Abstract:
Classical machine learning has extensively been used to treat bio-signals to detect and classify multiple diseases, and this approach has also extended to deep learning,
offering more sophisticated and accurate methods for analyzing complex physiological
data.
Photoplethysmography (PPG) offers a non-invasive method to study blood volume
changes in the vascular system. It transmits light through the skin into the vascular beds
and measures the changes in the intensity of the transmitted or reflected light caused by
blood volume changes, providing insights into cardiovascular functions. This method
has been shown to return a high-fidelity signal regarding heart rate and blood pressure,
making it a reliable and convenient way to study cardiovascular function and potentially
monitor diseases such as sleep apnea.
However, PPG measurement suffers from variability based on the location of
measurement, skin tone, and motion artifacts, which can lead to inaccurate readings.
This study investigates the impact of sensor location on photoplethysmography (PPG)
signal quality for sleep apnea detection.
Firstly, we plan to demonstrate how the signal varies depending on the properties of
different recording sites based on other factors such as skin thickness and vascular
density.
The feasibility of classifying apneic events as obstructive or central using only PPG
signals is examined PPG by designing intelligent systems to do so. We will then design
a flowchart that will utilize a combination of PPG signals, deep learning models and
algorithms, to differentiate between the two types of apneas based on the distinct
physiological markers.
Our approach seeks to enhance the accuracy and reliability of sleep apnea diagnosis and
classification, for a better diagnosis of sleep apnea while using minimal commonly used
wearable devices.
