Automatic detection of swallowing events based on surface electromyography

Clinical evaluation of dysphagia requires modeling different phases of swallowing. This process is expensive, invasive, subjective, and typically based on Videofluoroscopic Swallowing Study (VFSS) studies. This work evaluates the use of surface electromyography (sEMG) signals for detecting bolus passage through the mandible line (ManL) and upper esophageal sphincter (UES) in patients with oropharyngeal dysphagia. The paper introduces a novel method where GRU networks are fed with spectrograms resulting from processing sEMG signals. Using VFSS as the gold standard, our results yielded F1 scores up to 96% for binary, and 80% for tri-class classifications, outperforming state-of-the-art methods. The detection was more accurate with larger volumes and thinner consistencies. These results indicate that sEMG is a promising biosignal especially because of its non-invasiveness nature, making it an ideal complement to VFSS for dysphagia diagnosis and monitoring.