Method and system for phenotyping atrial fibrillation

Abstract

A computer-implemented method is described for determining a patient's atrial fibrillation classification comprising; extracting, from a simulation and / or measured data of the patient's atrial fibrillation, one or more parameters representing one or more conduction pattern metrics associated with the patient's atrial fibrillation; and using the one or more parameters to determine a classification of the patient's atrial fibrillation.

Description

[0001] Method and system for phenotyping atrial fibrillation

[0002] TECHNICAL FIELD OF THE DISCLOSURE

[0003] The present disclosure relates to a computer-implemented method for determining a patient’s atrial fibrillation classification.

[0004] BACKGROUND

[0005] Effective and personalized treatment for atrial fibrillation is required to improve patient outcomes and ease the major healthcare, social and economic burden associated with atrial fibrillation. Atrial fibrillation is the most common cardiac arrhythmia. Risk of developing atrial fibrillation increases with age and it is associated with increased long-term risks of stroke, heart failure, dementia and mortality. Treatments include drugs and ablation, which may limit the recurrence of atrial fibrillation. However, not every patient who receives treatment experiences improved cardiovascular outcomes.

[0006] Some patients experience atrial fibrillation in relation to exercise, some in relation to rest, some at increasing age, some at a young age. However, existing classification systems do not accurately reflect the differences between patients, and as such, treatments cannot be fully personalised to the patient. A patient may be classified with ‘paroxysmal’ atrial fibrillation if the duration of atrial fibrillation episodes is less than seven days, ‘persistent’ if the duration is greater than seven days or ‘longstanding persistent’ if the duration is greater than one year.

[0007] Clinical scoring systems have been used to predict the outcome of some therapies, for example, atrial fibrillation ablation. Clinical scores may be calculated considering risk factors such as CAAP-AF (Coronary artery disease, left Atrial diameter, Age, Persistent or long-standing atrial fibrillation, number of Antiarrhythmic drugs failed, Female sex). Predictions suggest that the higher the CAAP-AF score, the more likely atrial fibrillation will persist two years after ablation treatment. However, one disadvantage is that these risk factors do not accurately reflect the differences in cardiac electrophysiology between patients, but rather infer outcomes from a combination of clinical parameters.

[0008] It is the aim of at least one aspect of the present disclosure to provide a computer- implemented method for determining a patient’s atrial fibrillation classification, which ameliorates one or more of the described disadvantages. SUMMARY

[0009] In accordance with a first aspect of the disclosure there is provided a computer- implemented method for determining a patient’s atrial fibrillation classification comprising: extracting, from a simulation and / or measured data of the patient’s atrial fibrillation, one or more parameters representing one or more conduction pattern metrics associated with the patient’s atrial fibrillation; and using the one or more parameters to determine a classification of the patient’s atrial fibrillation.

[0010] Thus, embodiments of the invention provide a data-driven classification of the patient’s atrial fibrillation which is more specific than current classifications which may be based on duration of symptoms. The method can be used to classify patients at the onset of symptoms and does not require observation of symptoms over days or even years. The method can be used to differentiate between atrial fibrillation caused by different mechanisms.

[0011] Notably, the method may be used to classify the clinical arrhythmia experienced by the patient into different types (e.g. classes or subclasses) of atrial fibrillation.

[0012] The classification types may be determined by the parameters representing the conduction pattern metrics. For example, a first type may comprise a low number of phase singularities and a low number of electrical waves; a second type may comprise a low number of phase singularities and an intermediate number of electrical waves; a third type may comprise an intermediate number of phase singularities and an intermediate number of electrical waves; and a fourth type may comprise a high number of phase singularities and a high number of electrical waves. Other classification types may be determined based on the same or different parameters and, optionally, other factors such as patient demographics and / or symptoms. The method may comprise extracting, from a simulation of the patient’s atrial fibrillation, the one or more parameters.

[0013] The simulation may be based on one or more measurements of the patient’s atrial anatomy and / or one or more images of the patient’s atrial anatomy.

[0014] The method may comprise adjusting one or more features of the simulation to tune the simulation of the patient’s atrial fibrillation so that a simulated output matches measured patient data. Thus, the simulation is personalised for each patient. The measured patient data may comprise clinical data. Clinical data may comprise recordings of atrial electrical activity (e.g. comprising action potential recordings, electrograms or electrocardiograms).

[0015] The extracting of the one or more parameters representing one or more conduction pattern metrics associated with the patient’s atrial fibrillation may be extracted from the simulation.

[0016] The method may comprise using a trained machine learning algorithm to identify patterns and relationships between simulated output parameters and physiological variables from clinical data.

[0017] Notably, neither the source nor the precise mechanism causing the atrial fibrillation needs to be identified in the present method. On the contrary, through use of a simulation tuned to provide an output resembling a patient’s observed atrial fibrillation, properties of the simulation can be extracted and used to determine a classification of the patient’s type of atrial fibrillation based on the simulated conduction pattern metrics.

[0018] The one or more conduction pattern metrics may comprise a complexity and / or extent of the patient’s atrial fibrillation.

[0019] The one or more parameters may comprise a number of electrical waves propagating in a first predefined time period.

[0020] The one or more parameters may comprise a number of phase singularities detected in a second predefined time period.

[0021] The number of phase singularities and the number of electrical waves to may be used to determine the classification. For example, a k-means clustering machine learning algorithm may be employed to group patients into subgroups based on the number of phase singularities and the number of electrical waves propagating in a predefined time period. The step of using the one or more parameters to determine the classification may comprise determining a probability of the patient’s atrial fibrillation belonging to a particular classification.

[0022] The method may comprise determining a treatment based on the classification.

[0023] The method may comprise determining a likelihood of success of the treatment based on the classification; optionally wherein the likelihood of success is based on a probability of the patient’s atrial fibrillation belonging to the classification.

[0024] The method may comprise displaying a representation of possible treatment strategies.

[0025] The possible treatment strategies may be represented with any one or more of a graphic, text, a number and a code. The one or more parameters may comprise a frequency of electrical waves induced and / or sustained in a third predefined time period.

[0026] The one or more parameters may comprise a duration of electrical waves sustained in a fourth predefined time period.

[0027] The one or more parameters may comprise a duration of phase singularities sustained in a fifth predefined time period.

[0028] The method may comprise use of an algorithm to track a number of electrical waves propagating in a sixth predefined time period.

[0029] Any or all of the first to sixth predefined time periods may the same or different.

[0030] The one or more parameters may be determined for a left atria and / or a right atria.

[0031] The method may comprise creating the simulation of the patient’s atrial fibrillation.

[0032] Creating the simulation of the patient’s atrial fibrillation may comprise: obtaining one or more measurements of the patient’s atrial anatomy; processing the one or more measurements to create a three-dimensional (3D) model of the patient’s atrial anatomy; and simulating the patient’s atrial fibrillation using the 3D model.

[0033] Creating the simulation of the patient’s atrial fibrillation may comprise: obtaining one or more images of the patient’s atrial anatomy; processing the one or more images to create a three-dimensional model of the patient’s atrial anatomy; and simulating the patient’s atrial fibrillation using the 3D model.

[0034] In some cases, the creating the simulation of the patient’s atrial fibrillation may comprise: obtaining one or more measurements and / or one or more images of the patient’s atrial anatomy; processing the one or more measurements and / or one or more images to create a three-dimensional (3D) model of the patient’s atrial anatomy; and simulating the patient’s atrial fibrillation using the 3D model.

[0035] The step of obtaining one or more images of the patient’s atrial anatomy may comprise one or more of: magnetic resonance imaging; late gadolinium enhancement magnetic resonance imaging; diffusion tensor magnetic resonance imaging; computed tomography scanning; echocardiogram imaging; nuclear cardiac stress testing; singlephoton emission computed tomography; cardiac positron emission tomography; multigated acquisition scanning; a coronary angiogram; cardiac angiography; intracardiac echocardiography; electro-anatomic mapping or left heart catheterization.

[0036] The step of obtaining one or more images of the patient’s atrial anatomy may comprise input from an expert. The step of processing the one or more images to create the 3D model of the patient’s atrial anatomy may comprise one or more of: noise reduction, intensity normalization, or image registration.

[0037] The method may comprise determining an extent of atrial fibrosis and taking the extent of atrial fibrosis into account in the simulation of the patient’s atrial fibrillation by including the extent of atrial fibrosis in the 3D model.

