Signal quality measures for pulse oximetry and blood pressure signals acquired in unsupervised home telecare environments

Download files
Access & Terms of Use
open access
Copyright: Abdul Sukor, Jumadi
Altmetric
Abstract
Home telecare has emerged as an alternative solution to the problem of delivering primary healthcare services to an increasingly aged population, particularly those suffering chronic disease conditions. However, there are some major challenges relating to data management and interpretation facing the home telecare paradigm that degrade its efficacy and impede its adoption; a decision support system (DSS) promises to be one of the best solutions to address these challenges. Recently, DSSs have become more widely accepted as a support tool for use with telecare systems, helping monitoring clinicians to summarise and digest what would otherwise be an unmanageable volume of data. Such systems are expected to become increasingly prominent in primary care, as telecare systems provide additional clinical measurement capabilities, supported by improving internet infrastructure and penetration throughout the world. One of the pillars of a home telecare system is the performance of unsupervised physiological self-measurement by patients in their own homes. Such measurements are prone to error and noise artifact, often due to poor measurement technique and ignorance of the measurement and transduction principles at work. These errors can degrade the quality of the recorded signals and ultimately degrade the performance of the DSS system which is aiding the clinician in their management of the patient. This thesis focuses the development of algorithms capable of automatically assessing the quality of two physiological measurements commonly acquired by a home telecare system, namely pulse oximetry and blood pressure (BP), which are both prone to artifact-related noise and interference. In developing an algorithm for automated quality assessment of pulse oximetry signals, a novel method to detect movement related noise has been developed and verified with a manually annotated gold standard (GS), performed by experts. This noise detection method relies on morphological analysis of the photoplethysmogram (PPG) signal, since there is a morphological difference between reliable PPG pulses and noisy sections of the signal in a noise contaminated PPG. Furthermore, a novel approach has been applied to develop a GS which forms the basis of the manual annotation of these noisy recordings. The developed algorithm has been tested on noise contaminated PPG signals, acquired from healthy people in a controlled laboratory environment. Statistical results from the comparison between the output from the developed algorithm and the manually annotated GS have shown that about 77% and 89% of the actual noise and good sections respectively have been correctly identified, with this agreement summarised by a mean Cohen’s Kappa coefficient,  of 0.64±0.22. The work presented relating to noise detection in the second physiological measurement, BP, assesses the quality of the BP signal and decides whether the signal is of an acceptable quality for an attempt at deriving systolic and diastolic pressures to proceed successfully. The decision is based on the distance between identified noisy sections (if any) and the determined candidate systolic and diastolic events. A novel method has been developed for identifying noisy sections and determining the reliability of Korotkoff sounds. Again, the developed algorithm is validated using a manually annotated GS. The output from the validation process verified that the developed algorithm has correctly determined whether recordings were worthy of deriving a BP estimate (for both systolic and diastolic) for 93% of all signals. When the algorithm examines the signal quality at the sample level, it correctly detected the actual noise for 81% of all noisy samples in 100 pooled signals. Consequentially, the resulting systolic and diastolic pressure estimation errors are reduced to 0.37±3.31 mmHg and 3.10±5.46 mmHg, respectively. This is in comparison to the results of an existing published algorithm by Park et al. [1] which achieves errors of 0.24±35.23 mmHg for systolic pressure and -8.34±24.96 mmHg for diastolic pressure. The data used to test the developed algorithm was recorded in a laboratory environment, of which half of the data is purposely contaminated with motion artifact. In addition to testing with the data recorded in a controlled laboratory environment, the developed algorithms for automated quality assessment for pulse oximetry and BP signals were also tested retrospectively with data acquired from a trial which recorded signals in a semi-supervised environment. The trial involved four aged subjects who performed pulse oximetry and BP measurements by themselves at their home for ten days, three times per day. Simultaneous ECG was also recorded during these measurements as a reference signal, but the ECG leads were placed with supervision. A manually annotated GS was again used as the reference against which the developed algorithms were evaluated after analysing the recordings. Assessing (1) pulse oximetry using this trial data, 95% of good sections and 67% of noisy sections were correctly detected by the developed algorithm, and a Cohen’s Kappa coefficient,  of 0.58 was obtained in 120 pooled signals. Assessing (2) BP measurement, 75% of the actual noisy sections were correctly classified in 120 pooled signals, with 97% and 91% of the signals correctly identified as worthy of attempting systolic and/or diastolic pressure estimation, respectively, with a mean error and standard deviation of 2.53±4.20 mmHg and 1.46±5.29 mmHg when compared to a manually annotated GS. Finally, the potential of the developed algorithms to improve telecare patient outcomes when incorporated in a DSS, used for patient management and health data interpretation, is investigated by incorporating the algorithms as a pre-processing block, to clean extracted signal parameters before they are input to an offline DSS software prototype under development by our research group. This DSS prototype will automatically interpret longitudinal multi-parameter telehealth records to predict incipient exacerbations in the patient’s condition. All recorded biosignal data acquired from trials of a home monitoring device for managing chronic lung disease, including pulse oximetry and BP, were stored in a database, affording the application of the signal quality algorithms proposed in this thesis on real unsupervised home telecare recordings. The efficacy of the developed algorithms when employed by the DSS is measured by comparing the performance of the DSS in obtaining accurate diagnoses of patient health statuses when the developed algorithms are deployed in the system, or not, compared to the GS reference constructed from carer journal notes and patient health information (acquired through the use of a questionnaire). The DSS’s performance (when incorporating the signal quality algorithms to clean the health parameter data (BP, heart rate (HR) and oxygen saturation (SpO2)) has increased by 43% (Moreover, the capability of the DSS of correctly identifying correctly diagnosed unstable patients has also increased by 42%. Thus, good performance of the developed algorithms was observed when tested with data acquired from three different scenarios: (1) laboratory environment; (2) semi-supervised home environment, and; (3) real unsupervised home telecare environment. These results demonstrate the feasibility, and highlight the potential benefit, of incorporating automated signal quality assessment algorithms for pulse oximetry and BP recording within a DSS for telehealth patient management.
Persistent link to this record
Link to Publisher Version
Link to Open Access Version
Additional Link
Author(s)
Abdul Sukor, Jumadi
Supervisor(s)
Hamilton Lovell, Nigel
James Redmond, Stephen
Creator(s)
Editor(s)
Translator(s)
Curator(s)
Designer(s)
Arranger(s)
Composer(s)
Recordist(s)
Conference Proceedings Editor(s)
Other Contributor(s)
Corporate/Industry Contributor(s)
Publication Year
2012
Resource Type
Thesis
Degree Type
PhD Doctorate
UNSW Faculty
Files
download whole.pdf 1.49 MB Adobe Portable Document Format
Related dataset(s)