Modelling of Pharyngeal Laryngeal Effectiveness

The purpose of this study is to collect the signals of pharyngeal laryngeal activity through five non-invasive sensors (microphone, accelerometer, surface electromyography (EMG), nasal cannula and oximeter) in order to identify indicators of functional efficiency of swallowing, protection of the lower airways and phonation.

440 patients (subjects with swallowing disorders), spread over 4 centers and 80 healthy subjects spread over 2 centers will be recruited for the study in an interventional research study involving the prospective, multicentric and longitudinal.

Pharyngolaryngeal effectiveness will be measured from 6 indicators identified by examinations or reference tests grouped into 3 functions:

* swallowing: pharyngeal transport capacity (Yale Residue) and Penetration Aspiration Scale (PAS) rated by videofluoroscopy of swallowing (VFS) or flexible endoscopic evaluation of swallowing (FEES); * airway protection: cough trigger (citric acid test) and cough power (peak expiratory flow); * phonation: vocal efficiency (maximum phonation time) and velar efficiency (nasal scores).

The signals obtained from the 5 sensors will be annotated. Stochastic modelling based on hidden Markov models will be used initially and followed by the implementation of deep neural networks to model indicators. For the complication's prediction algorithm, deep neural networks will also be used to evaluate signal-based methods.

The expected benefits are to obtain automated recognition of pharyngeal-laryngeal effectiveness to diagnose swallowing disorders using objective and quantifiable indicators, non-invasive devices, to assess the severity of these disorders and to identify the risk of complications.