ABSTRACT
Meal supervision for post-stroke dysphagia patients significantly improves prognosis during rehabilitation. Aspiration often occurs during meals, which may further incur aspiration pneumonia. Therefore, it's necessary to know the patient's swallowing ability as well as the occurrence of cough. Recently, some researchers have detected swallowing or coughing with audio signals and have made remarkable achievements. However, the users need to stay in quiet environments or wear uncomfortable cervical auscultation devices because the signals generated by swallowing are weak. In this work, we present MealCoach, a system that utilizes a contact microphone to collect high-quality signals to identify the events during meals. We take advantage of the insensitivity of contact microphones to ambient noise for free-living environment supervision. After balancing the wearing experience and identification accuracy, we elaborately select the optimal site to leverage the unique characteristics of cricoid cartilage movement during meals to accurately identify swallowing, coughing, speaking, and other events during meals. We collected data from thirty PSD patients in the hospital and evaluated our system, and the results demonstrate that MealCoach achieves a mean classification accuracy of 95.4%.
