A Signal Processing Approach to the Detection of Obstructive Sleep - - PowerPoint PPT Presentation

A Signal Processing Approach to the Detection of Obstructive Sleep Apnea

NRP: EEE24A

Jovyn Tan Li Shyan Hwa Chong Institution

OSA: Obstructive Sleep Apnea

2

❏ Sleep disorder where breathing stops for ❏ at least 10 seconds, ❏ more than 5 times/hour

1 in 3 Singaporeans 1 billion people worldwide Heart disease High blood pressure Daytime fatigue Health Complications

3

Electromyogram Electro-oculogram Electromyogram Electrocardiogram

Current Form of Diagnosis Polysomnography (PSG)

4

Current Form of Diagnosis Polysomnography (PSG) Apnea Hypopnea Index (AHI) Problems Distorts OSA condition Manual data analysis Cannot analyse all data collected No differentiation

• f event severity

Aims of Research

An automated system without full PSG

1

Automated 3-class system

(Healthy, Hypopnea, Apnea) 2

Aims & Objectives

3-class Fisher’s Ratio

3

5

Electromyogram Electro-oculogram Electromyogram Electrocardiogram

OSA Diagnosis

Oro-nasal thermistor Respiratory effort belts

Proposed method

Data

6

14 channels D a t a

• b

s e r v a t i

• n

s S a m p l i n g r a t e : 6 4 H z 6 4 H z * 6 s * 6 m i n * 6 . 2 h = 1 4 2 8 4 8 d a t a p

• i

n t s

Methodology

7

Signals extracted from data

• Oro-nasal airflow
• Rib cage movement
• Abdomen movement

Normalisation Preparation of data for machine learning

Methodology

8

Signals extracted from data

• Oro-nasal airflow
• Rib cage movement
• Abdomen movement

Feature Extraction Segmentation

windows of 1024 data points

Normalisation Preparation of data for machine learning

1024 points = 1 window

Feature Extraction

9

❏ 15 features extracted from each signal ❏ e.g. mean peak prominence, number of peaks ❏ 5 peak processing thresholds

Preparation of data for machine learning

Methodology

10

Signals extracted from data

• Oro-nasal airflow
• Rib cage movement
• Abdomen movement

By box plot analysis Feature Selection By 3-class Fisher’s ratio

Conceptualising 3-class Fisher’s Ratio

Segmentation

windows of 1024 data points

Normalisation Feature Extraction

Fisher’s Ratio

11

❏ Measures discriminating power of a variable ❏ 2-Class FR:

3-Class Fisher’s Ratio

12

Feature Selection by Box Plots

13

Suitable feature Unsuitable feature

Preparation of data for machine learning

Methodology

14

Signals extracted from data

• Oro-nasal airflow
• Rib cage movement
• Abdomen movement

By box plot analysis Feature Selection By 3-class Fisher’s ratio Recursive Feature Elimination Feature Elimination Principal Component Analysis (PCA) Support Vector Machines (SVM) using Matlab

Conceptualising 3-class Fisher’s Ratio

Segmentation

windows of 1024 data points

Normalisation Feature Extraction

Classification Results

15

2 Classes 3 Classes

Results [2 classes]

16

Highest Accuracy: 94.8% Sensitivity: 96% Specificity: 93%

Cubic kernel 28/28 features No PCA Selection by FR

Healthy

Healthy Apnea

Apnea True Class Predicted Class

Ribcage and Abdomen Movements Oro-nasal Airflow

Standard Deviation of Peak Prominence of 2 Signals

Results [3 class]

17

Highest Accuracy: 83.4%

Medium Gaussian kernel 19/28 features Selection by box plots PCA enabled

Oro-nasal Airflow

Healthy Healthy Severe Apnea Severe Apnea Mild Apnea Mild Apnea True Class Predicted Class

Standard Deviation of Peak Prominence of 2 Signals

Ribcage and Abdomen Movements

Conclusion

18

An automated system without full PSG

1

An automated 3-class system (Healthy, Hypopnea, Apnea)

3

3-class Fisher’s Ratio

2