Optimizing cervical breath sound acquisition protocol in dysphagia patients: an orthogonal experimental study

Dysphagia is a highly prevalent condition among patients with stroke. Cervical auscultation serves as a crucial screening method for dysphagia detection. The advancement of artificial intelligence has enabled acoustic analysis to emerge as a viable alternative to manual auscultation. While high-quality breath sound recordings are fundamental for reliable acoustic analysis in dysphagia screening, current clinical practice lacks standardized protocols for acquiring cervical respiratory sounds during auscultation.

To investigate the optimal acquisition protocol for cervical breath sounds in dysphagia patients after stroke using orthogonal experimental design.

The experimental design incorporated three key factors (auscultation position, dressing material, and tape type) with three levels each, configured into an L₉(3³) orthogonal array. Cervical breath sound data were systematically acquired from 298 dysphagia patients across nine protocol variations. Acoustic signal processing quantified three parameters: signal-to-noise ratio (SNR), aggregate entropy, and average power.

Statistical analysis revealed that Factor A (auscultation position), Factor B (dressing material), and Factor C (tape type) demonstrated significant effects (p < 0.05) on signal-to-noise ratio (SNR), aggregate entropy, and average power. The optimal configuration (A1B1C3) achieved maximum SNR (37.35 dB), maximum average power (78.79 dB), and minimum aggregate entropy (1.51 bits).

The optimal acquisition protocol involves positioning the stethoscope on the lateral trachea immediately below the cricoid cartilage, maintaining direct skin contact without interposing dressing materials, and secured with medical-grade pressure-sensitive tape. These evidence-based specifications offer technical guidance and standardized procedures for clinical implementation of cervical auscultation in the assessment of dysphagia after stroke.