System and method for patient-specific cardiovascular risk assessment using medical imaging data

Abstract

The present disclosure relates to systems and methods for assessing cardiovascular risk in patients. Specifically, the present disclosure provides a computer implemented method for calculating patient-specific cardiovascular risk using angiographic imaging, the method comprising: extracting anatomical parameters of a patient's coronary arteries from angiographic imaging data of the patient's coronary arteries, the anatomical parameters including one or more of vessel geometry, curvature, vessel diameter, and vessel tortuosity; calculating blood flow characteristics from the angiographic imaging data, the blood flow characteristics including at least one of blood flow shear stress, pressure gradients, and helicity; applying a machine learning-based model to the extracted anatomical parameters and calculated blood flow characteristics to generate patient-specific risk markers; using the patient-specific risk markers to assess the likelihood of future adverse cardiovascular events; and outputting the assessment of cardiovascular risk for clinical decision-making.

Description

SYSTEM AND METHOD FOR PATIENT-SPECIFIC CARDIOVASCULAR RISK ASSESSMENT USING MEDICAL IMAGING DATATechnical field

[0001] This application claims priority to Australian Provisional Patent Application No.2025900304 filed February 05, 2025, titled “System and method for patient-specific cardiovascular risk assessment using medical imaging data” which is hereby incorporated by reference in its entirety

[0002] The present disclosure relates to systems and methods for assessing cardiovascular risk in patients. Specifically, it concerns techniques for analyzing angiographic imaging data, extracting anatomical and hemodynamic parameters, and utilizing machine learning models to predict the likelihood of adverse cardiovascular events. The present disclosure further relates to systems for integrating cardiovascular risk assessments into clinical workflows and generating actionable insights for patient management.Background

[0003] Cardiovascular diseases (CVDs) remain one of the leading causes of mortality worldwide, with heart attacks accounting for a significant proportion of these fatalities. Current methods for stratifying patient risk rely on outdated scoring tools, such as the Framingham Risk Score, which primarily consider demographic and clinical history factors such as age, cholesterol levels, blood pressure, smoking history, and diabetes status. These methods often fail to identify high-risk patients, particularly those with no traditional risk factors, leading to delayed interventions and poor outcomes.

[0004] Accurate prediction of cardiovascular risk remains a significant challenge for clinicians due to the complexity of factors involved, including anatomical features, hemodynamic conditions, and demographic variations. Existing diagnostic systems primarily rely on subjective visual assessment of angiographic images or static parameters, which often fail to provide personalized or predictive insights into a patient’s cardiovascular health.

[0005] Recent advances in medical imaging technologies, such as Computed Tomography Coronary Angiography (CTCA), have enabled the non-invasive, low-dosage, and three-dimensional visualization of coronary arteries. However, the translation of these imaging capabilities into predictive clinical tools remains limited due to the lack of robust and scalable techniques for analyzing the vast amount of data generated. In particular, the ability to evaluate anatomical and hemodynamic characteristics of coronary arteries, such as vessel geometry, wall shear stress, and plaque burden, has been constrained by computational challenges and variability in patient-specific features.

[0006] The integration of artificial intelligence (Al) and machine learning (ML) into medical imaging analysis presents an opportunity to overcome these challenges. By leveraging these technologies, it is possible to extract meaningful patterns from imaging data, correlate them with clinical outcomes, and generate patient-specific cardiovascular risk markers with unprecedented accuracy and speed. Such tools could revolutionize the prediction and prevention of heart attacks, improving outcomes through earlier interventions and tailored treatment strategies.

[0007] Despite these advances, critical technical gaps remain in achieving rapid and accurate risk assessments that are suitable for clinical adoption. These gaps include the need for advanced segmentation techniques, robust handling of imaging artifacts, and scalable methods for integrating anatomical and hemodynamic data with clinical variables. Furthermore, the ability to generate synthetic data for training machine learning models and to incorporate populationspecific variations remains underexplored. The current systems also often lack integration of key metrics, such as advanced hemodynamic parameters, population-specific adjustments, and longitudinal imaging data, which are crucial for improving predictive accuracy and clinical utility.Summary

[0008] It is an object of the present disclosure to substantially overcomes or at least ameliorates one or more disadvantages of existing arrangements.

[0009] The present disclosure addresses these challenges in some embodiments by providing a comprehensive system and method for calculating cardiovascular risk using angiographic imaging data. The arrangements described combine advanced image processing techniques,hemodynamic modelling, and machine learning to deliver accurate, patient-specific risk assessments and actionable recommendations for clinical decision-making.

[0010] The reference in this specification to any prior publication (or information derived from it), or to any matter which is known, is not, and should not be taken as, an acknowledgement or admission or any form of suggestion that that prior publication (or information derived from it) or known matter forms part of the common general knowledge in the field of endeavour to which this specification relates.

[0010] The present disclosure provides systems and methods for advanced cardiovascular risk assessment using angiographic imaging and machine learning. By leveraging patient-specific anatomical and hemodynamic parameters, the disclosure facilitates precise evaluation of cardiovascular risks, enabling informed clinical decision-making.

[0011] According to an aspect, the present disclosure may provide a computer implemented method for calculating patient-specific cardiovascular risk using angiographic imaging, the method comprising:extracting anatomical parameters of a patient’s coronary arteries from angiographic imaging data of the patient’s coronary arteries, the anatomical parameters including one or more of vessel geometry, curvature, vessel diameter, and vessel tortuosity;calculating blood flow characteristics from the angiographic imaging data, the blood flow characteristics including at least one of blood flow shear stress, pressure gradients, and helicity;applying a machine learning-based model to the extracted anatomical parameters and calculated blood flow characteristics to generate patient-specific risk markers;using the patient-specific risk markers to assess the likelihood of future adverse cardiovascular events; andoutputting the assessment of cardiovascular risk for clinical decision-making.

[0012] The anatomical parameters may include, but are not limited to, bifurcation angles, plaque burden, vessel remodelling index, twistedness of the coronary arteries, vessel tapering, centerline curvature. The blood flow characteristics may include, but are not limited to oscillatory shear index, relative residence time, and wall shear stress distribution throughout the cardiac cycle and other hemodynamic metrics such as velocity profiles and pressure oscillations.

[0013] In an embodiment, the patient-specific risk markers may be derived by combining anatomical parameters with hemodynamic factors using feature weighting determined by a predetermined machine learning-based technique, enabling the identification of critical parameters influencing the risk assessment. In an embodiment, the angiographic imaging data may be pre-processed using a segmentation algorithm based on a convolutional neural network to isolate the coronary arteries.

[0014] In an embodiment, the method may further comprise generating synthetic coronary artery data to augment training of the machine learning-based model. The synthetic data may comprise variations in vessel, lesion and segment-specific variations, including but not limited to curvature, bifurcation angles, diameters and stenosis severity. Such variations may be generated based on observed statistical patterns, including standard deviations from a training dataset, or through data augmentation techniques, such as generative adversarial networks (GANs), to simulate a training dataset.

[0015] In an embodiment, the likelihood of adverse cardiovascular events is assessed by identifying regions of haemodynamic and anatomical markers that have a significant association with clinical adverse events, including but not limited to low shear stress associated with increased plaque risk.

[0016] In an embodiment, the method may further comprise outputting visual overlays on the angiographic images to highlight regions of high risk based on the generated patient- and / or population-specific risk markers. The output may comprise a risk score that integrates traditional risk factors, such as cholesterol levels, smoking history, etc., and additional patient demographic data, such as clinical history, and the newly derived markers of risk, including anatomical parameters and hemodynamic factors.

[0017] In an embodiment, the method may further comprise outputting a clinical assessment, wherein the outputted assessment may provide recommendations to assist clinicians in patient management based on patient-specific risk markers, such as initiation of medication, lifestyle modifications, or surgical interventions.

[0018] In an alternate embodiment, the method may further comprise validating the machine learning-based model using a dataset of angiographic imaging data and corresponding groundtruth hemodynamic simulations generated by computational fluid dynamics. The validation may further comprise cross-referencing predictions of wall shear stress with transient and time-averaged computational fluid dynamics simulations. The machine learning-based model may be adapted to account for population-specific variations, including age, sex, ethnicity, and prevalent risk factors.

[0019] In an embodiment, the segmentation algorithm may comprise a U-Net architecture or similar methods trained on a dataset comprising coronary angiograms of varying disease severities, medical image follow up data and at times with clinical outcome data. The machine learning-based model uses a graph neural network to label and classify coronary artery segments. The machine learning-based model may be trained using a dataset comprising angiographic imaging data annotated with clinical outcomes and patient demographic information.

[0020] In an alternate embodiment, the machine learning-based model includes a multi-scale convolutional neural network trained to assess anatomical characteristics and predict wall shear stress and other hemodynamic metrics.

[0021] In an embodiment, the anatomical parameters and blood flow characteristics are computed for specific segments of the coronary tree, including but not limited to the proximal left anterior descending artery (LAD) and proximal left circumflex artery (LCX). The extracted anatomical parameters may include metrics computed using a tessellated mesh representation of the coronary tree, smoothed using a diffusion Laplacian filter.

[0022] In an embodiment, the method may further comprise generating a population-based coronary artery atlas to facilitate comparisons and predictions across different demographic groups. In an alternate embodiment, the method may further comprising assessing longitudinal changes in patient-specific risk markers using repeat angiographic imaging data over a defined follow-up period.

[0023] According to another aspect, the present disclosure may provide a system for calculating patient-specific cardiovascular risk using angiographic imaging:a feature extraction module implemented on a processor, configured to extract anatomical parameters of a patient’s coronary arteries from angiographic imaging data of the patient’scoronary arteries, the anatomical parameters extracted at lesion, vessel and segment-specific regions for the whole tree from the angiographic imaging data, the anatomical parameters including but not limited to one or more of vessel geometry, curvature, vessel diameter, and vessel tortuosity;a hemodynamic calculation module configured to assess blood flow characteristics from the angiographic imaging data, the blood flow characteristics including at least one of blood flow shear stress, pressure gradients, and helicity;a machine learning-based risk analysis module configured to apply a trained machine learning model to the extracted anatomical parameters and calculated blood flow characteristics to generate patient-specific risk markers;a risk assessment module configured to use the patient-specific risk markers to assess the likelihood of adverse cardiovascular events; andan output module configured to deliver the assessment of cardiovascular risk, including visual and numerical outputs for clinical decision-making.

[0024] A data acquisition module may be configured to receive coronary angiographic imaging data, which may involve either directly importing raw imaging datasets from CTCA imaging systems or retrieving previously recorded imaging data from secure databases. This acquired data may form the input for subsequent feature extraction and hemodynamic analysis.

[0025] In an embodiment, the hemodynamic calculation module may comprise applying traditional computational fluid dynamics (CFD) methods, including partial differential equation (PDE) solvers, to compute blood flow characteristics such as wall shear stress, pressure gradients, and helicity. These outputs, alongside anatomical parameters extracted from the imaging data, are input into the machine learning-based risk analysis module, which applies trained machine learning models to generate patient-specific risk markers.

[0026] In an embodiment, the system may further comprise a data preprocessing module configured to filter and select the imaging dataset obtained by the data acquisition module, selecting optimal imaging series based on parameters including resolution, axial spacing, field of view, and phase of the cardiac cycle, and a visualization module configured to produce color-coded overlays on the angiographic imaging data, highlighting regions of high cardiovascular risk.

[0027] In an embodiment, the data acquisition module retrieves raw angiographic imaging data, while the preprocessing module selects the most suitable imaging series from the acquired dataset for subsequent processing and analysis.

[0028] The machine learning-based risk analysis module may be configured to generate synthetic coronary artery data for model training, incorporating variations in vessel curvature, stenosis severity, and bifurcation angles, and the risk assessment module is configured to generate confidence scores or uncertainty ranges for the predicted risk markers.

