Scribd
Upload
100 views21 pages

1 Classification of Sleep EEG - GERLA

This document discusses classifying sleep stages using EEG signals. It describes the main sleep stages as wake, REM, and NREM stages 1-4. Features are extracted from EEG signals like power spectral density and spectrograms. The features are normalized and decision rules are applied, such as if EEG 0.5-3Hz is above 85% it is stage S4. Markov models also use transitions between stages. The automated classification achieves 80% accuracy with some errors in detecting stage S1.

Uploaded by

iqbalslimboy
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
100 views21 pages

1 Classification of Sleep EEG - GERLA

This document discusses classifying sleep stages using EEG signals. It describes the main sleep stages as wake, REM, and NREM stages 1-4. Features are extracted from EEG signals like power spectral density and spectrograms. The features are normalized and decision rules are applied, such as if EEG 0.5-3Hz is above 85% it is stage S4. Markov models also use transitions between stages. The automated classification achieves 80% accuracy with some errors in detecting stage S1.

Uploaded by

iqbalslimboy
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 21

Classification of Sleep EEG

Vclav Gerla ([email protected])


Gerstner laboratory, Department of Cybernetics Technick 2, 166 27 Prague, Czech Republic Faculty of Electrical Engineering, Czech Technical University in Prague

- Stages of Sleep - Sleep Disorders - Measuring Sleep in the Laboratory - Brain Wave Frequencies - Artifacts - Sleep stages analysis

Stages of Sleep, Hypnogram


1. Wake (wakefulness, waking stage) 2. REM (Rapid Eye Movements) // dreams 3. NREM 1 (shallow/drowsy sleep) 4. NREM 2 (light sleep) 5. NREM 3 (deepening sleep) 6. NREM 4 (deepest sleep) Hypnogram:

Sleep Disorders
Headaches Insomnia (sleep - -) - difficulty falling asleep - waking up frequently during the night - waking up too early in the morning - unrefreshing sleep
Sleepiness (sleep + +) - fall asleep while driving - concentrating at work, school, or home - have difficulty remembering Restless Legs Syndrome - sensations of discomfort in the legs during periods of inactivity Narcolepsy - sudden and irresistible onsets of sleep during normal waking hours Sleep apnea REM sleep disorders

Proportion of REM/NREM stages


%
40 35 30 25 20 15 10 5 0 3 18 40 70 REM NREM(3+4)

age (years)

The decrease of NREM sleeping is caused partially by decrease of delta waves. (does not meet criteria for delta waves)

Measuring Sleep in the Laboratory

Electroencephalogram (EEG): Measures electrical activity of the brain. Electrooculogram (EOG): Measures eye movements. An electrode placed near the eye will record a change in voltage as the eye moves. Electromyogram (EMG): Measures electrical activity of the muscles. In humans, sleep researchers usually record from under the chin, as this area undergoes dramatic changes during sleep.

EEG signal example


19 EEG signals, EKG signal (+50 Hz artifact)

Brain Wave Frequencies


Delta (0.1 to 3 Hz) deep / dreamless sleep, non-REM sleep Theta (4-8 Hz) connection with creativity, intuition, daydreaming, fantasizing Alpha (8-12 Hz) relaxation, mental work - thinking or calculating Beta (above 12 Hz) normal rhythm, absent or reduced in areas of cortical damage

Binaural Beat Frequencies


Example of frequencies: // sporadic 0.15-0.3 Hz - depression 4.5-6.5 Hz - wakeful dreaming, vivid images 4-8 Hz - dreaming sleep, deep meditation, subconscious mind 5.0-10.0 Hz - relaxation 5.8 Hz - dizziness 7 Hz - increased reaction time 7.83 Hz - earth resonance 8.6-9.8 Hz - induces sleep, tingling sensations 15.0-18.0 Hz - increased mental ability 18 Hz - significant improvements in memory 55 Hz - Tantric yoga

LEFT EAR 70Hz RIGHT EAR 74Hz Binaural Beat 4Hz

Brain Wave Generator: http://www.BWgen.com

Stage Wake

EEG:

- rhythmic alpha waves (8-12Hz) // only if the eyes are closed - beta waves (20-30Hz)

EOG:
EMG:

- eye movement (observation process)


- continual tonically activity of muscles

Stage REM

EEG:

- relatively low voltage - mixed frequency

EOG:
EMG:

- contains rapid eye movements


- tonically suppressed (Sleep Paralysis)

Stage NREM 1(shallow/drowsy sleep)

EEG:

- the absence of alpha activity - Vertex sharp waves

EOG:
EMG:

- slow eye movement


- relatively lower amplitude

Stage NREM 2 (light sleep)

EEG:

- sleep spindles (oscillating with the frequency between 12-15 Hz) - K-complexes (high voltage, sharp rising and sharp falling wave) - relatively low voltage mixed frequency - the absence eye movements - constant tonic activity

EOG: EMG:

Stage NREM 3 (deepening sleep)

EEG:

- consists of high-voltage (>=75uV) - slow delta activity (<=2 Hz) // electrodes Fpz-Cz or Pz-Oz

EOG:

- the absence eye movement - delta waves from EEG


- low tonic activities

EMG:

Stage NREM 4 (deepest sleep)

As NREM 3 + delta activity duration more than 50% for epoch

Artifacts
Muscle artifacts:

Other artifacts: - Eye Flutter, slow and rapid eye movements - ECG artifact - Sweat artifact - Metal contact (touching metal during recording) - Salt Bridge (between two electrodes) - Static electricity artifact - Glossokinetic (movements of tongue)

System Structure
reduce data quantity (speeds up total computing time) divide signal into 1 second segments

compute mean power density in individual frequency bands for each segment

Feature Extraction
Hypnogram (rate by expert)
EEG (Fpz-Cz)

1Hz Power spectral density .

EEG (Pz-Oz)

Spectrogram:

29 Hz

Feature Normalization
The features contain great number of peaks -> normalization

NREM4 stage detection:

Wake stage detection:

Decision Rules
Searching suitable decision rules: - convert all features of all patients to the Weka format. - Weka (http://www.cs.waikato.ac.nz/ml/weka) is a collection of machine learning algorithmus contains tools for datapreprocessing, classification, regression, clustering, association rules and visualization The most significant found rules:
EEG 16-30Hz > 20% WAKE

EEG 0.5-3Hz > 85%

true

S4

false
EEG 0.5-3Hz > 65% S3

EEG 13-15Hz < 15% and EOG 0.15-1.2Hz > 50%

REM

EEG 13-15Hz > 20%

true

S2

false
EEG 13-15Hz > 10% S1

Markov models (utilization of time-dependence)


Aplication to segments which: - all rules are false - more rules are true

Markov models use - contextual information in EEG signa - approximate knowledge of transitions probability

Results
- Final classification accuracy approximately 80% - Problem with detection S1 stage

You might also like