[0038] The method of determining the extent of atrial fibrosis may comprise one or more of: late gadolinium enhancement magnetic resonance imaging, electro-anatomic mapping, T1 mapping magnetic resonance imaging, positron emission tomography imaging or delayed enhancement computed tomography.

[0039] The method of determining the extent of atrial fibrosis may comprise identifying one or more regions of the left and / or right atria with electrical conduction properties which differ from electrical conduction properties of regions of the left and / or right atria with no or less severe atrial fibrosis.

[0040] The method may comprise segmenting the 3D model comprising delineating anatomical structures.

[0041] The step of segmenting the 3D model may comprise input from an expert.

[0042] The method may comprise one or more of: image segmentation, geometric modelling, mesh morphing, or statistical shape modelling.

[0043] The step of creating the 3D model of the patient’s atrial anatomy may optionally comprise any one or more of: creating a mesh model of the patient’s atrial anatomy; defining fibre orientation and / or distribution across the 3D model; defining a location and / or distribution of one or more inter-atrial connections in the 3D model; defining one or more atrial regions; defining one or more electrophysiological properties for the one or more regions; defining a number and / or distribution of pacing sites at which electrical propagation is simulated in the simulation of the patient’s atrial fibrillation; defining a pacing protocol comprising a pacing sequence and / or rate of pacing for stimulating the pacing sites in the simulation of the patient’s atrial fibrillation; and adjusting one or more defined features of the 3D model and / or simulation to tune the simulation of the patient’s atrial fibrillation to match measured patient data.

[0044] The step of processing the one or more images to create the 3D model of the patient’s atrial anatomy may comprise input from an expert.

[0045] The step of creating the 3D model of the patient’s atrial anatomy may comprise segmenting the one or more images of the patient’s atrial anatomy.

[0046] The step of defining fibre orientation and / or distribution across the 3D model may comprise utilising one or more of: Universal Atrial Coordinates; a human atrial fibre atlas; diffusion tensor MRI; a simplified isotropic or uniformly anisotropic atrial tissue model; a histological study; an empirical model; or a mathematical or algorithmic rule-based method.

[0047] The step of defining the location and / or distribution of one or more inter-atrial connections in the 3D model may comprise using a location of an anatomical connection; optionally, the anatomical connection may comprise one or more of: Bachmann’s bundle; an interatrial septum; a muscle bridge in a region of a coronary sinus; and a connection in a region of a posterior left atrium.

[0048] The step of defining the location and / or distribution of one or more inter-atrial connections in the 3D model may comprise input from an expert.

[0049] The step of defining the location and / or distribution of one or more inter-atrial connections may comprise mapping inter-atrial connectivity using a Universal Atrial Coordinate method.

[0050] The step of defining the one or more electrophysiological properties for the one or more regions may comprise utilising one or more of: a homogeneous model; an idealized heterogeneity; or clinical recording data.

[0051] The electrophysiological properties may comprise one or more of an action potential duration; a longitudinal conductivity; a transverse conductivity; an ion channel conductance; a transmural conductivity; a conductivity of an inter-atrial connection.

[0052] The electrophysiological properties may be determined by a choice of cell model.

[0053] The step of defining a number and / or distribution of pacing sites at which electrical propagation is simulated in the simulation of the patient’s atrial fibrillation may comprise one or more of: localised pacing; random pacing; or uniform pacing.

[0054] The number of pacing sites may be of the order of up to 5, up to 10, up to 15, up to 20, up to 30, up to 40, and up to 50 or more. The pacing sequence and / or rate of pacing for stimulating the pacing sites in the simulation of the patient’s atrial fibrillation may comprise one or more of: incremental pacing; rapid burst pacing; extrastimulus pacing or cross field activation. In some cases, atrial fibrillation may be induced in the simulation by a pre-defined re-entrant wave or a spiral wave.

[0055] The pacing sites may be in proximity to a block line to create regions of slow conduction or unidirectional block.

[0056] The step of adjusting one or more defined features of the 3D model and / or simulation to tune the simulation of the patient’s atrial fibrillation to match measured patient data may comprise optimising modelling parameters of the 3D model and / or simulation using one or more of: statistical modelling; a machine learning algorithm; a model fitting to experimental data; and Bayesian algorithm inference.

[0057] The method may comprise obtaining measured data of the patient’s atrial fibrillation.

[0058] The method may comprise measuring atrial fibrillation using electrodes affixed to the atria during open heart surgery.

[0059] The method may comprise using any one or more of: composite mapping; high- density contact mapping; panoramic mapping using body surface electrodes and computed tomography; non-contact charge density mapping; and non-invasive electrocardiographic imaging.

[0060] The method may comprise sequentially recording atrial fibrillation from spatially overlapping electrodes in order to form a composite wave propagation map of the atria.

[0061] The method may comprise predicting the wave propagation direction using a threshold to find a similarity between two consecutive reconstructed phase maps.

[0062] The method may comprise solving mathematical equations that govern cardiac propagation comprising utilising one or more pieces of software suitable for cardiac electrophysiology simulations.

[0063] The step of determining a treatment based on the classification may comprise determining a treatment of one or more of: catheter ablation and a pharmacological intervention.

[0064] In accordance with a second aspect of the disclosure there is provided a non- transitory computer-readable medium comprising computer-readable instructions that are executable to perform a method for determining a patient’s atrial fibrillation classification, the method comprising: extracting, from a simulation and / or measured data of the patient’s atrial fibrillation, one or more parameters representing one or more conduction pattern metrics associated with the patient’s atrial fibrillation; and using the one or more parameters to determine a classification of the patient’s atrial fibrillation.

[0065] In accordance with a third aspect of the disclosure there is provided a system for determining a patient’s atrial fibrillation classification, the system comprising processing circuitry configured to: extract, from a simulation and / or measured data of the patient’s atrial fibrillation, one or more parameters representing one or more conduction pattern metrics associated with the patient’s atrial fibrillation; and use the one or more parameters to determine a classification of the patient’s atrial fibrillation.

[0066] BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Embodiments will now be described by way of example only, and with reference to the accompanying drawings, of which:

[0068] Figure 1 shows a block diagram of a method for determining a patient’s atrial fibrillation classification;

[0069] Figure 2 shows a block diagram of a method for determining the patient’s atrial fibrillation classification based on the probability of the patient’s atrial fibrillation belonging to a particular classification;

[0070] Figures 3A and 3B show detailed block diagrams of methods for creating a simulation of the patient’s atrial fibrillation;

[0071] Figure 4 shows a block diagram of a method for processing the one or more images to create the three-dimensional (3D) model of the patient’s atrial anatomy;

[0072] Figure 5A shows a series of cross-sectional images depicting the patient’s atrial anatomy, obtained using contrast-enhanced magnetic resonance angiography (CE- MRA) where individual image slices are in transverse (axial) orientation to a patient’s body with a first slice positioned at an inferior extent of the atria and subsequent slices positioned in a direction extending superiorly from a previous slice.

[0073] Figure 5B shows a series of cross-sectional images depicting the patient’s atrial anatomy, using late gadolinium enhancement magnetic resonance imaging (LGE-MRI) where individual image slices are in transverse (axial) orientation to the patient’s body with a first slice positioned at the inferior extent of the atria and subsequent slices positioned in a direction extending superiorly from a previous slice.