[0029] In an embodiment, the output module may be configured to integrate the cardiovascular risk assessment into an electronic health record system for seamless clinical workflow. The output module may deliver real-time cardiovascular risk assessments during ongoing CTCA imaging and optionally provide treatment recommendations based on the predicted cardiovascular risk. The recommendations may include suggestions for medication adjustments, surgical interventions, or lifestyle modifications, which clinicians can review and incorporate into patient care, an embodiment, the hemodynamic calculation module may be configured to analyze transient blood flow characteristics throughout the cardiac cycle, and the feature extraction module comprises graph neural networks to label and classify coronary artery segments and branches.

[0030] In an embodiment, the system may further comprise a cloud-based data storage module configured to securely store and process patient data in compliance with regulatory requirements.

[0031] In an embodiment, the risk assessment module may be configured to compare patientspecific risk markers against a population-based coronary artery atlas to identify deviations from population norms. The visualization module may provide interactive visualizations, allowing an end user to explore individual coronary segments and associated risk markers.

[0032] In an alternate embodiment the system may further comprise an alert generation module configured to notify clinicians of high-risk regions requiring immediate attention based on the cardiovascular risk assessment.

[0033] The present disclosure contemplates the use of multimodal imaging data, such ultrasound, in conjunction with angiographic imaging to enhance risk predictions. It supportsreal-time processing, cloud-based integration, and population-specific adjustments to improve clinical utility across diverse demographics.

[0034] By combining advanced imaging, computational techniques, and machine learning, the present invention provides an innovative approach for accurately assessing and predicting cardiovascular risk, thereby improving patient outcomes and supporting clinical decisionmaking.

[0035] Further embodiments may include a reporting tool for generating actionable insights based on patient-specific risk markers. The tool comprises modules for data acquisition, feature extraction, machine learning-based analysis, and visual reporting, which collectively provide clinicians with detailed cardiovascular risk profiles and recommendations.

[0036] Various embodiments described herein enable enhanced prediction accuracy through the integration of synthetic coronary artery data into machine learning model training, populationspecific risk adjustments, and advanced visualization techniques. Additional features include real-time risk assessment during imaging procedures, longitudinal tracking of risk progression, and the incorporation of complementary imaging modalities such as MRI or ultrasound for a multimodal approach.

[0037] According to another aspect, the present disclosure may provide a system for cardiovascular risk assessment and reporting, the system comprising:a data acquisition module configured to receive angiographic imaging data from medical imaging devices;a segmentation module implemented on a processor, configured to identify and isolate coronary artery structures from the imaging data using a trained neural network;a computational module configured to:compute anatomical features of the coronary arteries, including vessel curvature, bifurcation angles, and twistedness; andcalculate hemodynamic characteristics, including shear stress, helicity, and flow recirculation zones;a machine learning module configured to process the anatomical and hemodynamic characteristics to generate patient-specific risk metrics; andan output module configured to display the risk metrics, including a classification of cardiovascular risk and associated clinical recommendations based on the generated risk markers.

[0038] According to another aspect, the present disclosure may provide a system for enhancing cardiovascular risk prediction, the system comprising:a synthetic data generator configured to create augmented datasets by introducing variations in coronary artery anatomy and hemodynamic parameters, the variations including vessel curvature, stenosis severity, bifurcation angles, and plaque burden;a machine learning module configured to be trained on the augmented datasets to analyse anatomical and hemodynamic variations and generate risk assessment parameter;a processing module configured to apply the trained machine learning model to patientspecific angiographic imaging data to compute cardiovascular risk markers; andan output module configured to deliver a cardiovascular risk assessment report, including risk predictions based on the enhanced model accuracy.

[0039] According to another aspect, the present disclosure may provide a reporting tool for cardiovascular risk assessment, the tool comprising:a data input interface configured to receive angiographic imaging data and associated patient demographic information;a processing module configured to:extract anatomical features of coronary arteries from the angiographic imaging data, including vessel geometry, curvature, and bifurcation angles;calculate hemodynamic parameters, including wall shear stress, oscillatory shear index, and helicity;generate patient-specific cardiovascular risk markers using a machine learning model trained on clinical and imaging datasets;a reporting module configured to:produce a visual representation of the coronary arteries annotated with risk markers;generate a summary of cardiovascular risk, including a numerical risk score and classification into predefined risk categories;provide recommendations for clinical decision-making based on the generated risk markers; andan output interface configured to deliver the report to an end-user in a format for use in patient management.

[0040] According to another aspect, the present disclosure may provide a system for facilitating cardiovascular risk prediction, the system comprising:a synthetic data generator configured to create augmented datasets by introducing variations in coronary artery anatomy and hemodynamic parameters, the variations including vessel curvature, stenosis severity, bifurcation angles, and plaque burden;a machine learning model trained on the augmented datasets and clinical outcomes to improve robustness and accuracy of risk predictions;a processing module configured to apply the trained machine learning model to patientspecific angiographic imaging data to compute cardiovascular risk markers; andan output module configured to deliver a cardiovascular risk assessment report, including risk predictions based on the enhanced model accuracy.

[0041] According to another aspect, the present disclosure may provide a computer implemented method for predicting cardiovascular risk using multimodal imaging data, the method comprising:receiving angiographic imaging data of a patient’s coronary arteries and additional non-invasive imaging data, including Magnetic Resonance Imaging (MRI) or ultrasound;integrating anatomical features derived from angiographic imaging with tissue-specific characteristics obtained from the additional imaging data;calculating combined anatomical, hemodynamic, and tissue-level risk markers; predicting cardiovascular risk using a machine learning model trained on datasets combining data from multiple imaging techniques; andoutputting a comprehensive risk profile for clinical use.

[0042] According to another aspect, the present disclosure may provide a computer implemented method for assessing cardiovascular risk, the method comprising:obtaining a series of longitudinal angiographic imaging data of a patient's coronary arteries acquired over multiple time points;extracting anatomical and hemodynamic parameters from each set of angiographic imaging data, the parameters including vessel geometry, curvature, bifurcation angles, wall shear stress, and oscillatory shear index;analyzing changes in the extracted parameters to monitor plaque progression and associated hemodynamic variations over time;predicting the likelihood of future cardiovascular events based on trends and deviations in the parameters across the longitudinal dataset; andoutputting a time-based cardiovascular risk profile, including projected risks and recommended follow-up intervals for clinical intervention and decision-making.

[0043] According to another aspect, the present disclosure may provide a method for generating predictive models of cardiovascular risk, the method comprising:generating synthetic models of coronary vessels with stenosis, the synthetic models including both axially symmetric and non-symmetric stenoses with a range of vessel area reductions;applying a remapping procedure to transform the synthetic stenosed vessel models into a rectangular grid representation;training a machine learning module using the transformed synthetic models as input data to predict hemodynamic parameters associated with cardiovascular risk;extending the remapping procedure to a coronary arterial network comprising interconnected branches of the coronary vessels, to generate predictive outputs for computational metrics, including a topological shear variation index; andstitching predictive outputs from branches of the coronary tree to generate a complete anatomical and hemodynamic risk profile.

[0044] According to another aspect, the present disclosure may provide a method for predicting cardiovascular risk metrics for a coronary arterial network, the method comprising:transforming anatomical features of the coronary arterial network, including synthetic stenosed vessel models for analysis by a machine learning model;training the machine learning model to predict hemodynamic parameters associated with cardiovascular risk, based on features of individual branches of the coronary arterial network and interconnections within the network;analyzing every branch of the coronary arterial network using the machine learning model, including generating branch-specific predictions for hemodynamic parameters and cardiovascular risk metrics;integrating branch-specific predictions to form a unified cardiovascular risk profile for the coronary arterial network, wherein the cardiovascular risk metrics comprise at least a topological shear variation index and indicators of plaque progression risk; andoutputting the unified cardiovascular risk profile as a complete assessment of cardiovascular health across the entire coronary arterial network.

[0045] The branch-specific predictions may comprise wall shear stress, oscillatory shear index, and flow separation zones.

[0046] The machine learning model may be a 2-dimensional convolutional neural network (CNN). The CNN may be trained using synthetic stenosed vessel models representing area reductions ranging from 25% to 75% and vessel models representing area reductions ranging from 0 to 75%.

[0047] The synthetic data may comprise representations of anatomical features and hemodynamic parameters across varying demographic populations, is used for training the CNN to account for anatomical and demographic variations.

[0048] The Topological Shear Variation Index may be computed across multiple branches of the coronary arterial network to detect systemic flow abnormalities.

[0049] According to another aspect, the present disclosure may provide a method for integrating cardiovascular risk assessment into clinical workflows comprising automatically processing angiographic imaging data, generating risk markers, and outputting actionable insights during routine clinical imaging.

[0050] These and other aspects of the disclosure address key challenges in cardiovascular risk assessment by providing innovative tools that integrate advanced imaging, machine learning, and clinical decision support to enhance patient outcomes.

[0051] The term “comprising” as used in the specification and claims means “consisting at least in part of.” When interpreting each statement in this specification that includes the term “comprising” features other than that or those prefaced by the term may also be present. Related terms “comprise” and “comprises” are to be interpreted in the same manner.Brief Description of the Drawings

[0052] The following drawings illustrate embodiments of the invention and its associated system components, workflows, and functionalities:

[0053] Figure 1: A block diagram illustrating the system architecture for calculating patientspecific cardiovascular risk.

[0054] Figure 2: A schematic diagram illustrating the end-to-end system workflow, including image acquisition, segmentation, feature extraction, machine learning-based risk assessment, and visualization of risk markers.

[0055] Figure 3a: A diagram of the machine learning-based risk assessment framework, highlighting the integration of anatomical and hemodynamic parameters, predictive model training, and clinical output generation.

[0056] Figure 3b: An illustration of the training stage of the machine learning model.

[0057] Figure 3c: An illustration of the network architecture of the machine learning model.

[0058] Figure 4: A data flow diagram illustrating the integration of data sources into the coronary artery disease risk prediction model. The figure includes input data from CTCA images, patient demographics, clinical outcomes, and follow-up records.

[0059] Figure 5: A schematic representation of the image preprocessing and feature extraction pipeline, including segmentation, mesh generation, and anatomical parameter extraction.

[0060] Figure 6: A diagram of the hemodynamic calculation workflow, showing the computation of metrics such as wall shear stress, oscillatory shear index, and helicity, integrated with machine learning analysis.

[0061] Figure 7a: A schematic representation illustrating the workflow for population-level profiling and grouping of anatomical shapes and hemodynamic risk factors, starting from CTCA images, segmentation, reconstruction, and shape measurement to computed hemodynamic risk using CFD and final population grouping.

[0062] Figure 7b: A diagram showcasing the machine learning-based analysis workflow, including CTCA imaging, ML-based shape and risk assessment, comparison with standard modelling approaches, and the detection of anatomical risk markers in the left anterior descending (LAD) coronary artery.

[0063] Figure 7c: A data flow diagram depicting the integration of linked data sources, including CTCA imaging, patient records, hospitalization and death records, and anatomical risk markers, culminating in the coronary artery disease risk prediction model.

[0064] Figure 8: A schematic diagram illustrating advanced data handling techniques, including synthetic data augmentation and longitudinal tracking of anatomical and hemodynamic changes over time.

[0065] Figure 9: An illustration of an integrated workflow for cardiovascular risk prediction and treatment recommendations highlighting four key stages (i) data workflow with (ii) full classification of anatomy, plaque, and blood flow, (iii) profile population variation, (iv) risk prediction machine learning-based weighting of known risk factors, new medical image-derived risk factors, and clinical outcome.

