Search

GOAL: Current methods for estimating respiratory rate (RR) from the photoplethysmogram (PPG) typically fail to distinguish between periods of high- and low-quality input data, and fail to perform well on independent "validation" datasets. The lack of robustness of existing methods directly results in a lack of penetration of such systems into clinical practice. The present work proposes an alternative method to improve the robustness of the estimation of RR from the PPG. METHODS: The proposed algorithm is based on the use of multiple autoregressive models of different orders for determining the dominant respiratory frequency in the three respiratory-induced variations (frequency, amplitude, and intensity) derived from the PPG. The algorithm was tested on two different datasets comprising 95 eight-minute PPG recordings (in total) acquired from both children and adults in different clinical settings, and its performance using two window sizes (32 and 64 seconds) was compared with that of existing methods in the literature. RESULTS: The proposed method achieved comparable accuracy to existing methods in the literature, with mean absolute errors (median, 25[Formula: see text]-75[Formula: see text] percentiles for a window size of 32 seconds) of 1.5 (0.3-3.3) and 4.0 (1.8-5.5) breaths per minute (for each dataset respectively), whilst providing RR estimates for a greater proportion of windows (over 90% of the input data are kept). CONCLUSION: Increased robustness of RR estimation by the proposed method was demonstrated. SIGNIFICANCE: This work demonstrates that the use of large publicly available datasets is essential for improving the robustness of wearable-monitoring algorithms for use in clinical practice.

Estimation of respiratory rate (RR) from photoplethysmography (PPG) signals has important applications in the healthcare sector, from assisting doctors onwards to monitoring patients in their own homes. The problem is still very challenging, particularly during the motion for large segments of data, where results from different methods often do not agree. The authors aim to propose a new technique which performs motion reduction from PPG signals with the help of simultaneous acceleration signals where the PPG and accelerometer sensors need to be embedded in the same sensor unit. This method also reconstructs motion corrupted PPG signals in the Hilbert domain. An auto-regressive (AR) based technique has been used to estimate the RR from reconstructed PPGs. The proposed method has provided promising results for the estimation of RRs and their variations from PPG signals corrupted with motion artefact. The proposed platform is able to contribute to continuous in-hospital and home-based monitoring of patients using PPG signals under various conditions such as rest and motion states.

Introduction The shape of the arterial pulse wave (PW) is a rich source of information on cardiovascular (CV) health, since it is influenced by both the heart and the vasculature. Consequently, many algorithms have been proposed to estimate clinical parameters from PWs. However, it is difficult and costly to acquire comprehensive datasets with which to assess their performance. Our aim was to create a database of simulated PWs under a range of CV conditions, representative of a healthy population. Methods Baseline PWs were simulated using 1D computational modelling [1]. CV model parameters were varied across normal healthy ranges to simulate a sample of subjects for each age decade from 25 to 75 years. The model was extended to simulate photoplethysmographic PWs at common measurement sites, in addition to the pressure, flow and diameter PWs produced by the model. Results & Discussion The database was verified by comparing simulated PWs with in vivo PWs. Good agreement was observed, with age-related changes in blood pressure and wave morphology well reproduced. The utility of the database was demonstrated by applying algorithms to estimate arterial stiffness, cardiac output and left ventricular ejection time to PWs. The relative performances of several algorithms were compared, and CV factors which reduced algorithm performance were identified. Conclusion The database is a valuable resource for development and pre-clinical assessment of PW analysis algorithms [2, 3]. It is particularly useful because it contains several types of PWs at multiple measurement sites, and the exact CV conditions which generated each PW are known. References Mynard JP, Smolich JJ. Ann Biomed Eng. 2015; 43(6): 1443-1460. Vennin S et al. Hypertension. 2017; 70(6): 1176-1182. Li Y et al. Hypertension. 2017; 69(2): 375-381. Acknowledgments This work was supported by the British Heart Foundation [PG/15/104/31913]; the Wellcome/EPSRC Centre for Medical Engineering [WT 203148/Z/16/Z]., This presentation was given at the BioMedEng18 conference, held at Imperial College, London, on 6-7 September 2018. A video recording of this presentation is available at: https://youtu.be/X8aPZFs8c08

This poster provides an overview of the work described in: P. H. Charlton, "Continuous respiratory rate monitoring to detect clinical deteriorations using wearable sensors," Ph.D. Thesis, King’s College London, 2017., This poster was first presented at the Bioengenuity Keynotes Conference, held on Monday 6th March at the University of Oxford., {"references":["P. H. Charlton, \"Continuous respiratory rate monitoring to detect clinical deteriorations using wearable sensors,\" Ph.D. Thesis, King's College London, 2017."]}

