A comparison of piezoelectric-based inertial sensing and audio-based detection of swallows

Background: Prior research has shown a correlation between poor dietary habits and countless negative health outcomes such as heart disease, diabetes, and certain cancers. Automatic monitoring of food intake in an unobtrusive, wearable form-factor can encourage healthy dietary choices by enabling individuals to regulate their eating habits. Methods: This paper presents an objective comparison of two of the most promising methods for digital dietary intake monitoring: piezoelectric swallow sensing by means of a smart necklace which monitors vibrations in the neck, and audio-based detection using a throat microphone. Results: Data was collected from twenty subjects with ages ranging from 22 to 40 as they consumed a variety of foods using both devices. In Experiment I, we distinguished sandwich, chips, and water. In Experiment II, we distinguished nuts, chocolate, and a meat patty. F-Measures for the audio based approach were 91.3% and 88.5% for the first and second experiments, respectively. In the piezo-based approach, F-measures were 75.3% and 79.4%. Conclusion: The accuracy of the audio-based approach was significantly higher for classifying between different foods. However, this accuracy comes at the expense of computational overhead increased power dissipation due to the higher sample rates required to process audio signals compared to inertial sensor data.