[0066] Figure 10a: A block diagram illustrating the integration of clinical decision support tools, multimodal imaging data, and secure pipelines for patient-specific risk assessment.

[0067] Figure 10b: A block diagram illustrating customizable clinical dashboards integrating risk assessments, predictive alerts, and treatment recommendations for improved patient management.

[0068] Figs. Ila and 1 lb form a schematic block diagram of a general-purpose computer system upon which arrangements described can be practiced.Detailed Description

[0069] The present disclosure relates to a system and method for improving cardiovascular risk assessment by leveraging advanced angiographic imaging techniques, machine learning algorithms, and patient-specific anatomical and hemodynamic data. The following disclosure addresses the long-standing limitations of existing risk stratification methods by introducing anintegrated, automated framework capable of identifying high-risk individuals with unprecedented accuracy and efficiency. The disclosure also enables early detection, accurate risk prediction, and tailored clinical recommendations, significantly improving patient outcomes.

[0070] The system comprises interconnected modules for data acquisition, feature extraction, hemodynamic analysis, and risk prediction, seamlessly integrated into existing clinical workflows. Machine learning models are employed to synthesize anatomical and hemodynamic parameters with demographic data, delivering actionable insights to clinicians in real time. This detailed description references specific figures to illustrate embodiments of the invention and its components.

[0071] In an example embodiment, Figure 1 illustrates a system architecture 100 for calculating patient-specific cardiovascular risk. The architecture comprises several interlinked modules, each performing specialized tasks to process angiographic imaging data and generate actionable clinical insights.

[0072] The data acquisition module (101) is responsible for collecting angiographic imaging data, which may involve either directly importing raw imaging datasets from CTCA imaging systems or retrieving previously recorded imaging data from secure databases. In addition to the CTCA images, the data acquisition module (101) may also collect complementary non-invasive imaging data, including MRI or ultrasound where available. Furthermore, the module (101) also integrates patient-specific clinical and demographic data, such as age, gender, ethnicity, and medical history, to create a comprehensive dataset for subsequent analysis.

[0073] The feature extraction module (102) processes the imaging data using advanced segmentation algorithms to isolate anatomical features at lesion, vessel, and segment-specific regions for the entire coronary tree. These features include, but are not limited to, vessel geometry, curvature, vessel diameter, bifurcation angles, and tortuosity. The extracted features are converted into a tessellated mesh representation (102a), preserving anatomical accuracy and enabling further computational analysis in both localized and global coronary trees.

[0074] The hemodynamic calculation module (103) computes critical blood flow characteristics, such as wall shear stress, oscillatory shear index, and pressure gradients. These computations are supported by flow simulation data (103a). Additionally, the module (103) identifies areas of lowshear stress that are indicative of plaque accumulation and elevated cardiovascular risk.Computational geometry techniques are employed for these analyses, while ray-casting algorithms, typically used for diameter measurement, are applied earlier in the feature extraction module (102).

[0075] Furthermore, the hemodynamic calculation module (103) performs computational simulations to derive quantitative metrics that describe blood flow behavior within the coronary arteries. These calculations utilize advanced numerical techniques, including computational geometry and ray-casting algorithms, to compute parameters such as wall shear stress, oscillatory shear index, and pressure gradients. These processes form the foundation for the hemodynamic analysis, enabling the identification of regions of elevated cardiovascular risk. The computational nature of these hemodynamic calculations ensures a high degree of accuracy and reproducibility in the analysis of patient-specific coronary blood flow dynamics.

[0076] The outputs of the hemodynamic calculation module (103) are referred to as "blood flow characteristics" and include specific parameters such as wall shear stress, pressure gradients, and helicity. These characteristics serve as inputs for the machine learning-based risk analysis module (104), where they are combined with extracted anatomical parameters to generate patient-specific risk markers. The distinction between the calculation process and its outputs ensures that the blood flow characteristics are accurately represented as measurable metrics derived from the hemodynamic simulations. This separation provides a clear and consistent framework for integrating computational results into the broader cardiovascular risk assessment pipeline.

[0077] The machine learning-based risk analysis module (104) utilizes a trained neural network or any other machine learning model (104a) to analyze the anatomical and hemodynamic parameters, generating patient-specific risk markers. The module 104 incorporates synthetic data augmentation to enhance prediction accuracy and robustness, accounting for population-specific variations such as age, ethnicity, and prevalent risk factors.

[0078] The risk assessment module (105) synthesizes the risk markers into a comprehensive risk profile (105a), which stratifies patients into clinically relevant risk categories. The module 105 also integrates its outputs with existing clinical scoring systems to augment traditional risk assessment practices.

[0079] In an embodiment, the patient-specific risk markers generated by the machine-learning based risk analysis module (104) comprise numerical values produced by computational processing of the extracted anatomical parameters and calculated hemodynamic characteristics. Such risk markers may be associated with specific anatomical locations, including individual coronary artery segments, branches, lesions, or discrete elements of a vessel mesh representation, rather than being limited to a single aggregate score. In this manner, the risk markers can represent spatially localised indicators of cardiovascular risk that are mapped to corresponding regions of the coronary anatomy for further processing, visualisation, or reporting.

[0080] The output module (106) delivers results in multiple formats, including visual overlays (106a) on imaging data, numerical risk scores, and actionable clinical recommendations. A cloud-based data storage module (150) ensures secure handling and scalable processing of sensitive patient information. The system 100 also includes an alert generation module (107), which notifies clinicians of high-risk regions requiring immediate attention.

[0081] The data storage module (150) is configured to facilitate seamless integration with various components of the system, including the machine learning-based risk analysis module (104). This connection enables the storage of both raw and processed data, such as intermediate hemodynamic parameters, anatomical features, and patient-specific risk markers. By maintaining a repository of historical datasets, including synthetic data used for model training, the module supports continuous learning and validation of the machine learning models. Additionally, it provides a feedback loop for refining predictive algorithms, ensuring that the system evolves to incorporate the latest clinical insights and demographic variations. The connection between the data storage module (150) and the machine learning-based risk analysis module (104) is pivotal for enabling an adaptive learning process, ensuring robust and clinically relevant risk assessments over time.

[0082] Figure 2 depicts an end-to-end workflow 200 implemented by the system 100, detailing the seamless flow of data from acquisition to clinical output. The process begins with image acquisition (210), where CTCA imaging data is collected using standard clinical imaging systems. Additional modalities, such as MRI or ultrasound, may be integrated to enhance the dataset. The image acquisition 210 may be performed by the module 101.

[0083] The acquired data undergoes segmentation and feature extraction (220), where a segmentation algorithm (222) isolates the coronary arteries from the imaging data. The algorithm generates a three-dimensional representation (222a) of the vascular structure and extracts key anatomical features, such as bifurcation angles and vessel curvature. A tessellated mesh of the coronary tree (222b) is created to enable detailed hemodynamic analysis. The segmentation and feature extraction 220 may be performed by the module 102.

[0084] In an embodiment, the segmentation algorithm employed within the feature extraction module (102) may comprise a U-Net architecture or similar machine learning-based methods. These algorithms may be trained on datasets comprising coronary angiograms with varying disease severities, incorporating follow-up imaging data and, in some instances, clinical outcome data. Furthermore, the machine learning-based model may utilize a graph neural network to label and classify specific coronary artery segments, enabling detailed anatomical and functional analysis. The model may be trained using datasets that include angiographic imaging data annotated with clinical outcomes and patient demographic information, ensuring robustness and clinical relevance in predictive risk assessments.

[0085] The next stage involves hemodynamic analysis (230), which calculates blood flow characteristics, including wall shear stress, helicity, and oscillatory shear index. These metrics identify regions of abnormal flow dynamics, often associated with plaque accumulation and cardiovascular risk. This step produces hemodynamic profiles (230a) that are fed into the machine learning framework. The analysis 230 can be implemented by the module 103.

[0086] The machine learning-based risk assessment module (240) processes these features determined at 220 and 230, generating patient-specific risk markers (240a) using a neural network model trained on both real-world and synthetic datasets. Advanced interpretability techniques, such as SHAP analysis (240b) or other machine learning techniques, may enhance the understanding of parameter importance. The machine learning-based risk assessment can be implemented by the module 104.

[0087] The workflow concludes with visualization and output generation (250), where results are presented as intuitive visual overlays highlighting high-risk regions, accompanied by numerical risk scores (250a) and clinical recommendations. Outputs are formatted for directintegration into existing clinical workflows, enabling real-time decision-making. The visualization and output generation 250 can be implemented by the module 106.

[0088] Figure 3a illustrates a machine learning framework 300 that drives the cardiovascular risk prediction system and results in the machine learning module 104 and trained model 104a. The framework integrates diverse data sources and advanced computational techniques to provide accurate and interpretable predictions.

[0089] The framework 300 begins with data integration (301), combining imaging-derived features, such as vessel geometry (301a) and hemodynamic parameters (301b), with patient demographics and clinical history. This holistic dataset of vessel geometry and hemodynamic parameters ensures that the predictions account for both anatomical and systemic risk factors. During model training (302), synthetic datasets (302a) are employed to augment the training process, capturing variations in coronary anatomy and hemodynamics across diverse demographic groups. SHAP analysis (302b) is used to identify and prioritize the most impactful features, enhancing both model performance and interpretability.

[0090] The risk prediction module (303) generates patient-specific risk markers (303a), including indices of plaque progression risk and regions of low wall shear stress. The markers 303a are validated against real -world clinical outcomes and longitudinal follow-up data to ensure reliability. An output generation module (304) synthesizes the predictions 303a into actionable insights (304a), such as risk profiles and treatment recommendations. Outputs are optimized for integration into clinical workflows, enabling immediate and practical application in patient care.

[0091] In an embodiment, the deep convolutional neural network (CNN) is configured with 3D layers and convolutions to process three-dimensional CTCA imaging data. The network consists of encoder and decoder stages, where each stage includes two convolutional layers followed by either a max pooling operation (encoder) or an up-sampling operation (decoder) to extract and refine spatial features from the imaging data. The use of skip connections between corresponding encoder and decoder layers ensures spatial information is preserved during reconstruction, allowing for more precise segmentation of coronary arteries.

[0092] The input to the network consists of patches extracted from the CTCA imaging dataset, with a patch size of 256^384x384 voxels. This patch-based approach allows for efficient trainingwhile maintaining fine anatomical details. The first max pooling operation is applied only along the in-plane axes, as the in-plane resolution is higher compared to the through-plane resolution and can support further downsampling without significant loss of structural fidelity.

[0093] The network is designed with a depth of seven layers, with the bottleneck layer comprising 4^3x3 voxels, allowing for the capture of both local and global anatomical features. Each convolutional layer employs non-linear activation functions, such as Rectified Linear Units (ReLU), to introduce feature selectivity and improve model learning efficiency. Batch normalization layers may be included to stabilize training and accelerate convergence, while dropout regularization is applied at specific stages to reduce overfitting and enhance generalization to unseen data.

[0094] The segmentation output generates binary or multi-class probability maps that delineate the coronary arteries from surrounding tissue. This output is post-processed to refine vessel boundaries and correct segmentation artifacts, ensuring high anatomical accuracy. The segmented coronary structures serve as a crucial input for downstream feature extraction and hemodynamic modeling, enabling the system to compute anatomical parameters such as vessel geometry, bifurcation angles, and tortuosity for cardiovascular risk assessment.

[0095] In an embodiment, the machine learning model is trained using a dataset of 500 patients, which includes comprehensive coronary angiographic imaging data along with corresponding clinical and demographic information. The dataset size has been determined based on statistical power calculations to ensure robust predictive performance.

