Re-entrant ventricular arrhythmias are responsible for over 80,000 instances of Sudden Cardiac Death (SCD) annually in the UK and Ischemic Cardiomyopathy (ICM) is a particular risk factor. Patients at risk of lethal ventricular arrhythmias often receive an Implanted Cardioverter Defibrillator (ICD) if it is determined that the benefits of the device outweigh the risk of procedural and long-term complications. However, the majority of SCD events occur in patients who have not had an ICD: their elevated risk was not correctly identified.
Structural imaging-derived biomarkers, such as fibrosis from Magnetic Resonance Imaging (MRI), and biomarkers from functional electrical measurements, such as repolarisation dispersion from Electrocardiogram (ECG), have shown strong associations with SCD risk. However, these biomarkers have yet to be sufficient enough for uptake into clinical practice.
The study therefore aims to develop a more sensitive and specific clinical prediction model that combines both structural and functional features, augmented by structure-function simulation-based features, to significantly enhance risk stratification for ICD requirement in ICM patients.
Cardiovascular Magnetic Resonance (CMR) images (n=2,701) representing 377 patients at The Royal Brompton Hospital Trust were processed to extract structural-derived metrics related to infarct scar morphology. Patient follow-up data extending up to 6 years was also processed to determine an arrhythmic composite endpoint. Finite element image-based models were constructed directly from patient data and used to develop simulations to identify susceptible re-entrant arrhythmia pathways based on structure-function interaction. A combined machine learning risk prediction model, including image-based structural features along with simulation structure-function features, demonstrated the added importance of patient-specific simulations to augment the power of risk prediction in this population. Future work involves further developing the model to include functional biomarker features from the patients into an enhanced machine learning risk prediction model.