Back to Projects

Image Computing and Computational Modelling (pre-2019)

Automated detection of cardiac arrhythmia locations to optimise treatments using a combined physics and machine-learning approach

Project ID: 2018_315

Student: Sofia Monaci

1st supervisor: Martin Bishop, King’s College London
2nd supervisor: Andy King, King’s College London

Cardiac arrhythmias are responsible for over 80,000 instances of sudden cardiac death annually in the UK. At risk patients often receive an implanted cardioverter defibrillator (ICD) which automatically detects arrhythmias and applies appropriate electrotherapy; however, ICDs represent a highly non-optimal therapy, and are often ineffective. Knowing exactly where the driving core of the arrhythmia is located may improve the delivery of electrotherapy. It is the goal of this work to develop a novel computational algorithm that utilises electrical signals recorded from ICDs to identify key anatomical locations of lethal arrhythmias. Simulations of arrhythmia episodes will be performed within detailed computational models to define and extract a series of physics- and physiological-based features from the corresponding simulated recorded electrical signals. These features will be the input to a deep-learning approach to allow automated detection of arrhythmia cores. Electrotherapy application within the models will then demonstrate how accurate localisation may facilitate tailored therapy and increase electrotherapy success.

Key Problem – In order for an ICD to apply electrotherapy closer to the core of an arrhythmia, knowledge of the anatomical location of the driving centre of the arrhythmia must first be obtained in real-time, prior to therapy delivery, from the sensing electrodes within the ICD.

Hypotheses – We hypothesise that intracardiac electrograms (iEGMs) recorded during arrhythmias contain key physics- and physiological-based features that can be extracted and used to determine the location of the organising centre of a reentrant arrhythmia. We further postulate that obtaining this knowledge in real-time during an arrhythmic episode will allow optimisation of electrotherapy that will increase efficacy.

Proposed Plan of Work
Yr 1 We have access to a high-quality and unique experimental dataset, consisting of both structural (imaging) and functional electrical measurements from a cohort of 8 infarcted pig hearts. This dataset will be used to construct a cohort of anatomically-detailed, ‘personalised’ porcine models. Advanced computational bidomain simulations will be performed to simulate both the effects of rapid pacing at different locations within the ventricles, as well as induced episodes of arrhythmias. Corresponding simulated extracellular potential signals from all regions surrounding the heart (i.e. simulated iEGMs), including those representative of ICD recording sites, will also be produced.

Yr 2 Simulated iEGMs during rapid pacing protocols will then be used to define and extract a series of physics-based (for example local/global electric field direction) and physiological (for example relative activation times) features from the iEGM signals. A machine learning approach will be used with these to identify the known pacing locations. Similar features will also be extracted during simulated arrhythmias, gaining an understanding of how they are changed during arrhythmias. Strategies for modifying the extracted features to include, for example arrhythmia-specific information such as regions of slowed conduction or fibrosis, will then be tackled, to allow identification of reentry cores.

Yr 3 A series of CT-derived whole torso computational models will then be constructed, including simulated episodes of cardiac arrhythmias within the embedded heart models, to understand how real iEGM signals may be affected by surrounding volume conductors as occurring within a real implanted device within a patient. This will allow modification and tuning of the feature extraction and machine learning approach. Electrotherapy will then be simulated in the models, guided by the known arrhythmia core location, and assessment performed as to improvements in efficacy.

Yr 4 Data will be obtained from our clinical collaborators at St Thomas’ to validate our novel predictive algorithm on iEGM recordings from catheter ablation patients. This data will be collected by Prof Rinaldi throughout the project, but will be analysed at this final stage. As a first step in validation, in such patients, a catheter can be used to pace in known locations whilst standard iEGMs are recorded from the ICD, testing the ability of our approach to accurately locate a pacing site. Secondly, arrhythmias will be induced (as common clinical practice in these procedures) and mapping performed to locate the driving centre, again whilst standard iEGMs are recorded, testing our ability to locate an arrhythmia core.

Back to Projects