A deep learning algorithm model to automatically score and grade obstructive sleep apnea in adult polysomnography

Area of Science:

• Sleep Medicine
• Artificial Intelligence
• Medical Diagnostics

Background:

• Polysomnography (PSG) is essential for diagnosing sleep disorders like obstructive sleep apnea (OSA).
• Manual scoring of PSG data is labor-intensive and time-consuming.
• Automated scoring systems are needed to improve efficiency and accuracy.

Purpose of the Study:

• To evaluate a deep learning-based automated scoring system for respiratory events in patients with sleep-disordered breathing.
• To assess the accuracy of the deep learning model in detecting apnea/hypopnea events across various OSA severities.

Main Methods:

• A deep learning algorithm was developed using 1000 polysomnography (PSG) datasets (700 training, 200 validation, 100 testing).
• The model architecture includes perceptron and long short-term memory layers.
• The model was validated on PSG data from 100 patients across different OSA severity groups.

Main Results:

• The deep learning algorithm achieved high sensitivity (98.06-98.51%) and specificity (95.46-97.79%) in detecting apnea/hypopnea events.
• Area under the curve values for predicting OSA were consistently high (0.9402-0.9442) across different apnea-hypopnea index thresholds.
• The model demonstrated no significant performance differences across mild, moderate, and severe OSA groups.

Conclusions:

• The deep learning algorithm accurately identifies apnea/hypopnea episodes and assesses OSA severity.
• Automated scoring using deep learning techniques can significantly enhance the efficiency and accuracy of PSG analysis.
• This technology holds promise for improving sleep disorder diagnosis and management.