Method and system for quantifying blood-flow stagnation in a heart using medical imaging

Abstract

The present invention relates to a computer-implemented method for quantifying blood- flow stagnation in a heart of a subject comprises receiving medical imaging data representing heart geometry, identifying anatomically distinct cardiac components, and determining component-specific intracardiac motion over a cardiac cycle from the 5 imaging data. A time-dependent intracardiac fluid-flow field is then determined based on the motion data, and a blood-flow stagnation value is computed as a physical quantity derived from the fluid-flow field, including at least one of a residence-time, wash-out, recirculation, low-velocity indicator, or passive-scalar transport metric. The stagnation value represents slow or recirculating blood flow in one or more identified 10 components and is provided as an output, optionally as a component-wise map or spatial distribution.

Description

[0001] METHOD AND SYSTEM FOR QUANTIFYING BLOOD-FLOW STAGNATION IN A

[0002] HEART USING MEDICAL IMAGING

[0003] Technical Field

[0004] The present invention relates to computational analysis of intracardiac blood flow. More particularly, the invention concerns computer-implemented methods, systems, and machine-readable media for determining stagnation-related flow parameters in cardiac chambers based on medical imaging data.

[0005] Background

[0006] Stroke is one of the leading causes of death and long-term disability worldwide. A substantial proportion of strokes and other cerebral embolic events are associated with thromboembolic phenomena, which may arise under conditions of altered blood flow, coagulation, or vascular pathology. Atrial fibrillation (AF) is recognized as a major contributor to such phenomena, as it is associated with impaired atrial contraction and altered intracardiac blood-flow patterns that can promote blood stasis and coagulation. The prevalence of AF is high and increases with age, making it a significant populationlevel contributor to thromboembolic-related cerebral events. Despite advances in clinical risk assessment, accurately characterizing flow-related contributors to thromboembolic phenomena remains challenging, particularly across different physiological or rhythm conditions.

[0007] Clinically, the likelihood of stroke in a subject with AF, independent of AF type, is often estimated using scoring instruments such as CHA2DS2-VASc. While such scores reliably identify low-risk individuals, they only moderately stratify subjects with higher susceptibility. To mitigate thromboembolic-related cerebral events, many subjects with AF receive anticoagulant therapy; however, anticoagulant use is associated with an increased risk of intracerebral bleeding. This highlights the need for improved and more individualized stratification based on underlying physiological and hemodynamic factors.

[0008] Thromboembolic material may form on cardiac walls and travel through the bloodstream to occlude cerebral arteries. In AF, thrombus formation has frequently been observed in the left atrial appendage (LAA). A prevailing hypothesis is that atrial fibrillation leads to suboptimal blood flow in the left atrium (LA), particularly in the LAA, resulting in regions of blood stasis that may promote thrombus formation. However, this hypothesis has limitations, and existing approaches do not reliably characterize thromboembolic-related flow abnormalities, even in subjects who are in sinus rhythm at the time of imaging.

[0009] Existing image-based cardiac flow analysis techniques vary in their modelling assumptions and levels of anatomical and temporal detail. Some approaches employ simplified geometric representations or motion models, treat the atrium as a single region, or derive indices related to blood stasis from anatomical or motion-based parameters. Depending on the modelling approach and data available, such techniques may represent atrial motion or intracardiac flow conditions with differing degrees of spatial or temporal resolution.

[0010] In certain anatomical regions of the atrium, blood motion can be sensitive to subtle geometric features or temporal variations in wall motion. Approaches that rely on coarse representations or aggregated regional measures may therefore provide flow characterisations that differ from those obtained using more detailed anatomical or dynamic representations.

[0011] There is ongoing interest in techniques for analysing atrial blood flow that take into account the dynamic nature of intracardiac physiology and the potential influence of anatomical and motion-related variations on flow behaviour.

[0012] Summary

[0013] An objective of the present inventive concept is to provide an improved method for accurately quantifying blood flow behaviour in the heart of a subject. A particular aim is to enable reliable identification of blood-flow patterns associated with stagnation or reduced wash-out, based on information derived from medical imaging data. Further aims include supporting automated analysis workflows and facilitating the generation of parameters that may be used in clinical evaluation and decision-making.

[0014] These and other objectives are achieved by the invention as defined in the independent claims. Preferred embodiments are set out in the dependent claims.

[0015] According to a first aspect of the present inventive concept there is provided a computer-implemented method for quantifying a blood-flow stagnation in a heart of a subject. The method comprises providing input data including or derived from medical imaging data representing the heart of the subject, identifying a plurality of anatomically distinct components in the imaging data, determining component-specific intracardiac motion data for the identified components based on a motion estimation applied to the input data, determining a time-dependent fluid-flow field based on the componentspecific intracardiac motion data, determining a blood-flow stagnation value based on the fluid-flow field, and providing the blood-flow stagnation value.

[0016] Hence, there is provided an improved method for quantifying blood-flow stagnation. The method considers blood-flow stagnation or stasis, which is relevant to coagulation according to Virchow’s triad.

[0017] Identification of anatomical structures may be applied to any suitable medical imaging data, for example by segmentation, atlas-based detection, machine-learning-based identification, or other techniques appropriate for the imaging modality. Motion estimation may be performed using approaches such as image registration, optical flow, biomechanical modelling, or machine learning. A time-dependent fluid-flow field may be determined using computational fluid dynamics (CFD), reduced-order modelling, physics-informed neural networks, or other numerical or learned approaches capable of approximating intracardiac hemodynamics.

[0018] As used herein, the term “component” refers to any anatomically distinct region of the heart, including chambers, appendages, valves, vessels, or substructures thereof.

[0019] The term “motion estimation” refers to any method of determining time-varying geometric or positional changes of anatomical structures, including but not limited to image registration, optical flow, biomechanical modelling, atlas-based modelling, machine-learning inference, or combinations thereof. Motion estimation may be based on fully time-resolved imaging data, sparsely sampled imaging data, or single-phase imaging data, and may comprise interpolation, extrapolation, or model-based inference to obtain motion information over a cardiac cycle.

[0020] The term “fluid-flow field” encompasses any numerical, surrogate, reduced-order, or learned representation of blood-flow velocity and / or pressure distributions over time within the heart.

[0021] The term “blood-flow stagnation value” refers to a physical quantity or derived parameter that is computationally obtained from the time-dependent fluid-flow field and that reflects the presence of slow, recirculating, or retained blood within a region of the heart. Such stagnation values may include, but are not limited to, residence-time measures, wash-out measures, recirculation indicators, passive-scalar-based measures, or low-velocity indices.

[0022] As used herein, the term “subject” refers to any individual from whom medical imaging data of the heart is obtained, including patients, healthy volunteers, or retrospective datasets.

[0023] In clinical practice, cardiac medical imaging data is not always acquired as a fully time- resolved dataset covering an entire cardiac cycle. Depending on imaging protocol, dose constraints, heart-rate variability, arrhythmia, or the clinical question, imaging modalities such as computed tomography may provide a single cardiac phase or a limited subset of cardiac phases. As a result, direct extraction of subject-specific cardiac wall motion over a full cardiac cycle may not be available from the imaging data alone.

[0024] In such situations, time-varying cardiac motion information may be obtained by combining the available imaging data with motion-estimation or motion-inference techniques, optionally together with auxiliary physiological information or modelling assumptions. These approaches may be used to complete or infer missing portions of cardiac motion over a cardiac cycle and to derive physiologically consistent motion representations suitable for subsequent determination of a time-dependent fluid-flow field.

[0025] In some embodiments, the method quantifies blood-flow stagnation based on a simulated or approximated time-dependent fluid-flow field, derived from image-based anatomical data and measured or inferred motion information. The blood-flow stagnation value may be automatically generated as a physical quantity derived from the time-dependent fluid-flow field, for example by computational post-processing that determines one or more of residence-time, wash-out, recirculation, low-velocity, or passive-scalar transport metrics. The stagnation value may, in examples, be combined with additional parameters for reporting or visualisation. Optional exploration of different physiological or anatomical scenarios may be supported, such as modifying heart rate or rhythm or simulating structural variations of the atrial anatomy. Motion estimation may be carried out using image registration, optical flow, biomechanical modelling, machine-learning deformation inference, or any equivalent method capable of providing time-varying geometric information, including by inference from single-phase or sparsely sampled imaging data. The fluid-flow field may be determined using numerical solvers, reduced-order models, physics-informed neural networks, surrogate models, or any other computational or learned approach providing an approximation of intracardiac hemodynamics.

[0026] Additional modelling inputs may include representations of flow inlets or outlets, such as pulmonary veins or the aortic outlet, which may be incorporated using pressurebased or flow-based boundary models informed by anatomical measurements, inferred motion, physiological parameters, and / or imaging-based measurements. Anatomical and motion information derived or inferred from the imaging data may in some embodiments be combined with supplementary data obtained from other modalities, such as time-resolved flow MRI or ultrasound-derived measurements.

[0027] In further embodiments, an anatomical representation of the heart may be obtained from the imaging data using identification techniques suited to the modality and resolution of the data. The imaging data may comprise a single cardiac phase, a subset of cardiac phases, or time-resolved imaging data. Motion estimation may then be performed to obtain time-varying geometric information over a cardiac cycle, which may serve as input to a hemodynamic model configured to determine a fluid-flow field from which a stagnation parameter may be derived. Where the imaging data does not directly depict motion over the full cardiac cycle, the time-varying geometric information may be inferred or completed using model-based estimation, interpolation, extrapolation, or learned motion representations.

[0028] Medical imaging data may include computed tomography, magnetic resonance imaging, 4D flow MRI, ultrasound imaging, or other imaging modalities capable of depicting cardiac anatomy and / or motion. Such imaging data may be processed to derive an anatomical representation of the heart and, where applicable, temporal information reflecting cardiac motion. Identification of cardiac components may be carried out using techniques such as segmentation, atlas-based detection, or machinelearning-based identification. Motion estimation may be performed using approaches such as image registration, optical-flow analysis, biomechanical modelling, or machine- learning-based inference, including inference from sparsely sampled or single-phase imaging data.

[0029] In embodiments where image registration is used for motion estimation, a non-rigid registration algorithm may be employed to align images representing cardiac anatomy acquired at different cardiac phases or under different conditions. In cases where multiple phases are available, such registration may be used to estimate cardiac deformation over time and to provide time-resolved motion fields for one or more identified cardiac components. Where imaging data comprises a single phase or a limited subset of phases, registration may be combined with model-based, atlas-based, or learned motion representations to infer time-varying motion fields consistent with the available data. Registration outputs may, in certain examples, be re-meshed into surface or volumetric mesh representations suitable for subsequent analysis or visualisation. These processed data may then be incorporated into simulation models to facilitate computation of a time-dependent fluid-flow field.

[0030] The identified components of the heart may include any one or more of the left atrium, left atrial appendage, left ventricle, ascending aorta, pulmonary veins, superior vena cava, inferior vena cava, pulmonary trunk, right ventricle, right atrium, or other cardiac structures. Optional modelling considerations may include the effects of papillary muscles or trabeculations, which may influence the determination of motion or the resulting fluid-flow field.

