Sleep snoring detection using multi-layer neural networks.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Snoring detection is crucial for diagnosing obstructive sleep apnea syndrome (OSAS) and other sleep-related respiratory disorders.
• Traditional snoring analysis relies on audio signal processing, focusing on frequency characteristics.
• Conventional multilayer neural networks (o-MLP) have limitations in snoring pattern classification.

Purpose of the Study:

• To evaluate the efficacy of a novel correlational filter Multilayer Perceptron (f-MLP) neural network for sleep snoring signal detection.
• To compare the performance of f-MLP against conventional o-MLP in classifying snoring patterns.
• To leverage frequency domain analysis and adaptive filter coefficients for enhanced snoring detection.

Main Methods:

• Implementation of a novel f-MLP architecture with a correlational filter operation in the first hidden layer.
• Utilizing the back-propagation learning algorithm to enable self-adaptation of filter coefficients within the correlational filter layer.
• Applying the f-MLP model to a dataset of sleep snoring signals for detection and classification tasks.

Main Results:

• The f-MLP achieved a high average detection rate of 96% for test patterns.
• Conventional o-MLP demonstrated a significantly lower average detection rate of 82%.
• The correlational filter operation in f-MLP enhanced the discrimination power for classification.

Conclusions:

• The f-MLP neural network offers superior performance for sleep snoring detection compared to conventional methods.
• Frequency information is a dominant feature for snoring classification, effectively utilized by f-MLP.
• This advanced neural network approach shows promise for improving the diagnosis of sleep-related respiratory disorders.