Integration Model of Clinical Variables and Computational Swallowing Biosignals in Neurogenic Oropharyngeal Dysphagia

Abstract

Clinical variables and biosignals can potentially be identified during the assessment, screening, and diagnostic characterization of neurogenic oropharyngeal dysphagia (NOD). This study aimed to develop an integration model to distinguish healthy individuals from patients with NOD by combining clinical variables with features extracted from surface electromyography (sEMG), laryngeal accelerometry (LA), and voice signals. These signals were recorded before and after swallowing different consistencies and volumes.

A case-control study was conducted, including 80 healthy individuals and 86 patients diagnosed with NOD, and 158 clinical variables and 5,080 non-invasive swallowing-related signal features were collected. After dimensionality reduction, the data were integrated using logistic regression models. Statistically significant differences were found in 88 clinical variables, 36 latent variables from sEMG, 72 combined features from sEMG and LA, and 61 from voice signals. The final model included five clinical and four biosignal variables: two background variables, three findings from the physical examination, one sEMG feature from the infrahyoid region during water swallowing, one LA feature in the mediolateral axis during yogurt swallowing, and two voice subfeatures reflecting changes observed during continuous articulation and sustained phonation of the vowel “a.” Together, these variables explained 90.6% of the variance in classifying individuals as NOD patients. The integration of computational swallowing methodologies using non-invasive signal processing with clinical variables may enhance screening and supplement gold-standard diagnostic tools in oropharyngeal dysphagia.