[0074] Figure 6 shows a series of images illustrating delineation of anatomical structures from medical imaging data using image segmentation;

[0075] Figure 7A shows an anterior view and Figure 7B shows a posterior view of a 3D model of atria illustrating the fibre orientation in the cardiac muscle;

[0076] Figure 8A shows an anterior view and Figure 8B shows a posterior view of a 3D model of an atrium defining atrial regions;

[0077] Figure 9A shows an anterior and posterior view of a 3D model of the patient’s atria, with pacing sites depicted as black stars;

[0078] Figure 9B shows consecutive snapshots in time of the simulation of the patient’s atrial fibrillation;

[0079] Figure 10 shows a graph of an applied stimulation signal for an electrode applied to an atria, illustrating incremental pacing;

[0080] Figure 11 A shows a series of partial images of an atria obtained using atrial late gadolinium enhancement magnetic resonance imaging (LGE-MRI) to represent various categories of qualitative assessment of atrial fibrosis (none, mild, moderate, severe);

[0081] Figure 11 B shows atrial late gadolinium enhancement (LGE) shells, reconstructed from atrial LGE imaging and visualised using an image processing toolkit for cardiovascular data, with each shell corresponding to the adjacent categories of qualitative assessment of atrial fibrosis as depicted in Fig. 11A;

[0082] Figure 11C shows electroanatomic voltage shells visualised using an electrophysiology data analysis tool, with each shell corresponding to the adjacent categories of qualitative assessment of atrial fibrosis as depicted in Fig. 11 A;

[0083] Figure 12A shows two images of a simulated atria with a number of identified atrial fibrillation waves being tracked in time from one image to the next;

[0084] Figure 12B shows two images of simulated atria with illustrating transmembrane potential for phase singularity detection, with phase singularity locations depicted as white spheres;

[0085] Figure 13A shows electrode locations for measuring atrial fibrillation and illustrates sequential measurements of atrial regions;

[0086] Figure 13B shows drivers in different locations of the atria;

[0087] Figure 14A shows a graph illustrating wave duration;

[0088] Figure 14B shows a graph illustrating the duration of phase singularities; Figure 15A shows a graph illustrating clustering of atrial fibrillation conduction pattern complexities quantified by number of waves and phase singularities;

[0089] Figure 15B shows a graph of atrial fibrillation initiation rate versus number of atrial fibrillation waves;

[0090] Figure 15C shows a graph depicting, for a number of patients, a predicted probability of the patient having atrial fibrillation according to four different classifications;

[0091] Figure 16 shows a block diagram of a system for determining a patient’s atrial fibrillation classification;

[0092] Figure 17 shows bi-atrial MRI segmentations used to generate patient-specific bi-atrial anatomy models;

[0093] Figures 18A and 18B show respectively the mean number of (A) phase singularities and (B) waves per simulation in patients with and without arrhythmia recurrence;

[0094] Figure 19 shows a graph illustrating clustering of atrial fibrillation conduction pattern complexities quantified by number of waves and phase singularities;

[0095] Figure 20 shows a table of the arrhythmia recurrence rate for each cluster shown in Figure 19;

[0096] Figure 21 A and 21 B show respectively rates of atrial fibrillation initiation and atrial fibrillation termination for patients with and without arrhythmia recurrence;

[0097] Figure 22 shows histograms illustrating the distribution of additional parameters extracted from simulations;

[0098] Figure 23 shows the results of Area Under the Receiver Operating Characteristic Curve (AUC-ROC) analysis for a model predicting arrhythmia recurrence risk;

[0099] Figure 24 shows the full results of the Mean Decrease in Accuracy plot to determine how much the model’s accuracy decreases when the parameter values, of the parameters described herein, are permuted.

[0100] DETAILED DESCRIPTION OF THE EMBODIMENTS

[0101] Atrial fibrillation is generally defined as an abnormal heart rhythm caused by electrical waves, which begin at one or more focal or localised source(s) in an atrium and propagate throughout the atria via atrial muscle and specialised conduction tissue. The electrical waves are apparently chaotic and travel across at least part of a first atrium of a heart. The electrical waves may travel across inter-atrial connections and continue to travel across at least part of a second atrium of the heart. The first atrium may be constituted by a right atrium and the second atrium may be constituted by a left atrium. Alternatively, the first atrium may be constituted by the left atrium and the second atrium may be constituted by the right atrium.

[0102] In the course of normal activation of the heart, a sinus node generates an electrical wave, which activates atria to contract. Subsequently, the electrical wave travels to an atrioventricular node, then to a bundle of conduction tissues cells known as a Bundle of His, then to ventricles which are activated and contract. The heart remains at rest between contraction of the ventricles and a next electrical wave generated at the sinus node. Drivers of atrial fibrillation cause atrial fibrillation to begin at a focal or localised source. The drivers may comprise re-entrant waves, which occur when the electrical wave re-enters cardiac tissue after causing the previous depolarisation.

[0103] Due to the apparently chaotic nature of atrial fibrillation, patients have historically been diagnosed based on clinical features, such as whether the abnormal heart rhythm occurs acutely or long-term, rather than using a data-led approach.

[0104] Figure 1 shows a block diagram of a method 100 for determining a patient’s atrial fibrillation classification. The method 100 comprises a step 102 of extracting, from a simulation and / or measured data of the patient’s atrial fibrillation, one or more parameters representing one or more conduction pattern metrics associated with the patient’s atrial fibrillation; and, in a step 104, using the one or more parameters to determine the classification of the patient’s atrial fibrillation. The one or more conduction pattern metrics may comprise a complexity and / or extent of the patient’s atrial fibrillation. Further details relating to step 102 are presented in Figure 12. Further details relating to step 104 are presented in Figure 2.

[0105] The method 100 may comprise a preceding step of obtaining the simulation of the patient’s atrial fibrillation and / or obtaining measured data of the patient’s atrial fibrillation. The step of obtaining the simulation of the patient’s atrial fibrillation may comprise obtaining the simulation from a memory or creating the simulation. The step of obtaining measured data of the patient’s atrial fibrillation may comprise obtaining the measured data from a memory or measuring the data by performing one or more tests on the patient and / or by analysis of patient test data.

[0106] Figures 3-12 relate to the case where simulated data is obtained and used to extract the parameters in accordance with step 102.

[0107] Figure 2 shows a block diagram of a method 200 for determining the patient’s atrial fibrillation classification based on the probability of the patient’s atrial fibrillation belonging to a particular classification. The method 200 comprises: determining a probability of the patient’s atrial fibrillation belonging to a particular classification in a step 202; and determining the classification of the patient’s atrial fibrillation based on the probability in step 204.

[0108] Optionally the method 100 and / or the method 200 may further comprise determining a treatment based on the classification in a step 206; and optionally determining a likelihood of success of the treatment based on the classification in a step 208.

[0109] Subtyping atrial fibrillation patients based on conduction pattern complexity metrics enables the development of tailored therapeutic strategies. In the example shown, step 206 may comprise determining the treatment based on the classification and recommending the determined treatment to a clinician who may then use the determined treatment to customize a treatment approach to address the specific arrhythmia characteristics of the patient. For example, the determined treatment may be to clinically ablate specific areas of the atria, prescribe pharmacological treatment to control atrial fibrillation, pharmacological treatment to control a ventricular rate in atrial fibrillation, or determine that atrial fibrillation cannot be controlled by currently available strategies and alternatively consider pacemaker and atrioventricular nodal ablation. Determining the patient’s atrial fibrillation subtype may also allow personalised therapy to modify risks of dementia, heart failure or mortality associated with patient-specific atrial fibrillation mechanisms.

[0110] Figure 3A shows a detailed block diagram of a method 300 for creating the simulation of the patient’s atrial fibrillation. The method 300 comprises: obtaining one or more images of a patient’s atrial anatomy in a step 302; processing the images to create a three-dimensional (3D) model of the patient’s atrial anatomy in a step 304; optionally determining an extent of atrial fibrosis and including the extent of atrial fibrosis in the 3D model in a step 306; and simulating the patient’s atrial fibrillation using the 3D model in a step 308.

[0111] The step of obtaining one or more images of the patient’s atrial anatomy 302 may comprise using one or more imaging techniques such as but not limited to: magnetic resonance imaging; late gadolinium enhancement magnetic resonance imaging; diffusion tensor magnetic resonance imaging; computed tomography (CT) scanning; echocardiogram imaging; nuclear cardiac stress testing; single-photon emission computed tomography; cardiac positron emission tomography; multigated acquisition scanning; a coronary angiogram; cardiac angiography; intracardiac echoangiography; electro-anatomic mapping or left heart catheterization. The imaging technique may be chosen based on a compromise between spatial resolution, soft tissue contrast, use of ionizing radiation, use of contrast agents, and level of invasiveness. In some cases, the images may have been previously obtained and stored in a memory device such that the step of obtaining the one or more images may simply comprise retrieving the one or images from the memory device. Figures 5A and 5B show example images obtained in accordance with step 302.

[0112] In some cases, electro-anatomic mapping is considered an imaging technique. In other cases, electro-anatomic mapping is considered a measurement technique as discussed in relation to step 322 below.

[0113] The step of processing the images to create the 3D model of the patient’s atrial anatomy 304 may optionally comprise pre-processing to improve image quality and consistency. Pre-processing may comprise any one or more of: noise reduction; intensity normalization; and image registration.

[0114] In the example shown, the step of creating the 3D model of the patient’s atrial anatomy based on the one or more images of a patient’s atrial anatomy 304 comprises segmenting the one or more images of the patient’s atrial anatomy. Further details relating to step 304 are presented in Figures 4 and 6 and described in more detail below.

