Development of a medical system to indicate risk of cardiovascular disease

Sooriamoorthy, Denesh (2022) Development of a medical system to indicate risk of cardiovascular disease. PhD thesis, University of Nottingham.

[img] PDF (Thesis - as examined) - Repository staff only until 23 July 2024. Subsequently available to Repository staff only - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Download (8MB)

Abstract

One significant issue that is encountered by individuals who experience ill effects of cardiovascular sickness, is not having the option to distinguish their illness until the manifestations give an evident sign, generally at the critical stage, which brings about a high threat of death. Furthermore, the most common method of treating severe cases is by invasive medical treatment, which is agonizing to patients. As a means to reduce this threat of severe cases or abrupt demise among individuals today due to cardiovascular disease, this research proposes to create a medical system focusing on the upstream blood pressure waveform and artificial intelligence, which would indicate the risk of cardiovascular disorder. The work in this thesis lays the groundwork for a novel risk indication system with three key sub systems. First a data acquisition system that uses a human wrist wearable device for acquiring data non-invasively, next a signal conversion system to transform radial waveforms to aortic waveforms, and finally the risk prediction system that uses a combination of a Convolutional Neural Network (CNN) and a zero-dimensional cardiovascular model’s parameters. In today's world, this combination of methods to indicate the risk of cardiovascular diseases has yet to be explored and developed, giving rise to a new pathway of risk indication. This thesis shows the details of the proposed medical system, along with testing of the hardware and the various sub systems.

For data acquisition, as currently marketed devices could not be used for this project, a wearable device with an embedded pressure sensor (Honeywell FSS005WNSB) was selected to acquire radial waveforms from the wearer’s wrist. As this is yet to be developed into a fully wearable device, pairs of radial and aortic signals were obtained from two databases (PhysioNet and HaeMod) for this research. These radial signals are then converted to aortic signals with the use of the newly developed Electrical Impedance Function (EIF), which is then compared to current conversion methods such as the Generalised Transfer Function (GTF), N-Point Moving Average (NPMA), and the Adaptive Transfer Function (ATF). Waveforms produced by the EIF have an average RMSE of 9.4838 and MAPE of 0.0661, with a peak difference of 6.35mmHg and 0.0129ms computational time, demonstrating a comparable performance with the GTF and a better estimation approach when compared with NPMA and ATF.

The transformed signals were then used for risk indication, utilizing Vincent Rideout’s cardiovascular model to produce data for the CNN. From the iterative investigation of the Vincent Rideout model, it is discovered that there are 16 parameters that significantly influences the model's aortic wave. Next a regression-based CNN is trained, with aortic waveforms as inputs, and their corresponding 16 parameters as outputs. When the trained CNN is tested with cardiovascular disease aortic pulse waveforms, which were converted from radial pulse waveforms utilizing both transfer functions (EIF and GTF) separately, it is observed that 2 key parameters out of the 16 could be used for indicating cardiovascular diseases. The two parameters - Pulmonary Vein 2 (RL2) and Systemic Aortic Artery 1 (RA1) – could be related biologically, as it can be postulated that they relax concurrently to permit the blood to flow smoothly in its closed-loop framework resulting in the decrease of the resistance value.

From the experiments conducted, the values of RL2 and RA1 when it acclimates to cardiovascular conditions is equal to or beneath 10.640691 g·s/cm^4and 9.7667933 g·s/cm^4 respectively when using EIF as the transfer function. On the other hand, by using GTF as the transfer function, the values of RL2 and RA1 when it acclimates to these cardiovascular conditions is equal to or beneath 10.530969 g·s/cm^4 and 9.8313036 g·s/cm^4 respectively. An 80.0% and 82.5% classification accuracy was obtained when these limits were used as identifiers on cardiovascular disease data obtained from Hospital Sultanah Bahiyah using EIF and GTF respectively as the transfer function.

Overall, this research shows that the proposed medical system can provide a minimum identification accuracy of 80% for cardiovascular disease, which can be considered a reasonably good performance, increasing the number of early detections and thus helping those at risk of cardiovascular disease on time.

Item Type: Thesis (University of Nottingham only) (PhD)
Supervisors: Nafea, Marwan
Shanmugam, Anandan
Keywords: cardiovascular system, blood pressure, parameters estimation, aortic pressure, 0D modeling, pulse wave, generalized transfer function, N-point moving average, adaptive transfer function, electrical impedance function, mathematical function, blood pressure waveform
Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering
Faculties/Schools: University of Nottingham, Malaysia > Faculty of Science and Engineering — Engineering > Department of Electrical and Electronic Engineering
Item ID: 69305
Depositing User: Sooriamoorthy, Denesh
Date Deposited: 24 Jul 2022 04:40
Last Modified: 24 Jul 2022 04:40
URI: https://eprints.nottingham.ac.uk/id/eprint/69305

Actions (Archive Staff Only)

Edit View Edit View