Introduction
Related works
Sleep stage classification
Domain generalization
Data manipulation
Representation learning
Learning strategies
Gaussian transformations for EEG signals
Methods
Problem formulation
Cluster-then-label algorithm
TinySleepNet classifier plus k-means clustering
LDA classifier plus GMM clustering, part I: multitaper spectrogram
LDA classifier plus GMM clustering, part II: Gaussian transformation
Assembled-fixed transformation
Neural network based gaussian transformation
Results
Cluster-then-label using TinySleepNet and k-means
Fold no. | 1 | 2 | 3 | 4 | 5 | 6 |
---|---|---|---|---|---|---|
Before retraining | 74.6 | 74.2 | 82.4 | 84.9 | 89.6 | 83.4 |
After retraining | 77.2 | 78.1 | 84.7 | 87.3 | 90.4 | 83.8 |
Fold no. | 1 | 2 | 3 | 4 | 5 | 6 |
---|---|---|---|---|---|---|
Before retraining | 55.1 | 49.7 | 68.0 | 68.4 | 73.6 | 69.5 |
After retraining | 58.4 | 54.8 | 69.5 | 70.7 | 73.6 | 69.5 |
Cluster-then-label using LDA and GMM
Fold no. | 1 | 2 | 3 | 4 | 5 | 6 |
---|---|---|---|---|---|---|
Before retraining | 0.57 | 0.72 | 0.63 | 0.83 | 0.62 | 0.65 |
After retraining | 0.72 | 0.73 | 0.76 | 0.89 | 0.60 | 0.75 |
Fold no. | 1 | 2 | 3 | 4 | 5 | 6 |
---|---|---|---|---|---|---|
Before retraining | 0.51 | 0.62 | 0.57 | 0.54 | 0.46 | 0.70 |
After retraining | 0.65 | 0.64 | 0.73 | 0.63 | 0.50 | 0.80 |
Assembled-fixed Gaussian transformation
Band | Beta | Alpha | Theta | Delta |
---|---|---|---|---|
Before transformation | ||||
Wake | 137.09 | 252.17 | 93.45 | 178.01 |
N1 | 4.15 | 6.39 | 2.14 | 4.86 |
N2 | 1.81 | 30.19 | 0.48 | 10.01 |
N3 | 2.49 | 2.70 | 1.16 | 0.22 |
REM | 5.92 | 4.09 | 2.07 | 1.47 |
After transformation | ||||
Wake | 156.99 | 223.13 | 52.09 | 49.78 |
N1 | 2.26 | 3.96 | 2.87 | 2.40 |
N2 | 0.73 | 18.64 | 1.26 | 4.09 |
N3 | 1.19 | 2.44 | 1.54 | 0.38 |
REM | 3.67 | 7.44 | 1.27 | 0.55 |
Sleep stages | Wake | N1 | N2 | N3 | REM |
---|---|---|---|---|---|
Before transformation | 20.13 | 1.16 | 6.91 | 0.42 | 0.75 |
After transformation | 26.70 | 0.82 | 1.44 | 0.28 | 0.21 |