[0115] Figure 3B shows a detailed block diagram of a method 320 for creating the simulation of the patient’s atrial fibrillation. The method 320 comprises: obtaining one or more measurements of a patient’s atrial anatomy in a step 322; processing the measurements to create the three-dimensional model of the patient’s atrial anatomy in a step 324; optionally determining the extent of atrial fibrosis and including the extent of atrial fibrosis in the 3D model in the step 306; and simulating the patient’s atrial fibrillation using the 3D model in the step 308.

[0116] In the example shown, the step of obtaining one or more measurements of a patient’s atrial anatomy in the step 322 comprises electro-anatomic mapping.

[0117] In other examples, step 322 may comprise one or more of: echocardiography and inferring atrial size from anthropometry measurements.

[0118] In the example shown, the step of creating the 3D model of the patient’s atrial anatomy based on the one or more measurements of a patient’s atrial anatomy 324 comprises inferring atrial anatomy using a pre-trained statistical shape model.

[0119] In other examples, step 324 may comprise any one or more of: rigid scaling of a predefined atrial anatomical model; and deformable scaling of a pre-defined atrial anatomical model.

[0120] The methods 300 and 320 depicted in Figures 3A and 3B respectively may comprise an optional step of determining the extent of atrial fibrosis 306. In the present example step 306 comprises using late gadolinium enhancement magnetic resonance imaging (LGE-MRI) or electro-anatomic mapping. Incorporating LGE-MRI data in the simulation of the patient’s atrial fibrillation allows for the projection of fibrosis distribution, which may be important for understanding arrhythmia mechanisms and guiding therapeutic interventions.

[0121] In the present example, the step 308 of simulating the patient’s atrial fibrillation using the 3D model comprises initiating atrial fibrillation at pacing sites using a pacing protocol. In the example shown, step 308 comprises calculating electrical propagation in the 3D model of the patient’s atrial anatomy using a known technique to determine localised timing of atrial depolarisation across the 3D model, which may include solving partial differential equations to define atrial cell activation and cell-to-cell propagation in both intracellular and extracellular spaces, solving partial differential equations describing atrial cell activation and cell-to-cell propagation only in the intracellular space, or using conduction velocities to determine relative activation timing of atrial myocytes. Parameters for each approach are selected according to previous literature or determined according to experimental studies or clinical recordings. Advantageously, the simulation should incorporate detailed biophysical models and physiological parameterizations to accurately represent cellular action potentials, tissue conduction properties, and organ-level dynamics. The simulation may be configured to accurately represent physiological processes. In some cases, the simulation may be configured to enable a skilled operator to study arrhythmia mechanisms, drug effects, and cardiac therapies to at least reduce or mitigate against atrial fibrillation. The simulation may be configured for large-scale anatomical models and complex spatiotemporal dynamics.

[0122] Further details relating to step 308 are described below in relation to Figures 9A, 9B, and 10.

[0123] Figure 4 shows a block diagram of a method 400 for processing the one or more images to create the 3D model of the patient’s atrial anatomy. The method 400 comprises any one or more of: creating a mesh model of the patient’s atrial anatomy 401 ; defining fibre orientation and / or distribution across the 3D model in a step 402; defining a location and / or distribution of one or more inter-atrial connections in the 3D model in a step 404; defining one or more atrial regions in a step 406; defining one or more electrophysiological properties for the one or more atrial regions in a step 408; defining a number and / or distribution of pacing sites at which electrical propagation is simulated in the simulation in a step 410; defining the pacing protocol comprising a pacing sequence and / or rate of pacing for stimulating the pacing sites in the simulation in a step 412; and adjusting one or more defined features of the 3D model and / or simulation to tune the simulation of the patient’s atrial fibrillation to match measured patient data in a step 414.

[0124] In the example shown, the step of creating a mesh model of the patient’s atrial anatomy 401 comprises geometric modelling to create a simplified 3D model of the patient’s atrial anatomy comprising spheres. In other examples, the simplified 3D model of the patient’s atrial anatomy comprises any one or more of: spheroids; ellipsoid; cylinders; and generalized atrial models.

[0125] In other examples, the mesh model may be generated using mesh morphing which comprises: extracting features from images of the patient’s atrial anatomy or statistical shape models; calculating deformation parameters based on the extracted features; and deforming a pre-existing computational mesh model of a generic atrial anatomy to create a customized mesh model of the patient’s atrial anatomy.

[0126] Further details relating to steps 402 and 404 are presented in Figure 7 and described in more detail below. Further details relating to steps 406 and 408 are presented in Figure 8 and described in more detail below. Further details relating to step 410 are presented in Figure 9A and described in more detail below. Further details relating to steps 412 and 414 are presented in Figure 10 and described in more detail below.

[0127] Figure 5A (i-v) shows example medical images obtained using contrast-enhanced magnetic resonance angiography (CE-MRA).

[0128] Figure 5A shows a series of cross-sectional images depicting the patient’s atrial anatomy, obtained using contrast-enhanced magnetic resonance angiography (CE- MRA) where individual image slices are in transverse (axial) orientation to a patient’s body with a first slice (i) positioned at an inferior extent of the atria and subsequent slices (ii-v) positioned in a direction extending superiorly from a previous slice.

[0129] In the example shown, the patient’s atrial anatomy is defined using CE-MRA cross- sectional imaging. CE-MRA involves the intravenous administration of gadolinium-based contrast agents to the patient before CT angiography is performed. Blood vessels and cardiac chambers appear bright on the images in Figure 5A and signal intensity is not impacted by characteristics of blood flow such as velocity or flow direction. The images depict a right ventricle 502, a left ventricle 504, a right atrium 506 and a left atrium 508.

[0130] Figure 5B (i-v) shows example medical images obtained using LGE-MRI.

[0131] Figure 5B shows a series of cross-sectional images depicting the patient’s atrial anatomy, using late gadolinium enhancement magnetic resonance imaging (LGE-MRI) where individual image slices are in transverse (axial) orientation to the patient’s body with a first slice (i) positioned at the inferior extent of the atria and subsequent slices (ii- v) positioned in a direction extending superiorly from a previous slice.

[0132] In the example shown, the patient’s atrial anatomy is defined using LGE-MRI cross- sectional imaging. LGE-MRI involves intravenous administration of gadolinium-based contrast agents to the patient before the MRI is performed. The images depict a right ventricle 502, a left ventricle 504, a right atrium 506 and a left atrium 508.

[0133] In other embodiments, the patient’s atrial anatomy may be defined using one or more other imaging techniques such as 3D medical images.

[0134] Figure 6 shows a series of images labelled 1-5, illustrating delineation of anatomical structures from medical imaging data using image segmentation. Images 1-5 highlight the following steps: (1) left atrial segmentation (602) from a CE-MRA image; (2) pulmonary vein (604) and left atrial appendage (606) clipping; (3) mitral valve (608) clipping; (4) registration of segmentation (610) onto a LGE-MRI image; (5) final three- dimensional left atrial segmentation.

[0135] In the example shown, image segmentation may be a fully-automatic process or may be a semi-automatic process performed by a skilled operator comprising using image processing software to delineate the patient’s atrial anatomy from surrounding tissue to produce a segmented image of the patient’s atrial anatomy.

[0136] The skilled operator may perform an image-to-mesh conversion to create a computational mesh from the segmented image. In the example shown, the image processing software uses threshold based segmentation with operator adjustment to produce a segmented and labelled image.

[0137] In some cases, image segmentation may be performed manually or automatically. In some cases, the image processing software may use a deep deconvolution neural network. In other embodiments, the image segmentation may comprise any one or more of: pixel-wise annotation; 2D patch-wise classification; region growing; and threshold segmentation.

[0138] Figures 7A and 7B show, respectively, an anterior and a posterior view of a 3D model of atria illustrating fibre orientation and anatomical connections 702a, 702b in the cardiac muscle.

[0139] Figure 7A depicts a right superior pulmonary vein 702c, a left superior pulmonary vein 702d, a mitral valve 702e and a tricuspid valve 702f.

[0140] Figure 7B also depicts the left superior pulmonary vein 702d and a left inferior pulmonary vein 702g. In the example shown, the step 402 of defining the fibre orientation and / or distribution across the 3D model comprises using a human atrial fibre atlas. Conduction is more rapid along fibre bundles than across fibre bundles. Fibre orientation is mapped from the human atrial fibre atlas to the 3D model in order to better approximate the electrophysiological properties of the atrial tissue.

[0141] In some cases, step 402 comprises using a simplified isotropic or uniformly anisotropic atrial tissue model, which assumes all atrial fibres have the same anisotropy ratio.

