Method, computer, and storage medium for estimating one or more cardiovascular parameters

Abstract

A computer-implemented method of estimating one or more cardiovascular parameters of a patient The method comprising a step of receiving: a flow value, which is indicative of the time required for contrast injected into the patient to reach a predefined distal landmark, and a blood pressure value for the patient. The method further comprising predicting a first value by providing the flow value as a predictor to a first machine-learning model; predicting a second value by providing the blood pressure value as a predictor to a second machine-learning model; and estimating the one or more cardiovascular parameters using the predicted first value and the predicted second value.

Description

[0001] METHOD, COMPUTER, AND STORAGE MEDIUM FOR ESTIMATING ONE OR MORE CARDIOVASCULAR PARAMETERS

[0002] TECHNICAL FIELD

[0003] The present invention relates to a computer-implemented method of estimating one or more cardiovascular parameters, a computer configured to do the same, and a storage medium containing executable instructions for the same.

[0004] BACKGROUND

[0005] ‘Small vessels’ provide the final common path for oxygenated arterial blood to the heart. Cardiac small vessel disease can cause a blood supply:demand mismatch leading to angina, myocardial infarction, and heart failure. However, coronary microcirculation is invisible on a standard coronary angiogram. This leads to cardiac small vessel diseased often being undiagnosed, which represents a missed opportunity for treatment. The index of microcirculatory resistance (IMR) is a pressuretemperature sensor guidewire-based measurement, typically performed during cardiac catheterization, of the minimum microcirculatory resistance in a target coronary artery territory. It provides a quantitative method to assess the coronary microvasculature in the clinical setting. Coronary flow reserve (CFR) is the maximum increase in blood flow through the coronary arteries above the normal resting volume, and its measurement is often used to assist in the treatment of conditions affecting the coronary arteries. An IMR is considered ‘normal’ if it is less than 25, and ‘abnormal’ if > 25.

[0006] An IMR and / or CFR can be measured through invasive coronary physiology, which is the reference method for the diagnosis of small vessel disease in the heart. This involves a trained cardiologist passing a guidewire into a coronary artery. The guidewire sensor can then detect flow (utilising a Doppler scanner), pressure, and temperature (thermodilution) of the flowing blood. Coronary instrumentation carries risks as well as costs. Whilst uncommon, the placement of a wire within a coronary vessel may cause damage or perforation to the artery, and ultimately cause severe complications up to and including a risk of death. As a result, invasive physiological testing is performed predominantly by tertiary referral centres and, despite being guideline directed for the investigation of coronary microvascular dysfunction in patients with ischaemia and no obstructive coronary arteries, is often underutilised.

[0007] Wire-free approaches have been investigated, both at point of care during an angiogram and retrospectively. Current platforms are typically stenosis-centric and focus on the significance of obstructing atherosclerosis lesions within the coronary artery. Existing software relies on computational fluid dynamics, which has theoretical and computational limits in reference to applicability to coronary arteries, such as the behaviour of blood as a fluid, and reliance on accurate three-dimensional reconstruction of the vessel. At the time of writing, there is no clinically indicated platform to assess the coronary microvasculature in patients with non-obstructive coronary arteries, or following revascularisation.

[0008] The present invention was arrived at in light of the above considerations.

[0009] SUMMARY

[0010] In a first aspect, embodiments of the present invention provide a computer-implemented method of estimating one or more cardiovascular parameters of a patient, the method comprising steps of: receiving: a flow value, which is indicative of the time required for contrast injected into the patient to reach a predefined distal landmark, and a blood pressure value for the patient; predicting a first value by providing the flow value as a predictor to a first machine-learning model; predicting a second value by providing the blood pressure value as a predictor to a second machine-learning model; and estimating the one or more cardiovascular parameters using the predicted first value and the predicted second value.

[0011] Such a method has been found by the inventors to provide useful estimations of the one or more cardiovascular parameters, and without requiring invasive measurement techniques.

[0012] The flow value may be a TIMI count, indictive of the number of frames taken for the contrast injected into the patient to reach a predefined distal landmark. The flow value may further include a corresponding shutter speed. The TIMI count may be derived from an angiogram, for example a coronary angiogram, using a reference point of the pressure wire sensor as the end frame. The reference point may have been positioned so that the sensor was placed in the distal two thirds of the vessel, or >6 cm from the ostium, whichever is greater.

