Non-invasive Detection of Adenoid Hypertrophy Using Deep Learning Based on Heart-Lung Sounds

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Pediatric Respiratory Medicine

Background:

• Adenoid hypertrophy is a prevalent childhood upper respiratory disorder causing symptoms like nasal congestion and sleep apnea.
• Current diagnostic methods (CT scans, endoscopy) are invasive, use radiation, and are unsuitable for continuous monitoring.
• There is a clinical need for non-invasive, accessible methods for diagnosing and monitoring adenoid hypertrophy.

Purpose of the Study:

• To develop and validate a novel deep learning approach for non-invasive detection of adenoid hypertrophy using heart-lung sounds.
• To explore the correlation between acoustic signals from heart-lung sounds and adenoid size.
• To establish a deep learning framework for classifying adenoid hypertrophy severity and predicting adenoid size.

Main Methods:

• A heart-lung sound database was created with labeled data correlating sounds with adenoid size.
• Three deep learning tasks were implemented: binary classification (normal vs. abnormal), four-grade classification (severity), and regression (size prediction).
• Models were trained and evaluated to assess their efficacy in detecting adenoid hypertrophy from acoustic data.

Main Results:

• Deep learning models demonstrated significant effectiveness in predicting adenoid hypertrophy based on heart-lung sounds.
• The proposed methods accurately classified the severity and predicted the size of adenoids.
• The approach showed promise for reliable, non-invasive assessment of adenoid hypertrophy.

Conclusions:

• Deep learning analysis of heart-lung sounds provides a viable non-invasive method for detecting adenoid hypertrophy.
• This approach can simplify diagnosis, reduce healthcare costs, and enable remote self-screening in resource-limited settings.
• The findings support the potential of acoustic monitoring for pediatric respiratory health assessments.