[0142] In some cases, step 402 comprises using a histological study to visualise the atrial fibres of ex vivo tissue, which may determine the fibre orientations for input to the 3D model.

[0143] In some cases, step 402 comprises using an empirical model which employs mathematical formulations or empirical relationships to describe electrical conduction patterns in atrial tissue without explicit consideration of fibre orientation.

[0144] In some cases, step 402 comprises using a mathematical or algorithmic rule-based method to define fibre orientations based on predefined rules or algorithms.

[0145] In some cases, step 402 comprises using Universal Atrial Coordinates to define a standardized coordinate system across atrial geometries, allowing for consistent fibre orientation mapping regardless of the patient’s atrial anatomy.

[0146] In the example shown, the step of defining a location and / or distribution of one or more inter-atrial connections in the 3D model in step 404 is performed manually by a skilled operator who uses a cardiac atlas to define the location and distribution of the anatomical connections 702a, 702b between the left and right atria. The skilled operator maps the cardiac atlas to the patient’s atrial anatomy using a Universal Coordinate Method. The anatomical connections 702a, 702b comprise: a Bachmann’s bundle, a specialised tract of atrial myocardium that conducts impulses from the right atrium to the left atrium; and septal connections comprising muscle bridges, which are areas of myocardial tissue that span an interatrial septum and electrically connect the left and right atria.

[0147] In some cases, the anatomical connections 702a, 702b comprise any one or more of: Bachmann’s bundle; an interatrial septum; a muscle bridge in a region of a coronary sinus; and a connection in a region of the posterior left atrium.

[0148] Figures 8A and 8B show, respectively, an anterior and a posterior view of a 3D model of an atrium with defined atrial regions (802a-802h). In the example shown, one or more atrial regions are defined in the step 406 by initially specifying regions which are likely to have similar electrophysiological properties based on a known body of literature. The step 406 further and optionally comprises performing a sensitivity analysis comprising determining how the atrial regions specified modify the one more parameters representing the one or more conduction pattern metrics associated with the patient’s atrial fibrillation. In the example shown, step 406 comprises defining one or more atrial regions which minimise an error value between recorded parameters and simulated parameters and / or optimise the similarity between the recorded parameters and measured parameters.

[0149] The step 408 of defining one or more electrophysiological properties for the one or more atrial regions comprises analysing clinical recording data to produce a patientspecific model of electrophysiological heterogeneities. Step 408 may further comprise a sensitivity analysis comprising determining how the electrophysiological properties modify the one more parameters representing one or more conduction pattern metrics associated with the patient’s atrial fibrillation. In the example shown, the electrophysiological properties are defined independently for the one or more atrial regions, and the electrophysiological properties comprise: action potential duration (duration of a cell model action potential); effective refractory period (longest coupled extrastimulus which fails to capture the tissue); ion channel expression (which are the conductances of ion channels in the model); transverse conductivities and longitudinal conductivities (which are conductivities of the tissue in across-fibre and along-fibre directions respectively).

[0150] In some cases step 408 comprises using a homogeneous model comprising uniform electrophysiological properties for the one or more regions.

[0151] In some cases step 408 comprises using an idealised heterogeneous model comprising predefined variations in electrophysiological parameters within one or more atrial regions. In some cases step 408 comprises using electrophysiological measurements of the patient’s tissue to determine electrophysiological properties for the one or more regions.

[0152] Figure 9A shows an anterior and posterior view of the 3D model of the patient’s atria, with pacing sites depicted as black stars. Figure 9A depicts the left inferior pulmonary vein 702g and left superior pulmonary vein 702d, the mitral valve 702e and the inferior vena cava 902. In the example shown, the step 410 of defining the number and / or distribution of pacing sites at which electrical propagation is simulated in the simulation comprises using an algorithm to select an arbitrary number of uniformly distributed pacing sites.

[0153] Uniformly distributed pacing involves placing pacing sites at regular distances or grid points across the atria, providing systematic stimulation of atrial tissue and facilitating the initiation of atrial fibrillation through widespread activation. Uniformly distributed pacing is reproducible, consistent, and scalable, enabling robust comparisons between simulations. Uniformly distributed pacing also enables initiation maps to be generated. Initiation maps provide valuable insights into the spatiotemporal patterns of atrial fibrillation onset, offering a visual representation of the regions within the atria where arrhythmia initiation is most likely to occur. Step 410 may comprise using targeted pacing comprising pinpointing anatomical or functional sites of interest and delivering controlled electrical stimuli to investigate the role such sites in arrhythmia initiation, propagation, and maintenance.

[0154] In some cases step 410 comprises using localised pacing comprising stimulating specific regions within the atria comprising any one or more of: the pulmonary veins; atrial appendages; and regions of fibrosis. These regions within the atria are known to harbour drivers of atrial fibrillation and are commonly targeted during clinical ablation procedures.

[0155] In some cases step 410 comprises using random pacing comprising stimulating random sites throughout the atria without predefined spatial constraints. Random pacing mimics the stochastic nature of ectopic activity and wavefront propagation observed in atrial fibrillation. Random pacing introduces variability in atrial fibrillation initiation patterns and may capture the spontaneous onset of atrial fibrillation in certain regions of the atria.

[0156] Figure 9B shows consecutive snapshots of the simulation of the patient’s atrial fibrillation. In the example shown, the pacing sites are activated at ‘t=0’. The electrical waves propagate across the atria and are measured at 1000 ms increments for up to at least 5000 ms.

[0157] In some cases, the electrical waves are measured at increments of the order of 10 ms, 50 ms, up to 100 ms, up to 500 ms, or more than 1000 ms.

[0158] In some cases, the electrical waves are measured for up to at least of the order of 100 ms, up to 1000 ms, up to 2000 ms, or more than 5000 ms.

[0159] Figure 10 shows a graph of an applied stimulation signal for an electrode applied to an atria, illustrating incremental pacing. In the example shown, the step 412 of defining the pacing protocol comprising the pacing sequence and / or rate of pacing for stimulating the pacing sites in the simulation comprises using incremental pacing to initiate atrial fibrillation. Incremental pacing comprises applying electrical stimuli to pacing sites at progressively increasing rates, leading to a stepwise acceleration of activation. Incremental pacing closely mimics the physiological progression towards atrial fibrillation onset by gradually increasing the rate of stimulation, allowing for a more nuanced exploration of arrhythmia susceptibility. Incremental pacing provides for a dynamic assessment of atrial electrophysiology, revealing rate-dependent alterations in effective refractory periods (the longest coupling interval which fails to capture the tissue), conduction velocity, and tissue excitability. This allows fora comprehensive understanding of the mechanisms underlying atrial fibrillation initiation. Incremental pacing offers flexibility in tailoring the pacing protocol to individual patient characteristics, enabling personalized assessments of arrhythmia vulnerability.

[0160] In some cases, step 412 comprises placing a temporary block line in the model of the atria to create a region of slow conduction or a unidirectional block in a location close to a pacing site, to initiate a re-entrant wave which arises when an impulse is blocked and can only travel around one side of a blocked area, forming a re-entry loop around the blocked area which the impulse continues to travel around.

[0161] In some cases, step 412 comprises placing one or more predefined spiral waves at a predetermined location in the model of the atria to trigger the onset of atrial fibrillation.

[0162] Temporary block lines and predefined spiral waves cause abrupt perturbations to atrial electrical activity, which may not fully capture the gradual changes seen in clinical scenarios, and may not account for patient-specific factors influencing AF susceptibility and response to pacing stimuli.

[0163] The step 414 of adjusting one or more defined features of the 3D model and / or simulation to tune the simulation of the patient’s atrial fibrillation to match measured patient data may comprise using a Bayesian inference optimisation algorithm and iteratively adjusting model parameters to minimize a discrepancy between simulated and recorded clinical data. Bayesian algorithm inference combines experimental data with prior knowledge about model parameters to infer posterior probability distributions, allowing for uncertainty quantification and robust estimation of atrial model parameters.

[0164] In some cases, step 414 comprises using statistical modelling techniques to relate model parameters to key physiological variables or experimental observations. Statistical modelling techniques may comprise regression analysis or generalized linear models. In some cases, step 414 comprises using a trained machine learning algorithm to identify patterns and relationships between model parameters and physiological variables from clinical data. The machine learning algorithm may comprise one or more of an artificial neural network, a support vector machine, and a random forest.