[0013] The blood pressure value may be a resting proximal aortic pressure Pa. It may be a measured arterial pressure, for example taken during an angiogram. The resting proximal aortic pressure may be a calculated arterial pressure, by taking an average of the systolic and diastolic blood pressures. The resting proximal aortic pressure may be an invasively transduced calculated arterial pressure, by taking an average of the systolic and diastolic blood pressures, or non-invasively obtained by sphygmomanometer, by taking an average of the systolic and diastolic blood pressures. The blood pressure value may be a mean arterial pressure.

[0014] The first value may be a resting coronary transit time, Resting Tmn. The second value may be a hyperaemic distal coronary pressure, Pd. The first machine-learning model may be a linear regression model (e.g., a feature scaled linear regression model), trained on data comprising pairs of flow values and resting coronary transit times, so as to predict the first value when provided with the flow value.

[0015] The second machine-learning model may be a multi-layer perceptron neural network regression model, trained on data comprising: hyperaemic distal coronary pressures and resting proximal aortic pressures, so as to predict the second value when provided with a resting proximal aortic pressure. The second machine-learning model may be a polynomial regression model or a random forest regression model. The second machine-learning model may also utilise a fractional flow reserve value (FFR) provided by the user in it’s prediction.

[0016] The method may further comprise a step of predicting a third value, by providing the flow value and the first value to a third machine-learning model. The third value may be a hyperaemic transit time, Hyperaemic Tmn. The third machine-learning model may be a multi-layer perceptron neural network regression model, trained on data comprising: resting coronary transit times and flow values, so as to predict the third value when provided with the flow value and the first value. The third machinelearning model may take, as further inputs, any one or more of: vessel studied, sex of patient, and coronary dominance. The vessel studied may be the left anterior descending coronary artery and branches thereof, left circumflex artery and branches thereof, or right coronary artery and branches thereof. The coronary dominance may be right coronary dominant, left coronary dominant, or codominant.

[0017] When the third machine-learning model incorporated these further inputs, the explained variance was found to increase from approximately 0.33 to 0.49.

[0018] The method may further comprise receiving a selection of the coronary artery. The models used may be selected on the basis of the received selected coronary artery. The selected coronary artery may be provided as a further predictor to one or more of the machine-learning models.

[0019] The method may further comprise receiving a fractional flow reserve (FFR) value for the patient, and the received FFR may be provided as a further predictor to one or more of the machine-learning models.

[0020] The one or more cardiovascular parameters may include an index of microvascular resistance, IMR, and / or a coronary flow reserve, CFR. The index of microvascular resistance may be estimated using the second value and the third value. The IMR may be estimated using the equation IMR = Pdx Hyperaemic Tmn, where Pdis the predicted hyperaemic distal coronary pressure and Hyperaemic Tmnis the predicted hyperaemic transit time. The coronary flow reserve may be estimated using the first value and the third value. The coronary flow reserve may be estimated using the equation CFR = HyRpeesrtainemTic Tmn Receiving the flow value may include processing an angiogram cine to calculate the flow value.

[0021] Calculating the flow value may include applying an edge detection algorithm to the angiogram cine.

[0022] In a second aspect, embodiments of the invention provide a computer, comprising a processor and memory, the memory containing machine-executable instructions which, when executed on the processor, cause the processor to perform the method of the first aspect optionally including any one, or any combination insofar as they are compatible, of the optional features set out with reference thereto.

[0023] In a third aspect, embodiments of the invention provide a non-transitory computer-readable storage medium containing machine-executable instructions which, when executed on a processor, cause the processor to perform the method of the first aspect optionally including any one, or any combination insofar as they are compatible, of the optional features set out with reference thereto.

[0024] In a fourth aspect, embodiments of the invention provide a method of training a first machine-learning model to predict a resting coronary transit time, by providing a training data set comprising pairs of flow values and resting coronary transit times, and training the model on training data set. The first machine-learning model may include any one, or any combination insofar as they are compatible, of the optional features set out with reference to any preceding aspect.

[0025] In a fifth aspect, embodiments of the invention provide a method of training a second machinelearning model to predict a hyperaemic distal coronary pressure, by providing a training set comprising: hyperaemic distal coronary pressures and resting proximal aortic pressures. The second machine-learning model may include any one, or any combination insofar as they are compatible, of the optional features set out with reference to any preceding aspect.

