Table 2.
Type | Dataset | Records | Wake | N1 | N2 | N3 | REM | Mean |
---|---|---|---|---|---|---|---|---|
Internal - Train/Test | ABC | 20 | 0.87 | 0.53 | 0.84 | 0.72 | 0.90 | 0.77 |
CCSHS | 78 | 0.93 | 0.63 | 0.91 | 0.88 | 0.93 | 0.85 | |
CFS | 92 | 0.93 | 0.52 | 0.89 | 0.84 | 0.91 | 0.82 | |
CHAT | 128 | 0.93 | 0.64 | 0.87 | 0.90 | 0.90 | 0.85 | |
DCSM | 39 | 0.97 | 0.48 | 0.86 | 0.83 | 0.89 | 0.81 | |
HPAP | 36 | 0.91 | 0.48 | 0.84 | 0.78 | 0.90 | 0.78 | |
MESA | 100 | 0.92 | 0.59 | 0.87 | 0.65 | 0.90 | 0.79 | |
MROS | 134 | 0.93 | 0.46 | 0.87 | 0.68 | 0.88 | 0.77 | |
PHYS | 100 | 0.84 | 0.60 | 0.84 | 0.81 | 0.87 | 0.79 | |
SEDF-SC | 23 | 0.93 | 0.57 | 0.86 | 0.71 | 0.88 | 0.79 | |
SEDF-ST | 8 | 0.80 | 0.58 | 0.88 | 0.64 | 0.91 | 0.76 | |
SHHS | 140 | 0.93 | 0.51 | 0.87 | 0.76 | 0.92 | 0.80 | |
SOF | 68 | 0.93 | 0.45 | 0.86 | 0.77 | 0.92 | 0.78 | |
Hold-Out | ISRUC-SG1 | 100 | 0.89 | 0.52 | 0.79 | 0.77 | 0.88 | 0.77 |
ISRUC-SG2 | 16 | 0.85 | 0.49 | 0.78 | 0.83 | 0.86 | 0.76 | |
ISRUC-SG3 | 10 | 0.90 | 0.55 | 0.78 | 0.74 | 0.85 | 0.77 | |
MASS-C1 | 53 | 0.94 | 0.41 | 0.81 | 0.61 | 0.88 | 0.73 | |
MASS-C3 | 62 | 0.93 | 0.54 | 0.87 | 0.75 | 0.91 | 0.80 | |
SVUH | 25 | 0.80 | 0.37 | 0.81 | 0.78 | 0.88 | 0.73 | |
DOD-H | 25 | 0.91 | 0.60 | 0.87 | 0.79 | 0.94 | 0.82 | |
DOD-O | 55 | 0.90 | 0.52 | 0.86 | 0.74 | 0.92 | 0.79 | |
Mean (weighted) | 0.91 | 0.53 | 0.86 | 0.77 | 0.90 | 0.79 | ||
STD (weighted) | 0.03 | 0.07 | 0.03 | 0.08 | 0.02 | 0.03 |
For each record in each dataset, U-Sleep generated a hynogram using all possible combinations of 1 EEG and 1 EOG channel. Results reported here are from the majority voted hypnograms across all such combinations as described in the Methods section. We refer to Supplementary Tables 4–25 for per-channel results. Here we report the global F1 scores across all subjects in each dataset. The reported Mean (weighted) and STD (weighted) statistics are computed across datasets in each column weighted by the number of PSG records in each row.