[0096] To establish statistical significance in cardiovascular risk prediction, the model requires a minimum sample size of approximately 445 patients to demonstrate that a c-statistic of 0.80 is significantly different from the null hypothesis, assuming:c-statistic of 0.71 for conventional risk assessment models,p-value threshold of 0.05,statistical power of 80%, andevent rate of 20%.

[0097] Additionally, the dataset includes a sufficient number of patient cases to support the prediction of plaque progression requiring intervention, ensuring that the trained model cangeneralize effectively across different clinical scenarios. The dataset provides sufficient power for meaningful statistical analysis and validation of risk prediction outputs.

[0098] In alternative embodiments, dataset sizes may vary depending on the availability of annotated angiographic imaging data and the statistical confidence required for specific clinical applications. Larger datasets may be incorporated to further improve the generalizability of the model, particularly when accounting for variations in patient demographics, disease prevalence, and imaging acquisition protocols.

[0099] In an embodiment, the system incorporates longitudinal data analysis to track changes in cardiovascular risk over time, enabling predictive modeling of future adverse events. The machine learning framework is designed to analyze time-dependent hemodynamic variations, leveraging data from sequential imaging studies to identify evolving risk markers.

[0100] For hemodynamic time-dependent prediction, a convolutional neural network (CNN) is employed, as it effectively models fixed temporal relationships due to the assumption that all patients have a consistent heart rate during the imaging acquisition process. This approach enables robust estimation of wall shear stress, oscillatory shear index, and other timevarying hemodynamic factors across multiple cardiac cycles.

[0100] In alternative embodiments, a recurrent neural network (RNN) architecture may be used to predict hemodynamics under conditions where the timing of imaging acquisitions varies across patients. RNNs allow the model to accommodate variable time intervals and heart rate fluctuations, enabling more flexible temporal modelling of disease progression.

[0101] The feasibility of this approach was validated in an experimental study focusing on the left main bifurcation, which accounts for approximately 70% of all major adverse cardiovascular events (MACE). The model demonstrated an area under the curve (AUC) of 0.72, successfully identifying high-risk events in patients without known risk factors. The median follow-up period for the dataset was 3.2 years, confirming the model’s ability to track and predict long-term changes in cardiovascular risk. By incorporating longitudinal imaging data and predictive modelling techniques, the system enhances risk stratification and supports proactive clinical decision-making, particularly in asymptomatic patients who may otherwise remain undiagnosed.

[0102] In an embodiment, the system also incorporates demographic and population-based variations into the machine learning training process to improve the generalizability of cardiovascular risk predictions across diverse patient cohorts. The dataset used for training the machine learning model includes key clinical comorbidities, such as hypertension, diabetes, and sex-based variations, which have been identified as significant predictors of adverse cardiovascular events.

[0103] The current longitudinal clinical dataset predominantly consists of a homogeneous patient population in terms of ethnicity and socioeconomic status. To address this limitation, the system explores the generation of synthetic datasets to introduce demographic diversity into model training. These synthetic datasets are designed to reflect variations in coronary artery morphology, hemodynamic characteristics, and disease progression patterns observed across different ethnic and socioeconomic groups. This approach aims to enhance the robustness of the machine learning model by ensuring it remains effective across varied populations.

[0104] During the data preprocessing phase, sampling techniques are applied to optimize the efficiency of model training. In an example embodiment, at least half of the training patches are selected to contain at least one foreground voxel, ensuring that meaningful anatomical features are present. This strategy prioritizes regions of interest and accelerates model convergence by reducing the influence of background-only patches, which offer limited predictive value.

[0105] By integrating population variations, comorbidity considerations, and synthetic data augmentation, the system enhances the reliability of risk predictions, ensuring that the model remains clinically relevant for a broader patient population.

[0106] Figure 3b provides a detailed view of a neural network architecture 300b used within the machine learning framework 300. This architecture processes both local and global features for comprehensive risk prediction.

[0107] Input features (310) of the architecture 300b include local parameters, such as longitudinal and angular coordinates (310a), centreline curvature (310b), and vessel tortuosity (310c), as well as global metrics, such as bifurcation angles (310d), inflow angle (310e), vessel radii (3 lOf), and average flow rate (310g). These inputs are processed through a machine learning model. In an embodiment, the machine learning model could be a multi-scaleconvolutional neural network (CNN) (311), which captures complex interactions between anatomical and hemodynamic characteristics. During training (312), separate CNN models are developed for specific fluid paths, such as left main to left anterior descending (LM-LAD) and left main to left circumflex (LM-LCX) blood flow paths. Results from overlapping regions are averaged during inference (312a) to ensure accuracy. Shear stress predictions (312b) are interpolated using bivariate third-degree splines, ensuring physically meaningful outputs that are remapped to the original coronary artery mesh (312c) for visualization. The training 312 results in the trained machine learning model 104a.

[0108] In an example embodiment, synthetic stenosed versions of the left main branch (LMB) are generated based on existing patient data to enhance the robustness and versatility of the training process. The remapping procedure employed facilitates the transformation of diseased vessels into a rectangular grid representation suitable for input into the convolutional neural network (CNN) 311. This approach is further extended to whole coronary trees, enabling predictions of additional metrics, such as the Topological Shear Variation Index. Consistent methodologies are employed for branches of the coronary tree, with CNN-derived results being stitched together to generate comprehensive and anatomically accurate risk assessments.

[0109] The architecture 300b incorporates dynamic training processes using synthetic stenosis models (313), which simulate variations in coronary artery geometry and flow characteristics under diseased conditions. These synthetic models enhance the network's ability to predict hemodynamic parameters in complex clinical scenarios.

[0110] The architecture 300b dynamically updates its weights and biases based on new patient data (314), ensuring continuous improvement in predictive performance. This adaptability, combined with multi-scale convolutional layers and advanced remapping techniques, ensures robust, interpretable, and clinically relevant risk predictions. The overall prediction score (315) is then displayed to an end-user, for example as output by the module 106.

[0111] The framework 300 leverages a diverse dataset comprising 500 patient records, as will be illustrated in relation to the Figure 9 embodiment, which serves as a robust baseline. These records include imaging-derived features, clinical outcomes, and demographic variations.However, this size is not an absolute minimum but an example dataset, as larger datasets would improve the model's generalizability and robustness. To enhance the training process, syntheticdatasets are generated, as described in Figure 3b, by introducing variations in coronary artery anatomy, including vessel curvature, bifurcation angles, and stenosis severity. These augmentations simulate real-world variability across populations, ensuring the model accounts for anatomical and demographic diversity.

[0112] Figure 3c illustrates a network architecture 300c for a multi-scale U-Net structure employed for the prediction network to facilitate cardiovascular risk assessment. The architecture 300c generates half- and quarter-scale versions of the input data through Max Pooling operations, enabling hierarchical feature extraction. Global features are transformed into two-dimensional feature maps using a locally connected layer and integrated with the quarterscale input data to enhance predictive accuracy. To mitigate overfitting and improve the generalization capability of the model, spatial dropout techniques are employed.

[0113] The neural network models 311 according to the example described herein are trained using an NVIDIA VI 00 GPU, ensuring computational efficiency for large-scale datasets. The training process 312 optimizes the model using the mean absolute error as the loss function, which provides an effective metric for minimizing prediction discrepancies. An Adam optimizer is employed during training 312 for its adaptive learning rate capabilities, facilitating efficient convergence during the training process. The model in this example embodiment underwent training for 1000 epochs to achieve optimal performance, validated through predefined benchmarking.

[0114] Figure 4 illustrates a data flow architecture 400 of the system 100, emphasizing the integration of diverse data sources into the coronary artery disease risk prediction model. The architecture 400 highlights the seamless aggregation of imaging data, patient records, and clinical outcomes to produce actionable insights.

[0115] The data acquisition stage (401) serves as the starting point, collecting CTCA imaging data from CTA scanning (401a), and other and supplementary imaging modalities such as MRI or ultrasound (401b). The imaging datasets 401a and 401b are accompanied by patient demographics (402), including age, gender, ethnicity, and lifestyle factors, as well as clinical history (403) containing details of past medical interventions, medications, and known comorbidities. The system also incorporates clinical outcomes (404), derived from longitudinal follow-up records (404a) and hospital admission data (404b). These outcomes provide groundtruth data for training and validating the risk prediction model 104a. The stages 401 to 404 may be implemented by the module 101.

[0116] The aggregated dataset is processed through the data integration module (405), which merges imaging-derived parameters (such as 102a and 103a) with patient-specific factors from 401 to 404 to form a unified dataset. The integrated data is then passed to the coronary artery disease risk prediction model (406, corresponding to 104a), which generates patient-specific risk markers (406a) and comprehensive risk profiles (406b). These outputs are designed to support evidence-based clinical decision-making. For example, the outputs may be in form of the overlays 106a or in some other form.

[0117] The system architecture 400 ensures compatibility with existing electronic health record (EHR) systems, facilitating seamless integration into clinical workflows. The data flow diagram emphasizes the robustness of the system in handling diverse data types and its capacity to provide accurate and interpretable cardiovascular risk assessments.

[0118] In addition to the core components described in Figure 4, the data flow incorporates advanced synthetic data generation through the synthetic data generator, as detailed in Figure 3b. This module creates augmented datasets to enhance model training, capturing diverse anatomical and demographic variations. The generated synthetic datasets are seamlessly integrated with real-world data in the data preprocessing pipeline, ensuring a robust and representative training set. A critical feature of the coronary artery disease risk prediction model is the inclusion of SHapley Additive exPlanations (SHAP) analysis to assign feature importance. This allows the model to evaluate the contribution of individual features, such as vessel curvature and wall shear stress, to the final risk prediction, enhancing interpretability for clinical use.

[0119] The feedback loop mechanism (407) refines the prediction model 406 using real -world clinical outcomes. For example, risk scores generated during earlier assessments are cross-referenced with subsequent follow-up records to validate and improve model performance. This iterative approach ensures continuous enhancement of predictive accuracy. The data integration step is designed to preprocess diverse clinical endpoints, such as major adverse cardiovascular events (MACE), into a standardized format. This ensures consistent input for the machine learning framework while maintaining compatibility with varied data source.

[0120] Figure 5 provides a schematic view of an image preprocessing and feature extraction pipeline 500, showcasing the transformation of raw imaging data into meaningful anatomical and hemodynamic parameters. The pipeline 500 may be performed by the modules 102 and 103.

[0121] The pipeline 500 begins with image acquisition (501), where raw CTCA imaging data is imported into the system 100. The data undergoes segmentation (502), performed by a machine learning model. In an embodiment, the machine learning model may be convolutional neural network, such as a U-Net model. This segmentation isolates the coronary arteries and generates a three-dimensional representation of the vascular structure. Next, the segmented data is processed through the mesh generation module (503), where a tessellated mesh representation (503a) of the coronary tree is created. This step includes the application of a diffusion Laplacian smoothing filter (503b), which refines the mesh by eliminating spatial anomalies and ensuring anatomical consistency. The mesh generation 503 may be performed by the module 102 and outputs 503a corresponding to 102a.

[0122] In an embodiment, the machine learning model may be implemented as a two-dimensional convolutional neural network (CNN). The CNN may be trained using a combination of synthetic stenosed vessel models and non-stenosed vessel models to enhance predictive accuracy. The stenosed vessel models may represent area reductions ranging from 25% to 75%, while the non-stenosed vessel models may cover area reductions from 0% to 75%. This comprehensive training dataset allows the model to capture a wide spectrum of anatomical variations, enabling robust generalization across different patient populations. By incorporating both stenosed and non-stenosed vessel models, the CNN can effectively distinguish between normal and diseased vessel structures, improving the reliability of cardiovascular risk assessments.

[0123] The pipeline 500 then proceeds to anatomical parameter extraction (504), where key features, such as vessel geometry (504a), bifurcation angles (504b), and vessel curvature (504c), are computed. These parameters provide inputs for subsequent hemodynamic analysis and machine learning-based risk assessment.