[0031] In different embodiments, the time-dependent fluid-flow field may be determined for any selected region or combination of regions of the heart, for example only the left atrium and left atrial appendage. Boundary conditions between regions may be based on anatomical measurements, motion-derived or inferred parameters, or other modelling assumptions appropriate to the selected hemodynamic model. The computed fluid-flow field may be laminar, turbulent, or transitional depending on the configuration of the model. Blood viscosity may be represented using Newtonian or non-Newtonian formulations, with density and viscosity set to predetermined or subjectspecific values.

[0032] In one embodiment, the computer-implemented method receives medical imaging data comprising computed tomography (CT), magnetic resonance imaging (MRI), 4D flow MRI, ultrasound imaging, or any combination thereof. In examples, CT data provides high-resolution anatomical geometry and may be acquired as time-resolved CT or as a single-phase, gated acquisition; cine MRI provides time-resolved anatomical motion information; 4D flow MRI provides time-resolved velocity measurements that may be used directly as flow-field input, boundary-condition constraints, and / or for validation; and ultrasound (e.g., transthoracic or transesophageal echocardiography) provides anatomical and / or Doppler-derived flow waveforms that may be used to inform inflow / outflow profiles or motion inference. In further embodiments, multimodal combinations are used, for example CT for anatomy together with ultrasound or 4D flow MRI for waveform and / or flow constraints.

[0033] In one embodiment, the computer-implemented method receives medical imaging data comprising computed tomography (CT), magnetic resonance imaging (MRI), 4D flow MRI, ultrasound imaging, or any combination thereof. In examples, CT data provides high-resolution anatomical geometry and may be acquired as time-resolved CT or as a single-phase, gated acquisition; cine MRI provides time-resolved anatomical motion information; 4D flow MRI provides time-resolved velocity measurements that may be used directly as flow-field input, boundary-condition constraints, and / or for validation; and ultrasound (e.g., transthoracic or transesophageal echocardiography) provides anatomical and / or Doppler-derived flow waveforms that may be used to inform inflow / outflow profiles or motion inference. In further embodiments, multimodal combinations are used, for example CT for anatomy together with ultrasound or 4D flow MRI for waveform and / or flow constraints.

[0034] In one embodiment, the medical imaging data comprises a single cardiac phase or a subset of cardiac phases, and the computer-implemented method infers the component-specific intracardiac motion data over the cardiac cycle from the available imaging data. The inference may be performed using a model-based, atlas-based, and / or machine-learning-based motion estimation procedure. In examples, a subjectspecific anatomical model obtained from a single-phase CT is registered to a reference motion atlas representing a cardiac cycle, and the resulting mapping is used to generate time-varying motion fields consistent with the subject’s anatomy. In further examples, a machine-learning model predicts time-resolved deformation fields from sparse phases by conditioning on the subject’s segmented anatomy and one or more physiological parameters, thereby producing motion data suitable for driving the subsequent time-dependent fluid-flow-field determination. In one embodiment, inferring the component-specific intracardiac motion data comprises interpolating, extrapolating, or completing missing motion phases based on one or more of anatomical priors, biomechanical constraints, periodicity constraints, and / or learned motion representations trained on time-resolved cardiac imaging data. In examples, anatomical priors constrain the motion to preserve topology and maintain physiologically plausible chamber wall thickness and valve-plane motion; biomechanical constraints constrain deformation fields to satisfy material or incompressibility properties of myocardium; periodicity constraints enforce cycle consistency at the start and end of a cardiac cycle; and learned motion representations provide a low-dimensional latent motion basis fitted to the subject geometry. In some embodiments, the method produces a temporally smooth motion field by applying spline-based interpolation between known phases, optionally combined with constraintbased optimization to enforce plausibility and continuity.

[0035] In some embodiments, instead of (or in addition to) applying component-specific motion as moving-wall boundary conditions, the hemodynamic effect of cardiac motion may be represented by prescribing one or more time-dependent inflow / outflow profiles (e.g., flow-rate waveforms or velocity profiles) at one or more boundaries, thereby enabling time-dependent flow-field determination even when fully time-resolved motion data is unavailable.

[0036] In one embodiment, identifying the plurality of anatomically distinct components comprises segmenting cardiac structures based on the medical imaging data. Segmentation may be performed using manual, semi-automatic, or fully automatic techniques, including threshold-based methods, region-growing methods, atlas-based segmentation, and / or neural-network-based segmentation. In examples, the method generates component masks or labels for one or more of: left atrium, left atrial appendage, left ventricle, pulmonary veins, mitral annulus region, aortic root or ascending aorta, and / or additional structures relevant to the selected flow domain. In further embodiments, the segmentation output is post-processed to produce surface meshes and / or volumetric representations suitable for discretization and flow-field computation.

[0037] In one embodiment, the medical imaging data comprises time-resolved imaging data, and segmentation is performed on one or more cardiac phases. In examples, segmentation is performed on each available phase to provide phase-specific component geometries and to support direct motion estimation via inter-phase registration. In other examples, segmentation is performed on a selected reference phase (e.g., end-diastole), and phase-to-phase motion fields derived by registration are used to propagate the segmentation through time, thereby generating time-resolved component geometries while reducing computational load. In further embodiments, segmentation and motion estimation are jointly refined, for example by iteratively updating segmentation boundaries to maintain temporal consistency across phases.

[0038] In one embodiment, the computer-implemented method receives the imaging data as medical standard data, including DICOM data, via a Hospital Information System (HIS) comprising a Picture Archiving and Communication System (PACS). The HIS may be implemented in a cloud-based infrastructure, an on-premise infrastructure, or a hybrid infrastructure. The PACS may store, retrieve, manage, and distribute medical images and related metadata, and may provide the input data to the method automatically or upon user request, thereby enabling integration of the method into established clinical imaging workflows.

[0039] In one embodiment, DICOM (Digital Imaging and Communications in Medicine) is used as a standard format for storing and transmitting medical imaging data and associated information. DICOM may be used for images acquired by multiple modalities including, for example, CT, MRI, ultrasound, and X-ray systems. In some embodiments, the input data comprises DICOM image series together with relevant acquisition metadata (e.g., gating information, timing, and spatial calibration) and may be retrieved from the PACS using standard identifiers. However, it is appreciated that other medical standards or data formats than DICOM may be used, and the method is not limited to any particular standard unless explicitly stated.

[0040] In one embodiment, the method is performed using a computer program product integrated with the PACS within or outside the HIS, wherein the steps of the method are executed on a data platform. The data platform may be deployed locally, onpremise, in the cloud, or in a hybrid configuration, and may comprise one or more processors and memory for executing the segmentation, motion estimation or motion inference, flow-field determination, and stagnation-value determination. In such embodiments, input and output are communicated via the PACS, for example by retrieving input image data from the PACS and returning results to the PACS for review, storage, or further processing. In one embodiment, the input data and / or intermediate results are processed within an established medical imaging environment, such as the HIS and / or a laboratory information system (LIS). The environment may enable displaying, storing, and / or analysing the blood-flow stagnation value and related outputs (e.g., component-wise statistics, time-resolved maps, and visualizations) within the PACS environment or in a connected viewing system. Centralizing input and output operations through the PACS may simplify data workflows, reduce the risk of manual handling errors, and improve data accessibility and security within a healthcare environment.

[0041] In one embodiment, the output data, including the blood-flow stagnation value and / or associated maps, is generated and stored in a format compatible with medical imaging workflows. In examples, the output comprises (i) one or more numerical values and / or a structured report associated with the imaging study, and / or (ii) a spatially resolved stagnation map stored as parametric data aligned with the anatomy, optionally in a DICOM-compatible form suitable for archival and review.

[0042] In one embodiment, the computer-implemented method identifies a plurality of anatomically distinct components that comprise one or more of: the left atrium, left atrial appendage, left ventricle, pulmonary veins, mitral valve region, and / or the aorta. The identified components may be represented as labelled regions in image space (e.g., voxel masks), as surface representations (e.g., triangulated meshes), and / or as volumetric representations (e.g., a discretized flow domain) suitable for motion estimation and time-dependent fluid-flow-field determination.

[0043] In one embodiment, the method focuses on atrial flow behaviour by identifying at least the left atrium and the left atrial appendage, optionally together with pulmonary veins and the mitral valve region to define physiologically meaningful inflow and outflow boundaries. In another embodiment, the method focuses on ventricular and outflow haemodynamics by identifying the left ventricle together with the mitral valve region and the aorta, optionally including the aortic root and / or ascending aorta. In further embodiments, additional cardiac structures are identified to refine the anatomical model and boundary definitions, for example by including portions of adjacent vessels or valve-adjacent regions as separate components to enable component-specific motion and boundary-condition assignment.

[0044] In one embodiment, the computer-implemented method determines the timedependent fluid-flow field by generating a volumetric flow domain from the identified anatomical components using a mesh-based or voxel-based discretization of the heart geometry. The discretization may be formed from segmented image masks, surface representations, and / or implicit geometry descriptions. In examples, the volumetric flow domain comprises a tetrahedral or hexahedral finite-volume / finite-element mesh of one or more cardiac chambers and connected vessels, or a voxel grid suitable for latticebased or finite-difference formulations. In some embodiments, the method performs surface conditioning prior to volumetric discretization, such as smoothing, hole closing, topological correction, trimming at anatomical openings, and / or remeshing to achieve target element quality. In further embodiments, mesh resolution is spatially varying, for example using local refinement near valves, inlets / outlets, or regions expected to exhibit recirculation, and / or using adaptive refinement based on one or more flow-field criteria.

[0045] In one embodiment, the computer-implemented method determines the timedependent fluid-flow field using a physics-based simulation, a reduced-order flow model, and / or a machine-learning model trained to approximate intracardiac blood flow dynamics. In examples, a physics-based simulation comprises computational fluid dynamics (CFD) solving the incompressible Navier-Stokes equations on a deforming or moving domain, optionally with Newtonian or non-Newtonian blood rheology and laminar, transitional, or turbulence modelling. In other examples, the reduced-order model comprises a lumped-parameter, quasi-1 D, or other surrogate representation configured to produce a time-dependent approximation of velocities, pressures, and / or flow rates consistent with boundary conditions and cardiac motion. In further examples, the machine-learning model comprises a neural network trained on simulated and / or measured flow fields to predict time-dependent velocity and / or pressure distributions within the identified components, optionally conditioned on anatomy, motion, and boundary-condition parameters. In some embodiments, hybrid approaches are used, for example a reduced-order model provides boundary conditions to a CFD solver, or a machine-learning surrogate accelerates a physics-based solver or directly outputs the flow field subject to physics-informed constraints.

[0046] In one embodiment, the computer-implemented method applies distinct inflow and / or outflow profiles to two or more of the identified anatomical components when determining the time-dependent fluid-flow field. The profiles may be defined as timedependent pressure, velocity, or flow-rate boundary conditions, and may be assigned independently to different inlets or outlets. In examples, distinct inflow profiles are applied to different pulmonary veins, and a distinct outflow profile is applied at the mitral valve plane and / or aortic outlet, thereby enabling asymmetric or physiologically differentiated boundary conditions. In some embodiments, the distinct profiles are derived from subject-specific measurements (e.g., Doppler ultrasound, phase-contrast MRI, 4D flow M Rl), inferred from anatomy and physiological parameters, and / or determined by parameterized models (e.g., Windkessel or compliance-resistance models). In further embodiments, the method enforces global consistency constraints such as mass conservation across multiple inlets / outlets while preserving componentspecific waveform differences.

