Author Identifier

Zhuoqiao He: https://orcid.org/0009-0000-5016-0684

Date of Award

2026

Document Type

Thesis - ECU Access Only

Publisher

Edith Cowan University

Degree Name

Doctor of Philosophy

School

School of Medical and Health Sciences

First Supervisor

Wei Wang

Second Supervisor

Xuerui Tan

Third Supervisor

Manshu Song

Abstract

Right and left ventricular outflow tracts (RVOT and LVOT) are common origins of premature ventricular complexes (PVCs) and repetitive ventricular tachycardia (VT) in structurally normal hearts, accounting for up to two-thirds of idiopathic ventricular arrhythmias (VAs). Frequent PVCs may lead to PVC-induced cardiomyopathy, thereby increasing cardiovascular risk and mortality. Radiofrequency catheter ablation (RFCA) has been established as an effective and potentially curative therapy for outflow tract ventricular arrhythmias (OTVAs). Accurate pre-procedural identification of the arrhythmia origin is therefore critical for guiding decisions on vascular access and treatment strategy. Although several electrocardiogram (ECG) algorithms, including the transitional zone index (TZI) and V2S/V3R index, have been proposed for distinguishing LVOT from RVOT arrhythmias, their diagnostic accuracy is inconsistent and often limited by poor interobserver agreement, small sample sizes, and reliance on a restricted number of ECG leads. Given the anatomical proximity of the RVOT and LVOT, differentiation remains challenging when based solely on traditional ECG indices. Integrating patient characteristics, such as age, sex, body mass index, comorbidities, and ventricular wall thickness, may provide additional diagnostic value. Furthermore, the growing maturity of machine learning (ML) techniques offers opportunities to model complex, non-linear relationships, identify high-dimensional predictive patterns, and develop robust individualized diagnostic models.

This thesis comprises six chapters: a comprehensive literature review (Chapter 1), three results chapters (Chapters 2, 3, and 5), a methodological chapter (Chapter 4), and a final chapter with a general discussion of the main findings, future directions, and conclusions (Chapter 6). Chapter 1 reviews the clinical significance of OTVA origins differentiation, limitations of existing ECG-based algorithms, and the emerging role of ML approaches. Chapter 2 revisits the correction of transitional zone index (CTZI), previously proposed in earlier work by this research group as a refinement of the traditional TZI, to evaluate whether it improves diagnostic performance. Despite its theoretical advantages, CTZI did not outperform TZI while introducing unnecessary complexity, underscoring the need for alternative approaches. Chapter 3 presents a systematic review and network meta-analysis to benchmark existing ECG algorithms, identifying the most effective existing algorithm while highlighting persistent limitations while reinforcing the need for more comprehensive and data-driven methods.

Chapter 4 emphasizes methodological rigor, by critically examining the use of Receiver Operating Characteristic (ROC) analysis, one of the most commonly used methods in diagnostic research. Four common yet under-recognized pitfalls in ROC evaluation were identified, and practical guidance was provided to improve the accuracy and transparency of diagnostic performance assessments. This enhanced understanding of methodological errors established a more reliable foundation for the subsequent comparison and evaluation of ML models. Building on these foundations, Chapter 5 describes the development of a novel ML-based diagnostic model based on 145 independent variables from ECG features and clinical characteristics. A refined support vector machine (SVM) model using 24 features demonstrated high predictive accuracy, and a simplified SVM model using only 10 features further outperformed conventional ECG algorithms, combining diagnostic precision with clinical practicality. Chapter 6 synthesizes the findings, discusses their implications for both clinical decision-making and methodological standards, and outlines future research directions and conclusions.

Collectively, this thesis advances the field by critically reassessing existing algorithms, addressing methodological challenges, and introducing a novel ML-driven diagnostic model. By leveraging advanced computational techniques and integrating diverse clinical and ECG data, it facilitates more accurate, reliable, and clinically applicable differentiation of OTVAs, ultimately improving outcomes for patients undergoing catheter ablation.

Access Note

Access to this thesis is embargoed until 13th February 2031 

DOI

10.25958/79vv-4m13

Download

Available for download on Thursday, February 13, 2031

Share

 
COinS