A Multi-Feature Fusion Framework for Automated Classification of Obstructive and Central Hypopneas in Polysomnography

Area of Science:

• Sleep Medicine
• Artificial Intelligence
• Biomedical Engineering

Background:

• Manual scoring of hypopneas in polysomnography (PSG) is subjective and inconsistent.
• Accurate classification of obstructive hypopnea (OH) and central hypopnea (CH) is crucial for effective sleep-disordered breathing treatment.
• Existing methods lack automation and interpretability.

Purpose of the Study:

• To develop and validate a fully automated, interpretable machine learning system for precise OH and CH classification from PSG.
• To address the subjectivity and inconsistency inherent in manual hypopnea scoring.
• To enhance diagnostic precision and standardize scoring for sleep-disordered breathing.

Main Methods:

• Operationalized clinical standards (AASM & Randerath) using 15+ multi-modal PSG features.
• Extracted features capturing inspiratory flow limitation (IFL), thoraco-abdominal paradox, and event characteristics.
• Developed an interpretable confidence-guided hierarchical classification (CGHC) framework for binary hypopnea classification.

Main Results:

• Evaluated on 2,509 hypopnea events, achieving 85.29% accuracy and Cohen's kappa of 0.71.
• Outperformed manual scoring consistency and demonstrated superior performance against expert consensus.
• Feature importance analysis confirmed IFL as the primary predictor, validating the clinical decision hierarchy.

Conclusions:

• The automated system accurately and reliably replicates expert clinical judgment for hypopnea subtyping.
• This represents the first robust, interpretable solution for automated hypopnea classification.
• The tool promises to enhance diagnostic precision, standardize scoring, and facilitate personalized sleep-disordered breathing therapy.