Artificial intelligence based platform for the automatic and simultaneous explainable detection of apnoea, oxygen...

Area of Science:

• Biomedical engineering
• Artificial intelligence in healthcare
• Pediatric sleep medicine

Background:

• Polysomnography is the gold standard for sleep apnoea (SA) diagnosis but is costly and inaccessible.
• Airflow and pulse oximetry signals offer a simpler, more accessible alternative for SA detection.
• Deep learning shows promise in analyzing these signals for pediatric SA diagnosis.

Purpose of the Study:

• To introduce REST, a novel platform for simultaneous detection of apnoea, desaturation, and artefacts.
• To develop a 1D deep neural network architecture for analyzing pediatric sleep signals.
• To improve the accuracy and reliability of SA diagnosis in children.

Main Methods:

• Developed a novel 1D deep neural network architecture for signal analysis.
• Trained, validated, and tested the REST platform on data from 86 pediatric patients.
• Utilized gradient-weighted class activation mapping (grad-CAM) for decision process explanation.

Main Results:

• Achieved high balanced classification accuracies: 92.50% for apnoea, 98.30% for desaturation, and 97.59% for artefacts.
• Outperformed three other literature approaches in detecting target events.
• Provided a confidence score to highlight samples for physician review, enhancing overall performance.

Conclusions:

• The REST platform demonstrates superior performance in detecting SA-related events and artefacts in pediatric patients.
• The AI's explainability feature (grad-CAM) increases user trust and reliability.
• REST offers a promising, accurate, and potentially more accessible tool for pediatric sleep apnoea diagnosis.