Table 1 Summary of the active electrode classification performance.

From: Unsupervised machine-learning classification of electrophysiologically active electrodes during human cognitive task performance

 

GC

GC1

SS

IP

\({{\rm{IP}}}_{{{\rm{\gamma }}}_{{\rm{h}}}}\)

All

Thresholding step

Sensitivity

87.83 (4.17)

94.60 (13.48)

16.90 (1.57)

93.39 (0.14)

96.97 (0.01)

93.19 (0.01)

100

Specificity

65.57 (7.28)

79.36 (1.78)

60.51 (6.78)

84.48 (0.06)

92.93 (0.02)

83.36 (0.02)

47.15

AUC

0.855 (0.02)

0.909 (0.123)

0.362 (0.080)

0.945 (0.001)

0.979 (0.001)

0.938 (0.001)

0.923

  1. Classification performance of various features, metrics and combination of metrics for finding active electrodes are provided. They are also compared with the “thresholding step”4 used in identifying the “ground truth” active electrodes. The thresholding step involved finding electrodes for which the standard deviation of the high-gamma power was greater than 0.05. The mean sensitivity, specificity and AUC values are averaged over 100 runs of GMM clustering and provided with a standard deviation (in parentheses). List of abbreviations: GC = Gamma Consistency metric; GC1 = first feature of GC metric; SS = Smoothness Score metric; IP = Induced Power metric; \({{\rm{IP}}}_{{{\rm{\gamma }}}_{{\rm{h}}}}\)= induced power in high-gamma feature; and All = all metrics combined.
Back to article page