Table 2 SS and KC detection performance.

From: A robust deep learning detector for sleep spindles and K-complexes: towards population norms

Dataset

Detector

F1-score (%)

mIoU (%)

Mean ± SD

p-value

Mean ± SD

p-value

MASS2-SS-E1 (15 subjects)

SEED

80.8 ± 2.1

 

84.8 ± 1.2

 

DOSED

76.8 ± 2.9

 < 0.001

74.7 ± 2.1

 < 0.001

A7

73.0 ± 3.4

 < 0.001

73.9 ± 1.0

 < 0.001

MASS2-SS-E1 (19 subjects)

SEED

80.5 ± 2.1

 

84.7 ± 1.0

 

SpindleU-Net

80.3 ± 1.9*

0.848

73.5

-

MASS2-SS-E2 (15 subjects)

SEED

86.1 ± 2.0

 

78.7 ± 1.1

 

DOSED

82.5 ± 2.5

 < 0.001

73.1 ± 1.1

 < 0.001

A7

74.9 ± 2.8

 < 0.001

74.7 ± 1.1

 < 0.001

SpindleU-Net

85.4 ± 2.7*

0.615

N.A

-

SpindleNet

83.0 ± 2.0

0.020

N.A

-

MODA

SEED

81.8 ± 1.4

 

83.4 ± 0.5

 

DOSED

77.5 ± 1.7

 < 0.001

71.4 ± 1.1

 < 0.001

A7

73.3 ± 1.9

 < 0.001

71.0 ± 0.9

 < 0.001

MASS2-KC (15 subjects)

SEED

83.7 ± 1.5

 

90.6 ± 0.6

 

DOSED

78.1 ± 2.2

 < 0.001

72.3 ± 1.4

 < 0.001

Spinky

63.1 ± 3.8

 < 0.001

41.2 ± 1.6

 < 0.001

MASS2-KC (19 subjects)

SEED

83.6 ± 1.7

 

90.4 ± 0.4

 

DKL-KC

78.0 ± 2.0

 < 0.001

N.A

-

  1. mIoU: mean Intersection over Union; N.A.: not available. Metrics of SEED (proposed detector), DOSED, A7 and Spinky were obtained using open-source implementations, whereas metrics of SpindleU-Net, SpindleNet and DKL-KC were obtained from their original publications. P-values are defined against SEED’s performance. * These standard deviations between partitions are not reported in the original publication; these estimations are based on the reported by-subject F1-score (see Methods).
Back to article page