[0124] Additionally, the preprocessing pipeline incorporates boundary condition estimation (505), where inflow rates (505a) and bifurcation flow splits (505b) are computed to support hemodynamic simulations. This ensures that the extracted features are biologically andphysically accurate, laying a strong foundation for downstream analysis. The image preprocessing pipeline incorporates longitudinal imaging data analysis as part of the segmentation module (502). By processing multiple time-point datasets, the system can monitor disease progression overtime. This functionality is used fortracking plaque development, changes in vessel geometry, and hemodynamic parameter evolution.

[0125] During the anatomical parameter extraction phase (504), the system 100 employs a population-based coronary artery atlas to identify deviations in patient-specific metrics from demographic norms. For instance, bifurcation angles and vessel tortuosity are compared to population averages, providing an additional layer of risk stratification. Noise reduction is further enhanced by adaptive filtering within the segmentation process (502). This step eliminates artifacts caused by imaging inconsistencies, ensuring clean and accurate anatomical representations.

[0126] Additionally, the system 100 integrates Al-assisted boundary identification using a machine learning-based module (506). The module 506 refines boundary conditions for inflow and bifurcation calculations, improving the accuracy of hemodynamic simulations. The image preprocessing pipeline incorporates longitudinal imaging data analysis as part of the segmentation module (503). By processing multiple time-point datasets, the system can monitor disease progression overtime. This functionality is used fortracking plaque development, changes in vessel geometry, and hemodynamic parameter evolution.

[0127] During the anatomical parameter extraction phase (504), the system employs a population-based coronary artery atlas to identify deviations in patient-specific metrics from demographic norms. For instance, bifurcation angles and vessel tortuosity are compared to population averages, providing an additional layer of risk stratification. Noise reduction is further enhanced by adaptive filtering within the segmentation process (505). This step eliminates artifacts caused by imaging inconsistencies, ensuring clean and accurate anatomical representations.

[0128] In implementations that include assessment of plaque burden or vessel remodelling, such parameters are quantitatively derived from the angiographic imaging data and associated anatomical representations. In an embodiment, plaque burden or remodelling indices may be determined based on one or more of lumen geometry, vessel diameter variation, wall thickness,changes in vessel cross-sectional area, attenuation or intensity characteristics within the imaging data, or changes observed between longitudinal imaging data acquired at different time points. These parameters are computed using image processing and computational analysis applied to the segmented coronary anatomy.

[0129] Additionally, the system integrates Al-assisted boundary identification using a machine learning-based module (506). This module refines boundary conditions for inflow and bifurcation calculations, improving the accuracy of hemodynamic simulations.

[0130] The extraction 504 and estimation 505 and module 506 can be implemented by the module 103 and operate to generate the flow simulation data 103a (for example as output from 506).

[0131] Figure 6 depicts a hemodynamic calculation workflow 600, detailing the computation of key metrics and their integration into the machine learning framework for cardiovascular risk assessment. The workflow 600 begins with input preparation (601), where anatomical features, such as vessel diameter (601a), curvature (601b), and bifurcation angles (601c), generated in the anatomical parameter extraction phase (504), are combined with flow boundary conditions, such as inflow rates (505a) and bifurcation flow splits (505b) computed during preprocessing in the boundary condition estimation module (505). These inputs are subsequently processed to generate initial hemodynamic parameters, such as wall shear stress and oscillatory shear index, which form the basis for downstream analysis in the machine learning model. The mesh representation (503a) created in the mesh generation module (503) of Figure 5 ensures that the spatial accuracy and consistency of these parameters are preserved throughout the simulation.

[0132] The hemodynamic calculation module (602) computes advanced blood flow metrics, including wall shear stress (602a), oscillatory shear index (OSI) (602b), and helicity (602c). These metrics are derived using ray-casting and computational geometry methods, ensuring high precision and accuracy. The hemodynamic calculation module also incorporates a validation framework (602d) that cross-references computed metrics with ground truth simulations generated by computational fluid dynamics (CFD) for accuracy and reliability. The computed parameters are fed into the risk analysis pipeline (603), where they are integrated with imaging-derived anatomical features. A machine learning framework (603a) processes these combinedinputs to generate patient-specific risk markers (603b). These markers include indices of plaque progression risk, regions of low wall shear stress, and other critical indicators.

[0133] The workflow 600 includes a visualization module (604), which presents the computed hemodynamic metrics as overlays 604a on imaging data. These visualizations (604a) provide an intuitive understanding of the patient’s cardiovascular risk profile. The hemodynamic calculation workflow 600 is designed to operate in real-time, ensuring rapid availability of results for clinical decision-making.

[0134] The workflow 600 further integrates advanced machine learning techniques within the risk analysis module (603). Specifically, graph neural networks are used to classify coronary artery segments and branches, enabling precise identification of anatomical and hemodynamic features.

[0135] The workflow 600 includes a real-time adjustment mechanism (605) to account for patient-specific factors such as transient blood pressure changes and heart rate variability. This dynamic adjustment ensures the relevance and accuracy of the computed parameters during clinical assessments. Confidence scoring (605a) is a key feature of the visualization module, providing clinicians with additional context for interpreting hemodynamic results. These scores reflect the reliability of metrics such as wall shear stress and oscillatory shear index, enhancing decision-making in clinical workflows.

[0136] Additional hemodynamic metrics, including relative residence time (RRT) and time-averaged wall shear stress (TAWSS), are calculated and integrated into the analysis pipeline (603). These metrics, alongside helicity and oscillatory shear index, offer a comprehensive view of flow dynamics and associated risks. To validate computed metrics, the workflow 600 incorporates cross-referencing with ground truth simulations generated by computational fluid dynamics (CFD). This validation step ensures the accuracy and reliability of the machine learning-based predictions, bolstering clinical confidence in the results.

[0137] In an embodiment, cardiovascular risk assessment outcomes are computed using a population-based coronary artery atlas. The atlas incorporates anatomical and hemodynamic data from a broad demographic, enabling comparisons against patient-specific metrics. Regions of high risk are visually highlighted with color-coded overlays, such as red for high-risk areas andgreen for low-risk regions. These overlays are based on computed parameters such as wall shear stress, oscillatory shear index, and plaque progression indices.

[0138] The system (100) leverages advanced data visualization techniques to enhance clinical interpretability. For example, interactive visual overlays 106a allow clinicians to explore specific coronary segments and access detailed numerical data associated with each region. This enables targeted analysis of high-risk zones, such as bifurcation points or areas of low shear stress. The visualization module integrates seamlessly with electronic health records (EHR), ensuring that results are presented within the clinical context of the patient. The system (100) also supports multi-patient visualization, providing comparative insights across different cases for populationlevel research or cohort-specific risk evaluation.

[0139] The system (100) also enables interactive overlays (such as 106a) that compare patientspecific risk markers with population-based averages stored in the coronary artery atlas. This functionality highlights deviations from normal anatomical and hemodynamic patterns, providing clinicians with a precise understanding of patient-specific risks. The visualization module 604 supports toggling between anatomical risk markers, such as bifurcation angles, and hemodynamic markers, such as regions of low wall shear stress. These overlays can be rendered in both 2D and 3D formats, allowing clinicians to visualize spatial risk distributions across coronary segments comprehensively. Furthermore, the system (100) facilitates exporting these visualizations for use in multidisciplinary team discussions, enhancing collaborative decisionmaking. The visualization module 604 supports interactive exploration, enabling clinicians to drill down into specific coronary segments and associated risk factors. This feature facilitates personalized treatment planning by correlating anatomical abnormalities with observed clinical outcomes from the population dataset.

[0140] In some implementations, the machine-learning based models described herein are configured to approximate or replace repeated computational fluid dynamics simulations for estimation of hemodynamic characteristics. This approach enables substantially reduced computational complexity and processing time compared to full physics-based simulations, thereby allowing near-real-time generation of hemodynamic parameters and associated risk markers from angiographic imaging data. Such reductions in computational burden facilitate integration of the system into clinical imaging workflows and real-time analysis environments.

[0141] Figure 7a illustrates a visual representation of a workflow (700a) for deriving populationlevel cardiovascular risk profiles from CTCA imaging data. The process (700a) begins with CTCA image acquisition (701), corresponding to the data acquisition module (101), where detailed coronary artery imaging is obtained for analysis. The imaging data undergoes segmentation and reconstruction (702), corresponding to the feature extraction module (102), wherein a detailed three-dimensional model of the coronary arteries is generated, isolating anatomical structures of interest. Subsequently, shape measurements (703) are extracted, including parameters such as vessel geometry, bifurcation angles, and curvature, tied to the anatomical parameter extraction step (504). These anatomical features are utilized in the next stage, which involves computed hemodynamic risk assessment using computational fluid dynamics (CFD) (704), tied to the hemodynamic calculation module (103). This step evaluates critical blood flow characteristics, such as wall shear stress and oscillatory shear index, to identify regions of elevated cardiovascular risk.

[0142] Finally, the workflow transitions to population profiling and grouping of shape and risk markers (705), where anatomical and hemodynamic features are analyzed across a cohort.Statistical models are applied to group individuals based on shape measures and calculated risks, enabling insights into the distribution of risk factors within the population. This comprehensive process allows for the identification of patient-specific and population-wide cardiovascular risk trends, aiding in early diagnosis and targeted interventions.

[0143] Referring to Figure 7b, the illustration includes an advanced workflow (700b) for cardiovascular risk assessment combining machine learning (ML) and CTCA imaging data. The workflow (700b) begins with CTCA image acquisition (701), tied to the data acquisition module 101, where high-resolution coronary artery imaging data is collected. In an example embodiment, this data is processed using a machine learning-based shape and risk assessment module (706), corresponding to the machine learning based risk assessment module (104) which significantly accelerates risk prediction, achieving results in approximately 2 seconds compared to the 5-hour standard workflow. The ML module computes hemodynamic parameters such as wall shear stress and identifies regions of elevated risk with high precision. The results of the ML-based workflow are validated against standard computational fluid dynamics (CFD) models (707), tied to the validation step (503a), achieving a high correlation coefficient (R2= 0.92), thereby confirming the accuracy and reliability of the ML predictions. The workflow 700b further identifies anatomical risk markers detectable in CTCA imaging (708), including specificbifurcation angles and regions prone to abnormal blood flow, aligned with the anatomical feature extraction module in (504). These anatomical markers, such as vessel angles and curvature, are critical for assessing cardiovascular risk.

[0144] Finally, Figure 7b illustrates a 3D representation of the left anterior descending artery (LAD) (709) and other coronary branches, with labeled angles (Angle A, Angle B, etc.) and corresponding metrics to highlight high-risk regions. This detailed anatomical analysis facilitates precise detection of cardiovascular abnormalities, supporting early interventions and personalized treatment strategies.

[0145] Figure 7c illustrates a comprehensive data integration workflow (700c) for the coronary artery disease risk prediction model. The workflow (700c) begins with data linkage (710), corresponding to the data acquisition module (101), which integrates CTCA images and reports with patient and death records and a dataset of retrospective cases. This data linkage step (710) forms a foundational dataset for predictive modelling, which feeds into a machine learning-based analysis in Figure 1 (module 104).

[0146] The next component involves the aggregation of patient records (711), which include diverse parameters such as age, ethnicity, socioeconomic status, health history, sex, height, body mass index (BMI), and biomarkers like blood pressure (BP) and low-density lipoprotein (LDL) levels. These records are complemented by hospitalization and death records (712), which contain clinical information, including calcium scoring, clinical diagnoses, clinical endpoints (e.g., myocardial infarction (MI), percutaneous coronary intervention (PCI), coronary artery bypass grafting (CABG), and death), and associated comorbidities, tied to the data storage module (150).

