Deep Learning Algorithm Update Using Real Patients for Out-of-hospital Obstructive Sleep Apnea Monitoring

Patients undergoing Level 1 polysomnography are equipped with the CART-I PLUS device, for collecting polysomnography data alongside concurrent photoplethysmography (PPG) signals.

The collected data is categorized into apnea, hypopnea, and normal segments based on the polysomnography results. Utilizing the PPG and accelerometer (ACC) signals from the CART-I PLUS, metrics such as SaO2 (oxygen saturation), respiratory rate, heart rate (HR), heart rate variability (HRV), and body movement are calculated for each segment. These metrics, along with the PPG and ACC signals, are then used to develop a deep learning model that classifies the segments into apnea, hypopnea, or normal.

Participants are divided into training and validation sets. The deep learning model is trained on data from the participants in the training set, and its performance is evaluated using the validation set.

The algorithm is constructed using convolutional neural networks (CNN), recurrent neural networks (RNN), attention mechanisms, and other advanced techniques recognized for their efficacy in classification tasks, specifically for identifying apnea, hypopnea, and normal segments.