E-Thesis 276 views 190 downloads
Application of machine learning classifiers to arterial disease detection, utilising virtual patient databases / GARETH JONES
Swansea University Author: GARETH JONES
DOI (Published version): 10.23889/SUthesis.58404
Abstract
Two of the most common forms of arterial disease are stenosis and aneurysm, estimated to affect between 1% and 20% of the population. Ruptured abdominal aortic aneurysms alone are estimated to be the cause of between 6,000 and 8,000 deaths a year within the United Kingdom. Patients with stenosis have...
| Published: |
Swansea
2021
|
|---|---|
| Institution: | Swansea University |
| Degree level: | Doctoral |
| Degree name: | Ph.D |
| Supervisor: | Pant, S., ; Nithiarasu, P. ; Parr, J. |
| URI: | https://cronfa.swan.ac.uk/Record/cronfa58404 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Abstract: |
Two of the most common forms of arterial disease are stenosis and aneurysm, estimated to affect between 1% and 20% of the population. Ruptured abdominal aortic aneurysms alone are estimated to be the cause of between 6,000 and 8,000 deaths a year within the United Kingdom. Patients with stenosis have been shown to have a mortality hazard ratio of 1.42 compared to a control population [2], and an unadjusted death rate of 3.35 per 100 person-years compared to 1.23 per 100 person-years in a control population [97]. Current methods for the detection of arterial disease are generally impractical for large scale screening, expensive, or both. If an inexpensive method for the detection of both stenosis and aneurysm is created, that minimises the need for invasive measurements, the cost effectiveness of large scale screening could be improved making both continuous monitoring and screening feasible. One such method is to use easily acquirable haemodynamic measurements at accessible peripheral locations within the circulatory system for diagnosis. Within this thesis an initial exploratory study into the potential of using machine learning classification algorithms to detect arterial disease from such measurements is presented.It is likely that the indicative biomarkers of arterial disease held within pressure and flow-rate profiles consist of micro inter- and intra- measurement details. To facilitate the use of a data driven approach to the discovery of any biomarkers a framework for the creation of virtual patients, through the employment of a mathematical model of blood flow, is presented. This framework is utilised to create a series of virtual patient databases, as the balance between simplicity and realism progresses through the thesis. The most realistic of these databases is made publicly available (https://doi.org/10.5281/zenodo.4549764). The aforementioned framework for the creation of virtual patients is a major contribution of this thesis, and can be applied to a wide range of biological systems given a mathematical description.The synthetic data sets are used to train and subsequently test a series of machine learning classifiers, to predict the presence of both stenosis and aneurysm, using various combinations of pressure and flow-rate measurements. It is shown that the inclusion of a diseased vessel (either stenosis or aneurysm) produces consistent and significant biomarkers in haemodynamic profiles, irrespective of a patients unique underlying arterial network. These biomarkers are found to be differentiable from the natural variability present across a large cohort of patients, showing that arterial disease has a clear and unique effect on pressure and flow-rate profiles. This suggests strong potential in the use of haemodynamic measurements to detect arterial disease. |
|---|---|
| Item Description: |
ORCiD identifier: https://orcid.org/0000-0002-3196-6663 |
| Keywords: |
Arterial disease diagnosis, machine learning, virtual patient database, pulse wave haemodynamics |
| College: |
Faculty of Science and Engineering |