[0165] In some cases, step 414 comprises directly fitting simulated model outputs to clinical data which comprises experimental recordings of atrial electrical activity. Experimental recordings of atrial electrical activity may comprise action potential recordings or electrograms.

[0166] Figure 11A shows medical images obtained using LGE-MRI to determine the extent of atrial fibrosis (1102) (none, mild, moderate, severe). Electroanatomic mapping is an invasive process which requires electrodes to be placed within the heart for intracardiac or epicardial electrocardiogram recording. Magnetic, electric or impedance fields may be used to map the geometry of the heart. Simulating electrical wave propagation through fibrotic tissue provides insights into arrhythmia mechanisms such as arrhythmia initiation, maintenance, and termination.

[0167] Figure 11 B shows atrial LGE shells reconstructed from atrial LGE imaging and visualised using an image processing toolkit for cardiovascular data, where the shells correspond to the categories of qualitative assessment of atrial fibrosis (1104) as depicted in Fig. 11 A. Atrial fibrosis results in changes in electrophysiological properties which influence the atrial fibrillation mechanism. Therefore, incorporating LGE-MRI data allows for the accurate projection of fibrosis distribution.

[0168] Figure 11C shows electroanatomic voltage shells visualised using an electrophysiology data analysis tool, where the shells correspond to the categories of qualitative assessment of atrial fibrosis as depicted in Fig. 11 A. In the example shown, the voltage maps identify heart regions exhibiting abnormal electrogram characteristics (1106) such as low voltage or fractionated signals, which are determined to correspond to fibrotic regions of the heart.

[0169] In some cases, so-called T1 mapping MRI can detect and characterize fibrosis within the atria, as a T1 relaxation time of tissue is greater for fibrotic tissue than healthy tissue. Delayed enhancement computed tomography involves acquiring CT images after the administration of contrast agents.

[0170] Figure 12A shows two images of the simulated atria with a number of identified atrial fibrillation waves (1202, 1024, 1206, 1208) being tracked in time from one image to the next. In the example shown, the step 102 of extracting, from the simulation and / or measured data of the patient’s atrial fibrillation, one or more parameters representing one or more conduction pattern metrics associated with the patient’s atrial fibrillation comprises extracting the number of waves. The number of waves reflects the spatial complexity of atrial activation during atrial fibrillation. Computational algorithms can be used to analyse the simulated conduction patterns to identify distinct wavefronts propagating through the atria. A higher number of waves (e.g. wavefronts) suggests increased spatial disorganisation and complexity in atrial activation patterns. In some cases, step 102 comprises using an algorithm to determine and / or track the number of waves.

[0171] Figure 12B shows two images of simulated atria illustrating transmembrane potential for phase singularity detection, with phase singularity locations depicted as white spheres (1210). In the example shown step 102 comprises extracting a number of phase singularities. These singularities serve as anchoring centres for re-entrant waves, contributing to the perpetuation of atrial fibrillation. Identification of phase singularities and tracking their location over time provides insights into the dynamics of re-entrant activity.

[0172] In some cases, step 102 comprises extracting the frequency of electrical waves in the atria.

[0173] In some cases, step 102 comprises extracting the duration of electrical waves and / or phase singularities in the atria which persist during atrial fibrillation. Monitoring the temporal evolution of electrical waves and / or phase singularities offers insights into the stability and sustainability of re-entrant waves. Prolonged durations may indicate more complex atrial fibrillation patterns.

[0174] Figures 13A and 13B relate to the case where measured data is obtained and used to extract the parameters in accordance with step 102.

[0175] Figure 13A shows electrode locations for measuring atrial fibrillation and illustrates sequential measurements of atrial regions. A set of electrode recording locations 1302 is shown in the left atrium, where each recording location has been recorded consecutively, but all recording locations are shown collectively. Figure 13A depicts the right superior pulmonary vein 702c, the left superior pulmonary vein 702d, and the mitral valve 702e.

[0176] In the example shown, the step of measuring atrial fibrillation comprises a surgeon or cardiologist performing a transvenous procedure to temporarily place catheters within the heart and electrodes on the endocardial surface of a patient’s atria. The measured data of the patient’s atrial fibrillation is used in step 102 of Figure 1 , which comprises extracting, from a simulation and / or measured data of the patient’s atrial fibrillation, one or more parameters representing one or more conduction pattern metrics associated with the patient’s atrial fibrillation.

[0177] In the example shown, repetitive atrial activation patterns are recorded during atrial fibrillation by sequentially recording from spatially overlapping regions of the atria and producing a composite map of the repetitive atrial activation patterns over time. Drivers of atrial fibrillation may be tracked over larger areas of the atria while maintaining high spatial resolution. This is achieved by comparing repetitiveness of conduction patterns with earlier recorded activation, where lower repetitiveness indicates a current location of atrial fibrillation drivers. A threshold is used to determine similarity between temporally consecutive composite maps to predict electrical wave propagation direction.

[0178] In some cases, the step of measuring atrial fibrillation comprises the surgeon performing left heart catheterisation, which involves a surgeon inserting a catheter into an artery in the wrist or groin and guiding the tube into the left side of the heart. The catheter may comprise the electrode.

[0179] In some cases, the step of measuring atrial fibrillation comprises the clinician performing non-contact charge density mapping, which involves mapping intracardiac electrograms to the model of the patient’s anatomy.

[0180] Figure 13B shows drivers of atrial fibrillation in different locations of the atria. In the example shown, electrical waves are propagated across the atria, with their propagation path determined at least in part by the conduction properties of the tissue. A subset of electrode recording locations 1302 from Figure 13A are shown in Figure 13B, where similarity of consecutive beats during atrial fibrillation has been used to identify the sequential activation pattern, highlighted by arrows. Figure 13B depicts the right superior pulmonary vein 702c, the left superior pulmonary vein 702d, and the mitral valve 702e.

[0181] Figure 14A shows a graph illustrating wave duration.

[0182] In the example shown, the graph depicts wave duration measurements between 0 and 3000 ms. In some cases, the wave duration may be measured for on the order of 1000 ms, up to 2000 ms, up to 5000 ms or more.

[0183] Figure 14B shows a graph illustrating the duration of phase singularities.

[0184] In the example shown, the graph depicts phase singularity duration measurements between 0 and 3000 ms. In some cases, the phase singularity duration may be measured for on the order of 1000 ms, up to 2000 ms, up to 5000 ms or more.

[0185] Figure 15A shows a graph illustrating clustering of atrial fibrillation conduction pattern complexities quantified by number of waves and number of phase singularities (PS). Patients with higher complexity metrics may represent distinct subpopulations with unique arrhythmia characteristics. In the example shown, the data is clustered into four classes. In the example shown, clustering is performed using a k-means clustering algorithm.

[0186] In the example shown, determining classification of the patient’s atrial fibrillation in the step 104 comprises determining the class of the atrial fibrillation conduction pattern complexities based on clustering.

[0187] In some cases clustering may be performed based on atrial fibrillation conduction pattern complexities quantified by any one or more of: number of waves, density of phase singularities, duration of one or more waves; and duration of one or more phase singularities.

[0188] In some cases, clustering may be performed by any one or more of: a MeanShift clustering algorithm; a DBSCAN clustering algorithm; a hierarchical clustering algorithm; a BIRCH clustering algorithm.

[0189] In some cases, a number of classes may be of the order of 2, up to 5, and up to 10 or more.

[0190] The parameters representing one or more conduction pattern metrics associated with the patient’s atrial fibrillation may comprise one or more of: number of electrical waves propagating in the simulation in a first predefined time period; number of phase singularities detected in the simulation in a second predefined time period; a frequency of electrical waves induced and / or sustained in a third predefined time period; a duration of electrical waves sustained in a fourth predefined time period; and a duration of phase singularities sustained in a fifth predefined time period. Any or all of the first to sixth predefined time periods may be the same or different, and may be chosen by any one or more of: the skilled operator; the clinician; an algorithm; and a machine learning model. Phase singularities may also be referred to as pivot points. Phase singularities represent points within the atria where a phase of an activation wavefront becomes undefined or undergoes an abrupt change.

[0191] Figure 15B shows a graph of atrial fibrillation initiation rate versus number of atrial fibrillation waves. In the example shown, as the atrial fibrillation rate increases from 40% to 80% in increments of 10%, the number of waves increases substantially linearly. The number of waves varies from approximately 2 waves to approximately 16 waves.

