Slide 05 · Data

Chosen Data

PhysioNet / CinC 2018 PSG dataset.

Why This Dataset

  • Large public PSG benchmark suited for controlled comparison.
  • Multi-channel biosignals and event annotations, including apnea-related labels.
  • Existing anonymized data, so we do not need to collect new clinical recordings.

Key point

One common public dataset lets us compare implementations under the same conditions.

Dataset Overview

Full-night PSG records with synchronized signals and annotations.

SourceMassachusetts General Hospital sleep lab
Total cohort
0
full-night PSG recordings
Coverage
13 signals · 3 label groups
F3-M2
Frontal EEG
F4-M1
Frontal EEG
C3-M2
Central EEG
C4-M1
Central EEG
O1-M2
Occipital EEG
O2-M1
Occipital EEG
EOG-L
Eye movement
CHIN EMG
Chin muscle
ABD
Abdominal effort
CHEST
Thoracic effort
AIRFLOW
Airflow
SaO2
Oxygen saturation
ECG
Heart rhythm