[0047] In one embodiment, the computer-implemented method determines the timedependent fluid-flow field by applying the component-specific intracardiac motion data as boundary conditions to a flow solver operating on a moving or deforming computational domain. In examples, the motion data defines time-varying wall positions, wall velocities, and / or deformation fields for one or more chamber walls or valve-adjacent surfaces, such that momentum transfer from moving cardiac structures to the blood is represented in the computed flow field. In some embodiments, the motion data is provided as a surface displacement field or volumetric deformation field over the cardiac cycle, and the flow solver uses an arbitrary Lagrangian-Eulerian (ALE) formulation, immersed boundary methods, moving-mesh methods, and / or other techniques for coupling wall motion to blood flow. In further embodiments, different motion fields are applied to different anatomical components (e.g., atrial wall motion and ventricular wall motion), optionally with constraints at interfaces or shared boundaries to ensure continuity and numerical stability.

[0048] In one embodiment, the computer-implemented method determines the blood-flow stagnation value separately for each of the plurality of anatomically distinct components, thereby generating a stagnation map associated with the anatomical components. The stagnation map may comprise, for each component, an aggregated stagnation metric (e.g., a mean, median, percentile, maximum, volume fraction above threshold, or histogram) and / or a spatially resolved field. In examples, distinct stagnation values may be determined for the left atrium, left atrial appendage, pulmonary veins, left ventricle, and / or ascending aorta, enabling automated comparison of stagnation behaviour across components. In some implementations, the stagnation map is stored as a structured dataset that associates each component label with corresponding stagnation statistics and / or a component-local scalar field for subsequent visualization or reporting.

[0049] In one embodiment, the computer-implemented method determines the blood-flow stagnation value as a time-dependent function over multiple cardiac cycles, such that stagnation is quantified not only instantaneously but also with respect to temporal persistence. The method may compute stagnation values at discrete time points, over temporal windows, or as cycle-averaged and / or cycle-to-cycle metrics. In examples, a stagnation metric may be evaluated over two or more cycles to capture periodicity effects, incomplete wash-out between cycles, or cycle-to-cycle variability. In some embodiments, the method determines (i) a per-cycle stagnation value, (ii) a cumulative stagnation measure over N cycles, and / or (iii) temporal stability measures such as variance, maxima, or the time fraction above a stagnation threshold. Where a passivescalar or residence-time formulation is used, the scalar may be advected over multiple cycles to quantify retained blood fractions and wash-out dynamics.

[0050] In one embodiment, the computer-implemented method subdivides at least one anatomical component into sub-regions, and determines a stagnation value for each sub-region. Subdivision may be based on anatomical landmarks (e.g., ostia locations, appendage lobes, annular planes), geometric partitioning (e.g., spatial clustering, Voronoi partitioning, centerline-based segmentation), functional partitioning (e.g., high- curvature regions, low-motion regions), or mesh / image-based partitioning (e.g., voxel blocks, surface patches, tetrahedral element groups). In examples, the left atrial appendage may be subdivided into proximal and distal regions, or into lobe-based regions, and separate stagnation values may be computed for each region to identify localized slow-flow behaviour. In some embodiments, the method outputs a subregional table of stagnation statistics and optionally a labelled sub-region mask to enable consistent cross-subject comparison.

[0051] In one embodiment, the computer-implemented method determines the blood-flow stagnation value as a plurality of pointwise stagnation values forming a spatial distribution within at least one identified anatomical component. The pointwise values may be defined on a volumetric grid (voxels), a computational mesh (nodes / elements), a surface representation (surface vertices / patches), or a particle representation, and may be provided as a scalar field aligned to the anatomical model. In examples, the spatial distribution may be a residence-time field, an age-of-blood field, a wash-out scalar field, a recirculation indicator field, or a low-velocity occupancy field, computed for each voxel or mesh element within the left atrium and / or left atrial appendage. In some implementations, the method stores the pointwise distribution in a format suitable for visualization, such as a DICOM-compatible parametric map, a mesh-associated attribute field, or a volumetric image with voxel values representing stagnation.

[0052] In one embodiment, the computer-implemented method derives the stagnation value from a fluid-flow velocity field, optionally together with pressure or other flow variables, by applying one or more velocity-based stagnation computations to the time-dependent flow field. In examples, stagnation may be quantified by identifying regions where the velocity magnitude falls below a predetermined threshold (absolute or relative), by computing the time fraction that local velocity remains below a threshold, by evaluating near-zero-flow persistence over time, and / or by computing recirculation proxies derived from velocity direction changes, vorticity, or streamline topology. In some embodiments, the method computes a low-velocity residence metric by accumulating, for each spatial location, an indicator function of whether the local velocity magnitude is below threshold over one or more cardiac cycles, thereby generating a spatial map of slow-flow persistence. In further examples, velocity-derived quantities may be combined with particle tracking or passive-scalar transport driven by the velocity field to produce residence-time or wash-out measures that remain “derived from the velocity field” as the primary flow descriptor.

[0053] In one embodiment, the blood-flow stagnation value is determined using one or a combination of several metrics derived from the time-dependent fluid-flow field. In particular, the blood-flow stagnation value may be obtained by computational postprocessing of the simulated or approximated flow field to quantify slow, recirculating, or retained blood within one or more regions of the heart.

[0054] In one embodiment, the stagnation value comprises or is related to a residence time indicating how long fluid elements remain within a region of the heart and / or how many cardiac cycles elapse before such elements are transported out of that region. In some implementations, residence time is represented using a passive scalar transported with the flow. The passive scalar may be computed using a transport equation and may optionally include a source term that increases the scalar value over time, such that regions of slow or recirculating flow obtain higher scalar values than regions exhibiting faster wash-out. In one embodiment, the stagnation value is represented using an age coefficient that increases as fluid elements persist within a region. Passive-scalar and / or agecoefficient formulations may be implemented using an advection equation, optionally with diffusion and optionally with a source term.

[0055] In one embodiment, the stagnation value is additionally or alternatively determined by trajectory-based analysis, for example by tracking massless particles introduced into the flow field. Particle trajectories may be analysed to identify recirculating behaviour or prolonged residence within specified regions. In examples, limited particle displacement, repeated looping trajectories, and / or extended retention over successive time steps or cardiac cycles indicates stagnation.

[0056] In one embodiment, the stagnation value is additionally or alternatively determined by analysing local velocities within the fluid-flow field. For example, regions where velocity magnitude falls below a predetermined absolute or relative threshold may be identified as exhibiting slow flow or reduced wash-out. In some implementations, the stagnation value comprises a time fraction that local velocity remains below threshold over one or more cardiac cycles.

[0057] In one embodiment, and / or as a complementary measure, the stagnation value is estimated using geometric or kinematic measures derived from the imaging data, such as chamber volumes or volume-change dynamics. For example, a geometric indicator may include a ratio between a maximum atrial volume and a ventricular stroke volume, or other ratios or indices derived from time-varying chamber geometry. Such geometric or kinematic indicators may correlate with patterns observed in flow-based stagnation metrics and may therefore be used as computationally efficient complementary measures and / or as proxies in cases where a full flow-field computation is limited.

[0058] In one embodiment, the blood-flow stagnation value is based on one or more of: passive-scalar transport metrics, age coefficients, particle-based residence estimations, velocity-threshold analyses, and / or geometric or kinematic indicators. Multiple metrics may be combined to form a composite stagnation measure, or a selected metric may be used depending on data availability, computational resources, or the desired level of detail. In one embodiment, providing the blood-flow stagnation value comprises presenting the stagnation value within at least one of the identified anatomical components of the heart. The stagnation value may be presented as a spatially resolved map overlaid on a surface model and / or volumetric model of the component, for example as a pointwise scalar field defined on a mesh or as a voxel-wise parametric map aligned with the anatomy. In some embodiments, the method outputs an interactive time-resolved visualization of the stagnation value and / or of the underlying time-dependent fluid-flow field, for example using streamlines, vectors, particle traces, contour slices, isosurfaces, or volume renderings of velocity, pressure, and / or stagnation-related scalar fields. Users may interact with the visualization by rotating, translating, clipping, zooming, thresholding, and / or adjusting temporal playback, thereby enabling inspection of stagnation behaviour over one or more cardiac cycles.

[0059] In one embodiment, providing the blood-flow stagnation value additionally or alternatively comprises outputting a standardized functional report. The report may include component-wise and / or sub-region-wise stagnation statistics (e.g., means, maxima, percentiles, volume fractions above threshold), time-resolved summaries over one or more cardiac cycles, and optionally supporting parameters derived from the flow field and / or geometry such as chamber volumes, temporal volume curves, flow rates, or other computationally determined quantities. The report may be stored and / or communicated in a structured format suitable for integration into clinical or research workflows.

[0060] In one embodiment, providing the blood-flow stagnation value comprises generating a categorical indicator representing one or more stagnation levels. The categorical indicator may be derived by applying one or more thresholds, quantiles, clustering procedures, or classification rules to the blood-flow stagnation value and / or to associated summary statistics. In examples, the method maps a continuous stagnation metric to discrete levels such as “low”, “moderate”, and “high” stagnation, or to a numeric grade on a predefined scale, optionally separately per anatomical component and / or per sub-region. In some embodiments, the categorical indicator is generated in a reproducible manner by using standardized thresholds and normalization schemes (e.g., component-volume normalization, cycle-averaging, or percentile-based normalization) to enable consistent comparison across subjects, scenarios, and imaging modalities. In one embodiment, the computer-implemented method further comprises providing data indicative of a propensity for cerebral events based on the blood-flow stagnation value. The data indicative of propensity may be generated as a computed output derived from one or more stagnation measures and / or their multi-scenario summaries, for example by applying a rule-based mapping, a scoring function, or a machinelearning model to the stagnation value and optionally to additional technical parameters derived from the flow field or geometry. In examples, the method outputs (i) a numerical index, (ii) a categorical level, and / or (iii) a structured feature set that can be stored and communicated together with the stagnation results for downstream use.

[0061] To avoid doubt, in one embodiment the method provides technical, flow-derived output data computed from imaging-derived models and does not, by itself, perform a medical diagnosis or prescribe treatment. The output data indicative of propensity may be used by external systems or workflows together with additional information to support broader assessment processes outside the claimed method.

[0062] In one embodiment, the method further comprises outputting the blood-flow stagnation value in one or more forms. In some embodiments, the stagnation value is presented within a visual representation of one or more identified components of the heart, for example as a colour-coded or otherwise annotated depiction overlaid on a geometric surface model and / or volumetric representation of the component.

[0063] In one embodiment, the method outputs an interactive, static, or time-resolved visualization of the time-dependent fluid-flow field and associated stagnation metrics. Flow-related visualisations may include streamlines, vectors, glyphs, particle traces, volume renderings, and / or contour slices of velocity, pressure, passive-scalar distributions, residence-time fields, wash-out fields, and / or other derived variables. Users may optionally interact with the visualization by rotating, translating, clipping, zooming, thresholding, and / or adjusting temporal playback and colour scales. The visualization may be provided in standardised anatomical views, for example views aligned with the left atrial appendage or other clinically relevant planes.

