Artificial Intelligence 3D Biventricular Scar Modelling to Guide Precision VT Ablation

Sudden cardiac death (SCD) is sudden, unexpected death caused by a change in heart rhythm and is responsible for half of all cardiac deaths. 75% of SCDs are related to a previous heart attack which occurs when a coronary artery supplying blood to the heart becomes blocked leading to scar tissue in the ventricles (the 2 bottom pumping chambers of the heart). This scar is often heterogeneous with a complex mix of dead and living cells causing abnormal and dangerous heart rhythms (ventricular arrhythmias-VAs). VAs can be divided into ventricular tachycardia (VT) & ventricular fibrillation (VF). Patients at high risk of VT/VF are offered a special device called an implantable cardioverter-defibrillator (ICD) which is inserted through the veins into the heart. ICDs have proven prognostic benefit as they deliver treatment eg a small, but powerful shock to revert the heart rhythm back to normal if a patient experiences a VA.
Although ICDs are life-saving they not prevent VAs from occurring which means recurrent, or even a single, painful shock from the device can lead to significant psychological and social stress to the patient. Concurrent use of medications is required to reduce or stop VAs from occurring however they can be ineffective or poorly tolerated by patients. In this scenario ablation of VT or VF can be performed by burning the abnormal living cells embedded in scar tissue responsible for VAs via catheters either placed inside the patient's heart to reach the inside of the heart muscle (endocardium) or through their chest wall to reach the outside of the heart (epicardium). A geometry of the ventricle can be created using 3D mapping systems in the clinical electrophysiology (EP) lab onto which data from the patient's heart can be displayed using parameters such as the voltage of the tissue thus allowing the display of the location of the scar and potential areas of living cells in the scar responsible for the VT/VF. This helps guide the operator to the regions which require ablation. However scar is a complex 3D architecture and unsurprisingly VT ablation procedures are often challenging and long (sometimes up to 8 hours) with recurrence rates of up to 40-50% at 2 years. In order to improve procedural and patient outcomes from VT ablation there is a global effort to improve the visualisation of scar, our understanding of the complexity of scar architecture and how it causes VT and to use this information to guide VT ablation.
Cardiac MRI scans (CMR) are able to identify ventricular scar and have become the cornerstone for scar assessment with excellent correlation of scar shown with histological examination in canine studies as well as in clinical human studies. Improving the resolution of CMR protocols and performing CMRs in patients with ICDs remain two significant challenges. We at Nottingham University Hospitals have overcome both of these and are now performing CMR studies in ICD patients with excellent scar resolution. Leveraging this capability with advancing knowledge and techniques in artificial intelligence we believe it is possible to improve our mechanistic understanding of VAs to explain why some, but not all patients with scar experience VA. Development of an AI 3D scar model which displays critical scar features can also help guide precision VT ablation.

Aims:
The overall aim of the study is to develop a fully automated artificial intelligence (AI) 3D computational biventricular scar model using Late Gadolinium Enhancement (LGE) cardiac magnetic resonance imaging (cMR) for descriptive representation of left ventricular scar.

The aim will be addressed through the following objectives:
1. Model development
2. Feasibility testing
3. Clinical Validation of the model

Sudden cardiac death (SCD) claims 450,000 lives annually in the US with a likely similar inferred incidence due to comparable cardiovascular risk profile in Western Europeans. The most common cause of SCD is ischaemic heart disease (IHD) with 75% of cases linked to previous myocardial infarction (MI) causing left ventricular scarring. This predisposes to ventricular arrhythmias (VA) eg ventricular tachycardia (VT) or ventricular fibrillation (VF) leading to SCD. Holter monitor studies have confirmed VT as a precursor to VF in up to 60% of all cases, in comparison to primary VF in 8%, suggesting VT as the predominant precursor of SCD. Implantable cardioverter-defibrillators (ICDs) decrease mortality and are the mainstay of treatment in these patients however whilst ICDs save lives they do not prevent VA occurrence. ICD shocks used to treat VAs are painful, lead to significant morbidity and do not offer complete protection against arrhythmic death. Concomitant use of anti-arrhythmic drugs to suppress VAs is common practice however these are often ineffective or poorly tolerated.
VT occurs due to slow abnormal conduction across living cells embedded within dead scar tissue leading to the formation of conduction channels (CCs) which perpetuate reentry. Radiofrequency ablation in the cardiac electrophysiology (EP) lab can be effective at transecting CCs in scar however due to the complex 3D architecture of scar VT ablation remains challenging with long procedure times and suboptimal recurrence rates of 40-50% at 2 years. Late Gadolinium enhanced cardiac MRI (LGE CMR) scans display high-resolution myocardial scar and there is an ample of published data confirming good correlation of CMR scar and intracardiac scar in canine histological and clinical human EP studies. With advances in artificial intelligence (AI) and our locally established cMR protocols performed in ICD patients we aim to develop an AI computational CMR derived 3D scar model to guide precision VT ablation