Use this URL to cite or link to this record in EThOS:
Title: Respiratory quality indices for automated monitoring of respiration from sensor data
Author: Birrenkott, Drew A.
ISNI:       0000 0004 7653 9010
Awarding Body: University of Oxford
Current Institution: University of Oxford
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Restricted access.
Access from Institution:
Abnormal respiratory rate (RR) is known to be one of the most clinically effective predictors of catastrophic decline. Despite this, RR is often the least monitored and most inaccurately measured vital sign. This is primarily because of the lack of a non-invasive, robust, automated method for estimating RR. It has previously been shown that the amplitude modulation (AM), frequency modulation (FM), and baseline wander (BW) of both the electrocardiogram (ECG) and photoplethysmogram (PPG) contain respiratory waveform information. However, these respiratory modulations are driven by the physiologic interrelationship between the cardiovascular and respiratory systems and may or may not be present based on a patient's characteristics and condition. Despite this, current methods for RR estimation from ECG and PPG do not account for this physiologic variability. The investigations in this thesis describe the development and evaluation of respiratory quality indices (RQIs), a novel method for evaluating the presence or absence of physiologically important respiratory information from the AM, FM, and BW extracted from the ECG and PPG. This work is conducted in three unique data sets, CapnoBase, MIMIC-III, and Dialysis III, all of which represent important, different patient populations. Five initial RQIs are described based on five signal processing techniques: fast Fourier transform (FFT), autocorrelation, cosine correlation, autoregression, and Hjorth parameters. Of these, the individual RQIs based on the FFT, autocorrelation, and autoregression are deemed to be good predictors of the presence of respiratory waveform data. The three individual RQIs are used to derive three fusion RQIs based on two supervised learning algorithms: linear regression and support vector regression (SVR) and one unsupervised learning algorithm: principal component analysis (PCA). Both the linear regression and PCA fusion RQIs are accurate and robust. The linear regression fusion RQI is used in the development of an RR estimation algorithm, termed RQIFusion, which achieves highly accurate and more complete RR estimates than existing methods. In the Dialysis III data set, implementation of RQIFusion improved RR estimation error by between 1.35 to 2.29 breaths per minute (brpm) to achieve RR estimation errors between 2.18 to 3.46 brpm, depending on the RR estimation algorithm employed. These results represent a marked improvement in RR estimation and indicate the importance of conducting respiratory quality analysis using RQIs on respiratory modulations extracted from ECG and PPG prior to RR estimation.
Supervisor: Clifton, David A. Sponsor: Rhodes Trust
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: Signal processing ; Biomedical engineering ; Computational health informatics