[0147] The integrated data is further enhanced by imaging-derived anatomical risk markers (713), tied to the anatomical parameter extraction step (504), which provide detailed insights into coronary artery structures and potential risk regions. These imaging features are fed into the coronary artery disease risk prediction model (714, corresponding to 104a). The model synthesizes 714 the combined datasets to produce actionable insights, enabling clinicians to make data-driven decisions regarding patient-specific cardiovascular risks.

[0148] In an embodiment, an overall predictive score can be generated by integrating the outputs of Figures 7, 7b, and 7c. The workflow described in Figures 7 to 7c provides detailed insights into population-level profiling and grouping of anatomical shapes and hemodynamic risk factors. The machine learning-based analysis shown in Figure 7b refines these results further by leveraging predictive modelling to identify anatomical risk markers detectable in CTCA imaging and providing a robust correlation to standard modelling approaches. The comprehensive data integration and linkage framework illustrated in Figure 7c combines patient records, imaging data, and hospitalization outcomes to feed into the coronary artery disease risk prediction model. By synthesizing the outputs of these complementary modules, the system can calculate a holistic predictive cardiovascular risk score, offering actionable insights for clinical decision-making and patient management.

[0149] Figure 8 illustrates a flow 800 of the advanced data handling techniques employed in the system to enhance the accuracy and robustness of cardiovascular risk assessment. The synthetic data augmentation module (801), corresponding to the synthetic stenosis model (313), generates artificial datasets with variations in anatomical and hemodynamic parameters, such as vessel curvature and flow velocity. This augmentation ensures that the machine learning model, corresponding to 104a in Figure 1, is trained on diverse scenarios, improving its generalizability to real-world cases. The flow 800 also incorporates longitudinal tracking capabilities (802), enabling the analysis of anatomical and hemodynamic changes over multiple time points. This tracking allows for an improved monitoring of the disease progression and evaluating the efficacy of therapeutic interventions. For example, changes in wall shear stress or plaque burden over time, tied to the outputs of the hemodynamic calculation module (601-604), are visualized through time-lapse overlays (803), providing a dynamic view of disease trajectory.

[0150] Data preprocessing pipelines (804) ensure consistency and standardization across datasets from different imaging modalities, such as CTCA, MRI, and ultrasound. These pipelines include automated data cleaning, alignment, and normalization steps, preparing the data for robust analysis. In an embodiment, the system may further comprise a data preprocessing module configured to filter and select the imaging dataset obtained by the data acquisition module, selecting optimal imaging series based on parameters including resolution, axial spacing, field of view, and phase of the cardiac cycle, and a visualization module configured to produce color-coded overlays on the angiographic imaging data, highlighting regions of high cardiovascular risk

[0151] The flow 800 also features a multi-source data integration module (805), linked to the data integration step (301), which combines imaging data with patient demographics, clinical history, and outcome records. This holistic approach ensures that risk predictions are both comprehensive and patient-specific. The data preprocessing pipeline (804) is configured to evaluate imaging series based on parameters such as axial spacing, which determines the resolution in the axial plane, and field of view, which defines the area captured in each image slice. These parameters ensure that the selected imaging series provides sufficient detail for accurate segmentation (502) and risk assessment (105).

[0152] Referring to Figure 8, the flow 800 employs advanced data handling techniques to harmonize data from multiple imaging modalities. A cross-modality harmonization module (801a), linked to the data acquisition module (401), ensures consistent preprocessing across CTCA, MRI, and ultrasound data, reducing variability introduced by differing imaging protocols. Synthetic data augmentation (802) further strengthens the robustness of machine learning models by introducing variations in stenosis severity, bifurcation angles, and hemodynamic disruptions. The longitudinal tracking module (803) incorporates statistical models to analyze trends in patient-specific risk markers, identifying patterns indicative of disease progression. This module dynamically updates predictions as new imaging and clinical data become available, ensuring continuous refinement of risk assessments. The data handling pipeline integrates seamlessly with the visualization and risk assessment modules, corresponding to the visualizations (604a) to provide clinicians with up-to-date, evidence-based insights.

[0153] Referring to Figure 9, the diagram illustrates an integrated framework (900) for cardiovascular risk prediction and treatment recommendations, focusing on repeated imaging, whole-tree modelling, population clustering, and individualized assessments. The workflow (900) begins with repeated CTCA imaging and Al-derived plaque and wall analysis, tied to the hemodynamic parameter extraction steps (601)-(604) in Figure 6, which incorporates a patient dataset, which in the example shown encompasses about 500 patients. This step utilizes advanced algorithms to identify anatomical abnormalities and derive critical metrics, such as wall thickness and plaque composition.

[0154] The next stage, whole-tree anatomy, plaque, and hemodynamic modelling, uses computational methods to reconstruct the entire coronary tree, enabling the analysis of flow dynamics, plaque burden, and structural variations across the coronary arteries. This stepprovides insights into the global interaction of anatomical features and hemodynamic parameters, building on the mesh generation (503) and anatomical parameter extraction steps (504) in Figure 5.

[0155] Following this, the cluster and classification module groups patients based on populationwide variations in anatomical and hemodynamic parameters. This step identifies subgroups with similar risk profiles or anatomical traits, enabling a targeted analysis of these patient populations.

[0156] The final module delivers risk assessments and treatment recommendations, integrating outputs from previous steps, including the risk markers (303a) generated by the machine learning / model training framework (302). The left-hand graph in this module highlights variable importance scores for risk factors such as gender, age, tortuosity, smoking, vessel narrowing and endothelial shear stress (ESS), which correspond to the Shapley Additive exPlanations (SHAP) analysis (302b). This graph provides a quantitative ranking of their predictive value. The adjacent graph, labelled function of models sensitivity and specificity, showcases the performance of the predictive models, demonstrating their ability to distinguish between high-and low-risk patients. This performance evaluation aids in validating the accuracy and reliability of the generated risk assessments. Together, these outputs enable the generation of tailored treatment recommendations, enhancing clinical decision-making through personalized risk stratification and targeted interventions.

[0157] Figure 10a presents a block diagram 1000 of the system’s (100) integration with clinical decision support (CDS) tools. The system 100 aggregates multimodal imaging data (1001), such as CTCA, MRI, and ultrasound, alongside patient-specific information, including demographics, medical history, and clinical endpoints. This data is securely processed through encryption-enabled pipelines (1002), ensuring compliance with regulatory standards for patient data protection.

[0158] The clinical decision support module (1003) synthesizes the data into actionable recommendations for clinicians. These recommendations include tailored treatment strategies (e.g., lifestyle modifications, medication adjustments, or interventional procedures) based on the patient’s risk profile. The module supports real-time decision-making by presenting insights in an intuitive interface.

[0159] The system also includes a cloud-based analytics engine (1004), which enables remote access to patient-specific risk assessments and facilitates collaborative deci si on -making among multidisciplinary teams. This engine leverages secure APIs to ensure compatibility with existing clinical software and workflows.

[0160] To enhance interpretability, the system incorporates explainable Al techniques (1005), tied to the prediction model (406), providing clinicians with detailed insights into how specific parameters contributed to the overall risk assessment. For example, the system may highlight the influence of low wall shear stress or bifurcation angles on the predicted risk score.

[0161] The integration of predictive modelling (1006) enables the system to forecast future cardiovascular events, offering proactive recommendations for follow-up care. This predictive capability is particularly valuable for high-risk patients, allowing for early intervention and improved clinical outcomes.

[0162] Referring to Figure 10b, in a system 1000b, the integration of clinical decision support tools is further enhanced through customizable dashboards (1011) tailored for different clinical roles. These dashboards consolidate multimodal imaging data, risk assessments, and treatment recommendations into intuitive, role-specific views. The system also incorporates an alert generation module (1012), which provides real-time notifications for high-risk patients requiring immediate attention. Proactive alerts (1013) are configured based on machine learning predictions, offering early warnings for acute events such as myocardial infarction. The secure data pipeline (1014) ensures compliance with health information standards, including HL7 and FHIR, enabling seamless integration with electronic health record (EHR) systems. Collaborative access (1015) allows care teams to review risk assessments and treatment recommendations securely, enhancing coordinated patient management. Additionally, the predictive modelling framework (1016) includes advanced algorithms to forecast future risk trajectories, supporting preventive care strategies.

[0163] Disclosed is a system and method for patient-specific cardiovascular risk assessment leveraging advanced image processing, machine learning, and predictive modelling to analyze anatomical and hemodynamic parameters of the coronary arterial network. The disclosure introduces novel aspects, including the use of whole-tree modelling for comprehensive analysis of the coronary arteries, integration of population-specific variations and synthetic dataaugmentation for improved model robustness, longitudinal tracking of patient-specific risk markers, and the generation of an overall predictive risk score. These innovations are further enhanced by the system’s ability to output actionable insights, such as visual overlays, numerical risk assessments, and treatment recommendations, ensuring seamless integration into clinical workflows and supporting preventive, personalized cardiovascular care.

[0164] Figures Ila and 1 lb depict a general -purpose computer system 1100, upon which the various arrangements described for the cardiovascular risk prediction system can be practiced.

[0165] As seen in Figure Ila, the computer system 1100 includes: a computer module 1101; input devices such as a keyboard 1102, a mouse pointer device 1003, a scanner 1126, a camera 1127, and a microphone 1180; and output devices including a printer 1115, a display device 1114, and loudspeakers 1117. An external Modulator-Demodulator (Modem) transceiver device 1116 may be used by the computer module 1101 for communicating to and from a communications network 1120 via a connection 1121. The communications network 1120 may be a wide-area network (WAN), such as the Internet, a cellular telecommunications network, or a private WAN. Where the connection 1121 is a telephone line, the modem 1116 may be a traditional “dial-up” modem. Alternatively, where the connection 1121 is a high-capacity (e.g., cable) connection, the modem 1116 may be a broadband modem. A wireless modem may also be used for wireless connection to the communications network 1120.

[0166] The computer module 1101 corresponds to components in the system architecture 100 (e.g., the data acquisition module 101, the feature extraction module 102, or the computing unit 120). The module 1101 typically includes at least one processor unit 1105, and a memory unit 1106. For example, the memory unit 1106 may have semiconductor random access memory (RAM) and semiconductor read-only memory (ROM). The computer module 1101 also includes several input / output (VO) interfaces, including: an audio-video interface 1107 that couples to the video display 1114, loudspeakers 1117, and microphone 1080; an I / O interface 1113 that couples to the keyboard 1102, mouse 1103, scanner 1126, camera 1127, and optionally a joystick or other human interface device (not illustrated); and an interface 1108 for the external modem 1116 and printer 1115. In some implementations, the modem 1116 may be incorporated within the computer module 1101, for example within the interface 1108.

[0167] The computer module 1101 also has a local network interface 1111, which permits coupling of the computer system 1100 via a connection 1123 to a local-area communications network 1122, known as a Local Area Network (LAN). As illustrated in Figure 1 la, the local communications network 1122 may also couple to the wide network 1120 via a connection 1124, which would typically include a so-called “firewall” device or device of similar functionality. The local network interface 1111 may comprise an Ethernet circuit card, a Bluetooth® wireless arrangement, or an IEEE 802.11 wireless arrangement; however, numerous other types of interfaces may be practiced for the interface 1111.

[0168] The I / O interfaces 1108 and 1113 may afford either or both of serial and parallel connectivity, the former typically being implemented according to the Universal Serial Bus (USB) standards and having corresponding USB connectors (not illustrated). Storage devices 1109 are provided and typically include a hard disk drive (HDD) 1110. Other storage devices such as a floppy disk drive and a magnetic tape drive (not illustrated) may also be used. An optical disk drive 1112 is typically provided to act as a non-volatile source of data. Portable memory devices, such as optical disks (e.g., CD-ROM, DVD, Blu-ray DiscTM), USB-RAM, portable external hard drives, and floppy disks, for example, may be used as appropriate sources of data to the system 1100.