[0064] In one embodiment, the method outputs a standardized report summarising one or more stagnation measures and / or additional derived parameters. Such outputs may include component-wise and / or sub-region-wise stagnation statistics, chamber volumes, temporal variations in volumes, flow-rate measures, and / or other computed parameters derived from the time-dependent fluid-flow field or from geometric or kinematic indicators.

[0065] In one embodiment, the blood-flow stagnation value is stored as a spatial parametric map aligned to the anatomy. In examples, the stagnation value is represented as a voxel-wise scalar field associated with the imaging study, optionally in a medical- standard-compatible representation, thereby enabling embedding of stagnation information within an anatomical context for review and archival.

[0066] In one embodiment, information derived from the blood-flow stagnation value is provided as data indicative of a propensity for cerebral events and / or as an indicator related to thromboembolic phenomena. Such data may be generated by applying a computational mapping to the stagnation value, for example a threshold rule, a scoring function, and / or a machine-learning model operating on stagnation features (e.g., component-wise maxima, percentiles, spatial extent of high-stagnation regions, and / or scenario-aggregated statistics).

[0067] In one embodiment, the indicator is produced together with additional technical parameters derived from the modelling pipeline, such as geometric or kinematic quantities (e.g., chamber volumes, timing measures), and / or fluid-dynamic quantities derived from the computed flow field (e.g., flow-rate measures or shear-related proxies). The outputs generated by the method may be incorporated into external analytical or decision-support systems that combine such outputs with additional subject-specific information.

[0068] To avoid doubt, in one embodiment the method provides technical parameters related to intracardiac flow behaviour and does not, by itself, perform a medical diagnosis or prescribe treatment. The output data may form part of broader assessment processes implemented outside the claimed method.

[0069] The input data may further comprise one or more physiological parameters related to the heart of the subject. Such parameters may include, for example, heart rate, heart rate variability, aortic or venous pressure, myocardial contractility, blood pressure, and parameters related to cardiac timing such as R-R interval variability derived from an electrocardiogram (ECG). These physiological parameters may be used, in some embodiments, as inputs to the modelling process, and the time-dependent fluid-flow field may be determined based on both the physiological parameters and the component-specific intracardiac motion data.

[0070] Variations in these physiological parameters may influence intracardiac blood-flow patterns, and such variations may, in certain embodiments, be intentionally explored. For example, changes in heart rate, blood pressure, or R-R interval variability may be applied to generate modified simulation conditions, thereby enabling study of how different physiological states affect the resulting fluid-flow field. The R-R interval corresponds to the duration of a cardiac cycle as measured between successive R- wave peaks on an ECG, and variation of this interval may be used to represent different degrees of rhythm irregularity.

[0071] In some embodiments, a new fluid-flow field may be determined for each variation of at least one physiological parameter. Such parameter variations may be used to investigate a range of flow responses associated with different cardiac rhythms or other physiological conditions. This may enable virtual assessment of how physiological variability influences flow behaviour or stagnation characteristics.

[0072] In one embodiment, the input data further comprises one or more physiological parameters of the subject, and determining the time-dependent fluid-flow field comprises incorporating the physiological parameter together with the componentspecific intracardiac motion data. The physiological parameters may be used to parameterize boundary conditions, timing of the cardiac cycle, compliance / resistance models, and / or motion-scaling or motion-timing models, thereby enabling determination of a time-dependent fluid-flow field that is consistent with the subject’s physiological state.

[0073] In one embodiment, the one or more physiological parameters comprise one or more of: heart rate, heart-rate variability, aortic pressure, venous pressure, myocardial contractility, and / or estimated treatment outcomes. In further examples, physiological parameters include blood pressure and parameters related to cardiac timing such as R- R interval variability derived from an electrocardiogram (ECG). The R-R interval corresponds to the duration of a cardiac cycle as measured between successive R- wave peaks on an ECG, and variation of this interval may be used to represent different degrees of rhythm irregularity. In some embodiments, the physiological parameters are obtained from the same imaging study metadata, from monitoring devices, from the patient record, and / or from user input. In one embodiment, variations in at least one physiological parameter are explored by determining a respective time-dependent fluid-flow field under modified parameter conditions. For example, changes in heart rate, pressure levels, contractility, and / or R- R interval variability may be applied to generate modified simulation or inference conditions, thereby enabling evaluation of how different physiological states affect the resulting flow field and derived stagnation values. In some embodiments, a new fluidflow field is determined for each parameter variation, and the corresponding stagnation values are computed using the same stagnation metric definition to enable consistent comparison across scenarios.

[0074] In one embodiment, the method is performed for a plurality of heart-rate scenarios of the subject, including sinus rhythm and one or more atrial or ventricular arrhythmias, and stagnation values are analysed across the plurality of scenarios. The arrhythmia scenarios may be represented, for example, by varying cycle-length sequences, waveform timing, or boundary-condition profiles to reflect rhythm irregularity. In some embodiments, the method generates an overall estimation in the form of a computed, multi-scenario indicator derived from the stagnation values across the scenarios, for example by computing one or more of: a maximum stagnation value across scenarios, an average or weighted average stagnation value, a variability measure, a percentilebased summary, and / or a categorical level representing stagnation severity across scenarios. Such an overall estimation may be provided as technical output for downstream use, for example to support comparison of simulated flow behaviour across rhythm conditions.

[0075] In one embodiment, the computer-implemented method is performed on a virtual representation of the heart after virtually modifying a behaviour and / or geometry of the heart to simulate a treatment. The virtual representation may comprise the identified anatomical components, the component-specific intracardiac motion data, and the corresponding volumetric flow domain used for determination of the time-dependent fluid-flow field. The modification may be applied to the anatomical model (geometry), to the motion model (behaviour), and / or to boundary-condition parameters, thereby enabling simulation of alternative anatomical or functional configurations without any physical intervention.

[0076] In one embodiment, the virtual modification comprises one or more structural or behavioural changes introduced to evaluate how the modified configuration influences the time-dependent fluid-flow field and the derived blood-flow stagnation value. In examples, the modification includes altering the size, shape, or connectivity of an anatomical component (e.g., modifying an appendage cavity or an ostium), modifying a wall-motion pattern or timing, and / or altering inflow or outflow boundary conditions consistent with the simulated configuration. The method may then determine a modified time-dependent fluid-flow field and one or more stagnation values for the modified virtual representation, and may provide outputs enabling comparison between a baseline virtual representation and one or more modified representations.

[0077] In one embodiment, the simulated treatment comprises one or more of: atrial ablation, atrial appendage removal, atrial appendage occlusion, electrical cardioversion, or medical cardioversion. In examples, atrial appendage occlusion is simulated by modifying the virtual geometry to reduce or block communication between the left atrium and the left atrial appendage, and / or by changing boundary conditions at an appendage ostium to represent an occluded state. In further examples, cardioversion is simulated by modifying rhythm-related parameters (e.g., cycle-length variability or waveform timing) and / or by modifying motion patterns to represent a transition toward a more regular contraction pattern. In some embodiments, ablation is simulated by modifying a region-specific motion contribution and / or by applying a regional change in atrial behaviour or timing consistent with altered conduction.

[0078] In one embodiment, the method outputs technical comparisons between baseline and modified simulations, for example differences in component-wise stagnation values, spatial stagnation maps, and / or time-dependent stagnation measures over one or more cardiac cycles. Such virtual evaluations can be performed without any physical procedure and provide technical information regarding how virtual modifications affect simulated intracardiac flow behaviour.

[0079] In one embodiment, identifying the components of the heart comprises applying a neural-network model trained to segment anatomical structures of the heart in imaging data. The neural-network model may operate on one or more of CT images, MRI images, 4D flow MRI-derived anatomical images, and / or ultrasound images, and may be configured to output one or more masks, labels, or probability maps delineating relevant cardiac structures. In examples, the neural network outputs separate component masks for one or more of the left atrium, left atrial appendage, left ventricle, pulmonary veins, mitral valve region, and / or aorta. In further examples, the model segments additional structures such as papillary muscles or larger trabeculations where such detail is used to refine the anatomical model or flow domain.

[0080] In one embodiment, one or more steps of the method is supported or performed by one or more neural networks trained on manually curated ground-truth data and / or other reference data. In examples, a neural network supports or performs motion estimation by predicting deformation fields, surface motion, or point trajectories for one or more identified components over a cardiac cycle, optionally conditioned on sparse-phase or single-phase imaging inputs and physiological parameters. In other examples, a neural network supports or performs determination of the time-dependent fluid-flow field, for example as a learned surrogate model trained to approximate intracardiac velocity and / or pressure fields based on anatomy and motion, optionally with physics-informed constraints. In further examples, a neural network supports or performs determination of the blood-flow stagnation value, for example by predicting a residence-time field, wash-out field, or categorical stagnation levels based on the computed flow field and / or intermediate representations produced during the workflow. The neural networks used for different steps may be separate models or may share parameters within a multi-task architecture.

[0081] In one embodiment, the use of machine-learning-based components improves computational efficiency, automation, robustness, and / or scalability of the workflow. In some embodiments, the machine-learning models are retrained or updated using additional training data, enabling improved performance without requiring manual adjustment of algorithmic parameters.

[0082] In one embodiment, the method further comprises performing quality assurance between one or more steps of the method using a machine-learning model. The machine-learning model may be configured to perform at least one of: (i) evaluating accuracy and / or plausibility of a result of a step of the method, (ii) checking that a step has been completed without errors, and / or (iii) generating a summary of a result of a step.

[0083] In one embodiment, the quality-assurance model evaluates intermediate results produced during anatomical identification / segmentation, motion estimation or motion inference, flow-domain discretization, flow-field determination, and / or stagnation-value computation. In examples, the model detects geometric inconsistencies such as holes, self-intersections, non-manifold regions, or unexpected topology in a surface mesh; checks that designated openings for relevant anatomical structures are present and correctly located; and / or verifies that mesh quality metrics fall within target ranges. In further examples, the model evaluates whether derived quantities such as chamber volumes, temporal volume curves, or motion-field continuity fall within expected physiological and numerical ranges, and flags anomalies for review or automatic correction.

[0084] In one embodiment, the quality-assurance model generates a structured output summarizing one or more intermediate results, for example a pass / fail indicator, confidence scores, identified anomaly types, and / or recommended corrective actions (e.g., re-running segmentation with adjusted parameters, refining a mesh, or re- estimating motion). In this manner, the workflow may be made more reliable and suitable for automated execution across large cohorts and heterogeneous imaging data.

[0085] In one embodiment, the method further comprises performing quality assurance for one or more steps of the method using at least one machine-learning model. The machinelearning model may be configured to evaluate accuracy and / or plausibility of an intermediate result, to check that a step has been completed without errors, and / or to generate a summary of one or more results. Such quality assurance may be associated with, for example, identification of anatomical components, motion estimation or motion inference, discretization of a flow domain, determination of a timedependent fluid-flow field, and / or determination of the blood-flow stagnation value.

