 

81.3 ± 7.1

89.8 ± 4.3

98.0 ± 1.9

93.7 ± 5.4

94.6 ± 2.4

81.9 ± 3.3

89.9 ± 6.9

 

N s

 

5

3

5

3

1

3

 
 

N m

 

1

2

2

2

1

2

 
 

Acc. Gain [%]

 

9.2

2.6

7.7

24.4

4.3

27.6

12.6 ± 10.7

 

Sens. Gain [%]

 

13.6

-0.1

12.3

16.3

7.6

23.1

12.1 ± 7.9

 

Spec. Gain [%]

 

5.3

5.4

3.4

32.7

1.2

32.3

13.4 ± 14.9

 

p-value1 Acc.

 

0.016 ∗

0.129

< 0.001 ∗∗

0.001 ∗∗

0.051

0.001 ∗∗

0.02 ∗

 

p-value1 Sens.

 

0.003 ∗∗

0.988

< 0.001 ∗∗

0.028 ∗

0.062

0.034 ∗

0.008 ∗∗

 

p-value1 Spec.

 

0.349

0.089

0.023 ∗

< 0.001 ∗∗

0.535

< 0.001 ∗∗

0.062

 

N T r i a l s

(60)

46

48

60

46

30 (60)

58

 
  1. Results of the binary classification of solely fNIRS data (upper part) and the combination of fNIRS data and biosignals (lower part). Reported are mean accuracy (Acc.), sensitivity (Sens.) and specificity (Spec.) with SD across 7 complete cross-validation runs for each subject. Further, average values across all subjects are given. For subject 6, session 2 and both sessions of subject 7, biosignals were not available. Numbers in parentheses are shown for completeness and were not used in the comparison between the two types of decoders. N s : Number of states in the HMM. N m : Number of Gaussians used to model the HMM’s emission probabilities. Gain: Relative change with respect to the fNIRS only case. N T r i a l s : total number of trials used for analysis (rest and active). 1: Paired t-tests on differences between the fNIRS only and the combined decoders on individual level (difference on group level for averages). 2: In brackets are the results when considering both sessions for the fNIRS only decoder. ∗ : p < 0.05. ∗ ∗ : p < 0.01. † : Accuracy significantly above chance.