What AI methods detect Obstructive Sleep Apnea (OSA) using Electroencephalography (EEG) data?

From the Guidelines

Several AI methods, including deep learning approaches like Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), have been developed to detect obstructive sleep apnea (OSA) using EEG data, offering a potential alternative to traditional polysomnography. These methods analyze changes in sleep architecture, particularly disruptions in slow-wave and REM sleep patterns that occur during apneic episodes 1. The use of EEG data for OSA detection is supported by studies that highlight the importance of accurate diagnosis and effective treatment of OSA to improve individual health, promote public safety, and reduce overall health care expenses 1.

Some of the key AI methods used for OSA detection include:

• Convolutional Neural Networks (CNNs) to analyze EEG spectral patterns associated with sleep disruptions
• Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks to capture temporal dependencies in EEG signals during apneic events
• Transfer learning techniques to improve detection accuracy by leveraging pre-trained models on large datasets
• Support Vector Machines (SVMs) and Random Forests to extract frequency-domain features from EEG data

The integration of EEG with other physiological signals like ECG, EMG, and oxygen saturation levels in multimodal approaches has further enhanced detection accuracy 1. However, it is essential to note that the minimum requirement for an acceptable study is a minimum of 4 hours of good quality data from HSAT recording, during the habitual sleep period, to diagnose OSA 1.

Overall, the development of AI methods for OSA detection using EEG data has the potential to enable home-based screening with portable EEG devices, though clinical validation remains ongoing. The use of these AI methods may offer advantages over traditional polysomnography, including increased convenience and reduced costs, while maintaining accurate diagnosis and effective treatment of OSA.

From the Research

AI Methods for OSA Detection

• The RAPIDEST framework uses EEG signals to detect OSA by analyzing the sequence of sleep stages and calculating a rarity score to capture unusual sleep stage transitions 2.
• A multi-layer feed-forward neural network (FNN) has been proposed to detect OSA using electrocardiogram (ECG), pulse oxygen saturation (SpO2), and body mass index (BMI) features, achieving an accuracy of 97.8% 3.
• Deep learning models, such as the multimodal signal fusion multiscale Transformer model, have been developed to detect OSA and assess its severity using ECG and SpO2 signals, with accuracy values of 91.38% and 96.08% for per-segment and per-recording detection, respectively 4.
• Machine learning algorithms, including Linear Discriminant Analysis (LDA), Artificial Neural Network (ANN), K-Nearest Neighbors (K-NN), and Support Vector Machine (SVM), have been used to detect OSA using frequency analysis of electrocardiographic RR interval, with the ANN achieving the highest performance with an accuracy of 84.64% 5.

EEG-Based OSA Detection

• The RAPIDEST framework is a notable example of an AI method that uses EEG signals to detect OSA, simplifying the signal collection process and reducing the complexity of severity determination 2.
• However, most studies have focused on using other signals, such as ECG and SpO2, for OSA detection, highlighting the need for further research on EEG-based methods.

Comparison of AI Methods

• A systematic review of AI-powered models for OSA screening and diagnosis found that AI algorithms demonstrated significant improvements in OSA detection, with accuracy, sensitivity, and specificity often exceeding traditional tools 6.
• The review also noted that anthropometric indexes were most widely used, especially in logistic regression-powered algorithms, while deep learning models showed great potential for OSA detection and severity assessment.