[0086] In one embodiment, the machine-learning model is trained on reference results, such as manually prepared anatomical annotations, previously validated simulations, and / or manually reviewed meshes. The model may be configured to detect inconsistencies in geometric representations, such as unintended holes, self-intersections, non-manifold regions, or deviations from expected anatomical topology. In examples, the model confirms that designated openings for relevant anatomical structures are present (e.g., pulmonary vein ostia, valvular planes) and / or that mesh connectivity, element counts, and mesh-quality measures remain within predefined ranges across geometric representations generated during motion estimation or motion inference.

[0087] In one embodiment, the machine-learning model assesses whether derived physiological quantities, such as chamber volumes, correspond to expected ranges, and / or whether their temporal evolution exhibits continuity and plausibility over a cardiac cycle. The model may operate on volumetric, surface-based, or image-based representations and may flag anomalies in geometry, motion fields, boundary conditions, and / or flow fields. One or more machine-learning models may be used, each tailored to a particular quality-assurance task, and may optionally be updated through additional training or feedback obtained during system operation.

[0088] In one embodiment, quality assurance comprises verifying that intermediate results used as inputs to subsequent modelling steps satisfy expected numerical constraints, for example convergence or stability indicators for a flow solver, mass-conservation tolerances, and / or timing / periodicity consistency over one or more cardiac cycles.

[0089] In one embodiment, the method is performed for a plurality of heart-rate and / or rhythm scenarios for the subject, including sinus rhythm and one or more atrial and / or ventricular arrhythmias. In such embodiments, a corresponding time-dependent fluidflow field and a corresponding blood-flow stagnation value may be determined for each scenario. The stagnation values obtained from multiple simulated or analysed scenarios may be compared, combined, or otherwise evaluated to characterise how variations in cardiac rhythm or cycle timing influence intracardiac flow behaviour.

[0090] In one embodiment, the overall analysis across scenarios comprises computing one or more aggregated outputs, for example a maximum stagnation value across scenarios, a weighted average stagnation value, a variability measure, and / or a categorical stagnation indicator representing scenario-robust stagnation levels. In this manner, stagnation information obtained across different scenarios may provide a more comprehensive technical characterisation of flow patterns for a given subject.

[0091] In a further aspect, the present disclosure relates to a blood-flow stagnation value obtained by the method as disclosed herein. The blood-flow stagnation value may be provided as a numeric quantity, a vector of quantities, and / or a spatially resolved field (e.g., a pointwise scalar field over a mesh or voxel grid) representing stagnation behaviour in one or more anatomical components of the heart. The blood-flow stagnation value may be stored in memory, transmitted as a signal, embedded in a file or message (e.g., in a medical imaging workflow), and / or associated with the corresponding imaging study and / or anatomical model. In some embodiments, the blood-flow stagnation value is accompanied by metadata describing the anatomical component, sub-region definitions, the cardiac-cycle interval(s) analysed, and / or the stagnation metric definition used (e.g., residence time, wash-out, recirculation, low- velocity, passive-scalar transport).

[0092] In a further aspect, the present disclosure relates to the use of a blood-flow stagnation value as disclosed herein for characterising one or more hemodynamic stagnation characteristics in the heart. Such hemodynamic stagnation characteristics may include, for example, slow-flow persistence, reduced wash-out, prolonged residence time, recirculatory flow patterns, and / or spatial distributions of stagnation within one or more cardiac components.

[0093] In one embodiment, the characterisation comprises computing component-wise and / or sub-region-wise summaries (e.g., percentiles, maxima, volume fractions above threshold), generating a stagnation map, and / or comparing stagnation measures across scenarios (e.g., across cardiac cycles or rhythm conditions), thereby providing a technical representation of intracardiac flow behaviour. In one embodiment, the use is for characterising regional blood stasis in one or more anatomical components of the heart, for example within the left atrium and / or left atrial appendage. In such embodiments, the regional characterisation may be based on spatially resolved stagnation fields and / or region-specific aggregates computed from the time-dependent fluid-flow field, optionally over multiple cardiac cycles.

[0094] In one embodiment, the blood-flow stagnation value and / or the derived characterisation results are provided in a form suitable for incorporation into analytical or decisionsupport processes. When combined with additional subject-specific information, such outputs may contribute to workflows that analyse physiological conditions or consider cerebral-event propensity, wherein such broader assessment frameworks are implemented outside the method disclosed herein. The output of the characterisation may be stored, displayed, and / or communicated as technical data associated with the anatomical model and / or imaging study.

[0095] In a further aspect, the present disclosure relates to a method for assessing thromboembolic phenomena in a subject, the method as disclosed herein comprising: (a) obtaining a blood-flow stagnation value according to the present disclosure; and (b) generating an indicator related to thromboembolic phenomena based on the blood-flow stagnation value. In one embodiment, the indicator is generated by applying a computational mapping to the blood-flow stagnation value, such as a thresholding rule, a scoring function, a statistical model, or a machine-learning model. The mapping may operate on a single stagnation metric (e.g., a residence-time value) or on a set of features derived from stagnation fields (e.g., component-wise maxima, percentiles, and / or spatial extent of high-stagnation regions). In some embodiments, the indicator is generated per anatomical component and / or per sub-region and is output as a numeric index, a categorical level, and / or a structured data object suitable for storage and transmission.

[0096] In one embodiment, the method is implemented as a computer-implemented data- processing workflow that generates technical output data derived from imaging-based modelling and does not, by itself, prescribe treatment. The indicator may be provided for downstream use, for example as an input to broader assessment workflows that may combine the indicator with other information.

[0097] In a further aspect, the present disclosure relates to a system for quantifying blood-flow stagnation in a heart of a subject, the system as disclosed herein comprising one or more processors and associated memory, the one or more processors being configured to execute one or more of the steps described in connection with the computer-implemented method of the present disclosure.

[0098] In one embodiment, the system is configured to receive input data comprising time- resolved medical imaging data representing the heart of the subject. The system may receive the input data from an imaging device, from a data repository, and / or via a hospital information system (HIS) comprising a picture archiving and communication system (PACS), for example in a medical standard format such as DICOM. The system may include one or more interfaces for receiving and transmitting data, including network interfaces enabling cloud-based, on-premise, or hybrid deployment.

[0099] In one embodiment, the system is configured to identify a plurality of anatomically distinct components of the heart from the input data. The identification may be performed by segmentation, atlas-based methods, and / or a machine-learning model (e.g., a neural network), producing labelled masks, surface models, and / or volumetric representations of one or more cardiac components. In one embodiment, the system is configured to determine component-specific intracardiac motion data for the identified components. The motion data may be determined using image registration, optical flow, model-based motion estimation, and / or machine-learning-based motion inference. The motion data may be represented as time-varying surface motion, displacement fields, deformation fields, or other timedependent motion representations over a cardiac cycle.

[0100] In one embodiment, the system is configured to determine a time-dependent fluid-flow field within the heart based on the component-specific motion data. The fluid-flow field may be determined using a physics-based simulation, a reduced-order model, a machine-learning model trained to approximate intracardiac flow dynamics, or hybrid combinations thereof. In examples, the component-specific motion data is applied as boundary conditions to a solver operating on a moving or deforming computational domain.

[0101] In one embodiment, the system is configured to determine a blood-flow stagnation value from the fluid-flow field, the stagnation value representing slow or recirculating blood flow in one or more identified components. The stagnation value may be computed by computational post-processing of the time-dependent fluid-flow field and may comprise, for example, residence-time measures, wash-out measures, recirculation indicators, low-velocity indices, and / or passive-scalar transport metrics, optionally as a spatial distribution and / or as component-wise summaries.

[0102] In one embodiment, the system is configured to provide the blood-flow stagnation value, for example by storing it, transmitting it, and / or presenting it within one or more identified anatomical components. The system may output an interactive time-resolved visualization and / or a standardized functional report, and may generate componentwise maps and / or categorical stagnation indicators as described herein. Output data may be communicated to external systems for storage or visualization, for example via HIS / PACS.

[0103] In one embodiment, the system comprises computing hardware including one or more CPUs, GPUs, and / or accelerators, and may execute the workflow in an automated manner. The system may be implemented on a workstation, server, cloud platform, or distributed computing environment, and may include quality-assurance components configured to evaluate plausibility and / or numerical consistency of intermediate results. In a further aspect, the present disclosure relates to a computer program comprising instructions which, when executed by one or more processors, cause the one or more processors to perform the method as disclosed herein, for example the method of any one of claims 1 -27. The computer program may be stored on a non-transitory computer-readable medium and / or distributed as a software package, containerized application, or cloud-deployed service.

[0104] In one embodiment, execution of the instructions causes the processor(s) to perform one or more of: receiving medical imaging input data, identifying anatomical components, determining component-specific motion data, determining a timedependent fluid-flow field, computing a blood-flow stagnation value derived from the flow field, and providing the stagnation value in one or more output forms as described herein. The computer program may be integrated into a medical imaging environment, for example with interfaces to HIS / PACS, and may implement optional qualityassurance checks between stages of the workflow.

[0105] In a further aspect, the present disclosure relates to a processing unit and / or system configured to perform the method as disclosed herein. The processing unit may be configured to receive input data including or derived from medical imaging data representing a heart of a subject, wherein the medical imaging data may comprise a single cardiac phase, a subset of cardiac phases, or time-resolved imaging data. The processing unit may be configured to execute one or more of the steps of: identifying a plurality of anatomically distinct components in the imaging data, determining component-specific intracardiac motion data using a motion-estimation and / or motioninference procedure applied to the input data, determining a time-dependent fluid-flow field influenced by the motion data, determining a blood-flow stagnation value derived from the fluid-flow field, and providing the stagnation value.

[0106] In one embodiment, the processing unit is implemented as an on-premise, cloudbased, or hybrid computational system. A cloud-based implementation may provide scalable computational resources and may support integration with additional services, data sources, and / or distributed processing nodes. The processing unit may comprise one or more processors, memory storing instructions and / or data, and one or more interfaces for communication with external devices, services, or users.

[0107] In one embodiment, the input data is uploaded by a user, received from a PACS within a healthcare information system, and / or transmitted directly from an imaging device. The processing unit may implement an automated computational framework in which one or more stages of the workflow, such as anatomical identification, motion estimation or motion inference, flow-field determination, and / or stagnation-value computation, are performed without manual intervention aside from providing the initial input data. One or more of these stages may be implemented using machine-learning models, and optional quality-assurance checks may be performed between stages as described herein.

[0108] In one embodiment, the output of the processing unit includes numerical stagnation values, component-wise summaries, categorical indicators, and / or spatially resolved stagnation maps, and may additionally include static or time-resolved renderings of flow-related variables. Flow-related visualisations may include streamlines, vectors, glyphs, particle traces, volume renderings, and / or contour slices of velocity, pressure, passive-scalar distributions, residence-time fields, wash-out fields, or other derived variables. Visualisation techniques may include clipping, thresholding, transparency, and / or other rendering operations to highlight selected regions or parameters of interest. The output may be communicated to a user or to an external system, for example a PACS, a display device, a reporting system, or an analytical platform.

[0109] In one embodiment, the processing unit may also receive contrast-enhanced images and / or other imaging markers that convey information about blood transport or distribution in the heart, and such information may be used to contextualise or visually correlate computed stagnation values with observed imaging features.

