Filters
Availability
Publication Year
Resource Type
Faculty
School/Institute
Publication Search Results
Sort
Results per page
1 - 3 of 3
-
(2022)ThesisThis thesis focuses on the development and applications of magnetic resonance electrical properties tomography (MREPT), which is an emerging imaging modality to noninvasively obtain the electrical properties of tissues, such as conductivity and permittivity. Chapter 2 describes the general information about human research ethics, MRI scanner, MR sequence and the method of phase-based MREPT implemented in this thesis. Chapter 3 examines the repeatability of phase-based MREPT in the brain conductivity measurement using balanced fast field echo (bFFE) and turbo spin echo (TSE) sequences, and investigate the effects of compressed SENSE, whole-head B_1 shimming and video watching during scan on the measurement precision. Chapter 4 investigates the conductivity signal in response to short-duration visual stimulus, compares the signal and functional activation pathway with that of BOLD, and tests the consistency of functional conductivity imaging (funCI) with visual stimulation across participants. Chapter 5 extends the use of functional conductivity imaging to somatosensory stimulation and trigeminal nerve stimulation to evaluate the consistency of functional conductivity activation across different types of stimuli. In addition, visual adaptation experiment is performed to test if the repetition suppression effect can be observed using funCI. Chapter 6 explores if resting state conductivity networks can be reliably constructed using resting state funCI, evaluates the consistency of persistent homology architectures, and compares the links between nodes in the whole brain. Chapter 7 investigates the feasibility of prostate conductivity imaging using MREPT, and distinctive features in the conductivity distribution between healthy participants and participants with suspected abnormalities.
-
(2022)ThesisCardiovascular disease (CVD) is the leading cause of global mortality. Two forms of CVD are acute coronary syndromes (ACS) and heart failure (HF). Patients with either are prone to repeat hospitalisations, which are detrimental to both patients and the healthcare system. Traditional care models are suboptimal in preventing readmissions. Mobile health interventions (MHIs) are attractive due to the computing power and convenience of the smartphone. Firstly, the literature regarding MHIs in CVD is systematically reviewed and meta-analysed. MHIs improved medication adherence in ACS patients and hospitalisation rates in HF patients. The review noted limitations of published trials and identified features of successful MHIs, which were incorporated into the design of a novel smartphone app-based model of care (TeleClinical Care, TCC). TCC allows home measurement of blood pressure, heart rate and weight by patients. The readings are automatically transmitted to a central server, where clinicians can identify abnormalities and intervene accordingly. A pilot RCT comparing TCC and usual care (UC) to UC alone was performed (n=164). Patients using TCC had fewer readmissions at 6 months (41 vs. 21, hazard ratio 0.51, P= 0.015), and were more likely to be adherent with medications (75% vs. 50%, P= 0.001) and complete cardiac rehabilitation (39% vs. 18%, odds ratio 2.9, P= 0.02) compared to patients in the control arm. A process evaluation of the RCT was subsequently undertaken, which identified several contributory factors to TCC’s success, such as a helpful orientation protocol for team members, and high background rates of HF outreach service and cardiologist follow-up in both trial arms. Via a series of interviews, methods to improve the future delivery of TCC were identified, particularly relating to its integration into mainstream healthcare. Patterns of smartphone ownership among cardiac inpatients were also examined. Age, sex, diagnosis, and private health insurance subscription influenced smartphone ownership. These data will help identify patients who may be excluded from MHIs. The thesis contains a cost-effectiveness model of TCC if applied widely. When enrolment exceeds 237 patients, TCC will reduce healthcare costs relative to UC, resultant to readmission prevention. Enrolment of 500 patients is projected to save $100,000 annually. In conclusion, TCC is demonstrated as a feasible, beneficial, safe, and cost-effective intervention for patients with CVD.
-
(2022)ThesisClinical data are highly complex and pose challenges to machine learning that can introduce bias or negatively affect performance. Clinical data are typically high-dimensional and of mixed types, they may contain correlated values and missing information and a large proportion of the data is often irrelevant. Clinical measurements are often repeated over time, and the data may be censored, meaning the disease of interest has not yet been observed. Alzheimer’s disease (AD) is a progressive neurodegenerative disease that is ultimately fatal and has no cure. There are more than 55 million people worldwide living with dementia today, with AD thought to account for 60-70% of those cases, and numbers are forecast to triple by 2050. The pathological processes leading to AD begin decades before overt symptoms appear, presenting an opportunity to determine early biomarkers that might help identify individuals at risk of developing AD. Traditionally the Cox proportional hazards model has been used to analyse censored data. But the Cox model does not scale well to high dimensions and is limited by some strict assumptions. Consequently, machine learning algorithms have been adapted to handle censored data. This thesis performs a thorough comparison of the performance and stability of the available machine learning and feature selection methods for survival analysis, identifying their strengths and weaknesses. Some of these methods can be unstable in the presence of high-dimensional or correlated data. This thesis examines the reasons for these instabilities and develops new ensemble feature selection frameworks to improve the stability of feature selection. Data-driven thresholds are also developed to automatically separate the important from the redundant features, and clustering is used to handle correlated features. Improvements in stability of up to 40% are achieved. Clinical data is often collected repeatedly over time. A novel temporal pattern mining algorithm is developed to analyse this temporal data and is combined with temporal abstraction to find patterns common to those who develop AD. Survival analysis shows that these patterns are predictive of AD, with a C-Index of up to 0.74, and a novel visualisation module displays the clinically relevant results in an easily interpretable way.