[0169] The components 1105 to 1113 of the computer module 1101 typically communicate via an interconnected bus 1104 and in a manner that results in a conventional mode of operation of the computer system 1100 known to those in the relevant art. For example, the processor 1005 is coupled to the system bus 1004 using a connection 1118. Likewise, the memory 1006 and optical disk drive 1112 are coupled to the system bus 1104 by connections 1119. Examples of computers on which the described arrangements can be practiced include IBM-PCs and compatibles, Sun Sparcstations, Apple Mac™, or like computer systems.

[0170] The method described herein may be implemented using the computer system 1100 wherein the processes of Figures 1 to 10, described above, may be implemented as one or more software application programs 1133 executable within the computer system 1100. In particular, the steps of the method of Figures 1 to 10 are effected by instructions 1131 (see Figure 1 lb) in the software 1133 that are carried out within the computer system 1100. The software instructions 1131 may be formed as one or more code modules, each for performing one or more particular tasks. The software may also be divided into two separate parts, in which a first partand the corresponding code modules perform the described methods and a second part and the corresponding code modules manage a user interface between the first part and the user.

[0171] The software may be stored in a computer-readable medium, including the storage devices described below, for example. The software is loaded into the computer system 1100 from the computer-readable medium and then executed by the computer system 1100. A computer-readable medium having such software or computer program recorded on the computer-readable medium is a computer program product. The use of the computer program product in the computer system 1100 preferably effects an advantageous apparatus for implementing the cardiovascular risk prediction system.

[0172] The software 1133 is typically stored in the HDD 1110 or the memory 1106. The software is loaded into the computer system 1000 from a computer-readable medium and executed by the computer system 1100. Thus, for example, the software 1133 may be stored on an optically readable disk storage medium (e.g., CD-ROM) 1025 that is read by the optical disk drive 1112. A computer-readable medium having such software or computer program recorded on it is a computer program product. The use of the computer program product in the computer system 1100 preferably effects an apparatus for performing the cardiovascular risk analysis processes.

[0173] In some instances, the application programs 1133 may be supplied to the user encoded on one or more CD-ROMs 1125 and read via the corresponding drive 1112, or alternatively may be read by the user from the networks 1120 or 1122. Still further, the software can also be loaded into the computer system 1100 from other computer-readable media. Computer-readable storage media refer to any non-transitory tangible storage medium that provides recorded instructions and / or data to the computer system 1100 for execution and / or processing. Examples of such storage media include floppy disks, magnetic tape, CD-ROM, DVD, Blu-ray TM Disc, a hard disk drive, a ROM or integrated circuit, USB memory, a magneto-optical disk, or a computer-readable card such as a PCMCIA card and the like, whether or not such devices are internal or external to the computer module 1101. Examples of transitory or non-tangible computer readable transmission media that may also participate in the provision of software, application programs, instructions and / or data to the computer module 1101 include radio or infra-red transmission channels as well as a network connection to another computer or networked device, and theInternet or Intranets including e-mail transmissions and information recorded on Websites and the like.

[0174] The second part of the application programs 1133 and the corresponding code modules mentioned above may be executed to implement one or more graphical user interfaces (GUIs) to be rendered or otherwise represented upon the display 1114. Through manipulation of typically the keyboard 1102 and the mouse 1103, a user of the computer system 1100 and the application may manipulate the interface in a functionally adaptable manner to provide controlling commands and / or input to the applications associated with the GUI(s). Other forms of functionally adaptable user interfaces may also be implemented, such as an audio interface utilizing speech prompts output via the loudspeakers 1117 and user voice commands input via the microphone 1080.

[0175] Figure 1 lb is a detailed schematic block diagram of the processor 1105 and a "memory" 1134. The memory 1134 represents a logical aggregation of all the memory modules (including the HDD 1110 and semiconductor memory 1106) that can be accessed by the computer module 1101 in Figure Ila.

[0176] When the computer module 1101 is initially powered up, a power-on self-test (POST) program 1150 executes. The POST program 1150 is typically stored in a ROM 1149 of the semiconductor memory 1106 of Figure 1 la. A hardware device such as the ROM 1149 storing software is sometimes referred to as firmware. The POST program 1150 examines hardware within the computer module 1101 to ensure proper functioning and typically checks the processor 1105, the memory 1134 (1109, 1106), and a basic input-output systems software (BIOS) module 1151, also typically stored in the ROM 1149, for correct operation. Once the POST program 1150 has run successfully, the BIOS 1011 activates the hard disk drive 1110 of Figure Ila. Activation of the hard disk drive 1110 causes a bootstrap loader program 1152 that is resident on the hard disk drive 1010 to execute via the processor 1105. This loads an operating system 1013 into the RAM memory 1106, upon which the operating system 1053 commences operation. The operating system 1153 is a system-level application, executable by the processor 1105, to fulfill various high-level functions, including processor management, memory management, device management, storage management, software application interface, and generic user interface.

[0177] The operating system 1153 manages the memory 1134 (1109, 1106) to ensure that each process or application running on the computer module 1101 has sufficient memory in which to execute without colliding with memory allocated to another process. Furthermore, the different types of memory available in the system 1100 of Figure Ila must be used properly so that each process can run effectively. Accordingly, the aggregated memory 1134 is not intended to illustrate how particular segments of memory are allocated (unless otherwise stated), but rather to provide a general view of the memory accessible by the computer system 1100 and how such is used.

[0178] As shown in Figure 1 lb, the processor 1105 includes a number of functional modules including a control unit 1139, an arithmetic logic unit (ALU) 1140, and a local or internal memory 1148, sometimes called a cache memory. The cache memory 1148 typically includes a number of storage registers 1144-1146 in a register section. One or more internal buses 1141 functionally interconnect these functional modules. The processor 1105 typically also has one or more interfaces 1142 for communicating with external devices via the system bus 1104, using a connection 1118. The memory 1134 is coupled to the bus 1104 using a connection 1119.

[0179] The application program 1133 includes a sequence of instructions 1131 that may include conditional branch and loop instructions. The program 1133 may also include data 1132 which is used in the execution of the program 1133. The instructions 1131 and the data 1132 are stored in memory locations 1128, 1129, 1130, and 1135, 1136, 1137, respectively. Depending upon the relative size of the instructions 1131 and the memory locations 1128-1130, a particular instruction may be stored in a single memory location as depicted by the instruction shown in the memory location 1130. Alternately, an instruction may be segmented into a number of parts each of which is stored in a separate memory location, as depicted by the instruction segments shown in the memory locations 1128 and 1129.

[0180] In general, the processor 1105 is given a set of instructions which are executed therein. The processor 1105 waits for a subsequent input, to which the processor 1105 reacts to by executing another set of instructions. Each input may be provided from one or more of a number of sources, including data generated by one or more of the input devices 1102, 1103, data received from an external source across one of the networks 1120, 1122, data retrieved from one of the storage devices 1106, 1110, or data retrieved from a storage medium 1125 inserted into the corresponding reader 1112, all depicted in Figure Ila. The execution of a set of theinstructions may in some cases result in the output of data. Execution may also involve storing data or variables to the memory 1134.

[0181] The disclosed arrangements use input variables 1154, which are stored in the memory 1134 in corresponding memory locations 1155, 1156, 1157. The disclosed arrangements produce output variables 1061, which are stored in the memory 1134 in corresponding memory locations 1162, 1163, 1164. Intermediate variables 1158 may be stored in memory locations. Referring to the processor 1105 of Figure 1 lb, the registers 1144, 1145, 1146, the arithmetic logic unit (ALU) 1140, and the control unit 1139 work together to perform sequences of micro-operations needed to perform "fetch, decode, and execute" cycles for every instruction in the instruction set making up the program 1133. Each fetch, decode, and execute cycle comprises:a fetch operation, which fetches or reads an instruction 1131 from a memory location 1128, 1129, 1130;a decode operation in which the control unit 1139 determines which instruction has been fetched; andan execute operation in which the control unit 1139 and / or the ALU 1140 execute the instruction.

[0182] Thereafter, a further fetch, decode, and execute cycle for the next instruction may be executed. Similarly, a store cycle may be performed by which the control unit 1139 stores or writes a value to a memory location 1132.

[0183] Each step or sub-process in the processes of Figures 1 to 10 is associated with one or more segments of the program 1133 and is performed by the register section 1144, 1145, 1147, the ALU 1140, and the control unit 1139 in the processor 1105 working together to perform the fetch, decode, and execute cycles for every instruction in the instruction set for the noted segments of the program 1133.

[0184] The methods described may alternatively be implemented in dedicated hardware such as one or more integrated circuits performing the functions or sub-functions of the cardiovascular risk prediction system. Such dedicated hardware may include graphic processors, digital signal processors, or one or more microprocessors and associated memories.

[0185] In this document, the terms “computer program medium” and “computer usable medium” are used to generally refer to transitory or non-transitory media. Such media may be, e.g., storage medium, storage devices and channel. These and other various forms of computer program media or computer usable media may be involved in carrying one or more sequences of one or more instructions to a processing device for execution. Such instructions embodied on the medium, are generally referred to as “computer program code” or a “computer program product” (which may be grouped in the form of computer programs or other groupings). When executed, such instructions might enable the computing component (e.g. processor 104) to perform features or functions of the present application as discussed herein.

[0186] The terms “a” and “an,” as used herein, are defined as one or more than one. The term “plurality,” as used herein, is defined as two or more than two. The term “another,” as used herein, is defined as at least a second or more. The terms “including” and / or “having,” as used herein, are defined as comprising (i.e. open language). The phrase “at least one of . . . and . . . .” as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items. As an example, the phrase “at least one of A, B, or C” includes A only, B only, C only, or any combination thereof (e.g. AB, AC, BC or ABC).

[0187] Although embodiments have been described with reference to a number of illustrative embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention. Many modifications will be apparent to those skilled in the art without departing from the scope of the present invention as herein described with reference to the accompanying drawings.

Claims

CLAIMS1. A computer implemented method for calculating patient-specific cardiovascular risk using angiographic imaging, the method comprising:extracting anatomical parameters of a patient’s coronary arteries from angiographic imaging data of the patient’s coronary arteries, the anatomical parameters including one or more of vessel geometry, curvature, vessel diameter, and vessel tortuosity;calculating blood flow characteristics from the angiographic imaging data, the blood flow characteristics including at least one of blood flow shear stress, pressure gradients, and helicity;applying a machine learning-based model to the extracted anatomical parameters and calculated blood flow characteristics to generate patient-specific risk markers;using the patient-specific risk markers to assess the likelihood of future adverse cardiovascular events; andoutputting the assessment of cardiovascular risk for clinical decision-making.

2. The method of claim 1, wherein the anatomical parameters comprise bifurcation angles, plaque burden, vessel remodelling index, twistedness of the coronary arteries, vessel tapering and centerline curvature.

3. The method of claim 1, wherein the blood flow characteristics comprise oscillatory shear index, relative residence time, and wall shear stress distribution throughout the cardiac cycle and other hemodynamic metrics such as velocity profiles and pressure oscillations.

4. The method of claim 1, wherein the patient-specific risk markers are derived by combining anatomical parameters with hemodynamic factors using feature weighting determined by a machine learning algorithm.

5. The method of claim 1, wherein the angiographic imaging data is pre-processed using a segmentation algorithm based on a convolutional neural network to isolate the coronary arteries.

6. The method of claim 1, further comprising generating synthetic coronary artery data to augment training of the machine learning-based model.