[0110] In a further aspect, the present disclosure relates to a computer program comprising instructions which, when executed by one or more processors, cause the one or more processors to perform the method as disclosed herein, for example the method of any one of claims 1 -27. The computer program may be stored on a non-transitory computer-readable medium and / or distributed as a software package, a containerized application, and / or a cloud-deployed service.

[0111] In one embodiment, execution of the instructions causes the processor(s) to perform one or more of: receiving medical imaging input data, identifying anatomical components, determining component-specific motion data, determining a timedependent fluid-flow field, computing a blood-flow stagnation value derived from the fluid-flow field, and providing the stagnation value in one or more output forms as described herein. Execution of the instructions may be carried out on any suitable computing architecture, including local hardware, cloud-based systems, distributed computing environments, or combinations thereof. The computer program may be integrated into a medical imaging environment, for example with interfaces to HIS / PACS, and may implement optional quality-assurance checks between stages of the workflow.

[0112] In a further aspect, the present disclosure relates to a computer-implemented method of training a neural network configured to determine a blood-flow stagnation value for a heart of a subject. The training method may comprise receiving training input data including or derived from medical imaging data representing the heart, wherein the medical imaging data may comprise a single cardiac phase, a subset of cardiac phases, or time-resolved imaging data.

[0113] In one embodiment, the training method further comprises providing target output data determined by a reference pipeline that performs one or more of: processing the imaging data to identify a plurality of anatomically distinct components, determining component-specific intracardiac motion data over a cardiac cycle (including by motion inference where applicable), determining a time-dependent fluid-flow field influenced by the motion data, and deriving a blood-flow stagnation value from the fluid-flow field. The neural network may be trained to map the training input data, and / or intermediate representations derived therefrom, to the corresponding target blood-flow stagnation value, for example as a component-wise value, a categorical indicator, and / or a spatially resolved stagnation map.

[0114] In one embodiment, the training input data comprises one or more of: imaging data, segmentation masks, anatomical meshes, inferred motion fields, boundary-condition parameters, and physiological parameters. In one embodiment, the target output data comprises one or more stagnation metrics derived from the time-dependent fluid-flow field, for example residence-time measures, wash-out measures, passive-scalar or age-coefficient measures, particle-based residence estimates, and / or velocity- threshold-based measures, optionally computed over multiple cardiac cycles.

[0115] In one embodiment, training the neural network enables efficient approximation of one or more steps of the method disclosed herein and may improve computational performance and / or automation for determining a stagnation value from new input data. The trained neural network may be deployed as part of the workflow, for example to support segmentation, motion estimation or inference, flow-field approximation, and / or direct inference of stagnation values.

[0116] Description of Figures

[0117] The above, as well as additional objects, features and advantages of the present inventive concept, will be more fully understood from the following illustrative and nonlimiting detailed description, taken together with the appended drawings. In the drawings, like reference numerals are used to denote like elements unless stated otherwise.

[0118] Fig. 1 illustrates a block diagram of an example computer-implemented method for determining a blood-flow stagnation value for a heart of a subject. The illustrated steps correspond to a general embodiment of the method and are not limited to any particular imaging modality, identification technique, motion-estimation approach, or flowmodelling method.

[0119] Fig. 2 illustrates a block diagram of an example computer-implemented method for determining a blood-flow stagnation value for a heart of a subject. The illustrated steps correspond to a general embodiment of the method and are not limited to any particular imaging modality, identification technique, motion-estimation approach, or flowmodelling method.

[0120] Fig. 3 schematically illustrates one example embodiment of a workflow for determining a blood-flow stagnation value using time-resolved computed tomography (CT) imaging. In this embodiment, CT images of the heart are acquired, anatomical components are segmented, intracardiac motion is estimated using image registration, a fluid-flow field is determined, and a stagnation value is computed. This CT-based workflow represents one specific implementation consistent with the general method shown in Fig. 1 .

[0121] Fig. 4 schematically illustrates an example anatomical cross-section of a human heart, showing selected cardiac components that may be identified in medical imaging data. The illustrated anatomy is provided for explanatory purposes and does not limit the components that may be identified in different embodiments or the temporal resolution of the imaging data.

[0122] Fig. 5 schematically illustrates an example of a system architecture for executing the method of Fig. 1 or Fig. 2. The illustrated imaging device may represent any modality capable of acquiring medical imaging data, including for example computed tomography (CT), magnetic resonance imaging (MRI), ultrasound imaging, or other imaging systems, and may provide single-phase, sparsely sampled, or time-resolved imaging data. The figure also illustrates a processing unit, server infrastructure, and a visualisation device configured to receive input data, process it automatically, determine a stagnation value, and output flow-related information. The scanner depiction is schematic and not intended to limit the modality or hardware configuration.

[0123] Detailed Description

[0124] The inventors have previously examined intracardiac blood-flow behaviour using medical imaging data together with computational hemodynamic modelling. Earlier studies demonstrated that flow patterns within the left atrium and left atrial appendage may vary substantially between individuals and across physiological conditions. Such studies, performed using modalities such as computed tomography and magnetic resonance imaging, including four-dimensional flow magnetic resonance imaging (4D Flow MRI), have shown that certain subjects may exhibit regions of reduced wash-out or prolonged residence of blood within atrial structures.

[0125] These investigations highlighted the potential value of subject-specific, image-based analysis of intracardiac flow dynamics. However, existing approaches often relied on labor-intensive workflows involving manual segmentation, manual motion delineation, and complex model preparation, which limited scalability and hindered broader adoption. To address these challenges, the inventors have developed a computational framework that integrates anatomical identification, motion-estimation or motioninference techniques, and numerical or learned hemodynamic modelling.

[0126] The framework enables efficient processing of medical imaging data, including singlephase, sparsely sampled, or time-resolved imaging data, to obtain component-specific motion information and to derive flow-related parameters, including measures associated with blood-flow stagnation. This allows for systematic examination of variations in atrial and ventricular flow behaviour across subjects or conditions. The aspects and examples described herein provide technical tools for analysing intracardiac flow patterns based on imaging-derived geometry and measured or inferred motion, enabling consistent evaluation of flow characteristics across a wide range of use scenarios. Fig. 1 illustrates an example of a computer-implemented method 1000 for determining a blood-flow stagnation value for a heart of a subject. The method may comprise the following steps, which correspond to the general workflow of the first aspect of the invention.

[0127] In step 1100, input data is provided, the input data including or derived from medical imaging data representing the heart of the subject. The medical imaging data may for example comprise a single cardiac phase, a subset of cardiac phases, or time-resolved imaging data.

[0128] In step 1200, a plurality of anatomically distinct cardiac components are identified in the input data. The identification may be performed using any suitable technique, including segmentation, atlas-based methods, machine-learning-based identification, or other approaches appropriate for the imaging modality.

[0129] In step 1300, component-specific intracardiac motion data is determined based on a motion-estimation or motion-inference procedure applied to the input data. Motion estimation or inference may be achieved using image registration, optical-flow methods, biomechanical modelling, machine learning, or other suitable techniques.

[0130] In step 1400, a time-dependent fluid-flow field is determined, the fluid-flow field being influenced by the component-specific intracardiac motion data. The flow field may be computed or approximated using numerical solvers, reduced-order models, physics- informed neural networks, machine-learning-based surrogate models, or other hemodynamic modelling techniques.

[0131] In step 1500, a blood-flow stagnation value is determined based on the fluid-flow field. The stagnation value may be derived using any suitable stagnation metric, including residence-time measures, wash-out indices, recirculation indicators, low-velocity metrics, or passive-scalar transport measures.

[0132] In step 1600, the blood-flow stagnation value is provided, for example for visualisation, storage, reporting, or downstream analytical use.

[0133] It will be appreciated that Fig. 1 represents a generic example of the method and that numerous variations of the individual steps may be used. Further embodiments may incorporate additional processing stages or may employ different imaging modalities, identification methods, motion-estimation or motion-inference techniques, flow-field modelling approaches, or stagnation metrics, while remaining within the scope of the invention as defined in the claims.

[0134] Fig. 2 illustrates one example embodiment 2000 of a computer-implemented method for determining a blood-flow stagnation value, shown here for an implementation based on time-resolved computed tomography (CT) imaging. The embodiment of Fig. 2 represents one specific workflow consistent with the general method described in Fig.

[0135] 1.

[0136] In step 2100, time-resolved CT images of the heart of the subject are provided as input data.

[0137] In step 2200, anatomical components of the heart are identified in the CT images, here illustrated by a segmentation procedure applied to the time-resolved CT data.

[0138] In step 2300, intracardiac motion of the identified components is determined by applying an image-registration algorithm to the segmented CT images to obtain timevarying deformation or surface-motion information.

[0139] In step 2400, a fluid-flow field is determined based on the motion data obtained in step 2300. In this embodiment, the flow field may be computed using a numerical hemodynamic model such as a computational fluid-dynamics solver.

[0140] In step 2500, a blood-flow stagnation value is determined from the fluid-flow field. The stagnation value may, for example, be based on a passive-scalar formulation, residence-time estimation, or other stagnation-related quantity derived from the simulated flow.

[0141] In step 2600, the blood-flow stagnation value is provided, for example by storing it, reporting it, or visualising it together with anatomical or flow-related information.

[0142] It will be appreciated that Fig. 2 illustrates only one particular embodiment of the method, corresponding to an implementation using time-resolved CT imaging, segmentation, image registration, and numerical flow simulation. Other embodiments may use different imaging modalities, identification techniques, motion-estimation approaches, hemodynamic modelling frameworks, or stagnation metrics, as shown more generally in Fig. 1 and described in the claims.

[0143] Fig. 3 presents a schematic illustration of a cross-section of a human heart 10. In the illustrated example, the heart 10 comprises several anatomical structures, including the right pulmonary artery, ascending aorta, superior vena cava, pulmonary trunk, right atrium, inferior vena cava, right ventricle, left pulmonary artery, pulmonary veins, left atrium, left atrial appendage, and left ventricle. It will be appreciated that additional cardiac structures may be present and that the depiction in Fig. 3 is simplified for illustrative purposes.

[0144] The components shown in Fig. 3 represent examples of anatomical regions that may be identified in medical imaging data for use in the methods described herein. Identification may be performed using segmentation, atlas-based labelling, machinelearning-based identification, or other suitable techniques depending on the imaging modality and desired level of detail. Other components not shown in Fig. 3 may likewise be identified if relevant for a given embodiment.

[0145] Fig. 4 schematically illustrates one example embodiment 200 of a system and workflow for determining a blood-flow stagnation value for a heart 10 of a subject 100. The illustrated embodiment represents one possible implementation of the general method described in Fig. 1 .

[0146] In the example shown, input data 200 is acquired from the subject 100 by an imaging device 210. The imaging device may be any modality capable of producing medical imaging data, including for example computed tomography (CT), magnetic resonance imaging (MRI), ultrasound imaging, or other imaging systems. The imaging device may provide single-phase, sparsely sampled, or time-resolved imaging data. The illustrated scanner structure is provided for schematic purposes only and is not intended to indicate a specific modality. The input data 200 may be provided in a standardized medical imaging format such as DICOM or a modality-specific format, and may be supplied to the processing workflow directly from the imaging device, via an intermediate acquisition workstation, or through a healthcare information system (HIS) including, for example, a picture archiving and communication system (PACS).

