Voluntary Cough Acoustic Signals Differentiate Disease State and Swallowing Safety Status: A Data Set in Healthy Controls vs. Motor Speech Disordered Patients

Objective: Changes in voluntary cough airflow are an established detector of swallowing safety in neurogenic dysphagia. The objective of the current project was to determine if acoustic measures of voluntary cough acoustic signals could be used to differentiate (1) healthy versus diseased state, (2) between disordered groups with Parkinson disease (PD) and Amyotrophic Lateral Sclerosis (ALS) and (3) those who have airway invasion during swallowing and those who do not (safe vs. unsafe swallows). Methods: Videofluoroscopic swallow studies (VFSS) and voluntary coughs were obtained from PD and ALS participants (n=10 and n=13 respectively) and age matched controls (n=10). Participants with PD and ALS were striated into safe and unsafe swallowing groups using the penetration aspiration scale (PAS) on VFSS. Cough acoustic waveforms were processed with Python 3.11.4 and audio signal processing library Librosa 0.10.0 for extracting Mel Frequency Cepstrum Coefficients (MFCCs) and related acoustic measures of zero crossing rate (ZCR), spectral centroid, spectral bandwidth, spectral roll off, and spectral contrast. Results: MFCC 5 emerged as the strongest acoustic discriminator between healthy and disordered populations with a receiver operating characteristic (ROC) area under the (AUC) = 0.80, sensitivity = 0.83, specificity = 0.73. MFCC 5 was also observed to successfully categorize healthy vs. PD subjects (ROC AUC = 0.84), while ZCR successfully categorized healthy vs. ALS subjects (AUC = 0.79). MFCC 2 successfully categorized ALS vs. PD subjects (AUC = 0.79), and MFCC 13 emerged as the best acoustic classifier of subjects with safe vs. unsafe swallow (AUC = 0.81). Conclusions: Cough creates a distinct sound pattern during the transition from a compression phase, or closure of the glottis, to expiration. These results demonstrate that, even with a conservative approach to acoustic analysis of voluntary cough signals, MFCC analyses, along with other acoustic features including ZCR, show the potential for distinguishing between disease states and swallowing safety status with strong classification accuracy. Further research on the physiologic etiology of these acoustic differences is needed to provide thorough interpretation of the characteristics of healthy vs. disordered coughs.