This presentation was given at the MEIbioeng / MPEC 2017 Conference on the 14th September 2017, in acceptance of the Martin Black Prize for the best paper published in Physiological Measurement in 2016. It provides an overview of the following paper: Charlton, P. H., Bonnici, T., Tarassenko, L., Clifton, D. A., Beale, R., & Watkinson, P. J. (2016). An assessment of algorithms to estimate respiratory rate from the electrocardiogram and photoplethysmogram. Physiological Measurement, 37(4), 610–26. DOI: https://doi.org/10.1088/0967-3334/37/4/610

An introduction to the Respiratory Rate Estimation (RRest) project, with a focus on the assessment of RR algorithms reported in this publication: Charlton P.H. et al. Extraction of respiratory signals from the electrocardiogram and photoplethysmogram: technical and physiological determinants, Physiological Measurement, 2017. DOI: http://doi.org/10.1088/1361-6579/aa670e This presentation is designed for use in Journal Club settings. Originally presented at the University of Oxford on 11th May 2016., This work was supported by the UK Engineering and Physical Sciences Research Council (EPSRC, Grant EP/H019944/1); the Oxford and King's College London Centres of Excellence in Medical Engineering funded by the Wellcome Trust and EPSRC under Grants WT88877/Z/09/Z and WT088641/Z/09/Z; the National Institute for Health Research (NIHR) Biomedical Research Centre based at Guy's & St Thomas' NHS Foundation Trust and King's College London; the NIHR Oxford Biomedical Research Centre Programme; a Royal Academy of Engineering Research Fellowship (RAEng) awarded to David A Clifton; and an EPSRC Challenge Award to David A Clifton. The views expressed are those of the authors and not necessarily those of the EPSRC, Wellcome Trust, NIHR, NHS, Department of Health or RAEng., {"references":["Charlton P.H. et al. Extraction of respiratory signals from the electrocardiogram and photoplethysmogram: technical and physiological determinants, Physiological Measurement, 2017. DOI: 10.1088/1361-6579/aa670e"]}

INTRODUCTION: Databases of physiological data from intensive care have facilitated the development of predictive models to improve processes and outcomes (1). Several physiological data recording systems have been implemented previously. However, their generalisability is limited since they require either in-house software development or have infrastructure requirements that rapidly increase with each additional monitored bed added to the system (2, 3). As part of a study to assess the feasibility of continuously monitoring patients throughout their post-operative hospital stay, we implemented a commercially available data recording system which overcomes these limitations. We report our experiences here.METHODS: BedMasterEx software (version 4.1.12, Excel Medical Electronics) was used to record both waveform and numerical data between November 2012 and January 2014. The system recorded data from bedside monitors (MP70, Philips Medical Systems) on five geographically disparate critical care units. Each unit had a “central station” (Intellivue Information Centre, Philips) with Web Service functionality enabled, an essential pre-requisite. The monitors have the facility to “admit” (enter demographics) and “discharge” patients (erase demographics) from the monitor. BedMasterEx was configured to start recording when a patient was admitted to a monitor and cease on discharge. Researchers verified system functionality daily. Times which patients spent in each bed were determined from multiple hospital information systems. Recorded data was reviewed manually to ensure that transfer times had been correctly identified. This avoided incorrectly attributing data from patients subsequently or previously staying in a particular bed to a study patient. Numeric parameters that are monitored continuously (e.g. heart rate) were recorded once per second. The proportion of time for which data was recorded was calculated as the proportion of seconds in which a physiologically-plausible numeric was recorded.RESULTS: 222 patients undergoing cardiac surgery were monitored. In total 24,839 patient hours were spent in monitored beds (critical care beds connected to BedMasterEx). The median time spent by each patient in monitored beds was 55 hours (IQR = 46 - 96). The median proportion of time in monitored beds for which data was recorded was 94% (IQR 88 - 97). All but two patients were transferred between monitored beds at least once during their stay (median transfers = 1, max = 8). Data losses were caused by delays in admitting or discharging patients from the monitors. These tasks may often be deferred in favour of clinical work. Additional data losses were caused by time-stamping errors due to network latency. Most time-stamps errors were corrected using software developed in-house. Some recordings required manual review to avoid attributing data to the wrong patient.CONCLUSIONS: The system facilitated data recording for a high proportion of each patient’s critical care stay. Unlike previous systems the hardware requirements do not rapidly scale according to the number of monitored beds, nor does it require in-house development of complex acquisition software. However, system maintenance and fault diagnosis required significant technical expertise, and data had to be manually reviewed. Further work will investigate solutions to these limitations to facilitate fully automated recording of patient physiology at scale.