[0192] Figure 15C shows a graph depicting, for a number of patients, a predicted probability of the patient having atrial fibrillation according to four different classifications. Subclassification of atrial fibrillation patients based on conduction pattern complexity metrics may uncover specific clinical phenotypes associated with different arrhythmia manifestations and outcomes.

[0193] In the example shown, the predicted probability is based on the confidence score of the classification. In some cases, the predicted probability is based on a threshold value for one or more of the atrial fibrillation conduction pattern complexities.

[0194] In the example shown, the majority of patients have a non-zero probability of belonging to more than one class. The majority of patients have a non-zero probability of belonging to two classes. An equal number of patients have a non-zero probability of belonging to one class as three classes.

[0195] In some cases, a patient may have a high probability of belonging to a particular class, and a lower probability of belonging to at least one other particular class.

[0196] In some cases, a patient may have an equal probability of belonging to two or more particular classes.

[0197] In some cases, the patients may be further grouped by the class which their atrial fibrillation has the highest probability of belonging to. In some cases, a probability of that class being correct may be determined.

[0198] In the example shown, determining classification of the patient’s atrial fibrillation in the step 104 comprises selecting the class which their atrial fibrillation has the highest probability of belonging to.

[0199] In some cases, step 104 comprises taking into account the probability that the atrial fibrillation conduction pattern complexities for the patient belong to one or more particular classes.

[0200] In the example shown, the patients are further grouped according to the known types of paroxysmal atrial fibrillation and persistent atrial fibrillation. This illustrates that within each of the known types of atrial fibrillation, further sub-classes are present.

[0201] In some cases, the patients may be further grouped by any one or more of: age, sex, previous treatment, left atrial diameter, and presence or absence of coronary artery disease.

[0202] In optional step 206, the treatment is determined based on the classification, using a machine learning model trained on patient classification, treatment, and treatment success data. In some cases, the treatment is determined based on entries in a lookup table comprised of treatments patients have historically responded well to, with reference to their classification. In some cases, the treatment is determined based on the group the patient belongs to. In optional step 208, the likelihood of success of the treatment is based on the classification. In the example shown, the likelihood of success is based on the probability of the patient’s atrial fibrillation belonging to a particular classification. In some cases, the likelihood of success is based on the particular classification, for example if the patient is classified as having atrial fibrillation which is known to respond well to a particular treatment. In some cases, the likelihood of success is based on the group the patient belongs to, for example it may be determined that patients who have responded well to a particular treatment in the past should receive that particular treatment.

[0203] Figure 16 shows a block diagram of a system 1600 for determining a patient’s atrial fibrillation classification.

[0204] In the example shown, processing circuitry 1602 is connected to memory 1604 and a display 1606. The memory 1604 is configured to store computer-readable instructions that are executable to perform the method for determining a patient’s atrial fibrillation classification in accordance with at least Figure 1. The processing circuitry 1602 is configured to execute the computer-readable instructions.

[0205] In the example shown, the processing circuitry 1602 is connected to the memory 1604 and the display 1606 via wired connections (1608, 1610). In some cases the processing circuitry 1602 may be connected to the memory 1604 and / or the display 1606 via a wireless connection. In some cases the processing circuitry 1602 is connected to the memory 1604 and the display 1606 via one or more intermediary devices (not shown). In some cases the memory 1604 is connected to the display 1606 via a wired (1608, 1610) or wireless connection, and / or via an intermediary device (not shown).

[0206] The step 202 of determining the probability of the patient’s atrial fibrillation belonging to a particular classification as depicted in Figure 2 may comprise displaying the probability on the display 1606. The probability may be displayed as a graphic on the display 1606.

[0207] The step 204 of determining the classification of the patient’s atrial fibrillation based on the probability of the patient’s atrial fibrillation belonging to a particular classification as depicted in Figure 2 may comprise displaying the classification on the display 1606. Displaying the classification may comprise plotting the patient within a full parameter space of clustering in a graphic representation to give a visual representation of where the patient ‘sits’ amongst all possible atrial fibrillation mechanisms.

[0208] The optional step of determining the treatment based on the classification of the patient’s atrial fibrillation 206 as depicted in Figure 2 may comprise displaying the treatment on the display 1606. The treatment may be displayed as a graphic on the display 1606.

[0209] The optional step of determining the likelihood of success of the treatment based on the classification 208 as depicted in Figure 2 may comprise displaying the likelihood on the display 1606. The likelihood of success may be displayed as a graphic on the display 1606.

[0210] In some cases, the probability is represented with any one or more of a text, a number and a code.

[0211] In some cases, the classification is represented with any one or more of a text, a number and a code.

[0212] In some cases, the treatment may be represented with any one or more of a text, a number and a code.

[0213] In some cases, the likelihood of success may be represented with any one or more of a text, a number and a code.

[0214] Example Study

[0215] At the time of writing, the Applicant is performing a prospective observational cohort study in 91 patients undergoing first-time atrial fibrillation catheter ablation procedures. The data presented here is a preliminary analysis of 21 patients (38% female) in whom follow-up data are available. Amongst these patients, 52% have experienced arrhythmia recurrence.

[0216] All study participants underwent pre-procedural atrial cardiac magnetic resonance imaging (MRI). As illustrated in Figure 17, bi-atrial MRI segmentations 1700 were used to generate patient-specific bi-atrial anatomy models using CEMRGApp, and postprocessing was performed with EP Workbench. The Atrial Modelling Toolkit, implemented through EP Workbench was used to generate bilayer bi-atrial models 1702, with fibre orientations mapped from a human fibre atlas using Universal Atrial Coordinates. The degree of atrial fibrosis was quantified through late gadolinium enhancement MRI analysis, and tissue conductivities were adjusted accordingly in fibrotic regions 1704 through EP Workbench. Computer models of human atrial tissue were constructed using the Courtemanche-Ramirez-Nattel model. For each patient, simulations were conducted from 20 distinct pacing sites 1706 uniformly distributed across the right and left atria. Parameters representing individualised conduction patterns were then extracted from 420 simulations performed for 21 patients, for purposes of patient classification. Further details of the study and the results are outlined below.

[0217] Computational modelling-based patient phenotyping

[0218] Previously, the Applicants demonstrated that four distinct mechanism-based subgroups could be identified according to the numbers of phase singularities and depolarisation waves in simulations (e.g. see Fig. 15B), and that these subgroups (class 1 to 4) correlated with arrhythmia duration-based atrial fibrillation classification. Further analysis of the prospective cohort reveals greater mechanistic complexity in patients with arrhythmia recurrence, as demonstrated by higher numbers of phase singularities (560 ± 89.1 versus 380 ± 68.1) and a trend towards high numbers of waves (910 ± 86.6 versus 846 ± 79.4) in simulations of patients with arrhythmia recurrence. This is illustrated in Figures 18A and 18B, respectively showing the mean number of (A) phase singularities and (B) waves per simulation in patients with and without arrhythmia recurrence. Error bars represent standard error of the mean.

[0219] As shown in Figure 19, four distinct mechanism-based atrial fibrillation subgroups (e.g. clusters) were identified from the prospective cohort using unsupervised machine learning employing K-means clustering. In this case, a first subgroup comprises a low number of phase singularities and a low number of electrical waves; a second subgroup comprises a low number of phase singularities and an intermediate number of electrical waves; a third subgroup comprises an intermediate number of phase singularities and an intermediate number of electrical waves; and a fourth subgroup comprises a high number of phase singularities and a high number of electrical waves.

[0220] In addition, patients in the higher atrial fibrillation complexity subgroups (Intermediate Waves / lntermediate Phase Singularities and High Waves / High Phase Singularities) were identified to have higher rates of arrhythmia recurrence (60.6% and 64.7%, respectively) compared with patients in the lower atrial fibrillation complexity subgroups (52.1% and 45.3%), as shown in the table of Figure 20.

[0221] Novel parameters

[0222] In addition to numbers of waves and phase singularities, the Applicants have extracted novel parameters. Rates of atrial fibrillation initiation (corresponding to sustained atrial activity for at least two cycles after pacing (assuming an atrial fibrillation cycle length of 150ms)) and atrial fibrillation termination (instances where no waves are detectable at the end of a simulation of successful atrial fibrillation initiation) have been calculated and plotted, respectively, in Figures 21 A and 21 B, for patients with and without arrhythmia recurrence.

[0223] The results demonstrate a trend toward higher atrial fibrillation initiation rates and lower atrial fibrillation termination rates in patients with arrhythmia recurrence.

