Analyzing Breathing Signals and Swallow Sequence Locality for Solid Food Intake MonitoringReport as inadecuate




Analyzing Breathing Signals and Swallow Sequence Locality for Solid Food Intake Monitoring - Download this document for free, or read online. Document in PDF available to download.

Journal of Medical and Biological Engineering

, Volume 36, Issue 6, pp 765–775

First Online: 09 December 2016Received: 18 September 2015Accepted: 23 May 2016

Abstract

Self-reported questionnaires are widely used by researchers for analyzing the dietary behavior of overweight and obese individuals. It has been established that questionnaire-based data collection often suffers from high errors due to its reporting subjectivity. Automatic swallow detection, as an alternative to questionnaires, is proposed in this paper to avoid such subjectivity. Existing approaches for swallow detection include the use of surface electromyography and sound to detect individual swallowing events. Many of these methods are generally too complicated and cumbersome for daily usage in a free-living setting. This paper presents a wearable solid food intake monitoring system that analyzes human breathing signals and swallow sequence locality. Food intake is identified by detecting swallow events. The system works based on a key observation that the otherwise continuous breathing process is interrupted by a short apnea during swallowing. A support vector machine SVM is first used for detecting such apneas in breathing signals collected from a wearable chest belt. The resulting swallow detection is then refined using a hidden Markov model HMM-based mechanism that leverages the known temporal locality in the sequence of human swallows. Temporal locality is based on the fact that people usually do not swallow in consecutive breathing cycles. The HMM model is used to model such temporal locality in order to refine the SVM results. Experiments were carried out on six healthy subjects wearing the proposed system. The proposed SVM method achieved up to 61% precision and 91% recall on average. Utilization of HMM in addition to SVM improved the overall performance to up to 75% precision and 86% recall.

KeywordsWearable sensors Swallow detection Food intake monitoring Support vector machine SVM Hidden Markov model  Download fulltext PDF



Author: Bo Dong - Subir Biswas

Source: https://link.springer.com/







Related documents