[0147] Although not explicitly depicted in Fig. 4, the input data may additionally include physiological parameters associated with the subject 100, such as heart rate, heartrate variability, pressure measurements, or other subject-specific information. Such parameters may optionally be incorporated into motion estimation or motion inference, and / or into flow-field computation. In the illustrated embodiment, the processing of the input data is performed by a processing unit implemented on a cloud-based infrastructure 300, an on-premise computing system, or a hybrid configuration. A computer program product integrated within the HIS or external infrastructure may facilitate transfer of input data and output results between systems.

[0148] The processing unit may automatically perform one or more stages of the method, including identifying anatomical components in the imaging data, determining component-specific motion using a motion-estimation or motion-inference procedure, determining a time-dependent fluid-flow field, and computing a blood-flow stagnation value. One or more of these stages may be implemented using machine-learning models or numerical modelling techniques.

[0149] The output of the method may include numerical stagnation measures, visual representations of stagnation values mapped onto anatomical structures, or static or time-resolved renderings of flow-related quantities. In Fig. 4, an example visualisation 410 is shown, which may be presented to a user through a visualisation device 400 such as a workstation, medical display, tablet, smartphone, or other suitable output device.

[0150] The processing unit may communicate with a server 500 and a visualisation device 400 to manage data flow, post-processing, and rendering. The illustrated architecture is one example, and alternative processing environments may be used, including distributed, containerised, or embedded systems.

[0151] It will be appreciated that Fig. 4 illustrates only one example embodiment. Other embodiments may employ different imaging modalities, processing arrangements, identification techniques, motion-estimation or motion-inference approaches, flow-field models, or output formats while remaining within the scope of the method as defined in Fig. 1 and the claims.

[0152] Examples

[0153] The following examples illustrate several possible implementations of the method. These examples are non-limiting and are provided to demonstrate the range of imaging modalities, identification approaches, motion-estimation methods, flow-field modelling techniques, and stagnation metrics that may be used. Other embodiments will be apparent to a person skilled in the art from the general description and the claims. Example 1 : CT-based Workflow for Determining a Fluid-Flow Field and Stagnation Value

[0154] The following example illustrates one possible implementation of determining a timedependent fluid-flow field and deriving a blood-flow stagnation value. This example is non-limiting and corresponds to a specific embodiment based on time-resolved computed tomography (CT) imaging.

[0155] A non-transitory computer-readable medium may store instructions which, when executed by a processor, implement the stages of this example.

[0156] From time-resolved CT images, anatomical structures such as the left ventricle (LV) and ascending aorta may be segmented at an end-diastolic phase of the cardiac cycle. Papillary muscles and larger trabeculations may be retained in the LV segmentation to preserve geometric fidelity. The resulting cardiac geometry may be exported in STL format for further processing.

[0157] A specialised program may be used to regularise the segmented surface and subsequently remesh it into triangular elements, for example equilateral second-order TRI6 elements of approximately 0.5 mm edge length.

[0158] To obtain motion information, image registration may be applied to the time-resolved CT images to extract time-varying deformation of the LV and aorta throughout the cardiac cycle. The deformation field may be interpolated in time using a piecewise cubic Hermite polynomial to achieve smooth motion between CT phases.

[0159] Based on the extracted motion, a fluid-flow field may be simulated using an immersed boundary (IB) method. The IB formulation may treat the fluid in Eulerian coordinates and the moving cardiac structure in Lagrangian coordinates, for example using:

[0160] 7 - u(x, t) = 0, (2) where u and p denote velocity and pressure.

[0161] Blood density and viscosity may be set to representative physiological values, such as p = 1060 kg / m3and p = 3.5x10“3Pa s. Adaptive mesh refinement may be used to increase resolution within cardiac cavities while reducing computational load in surrounding regions. A fixed time step (e.g., 2.5x10“5s) may be used, with Courant- Friedrichs-Lewy (CFL) conditions maintained within a range suitable for numerical stability.

[0162] Boundary conditions may be set to represent pulmonary venous inflow and aortic outflow. These may be implemented using pressure-based or flow-based models, such as three-element Windkessel models, or may incorporate measurement-derived values from other modalities.

[0163] A blood-flow stagnation value may then be computed from the resulting flow field. In this example, stagnation may be characterised using a passive scalar, residence-time estimation, or another stagnation-related metric derived from the simulated field.

[0164] This example is illustrative only. Other implementations may use alternative imaging modalities, identification methods, motion-estimation techniques, flow-field models, stagnation metrics, or computational frameworks while remaining within the scope of the inventive concept.

[0165] Example 2: MRI-Based Workflow Using Anatomical Identification and Surrogate Flow Modelling

[0166] In another example embodiment, the input data may comprise time-resolved magnetic resonance imaging, such as 4D flow MRI. Anatomical structures (e.g., the left atrium, left atrial appendage, pulmonary veins, and mitral annulus) may be identified using atlas-based techniques, segmentation algorithms adapted to MRI data, or machinelearning-based identification.

[0167] Motion estimation may be performed by analysing temporal changes in anatomical boundaries or by applying optical-flow or machine-learning models trained to infer deformation fields across the cardiac cycle.

[0168] A time-dependent fluid-flow field may be approximated using a surrogate model that integrates MRI-derived velocities with estimated boundary motions. For example, a reduced-order hemodynamic model or a physics-informed neural network (PINN) may be used to generate a flow-field approximation consistent with observed velocities. A stagnation value may then be determined using a wash-out metric, virtual-particle tracking, or a residence-time index applied to the approximated flow field.

[0169] This example illustrates that the method may be implemented with imaging modalities other than CT and without reliance on a numerical CFD solver.

[0170] Example 3: Machine-Learninq-Based Surrogate Workflow

[0171] In a further example embodiment, a machine-learning model may be trained to perform one or more stages of the method, including anatomical identification, motion estimation, flow-field determination, or stagnation estimation.

[0172] Time-resolved medical imaging data of various modalities (e.g., CT, MRI, ultrasound) may serve as input. A neural-network model may identify cardiac components directly from the imaging data without requiring explicit segmentation. Motion estimation may likewise be obtained by predicting deformation fields or point trajectories for the identified components.

[0173] The system may include a learned surrogate model trained to approximate flow-field quantities from anatomical and motion representations. Training data may be derived from numerical simulations, reduced-order models, or measurement-based reference datasets.

[0174] A stagnation metric, such as residence time, low-velocity indicator, or passive-scalar proxy, may be determined either by a further neural-network module or by applying a metric to the surrogate-predicted flow-field quantities.

[0175] This example demonstrates that the method can be implemented without numerical CFD or manual segmentation and may rely entirely on learned models.

[0176] These examples are illustrative and are not intended to limit the scope of the invention. A person skilled in the art will recognise that variations, substitutions, and combinations of components of the examples may be employed while remaining within the scope of the claims. Unless explicitly stated otherwise, any feature described in connection with an example may be combined with features of other examples and with the general method as set out in the claims.

[0177] 4: Static-lmaqe-Based Workflow with Inferred Cardiac Motion The following example illustrates an embodiment in which the input data comprises a single-phase medical image of the heart. This example is non-limiting and demonstrates that a blood-flow stagnation value may be determined even when time- resolved imaging data of the full cardiac cycle is not available.

[0178] In this example, the input data comprises a single cardiac phase acquired using computed tomography (CT), for example a prospectively gated end-diastolic CT image. Anatomical structures of interest, such as the left atrium, left atrial appendage, pulmonary veins, and adjacent cardiac structures, may be identified in the image using segmentation techniques, atlas-based labelling, or machine-learning-based anatomical identification. The resulting anatomical representation may be processed to generate a surface or volumetric model of the cardiac geometry.

[0179] Because the input imaging data does not directly depict motion over the cardiac cycle, component-specific intracardiac motion data may be inferred using a model-based motion estimation approach. In one embodiment, a parametric or atlas-based cardiac motion model may be employed, wherein the segmented anatomy from the static image is registered to a reference motion model representing a cardiac cycle. In another embodiment, a machine-learning model trained on time-resolved cardiac imaging data may be used to infer time-varying motion fields from the single-phase input geometry. The inferred motion may optionally be constrained using auxiliary physiological information, such as heart rate, stroke volume, atrial or ventricular volume change, or inflow and outflow measurements derived from ultrasound or other modalities.

[0180] Based on the inferred time-varying motion of the cardiac components, a timedependent fluid-flow field may be determined using a numerical solver, a reduced-order hemodynamic model, or a learned surrogate model. Boundary conditions representing inflow and outflow, such as pulmonary venous inflow or ventricular outflow, may be prescribed using physiological assumptions, measured parameters, or model-based constraints to obtain a physiologically consistent flow field over one or more cardiac cycles.

[0181] A blood-flow stagnation value may then be derived from the simulated or approximated fluid-flow field. In this example, stagnation may be characterised using a residencetime measure, a passive-scalar-based metric, a recirculation indicator, or a low-velocity index evaluated within the left atrium or left atrial appendage. The resulting stagnation value may be output as a numerical parameter, visualised on the anatomical model, or incorporated into a report or downstream analytical process.

[0182] This example demonstrates that the method may be applied using static medical imaging data by inferring cardiac motion and flow dynamics, and that determination of a blood-flow stagnation value does not require direct acquisition of fully time-resolved imaging data. Other variations, including alternative motion-inference models, flow solvers, or stagnation metrics, may be employed while remaining within the scope of the claims.

[0183] Items

[0184] 1 . A computer-implemented method (1000) for quantifying a blood flow stagnation in a heart (10) of a subject (100), the method (1000) comprising:

[0185] • providing input data (200), the input data (200) comprising time resolved computed tomography, CT, images (210) of the heart (10) of the subject (100),

[0186] • segmenting the time resolved CT images (210) to identify components of the heart (10) in the time resolved CT images (210),

[0187] • applying image registration on the segmented time resolved CT images to determine intra cardiac surface motion data of the identified components of the heart (10),

[0188] • determining a fluid flow field based on the determined intra cardiac surface motion data of the identified components of the heart (10),

[0189] • determining a blood flow stagnation value based on the fluid flow field, and outputting the blood flow stagnation value.

[0190] 2. The computer-implemented method according to item 1 , wherein the method further comprises: a. performing quality assurance between one or more steps of the method by using a machine learning model, wherein the machine learning model performs at least one of: b. evaluating accuracy and / or plausibility of a step of the method, c. checking that a step has been completed without errors, and d. generating a summary of a result of a step. 3. The computer-implemented method according to item 1 or 2, wherein the input data is provided in form of medical standard data such as DICOM data, the input data being received via a Hospital Information System, HIS, in a cloud-based and / or onpremise infrastructure, wherein the HIS comprises a Picture Archiving and Communication System, PACS.

[0191] 4. The computer-implemented method according to item 3, wherein the method is performed using a computer program product integrated with the PACS within or outside the HIS, wherein the steps of the method are executed on a data platform, and input / output is communicated via the PACS within the HIS.