7. The method of claim 6, wherein the synthetic data comprises variations in vessel, lesion and segment-specific variations, curvature, bifurcation angles, and stenosis severity based on standard deviations observed in a training dataset.

8. The method of claim 1, wherein the likelihood of adverse cardiovascular events is assessed by identifying regions of haemodynamic and anatomical markers that have a significant association with clinical adverse events including low shear stress associated with increased plaque risk.

9. The method of claim 1, further comprising outputting visual overlays on the angiographic images to highlight regions of high risk based on the generated patient and / or population specific risk markers.

10. The method of claim 1, wherein the output includes a risk score that integrates traditional risk factors and additional patient demographic data, including clinical history, the patient specific risk markers, anatomical parameters, and hemodynamic factors.

11. The method of claim 1, further comprising outputting a clinical assessment, wherein the outputted assessment includes recommendations to assist clinicians in patient management based on patient-specific risk markers, such as initiation of medication, lifestyle modifications, or surgical interventions.

12. The method of claim 1, further comprising validating the machine learning-based model using a dataset of angiographic imaging data and corresponding ground truth hemodynamic simulations generated by computational fluid dynamics.

13. The method of claim 12, wherein the validation comprises cross-referencing predictions of wall shear stress with transient and time-averaged computational fluid dynamics simulations.

14. The method of claim 1, wherein the anatomical parameters and blood flow characteristics are computed for specific segments of the coronary tree, including proximal left anterior descending artery (LAD) and proximal left circumflex artery (LCX).

15. The method of claim 1, wherein the machine learning-based model is adapted to account for population-specific variations, including age, sex, ethnicity, and prevalent risk factors.

16. The method of claim 1, further comprising incorporating non-invasive physiological measurements, including blood pressure and cholesterol levels, into the machine learning-based model.

17. The method of claim 1, wherein the extracted anatomical parameters include metrics computed using a tessellated mesh representation of the coronary tree, smoothed using a diffusion Laplacian filter.

18. The method of claim 1, wherein the machine learning-based model comprises a multiscale convolutional neural network trained to assess anatomical characteristics and predict wall shear stress and other hemodynamic metrics .

19. The method of claim 1, further comprising generating a population-based coronary artery atlas to facilitate comparisons and predictions across different demographic groups.

20. The method of claim 2, wherein the segmentation algorithm comprises a U-Net architecture trained on a dataset comprising coronary angiograms of varying disease severities.

21. The method of claim 1, further comprising assessing longitudinal changes in patientspecific risk markers using repeat angiographic imaging data over a defined follow-up period.

22. The method of claim 1, wherein the machine learning-based model is trained using a dataset comprising angiographic imaging data annotated with clinical outcomes and patient demographic information.

23. The method of claim 1, wherein the machine learning-based model includes a multi-scale convolutional neural network trained to predict wall shear stress and other hemodynamic metrics.

24. A system for calculating patient-specific cardiovascular risk using angiographic imaging, the system comprising:a feature extraction module implemented on a processor, configured to extract anatomical parameters of a patient’s coronary arteries from angiographic imaging data of the patient’s coronary arteries, the anatomical parameters extracted at lesion, vessel and segment-specific regions for the whole tree from the angiographic imaging data, the anatomical parameters including but not limited to one or more of vessel geometry, curvature, vessel diameter, and vessel tortuosity;a hemodynamic calculation module configured to assess blood flow characteristics from the angiographic imaging data, the blood flow characteristics including at least one of blood flow shear stress, pressure gradients, and helicity;a machine learning-based risk analysis module configured to apply a trained machine learning model to the extracted anatomical parameters and calculated blood flow characteristics to generate patient-specific risk markers;a risk assessment module configured to use the patient-specific risk markers to assess the likelihood of adverse cardiovascular events; andan output module configured to deliver the assessment of cardiovascular risk, including visual and numerical outputs for clinical decision-making.

25. The system of claim 24, further comprising a data preprocessing module configured to filter and select the optimal imaging series based on parameters including resolution, axial spacing, field of view, and phase of the cardiac cycle.

26. The system of claim 24, wherein the machine learning-based risk analysis module is configured to generate synthetic coronary artery data for model training, incorporating variations in vessel curvature, stenosis severity, and bifurcation angles.

27. The system of claim 24, further comprising a visualization module configured to produce color-coded overlays on the angiographic imaging data, highlighting regions of high cardiovascular risk.

28. The system of claim 24, wherein the risk assessment module is configured to generate confidence scores or uncertainty ranges for the predicted risk markers.

29. The system of claim 24, wherein the output module is configured to integrate the cardiovascular risk assessment into an electronic health record system for seamless clinical workflow.

30. The system of claim 24, wherein the output module is configured to generates treatment recommendations based on the predicted cardiovascular risk, including suggestions for medication adjustments, surgical interventions, or lifestyle modifications.

31. The system of claim 24, wherein the hemodynamic calculation module is configured to analyze transient blood flow characteristics throughout the cardiac cycle.

32. The system of claim 24, wherein the feature extraction module comprises graph neural networks to label and classify coronary artery segments and branches.

33. The system of claim 24, further comprising a cloud-based data storage module configured to securely store and process patient data in compliance with regulatory requirements.

34. The system of claim 24, wherein the output module is configured to deliver real-time cardiovascular risk assessments during ongoing CTCA imaging.

35. The system of claim 24, wherein the risk assessment module is configured to compare patient-specific risk markers against a population-based coronary artery atlas to identify deviations from population norms.

36. The system of claim 24, wherein the visualization module provides interactive visualizations, allowing an end user to explore individual coronary segments and associated risk markers.

37. The system of claim 24, further comprising an alert generation module configured to notify clinicians of high-risk regions requiring immediate attention based on the cardiovascular risk assessment.

38. A system for cardiovascular risk assessment and reporting, the system comprising:a data acquisition module configured to receive angiographic imaging data from medical imaging devices;a segmentation module implemented on a processor, configured to identify and isolate coronary artery structures from the imaging data using a trained neural network;a computational module configured to:compute anatomical features of the coronary arteries, including vessel curvature, bifurcation angles, and twistedness; andcalculate hemodynamic characteristics, including shear stress, helicity, and flow recirculation zones;a machine learning module configured to process the anatomical and hemodynamic characteristics to generate patient-specific risk metrics; andan output module configured to display the risk metrics, including a classification of cardiovascular risk and associated clinical recommendations based on the generated risk markers.

39. A system for enhancing cardiovascular risk prediction, the system comprising:a synthetic data generator configured to create augmented datasets by introducing variations in coronary artery anatomy and hemodynamic parameters, the variations including vessel curvature, stenosis severity, bifurcation angles, and plaque burden;a machine learning module configured to be trained on the augmented datasets to analyse anatomical and hemodynamic variations and generate risk assessment parameter;a processing module configured to apply the trained machine learning model to patientspecific angiographic imaging data to compute cardiovascular risk markers; andan output module configured to deliver a cardiovascular risk assessment report, including risk predictions based on the enhanced model accuracy.

40. A reporting tool for cardiovascular risk assessment, the tool comprising:a data input interface configured to receive angiographic imaging data and associated patient demographic information;a processing module configured to:extract anatomical features of coronary arteries from the angiographic imaging data, including vessel geometry, curvature, and bifurcation angles;calculate hemodynamic parameters, including wall shear stress, oscillatory shear index, and helicity;generate patient-specific cardiovascular risk markers using a machine learning model trained on clinical and imaging datasets;a reporting module configured to:produce a visual representation of the coronary arteries annotated with risk markers;generate a summary of cardiovascular risk, including a numerical risk score and classification into predefined risk categories;provide recommendations for clinical decision-making based on the generated risk markers; andan output interface configured to deliver the report to an end-user in a format for use in patient management.

41. A system for facilitating cardiovascular risk prediction, the system comprising:a synthetic data generator configured to create augmented datasets by introducing variations in coronary artery anatomy and hemodynamic parameters, the variations including vessel curvature, stenosis severity, bifurcation angles, and plaque burden;a machine learning model trained on the augmented datasets and clinical outcomes to improve robustness and accuracy of risk predictions;a processing module configured to apply the trained machine learning model to patientspecific angiographic imaging data to compute cardiovascular risk markers; andan output module configured to deliver a cardiovascular risk assessment report, including risk predictions based on the enhanced model accuracy.

42. A computer implemented method for predicting cardiovascular risk using multimodal imaging data, the method comprising:receiving angiographic imaging data of a patient’s coronary arteries and additional non-invasive imaging data, including Magnetic Resonance Imaging (MRI) or ultrasound;integrating anatomical features derived from angiographic imaging with tissue-specific characteristics obtained from the additional imaging data;calculating combined anatomical, hemodynamic, and tissue-level risk markers; predicting cardiovascular risk using a machine learning model trained on datasets combining data from multiple imaging techniques; andoutputting a comprehensive risk profile for clinical use.

43. A computer implemented method for assessing cardiovascular risk, the method comprising:receiving a series of longitudinal angiographic imaging data of a patient's coronary arteries acquired over multiple time points;extracting anatomical and hemodynamic parameters from each set of angiographic imaging data, the parameters including vessel geometry, curvature, bifurcation angles, wall shear stress, and oscillatory shear index;analyzing changes in the extracted parameters to monitor plaque progression and associated hemodynamic variations over time;predicting the likelihood of future cardiovascular events based on trends and deviations in the parameters across the longitudinal dataset; andoutputting a time-based cardiovascular risk profile, including projected risks and recommended follow-up intervals for clinical intervention and decision-making.

44. A method for generating predictive models of cardiovascular risk, the method comprising:generating synthetic models of coronary vessels with stenosis, the synthetic models including both axially symmetric and non-symmetric stenoses with a range of vessel area reductions;applying a remapping procedure to transform the synthetic stenosed vessel models into a rectangular grid representation;training a machine learning module using the transformed synthetic models as input data to predict hemodynamic parameters associated with cardiovascular risk;extending the remapping procedure to a coronary arterial network comprising interconnected branches of the coronary vessels, to generate predictive outputs for computational metrics, including a topological shear variation index; andstitching predictive outputs from branches of the coronary tree to generate a complete anatomical and hemodynamic risk profile.

45. A method for predicting cardiovascular risk metrics for a coronary arterial network, the method comprising:transforming anatomical features of the coronary arterial network, including synthetic stenosed vessel models for analysis by a machine learning model;training the machine learning model to predict hemodynamic parameters associated with cardiovascular risk, based on features of individual branches of the coronary arterial network and interconnections within the network;analyzing every branch of the coronary arterial network using the machine learning model, including generating branch-specific predictions for hemodynamic parameters and cardiovascular risk metrics;integrating branch-specific predictions to form a unified cardiovascular risk profile for the coronary arterial network, wherein the cardiovascular risk metrics comprise at least a topological shear variation index and indicators of plaque progression risk; andoutputting the unified cardiovascular risk profile as a complete assessment of cardiovascular health across the entire coronary arterial network.

46. The method of claim 45, wherein branch-specific predictions include wall shear stress, oscillatory shear index, and flow separation zones.

47. The method of claim 45, wherein in the machine learning model is a 2-dimensional convolutional neural network (CNN), and wherein the CNN is trained using synthetic stenosed vessel models representing area reductions ranging from 25% to 75%.

48. The method of claim 45, wherein the CNN is trained using vessel geometry vessel models representing area reductions ranging from 0% to 75%.

49. The method of claim 45, wherein synthetic data, comprising representations of anatomical features and hemodynamic parameters across varying demographic populations, is used for training the CNN to account for anatomical and demographic variations.

50. The method of claim 45, wherein the Topological Shear Variation Index is computed across multiple branches of the coronary arterial network to detect systemic flow abnormalities.

Images