Sleep Stage Identification

1
Sleep Stage Identification

• Jessie Y. Shen
• February 17, 2004.

2
Objective

• How Sleep Stage Identification fits into the
  Narcolepsy Project?
• Manual Sleep Staging Overview
• Review on Previous Automation Attempts
• Problems, Issues, and Solutions
• Work in Progress

3
Narcolepsy Project
Detection Algorithm
4
Detection Algorithm

• Goal
• Correctly identify the conscious level of subject
  while awake and the sleep stage while sleeping.
• Method
• Quantify brain activity
• Sleep staging automation

Sleep staging automation
5
Manual Sleep Staging

• Standard set by Rechtschaffen and Kales
• Awake, NREM I to IV, REM, MT
• Polysomnogram
• EEG
• EOG
• EMG

6
EEG
7
Previous Research

• Shimada 1998 NN at 80
• 1st ANN for EEG to characteristic waves
• 2nd ANN for characteristic waves to stage
• 3rd ANN for contextual correction
• Oropesa 1999 Wavelet NN at 77.6
• Flexer 2000 HMM at 80

8
FYDP
9
5 Issues

• 1. Stages often changes during epoch.
• 2. Changes are gradual.
• 3. Some features are only present some of the
  time.
• 4. Sleep staging rules are not intuitive.
• 5. Medical experts have an inter-observer
  agreement of less than 90.

10
Solutions

• Mimic medical experts actions.
• 1. Extract Feature Information (Activity Band
  Info, Characteristic Wave Info, and Other Info)
• 2. Establish Contextual Information (last stage,
  the duration in the current stage, etc.)
• 3. Determine Sleep Stage by processing the
  feature and contextual information with a
  complete rule based expert system.

11
Components
12
Extract Feature Information

• Mixed frequency activity
• Spectrogram
• Identify Awake and REM from other stages

13
Extract Feature Information
Awake
REM
sensitivity 93.51 specificity 94.60
14
Extract Feature Information
III
IV

• Delta band content
• Scalogram
• Differentiate NREM II, III, and IV

Stage II(90.23, 86.06), Stage III(98.60,
96.81), Stage IV(99.53, 98.03)
15
Establish Contextual Information
Standard Hypnogram
For Healthy Young Adults
16
Establish Contextual Information
17
Establish Contextual Information
Awake
Stage I
Stage II
Stage III
Stage IV
REM
18
Work in Progress

• Extract Feature Information
• Sleep spindles, K-complex, Saw-tooth waves, etc.
• Establish Contextual Information
• Consider duration of each stage, number of
  elapsed cycles, etc.
• Build Rule-based Inference System

19
Thank You!