[0192] 5. The computer-implemented method according to any one of the preceding items, wherein the input data further comprises a physiological parameter related to the heart of the subject, such as any one or more of: heart rate, heart rate variability, aortic and venous pressure, myocardial contractibility, and / or estimated treatment outcomes, wherein the fluid flow field is determined based on the physiological parameter and the determined intra cardiac surface motion data.

[0193] 6. The computer-implemented method according to any one of the preceding items, wherein the method is performed on a range of heart rate scenarios of the subject, including sinus rhythm, and various atrial or ventricular arrythmia, wherein an overall estimation of risk for cardioembolic events in the heart of the subject is obtained by analyzing the blood flow stagnation values from the different scenarios.

[0194] 7. The computer-implemented method according to any one of the preceding items, wherein the method is performed after interventions to a virtual representation of the heart behavior or geometry to virtually simulate a treatment such as atrial ablation, atrial appendage removal or occlusion, or electrical or medical conversion.

[0195] 8. The computer-implemented method according to any one of the preceding items, wherein the blood flow stagnation value comprises a blood flow residence time indicating a number of cardiac cycles for blood particles, and / or wherein the blood flow stagnation value is based on a fluid flow velocity in the fluid flow field. The computer-implemented method according to any one of the preceding items, wherein outputting of the blood flow stagnation value comprises presenting the blood flow stagnation value inside at least one of the components of the heart to a user, and / or wherein the method further comprises outputting an interactive time- resolved visualization and / or a standardized functional report. The computer-implemented method according to any one of the preceding items further comprising, based on the blood flow stagnation value, indicating a risk for cerebral events. The computer-implemented method according to any one of the preceding items, wherein the segmenting of the CT images comprises applying a neural network model trained to segment components of a heart in CT images, and / or wherein one or more of the steps of the method is supported and / or carried out by one or more neural networks, wherein the neural networks are trained on manually curated ground truth data. Use of the blood flow stagnation value determined by the computer-implemented method (1000) according to any one of item 1 -11 to determine a basis for estimation of a risk for cerebral events for the subject (100). A processing unit configured to receive input data (200) comprising time resolved computed tomography, CT, images (210) of a heart (10) of a subject (100), wherein the processing unit is configured to execute the steps of: a. segmenting the CT images (210) to identify components of the heart (10) in the CT images (210), b. applying image registration on the segmented CT images to determine intra cardiac surface motion data of the identified components of the heart (10), c. determining a fluid flow field based on the determined intra cardiac surface motion data of the identified components of the heart (10), d. determining a blood flow stagnation value based on the fluid flow field, and e. outputting the blood flow stagnation value. 14. The processing unit according to item 13, wherein the processing unit represents an automatic computational framework for examining patterns in flow markers in the CT images of the heart of the subject.

[0196] 15. A non-transitory computer-readable medium storing instructions thereon which, when executed by a processor, cause the processor to carry out the steps of the method (1000) according to item 1 .

[0197] 16. A method of training a neural network for quantifying a blood flow stagnation in a heart (10) of a subject (100) by receiving input data (200) comprising time resolved computed tomography, CT, images (210) of the heart (10) of the subject (100), and output a blood flow stagnation value, the method comprising training the neural network on data determined by: a. segmenting time resolved CT images (210) to identify components of the heart (10) in the time resolved CT images (210), b. applying image registration on the segmented CT images to determine intra cardiac surface motion data of the identified components of the heart (10), c. determining a fluid flow field based on the determined intra cardiac surface motion data of the identified components of the heart (10), and d. determining a blood flow stagnation value based on the fluid flow field.

Claims

45Claims1 . A computer-implemented method for quantifying blood flow stagnation in a heart of a subject, the method comprising: a) providing input data comprising medical imaging data representing the heart geometry of the subject; b) identifying, from the input data, a plurality of anatomically distinct components of the heart; c) determining component-specific intracardiac motion data for the identified components over a cardiac cycle based on motion estimation applied to the input data; d) determining a time-dependent fluid-flow field within the heart based on the component-specific intracardiac motion data; e) determining a blood-flow stagnation value defined as a physical quantity derived from the fluid-flow field, including at least one of a residence-time, wash-out, recirculation, low-velocity indicator, or passive-scalar transport metric, the blood-flow stagnation value representing slow or recirculating blood flow in one or more of the identified components; and f) providing the blood-flow stagnation value.

2. The computer-implemented method according to any one of the preceding claims, wherein the blood-flow stagnation value is determined for each of the plurality of anatomically distinct components, thereby generating a stagnation map for the anatomical components.

3. The computer-implemented method according to any one of the preceding claims, wherein the blood-flow stagnation value is determined as a timedependent function over multiple cardiac cycles.

4. The computer-implemented method according to any one of the preceding claims, wherein each anatomical component is subdivided into sub-regions, and a stagnation value is determined for each sub-region.

5. The computer-implemented method according to any one of the preceding claims, wherein the blood-flow stagnation value comprises a plurality of46 pointwise stagnation values forming a spatial distribution within at least one of the identified anatomical components.

6. The computer-implemented method according to any one of the preceding claims, wherein the stagnation value is derived from a fluid-flow velocity field.

7. The computer-implemented method according to any one of the preceding claims, wherein determining the time-dependent fluid-flow field comprises generating a volumetric flow domain from the identified components using a mesh-based or voxel-based discretization of the heart geometry.

8. The computer-implemented method according to any one of the preceding claims, wherein determining the time-dependent fluid-flow field comprises using a physics-based simulation, a reduced-order flow model, or a machine-learning model trained to approximate intracardiac blood flow dynamics.

9. The computer-implemented method according to any one of the preceding claims, wherein the method comprises applying distinct inflow or outflow profiles to two or more of the identified anatomical components.

10. The computer-implemented method according to any one of the preceding claims, wherein determining the time-dependent fluid-flow field comprises applying the component-specific motion data as boundary conditions to the flow solver.11 . The computer-implemented method according to any one of the preceding claims, wherein the medical imaging data comprises computed tomography, magnetic resonance imaging, 4D flow MRI, ultrasound imaging, or any combination thereof.

12. The computer-implemented method according to any one of the preceding claims, wherein determining the component-specific intracardiac motion data comprises using image registration, optical flow, model-based motion estimation, or a machine-learning model trained to infer motion fields.4713. The computer-implemented method according to any one of the preceding claims, wherein the medical imaging data comprises a single cardiac phase or a subset of cardiac phases, and wherein the component-specific intracardiac motion data over the cardiac cycle is inferred from the medical imaging data, such as by using a model-based, atlas-based, or machine-learning-based motion estimation procedure.

14. The computer-implemented method according to claim 13, wherein inferring the component-specific intracardiac motion data comprises interpolating, extrapolating, or completing missing motion phases based on one or more of: o anatomical priors, o biomechanical constraints, o periodicity constraints, or o learned motion representations trained on time-resolved cardiac imaging data.

15. The computer-implemented method according to any one of the preceding claims, wherein identifying the plurality of anatomically distinct components comprises segmenting cardiac structures based on the medical imaging data.

16. The computer-implemented method according to claim 15, wherein the medical imaging data comprises time-resolved imaging data and the segmentation is performed on one or more cardiac phases.

17. The computer-implemented method according to any one of the preceding claims, wherein the imaging data is provided as medical standard data, including DICOM data, received via a Hospital Information System (HIS) comprising a Picture Archiving and Communication System (PACS).

18. The computer-implemented method according to claim 14, wherein the method is performed using a computer program product integrated with the PACS within or outside the HIS, wherein the steps of the method are executed on a data platform, and wherein input and output are communicated via the PACS.

19. The computer-implemented method according to any one of the preceding claims, wherein the input data further comprises one or more physiological parameters of the subject, and wherein determining the fluid-flow field comprises incorporating the physiological parameter together with the component-specific motion data.

20. The computer-implemented method according to claim 16, wherein the one or more physiological parameters of the subject comprises one or more of: heart rate, heart-rate variability, aortic pressure, venous pressure, myocardial contractility, or estimated treatment outcomes.21 . The computer-implemented method according to any one of the preceding claims, wherein the method is performed for a plurality of heart-rate scenarios of the subject, including sinus rhythm and one or more atrial or ventricular arrhythmias, and wherein an overall estimation indicative of thromboembolic phenomena is obtained by analysing stagnation values across the plurality of scenarios.

22. The computer-implemented method according to any one of the preceding claims, wherein the method is performed on a virtual representation of the heart after virtually modifying a behaviour or geometry of the heart, to simulate a treatment.

23. The computer-implemented method according to claim 19 wherein the treatment comprises one or more of atrial ablation, atrial appendage removal, atrial appendage occlusion, electrical cardioversion, or medical cardioversion.

24. The computer-implemented method according to any one of the preceding claims, wherein the plurality of anatomically distinct components comprise one or more of: the left atrium, left atrial appendage, left ventricle, pulmonary veins, mitral valve region, or the aorta.

25. The computer-implemented method according to any one of the preceding claims, wherein providing the blood-flow stagnation value comprises presenting the stagnation value within at least one of the identified anatomical componentsof the heart, and / or outputting an interactive time-resolved visualization and / or a standardized functional report.

26. The computer-implemented method according to any one of the preceding claims, further comprising providing data indicative of a propensity for cerebral events based on the blood-flow stagnation value.

27. The computer-implemented method according to any one of the preceding claims, wherein providing the blood-flow stagnation value comprises generating a categorical indicator representing one or more stagnation levels.

28. The computer-implemented method according to any one of the preceding claims, wherein identifying the components of the heart comprises applying a neural-network model trained to segment anatomical structures of the heart in imaging data.

29. The computer-implemented method according to any one of the preceding claims, wherein one or more of the steps of the method is supported or performed by one or more neural networks trained on manually curated groundtruth data.

30. The computer-implemented method according to any one of the preceding claims, wherein the method further comprises performing quality assurance between one or more steps of the method using a machine learning model, the machine learning model being configured to perform at least one of: o evaluating accuracy and / or plausibility of a result of a step of the method; o checking that a step has been completed without errors; and o generating a summary of a result of a step.31 . A blood-flow stagnation value obtained by the method of any of Claims 1 -30.

32. Use of the blood-flow stagnation value of claim 31 for characterising one or more hemodynamic stagnation characteristics in the heart.

33. The use according to claim 32, wherein the hemodynamic stagnation characteristics comprise regional blood stasis in one or more anatomical components of the heart.

34. A method for assessing thromboembolic phenomena in a subject, comprising: a) obtaining a blood-flow stagnation value according to claim 31 ; and b) generating an indicator related to thromboembolic phenomena based on the blood-flow stagnation value.

35. A system for quantifying blood-flow stagnation in a heart of a subject, comprising one or more processors configured to: a) receive input data comprising time-resolved medical imaging data representing the heart of the subject; b) identify, from the input data, a plurality of anatomically distinct components of the heart; c) determine component-specific intracardiac motion data for the identified components; d) determine a time-dependent fluid-flow field within the heart based on the component-specific motion data; e) determine, from the fluid-flow field, a blood-flow stagnation value that represents slow or recirculating blood flow in one or more of the identified components; and f) providing the blood-flow stagnation value.

36. A computer program comprising instructions which, when executed by one or more processors, cause the one or more processors to perform the method of any one of claims 1-30.

Images