ECG-based convolutional neural network in pediatric obstructive sleep apnea diagnosis

Area of Science:

• Cardiology
• Pediatric Pulmonology
• Artificial Intelligence in Medicine

Background:

• Obstructive sleep apnea (OSA) is a common pediatric respiratory condition linked to cardiovascular risks.
• Current diagnostic methods like polysomnography (PSG) are complex, costly, and inaccessible, leading to underdiagnosis.
• A simpler, more accessible diagnostic tool for pediatric OSA is critically needed.

Purpose of the Study:

• To develop and validate a novel deep learning approach for diagnosing pediatric obstructive sleep apnea (OSA).
• To utilize raw electrocardiogram (ECG) data for predicting OSA severity based on the apnea-hypopnea index (AHI).
• To create a simplified, faster, and more accessible diagnostic method for pediatric OSA.

Main Methods:

• A convolutional neural network (CNN)-based regression model was developed using raw ECG tracings.
• The model was trained and validated on 1,610 overnight ECG recordings from the Childhood Adenotonsillectomy Trial (CHAT) database.
• The CNN model predicted the apnea-hypopnea index (AHI) and categorized OSA severity into 4 classes.

Main Results:

• The proposed CNN model demonstrated superior performance compared to previous ECG-derived feature algorithms (Cohen's kappa: 0.373 vs. 0.166).
• For AHI cutoffs of 1, 5, and 10 events/hour, the model achieved high accuracies (75.92%, 86.96%, 91.97%) and specificities (46.15%, 91.39%, 98.06%).
• The model achieved sensitivities of 84.19%, 76.67%, and 53.66% for the respective AHI cutoffs.

Conclusions:

• A novel CNN model effectively diagnoses pediatric obstructive sleep apnea (OSA) using only ECG data.
• This deep learning approach offers a simpler, faster, and more accessible alternative to traditional PSG for pediatric OSA diagnosis.
• The findings suggest this method can be readily implemented in clinical practice to improve pediatric OSA identification and management.