Multi-sensor Data Fusion for
Wearable Respiratory Rate Monitoring Microsystem

Senior Design Project

Problem Statement

Nowadays, respiratory rate is a vital sign used to monitor the progression of illness. An abnormal respiratory rate (RR) is an important marker of serious illness. Various methods have been explored for respiratory rate measurement. These include contact methods based on measurement of chest and abdominal movements, acoustic sounds and airflow, exhaled carbon dioxide and calculation of RR from electrocardiogram (ECG) or oxygen saturation (i.e., photoplethysmography or PPG).

Current Solutions

Movement Detection

Fluctuations between the angles of chest undulation can be measured using an inertial sensor. By placing the sensor as shown in the left figure above, the angle fluctuation curve as shown in the right figure above can be obtained. Through further processing, RR monitoring can be realized theoretically. It has the ability to get continuous accurate measurements, and can detect subtle thoraco-abdominal asynchrony related to specific respiratory disorders. Nevertheless, the disadvantages are obvious also. It has the limitation of sleep studies since violent movement and constant body movement introduce a great deal of uncertainty into the measurement of the inertial sensor. Moreover, different body types will also make it difficult to start from the angle measurement to carry out further data manipulation, especially for children.

Abstract

This project aims at improving accuracy of a wearable respiratory rate (RR) monitoring microsystem by data fusion of outputs from RR sensors based on microphone and oximeter. Different algorithms are evaluated and applied for situations with and without reference data. Kalman Filter, relevance coefficient, covariance, Mahalanobis distance and multiple linear regression are applied to realize an optimized algorithm for data fusion.