[0224] Additional parameters describing wave / phase singularity characteristics and information pertaining to spatial location have also been extracted. Figure 22 shows histograms illustrating the distribution of additional parameters extracted from simulations (where PS = phase singularity). In particular, distributions amongst simulations in patients with atrial fibrillation are provided of the following parameters: total number of phase singularities in both atria, left atrium and right atrium; average number of phase singularities per timestep in both atria, left atrium and right atrium; number of pulmonary vein phase singularities; proportion of left atrial phase singularities in the pulmonary veins; average region coverage of each phase singularity; total number of waves; average number of waves per timestep; and average wave area / mesh area. Any of these parameters (or others) may be used to determine a classification of the patient’s atrial fibrillation.

[0225] Treatment response prediction

[0226] To predict risk of arrhythmia recurrence, a random forest classifier has been developed using parameters extracted from simulations. Variables with collinearity were excluded. The model was trained with 500 trees, and its performance was evaluated using Area Under the Receiver Operating Characteristic Curve (AUC-ROC) analysis. The computational modelling-based model achieved an Area Under the Curve (AUC) of 0.717 (95% confidence level (Cl): 0.669-0.765) for predicting arrhythmia recurrence risk in the training dataset, as illustrated in Figure 23.

[0227] The Mean Decrease in Accuracy was used to determine the features that contributed most to model decision-making. Preliminary analysis identified the average wave count per timestep to be the most important parameter in evaluating model performance, followed by the average number of phase singularities per timestep in the left atrium. Figure 24 shows the full results of the Mean Decrease in Accuracy plot to determine how much the model’s accuracy decreases when the parameter values, of the parameters described herein, are permuted.

[0228] The initial results from the study demonstrate: 1) distinct computational modellingbased atrial fibrillation phenotypes which correlate with response to ablation (indicated by arrhythmia recurrence); 2) the identification of novel computational modeling parameters and their distribution across simulations in patients with atrial fibrillation; and 3) the feasibility of a model based on computational modelling-derived parameters to predict patient response to ablation.

[0229] Thus, it will be understood that the present disclosure provides an improved data- driven approach for classifying a patient’s atrial fibrillation and which can be used to determine more effective treatments and outcomes.

[0230] Although the disclosure has been described in terms of some specific examples as set forth above, it should be understood that these examples are illustrative only and that the claims are not limited to those examples. Those skilled in the art will be able to make modifications and alternatives in view of the disclosure, which are contemplated as falling within the scope of the appended claims. Each feature disclosed or illustrated in the present specification may be incorporated in any embodiments, whether alone or in any appropriate combination with any other feature disclosed or illustrated herein.

Claims

CLAIMS:

1. A computer-implemented method for determining a patient’s atrial fibrillation classification comprising; extracting, from a simulation and / or measured data of the patient’s atrial fibrillation, one or more parameters representing one or more conduction pattern metrics associated with the patient’s atrial fibrillation; and using the one or more parameters to determine a classification of the patient’s atrial fibrillation.

2. The method of claim 1 wherein the one or more conduction pattern metrics comprise a complexity and / or extent of the patient’s atrial fibrillation.

3. The method of claim 1 or 2 wherein the one or more parameters comprise a number of electrical waves propagating in a first predefined time period.

4. The method of any preceding claim wherein the one or more parameters comprise a number of phase singularities detected in a second predefined time period.

5. The method of claim 4, when dependent on claim 3, further comprising using the number of phase singularities and the number of electrical waves to determine the classification.

6. The method of any preceding claim wherein the step of using the one or more parameters to determine the classification comprises determining a probability of the patient’s atrial fibrillation belonging to a particular classification.

7. The method of any preceding claim further comprising determining a treatment based on the classification.

8. The method of claim 7 further comprising determining a likelihood of success of the treatment based on the classification; optionally wherein the likelihood of success is based on a probability of the patient’s atrial fibrillation belonging to the classification.

9. The method of any preceding claim wherein the one or more parameters comprise a frequency of electrical waves induced and / or sustained in a third predefined time period.

10. The method of any preceding claim wherein the one or more parameters comprise a duration of electrical waves sustained in a fourth predefined time period.11 . The method of any of: claim 10 when dependent on any one or more of claims 3, 4 and 9; claim 9 when dependent on any one or more of claims 3 and 4; claim 4 when dependent on claim 3; wherein any two or more of the first predefined time period, the second predefined time period, the third predefined time period and the fourth predefined time period have a different start time and / or have a different end time.

12. The method of any preceding claim further comprising creating the simulation of the patient’s atrial fibrillation.

13. The method of claim 12 wherein creating the simulation of the patient’s atrial fibrillation comprises: obtaining one or more measurements of the patient’s atrial anatomy; processing the one or more measurements to create a 3D model of the patient’s atrial anatomy; and simulating the patient’s atrial fibrillation using the 3D model.

14. The method of claim 13 wherein the step of obtaining one or more measurements of the patient’s atrial anatomy comprises one or more of: electro-anatomic mapping; echocardiography; and inferring atrial size from anthropometry measurements.

15. The method of claim 13 wherein the step of creating the 3D model of the patient’s anatomy comprises one or more of: rigid scaling of a pre-defined atrial anatomical model; deformable scaling of a pre-defined atrial anatomical model; inferring atrial anatomy using a pre-trained statistical shape model.

16. The method of claim 12 wherein creating the simulation of the patient’s atrial fibrillation comprises: obtaining one or more images of the patient’s atrial anatomy; processing the one or more images to create a 3D model of the patient’s atrial anatomy; and simulating the patient’s atrial fibrillation using the 3D model.

17. The method of claim 16 wherein the step of obtaining one or more images of the patient’s atrial anatomy comprises one or more of: magnetic resonance imaging; late gadolinium enhancement magnetic resonance imaging; diffusion tensor magnetic resonance imaging; computed tomography scanning; echocardiogram imaging; nuclear cardiac stress testing; single-photon emission computed tomography; cardiac positron emission tomography; multigated acquisition scanning; a coronary angiogram; cardiac angiography; intracardiac echocardiography; electro-anatomic mapping or left heart catheterization.

18. The method of claim 16 wherein creating the 3D model of the patient’s atrial anatomy comprises segmenting the one or more images of the patient’s atrial anatomy.

19. The method of any one of claims 13 to 18 wherein the step of creating the 3D model of the patient’s anatomy comprises any one or more of: creating a mesh model of the patient’s atrial anatomy; defining fibre orientation and / or distribution across the 3D model; defining a location and / or distribution of one or more intra-atrial connections in the 3D model; defining one or more atrial regions; defining one or more electrophysiological properties for the one or more regions; defining a number and / or distribution of pacing sites at which electrical propagation is simulated in the simulation; defining a pacing protocol comprising a pacing sequence and / or rate of pacing for stimulating the pacing sites in the simulation; and adjusting one or more defined features of the 3D model and / or simulation to tune the simulation of the patient’s atrial fibrillation to match measured patient data.

20. The method of claim 13 or 16 further comprising determining an extent of atrial fibrosis and taking the extent of atrial fibrosis into account in the simulation by including the extent of atrial fibrosis in the 3D model.

21. The method of claim 19 wherein defining the location and / or distribution of one or more intra-atrial connections in the 3D model comprises using a location of an anatomical connection; optionally, the anatomical connection may comprise one or more of: Bachmann’s bundle; an interatrial septum; a muscle bridge in a region of a coronary sinus; and a connection in a region of a posterior left atrium.

22. The method of claim 19 wherein the electrophysiological properties comprise one or more of: an action potential duration; a longitudinal conductivity; a transverse conductivity; an ion channel conductance; a transmural conductivity; a conductivity of an inter-atrial connection; a conduction velocity.

23. The method of any of claims 1 to 12 further comprising obtaining measured data of the patient’s atrial fibrillation.

24. A non-transitory computer-readable medium that comprises computer-readable instructions that are executable to perform a method for determining a patient’s atrial fibrillation classification, the method comprising: extracting, from a simulation and / or measured data of the patient’s atrial fibrillation, one or more parameters representing one or more conduction pattern metrics associated with the patient’s atrial fibrillation; and using the one or more parameters to determine a classification of the patient’s atrial fibrillation.

25. A system for determining a patient’s atrial fibrillation classification, the system comprising processing circuitry configured to: extract, from a simulation and / or measured data of the patient’s atrial fibrillation, one or more parameters representing one or more conduction pattern metrics associated with the patient’s atrial fibrillation; and use the one or more parameters to determine a classification of the patient’s atrial fibrillation.

Images