Accelerometer- and Gyroscope-Based Sensor System for Detecting Swallowing, Speaking, and Coughing Events in ICU Patients

(1) Background: Swallowing, coughing, and speaking are clinically relevant physiological events, but they are not routinely and continuously monitored in intensive care. This study evaluates the feasibility of detecting these events using inertial measurement unit (IMU) signals acquired from the cricoid and thoracic regions. (2) Methods: Data were collected from 29 patients in the intensive care unit (ICU) during predefined swallowing, coughing, speaking, and movement tasks. A classical machine learning pipeline was developed and evaluated using leave-one-patient-out cross-validation. (3) Results: Cricoid and thoracic accelerometer signals were informative for event detection, with the thoracic sensor particularly contributing to cough detection. Speaking (F1=0.80), coughing (F1=0.72), and aggregated swallowing (F1=0.74) were detected in this controlled experimental setting. (4) Conclusions: The results support the feasibility of IMU-based detection of prompted swallowing, coughing, and speaking events in ICU patients. Further validation under spontaneous and clinically heterogeneous conditions is required before continuous clinical monitoring can be assessed.