Sleep apnea Is a sleep disorder of shallow breathing or complete breathing stops that takes place during sleep and can last from few seconds to several minutes.
Abstract
Sleep apnea Is a sleep disorder of shallow breathing or complete breathing stops that takes place during sleep and can last from few seconds to several minutes.
While sleep apnea can affect all people regardless of sex or age it has a big impact on children development and health.
Sleep apnea is diagnosed in sleep study labs in which the patient would be wired-up to several monitors, and it is more difficult with children as they will have to sleep outside their home environment.
Sleep apnea can be treated by surgical operation or medication.
Motivation
- Provide a low cost and patient friendly solution for respiratory rate monitoring and sleep apnea detection using 3D imaging technologies and image & signal processing techniques
- Using this solution, sleep apnea diagnoses would be accessible in the home environment as 3D cameras are accessible there
Flowchart
Stage-1:Capturing 3D movie and pre-processing
Stage-2:Frames segmenting and respiratory motion detection
1. Divide each frame to a set of N non-overlapping blocks.
2. For each block calculate the power spectral density estimate.
3. For each block if the maximum of the PSD in the frequency window of [0.2, 2] Hz is greater than a certain threshold declare this block as a respiratory motion containing block.
4. Repeat steps 1-3 on the blocks declared in step 3 until we get a good set of signal blocks that’s characterized by a good SNR and a low data-consumption footprint.
Stage-2 Demonstration:
Non-signal blocks
respiratory motion containing blocks iteration 1
respiratory motion containing blocks iteration 2
PSD criteria demonstration:
Background block 
Respiratory block 
Stage-3: Respiratory rate quality enhancement
All declared signal blocks from previous stage should be synchronized and summed up to improve SNR and get high quality respiratory signal.
SNR Enhancement Demonstration:
Stage-4: Sleep apnea detection
- Divide the respiratory signal to N overlapping intervals
- Let I = Array of size N
- For each interval k in N:
- Calculate the power spectral density estimate
- if the maximum of the PSD in the frequency window of [0.2, 2] Hz is greater than a certain threshold :
- Set I[k] = 1
- Else:
- Set I[k] = 0
- Let M be the minimum number of overlapping intervals to declare a sleep apnea
- If there is a sequence of zeros in I with length greater or equal M declare this sequence as a sleep apnea
Complete Flow Demonstration: Stage-2: Frames segmenting and respiratory motion detection:
Non-signal blocks
respiratory motion containing blocks iteration 1
respiratory motion containing blocks iteration 2
Complete Flow Demonstration: Stage-4: Sleep apnea detection:
Enhancements
- Implement a real-time algorithm based on this
- Use machine learning to automatically determine the thresholds that we use throughout the algorithm
- Cover more corner cases that the basic algorithm does not cover
Summary
- In this project we had implemented two algorithms:
- Acquiring respiratory signal from 3D camera frames
- Detecting sleep apnea from the output of the algorithm mentioned above
- The implementation of the two algorithms mentioned above lays down a stable stepping stone towards further works and enhancements