[0026] In a sixth aspect, embodiments of the invention provide a method of training a third machine-learning model to predict a hyperaemic transit time, by providing a training set comprising: resting coronary transit times and flow values. The third machine-learning model may include any one, or any combination insofar as they are compatible, of the optional features set out with reference to any preceding aspect.

[0027] The invention includes the combination of the aspects and preferred features described except where such a combination is clearly impermissible or expressly avoided.

[0028] Further aspects of the present invention provide: a computer program comprising code which, when run on a computer, causes the computer to perform the method of the first aspect; a computer readable medium storing a computer program comprising code which, when run on a computer, causes the computer to perform the method of the first aspect; and a computer system programmed to perform the method of the first aspect.

[0029] BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Fig. 1 shows a method according to the present invention;

[0031] Fig. 2 shows a computer according to the present invention;

[0032] Fig. 3 shows a system;

[0033] Fig. 4 shows four Bland-Altman plots and

[0034] Fig. 5 shows an ROC curve for the classification of IMR.

[0035] DETAILED DESCRIPTION

[0036] Aspects and embodiments of the present invention will now be discussed with reference to the accompanying figures. Further aspects and embodiments will be apparent to those skilled in the art.

[0037] Methods

[0038] A dataset was constructed consisting of 300 independent records from patients who had undergone invasive coronary thermodilution physiology assessment using a dedicated wire. The records included values of resting proximal aortic pressure (Pu) in mmHg, Fractional Flow Reserve (FFR), TIMI frame count, shutter speed (frame / sec), invasive hyperaemic distal coronary pressure (Pd) in mmHg, invasive index of microvascular resistance (IMR), invasive hyperaemic transit time in seconds, invasive resting transit time in seconds, and the coronary flow reserve. The records were identified from existing studies from the School of Cardiovascular & Metabolic Health of the University of Glasgow, and included patients with angiographically unobstructed coronary arteries. A blinded assessment of uncorrected TIMI frame count was undertaken on coronary angiograms prior to the induction of hyperaemia, using the reference point of the pressure wire sensor as the end frame. All angiograms were acquired at 15 frames per second using a GE Healthcare Innova angiography system at The Golden Jubilee National University Hospital, Clydebank, UK, between 2018 and 2023.

[0039] The data was retrospectively assessed, and a feature scaled linear regression model was developed to predict resting coronary transit times from TIMI frame counts. Linear regression is a statistical method that models the relationship between a dependent variable and one or more independent variables. Herein, linear regression was used to predict coronary transit times from TIMI frame counts. A multi-layer perceptron (MLP) is a type of artificial neural network that uses multiple layers of nodes (neurons) to make predictions. The MLPRegressor model, a specific implementation of a MLP, was used to predict non-linear parameters including hyperaemic response to derive distal coronal pressure and coronary transit time. The MLPRegressor model is well-suited to this task, as it can model complex, non-linear relationships between inputs and outputs, which his often the case for physiological parameters. The parameters of the model, including the weight matrices and bias vectors, determine the strength of the connections between the neurons. The behaviour of the model can be presented in simplified mathematical notation as:

[0040] Predicted value = f(Wr■ f(W2■ ... (Wn■ X + b^ + b2^) ... ) + bn~)

[0041] Where () represents the activation function used by the model, W represents the weight matrices, X represents the input features, and b represents the bias vectors. The subscripts (1,2, ...,ri) indicate the different layers of the network.

[0042] The MLPRegressor model parameters were optimised using GridSearchCV, a module in Scikit-learn that performs exhaustive searching over specific parameter values for an estimator.

[0043] The 300 records identified above each included: a resting proximal aortic pressure (mmHg); a TIMI frame count; a hyperaemic distal coronary pressure (mmHg); a hyperaemic transit time (s); and a resting transit time(s). Three models were trained on some or all of the data: (i) a linear regression model which predicts resting coronary transit time in seconds from a TIMI frame count; (ii) a first MLPR model which predicts hyperaemic distal coronary pressure from resting proximal aortic pressure; and (iii) a second MLPR model which predicts hyperaemic transit time in seconds from: a predicted resting coronary transit time; training data including resting transit times; and TIMI frame counts.

[0044] The performance of the trained models was evaluated using metrics such as R2(coefficient of determination), mean absolute error (MAE), and root mean square error (RMSE). These metrics provide a quantitative measure of the models’ predictive performance, where higher R2and lower MAE and RMSE values indicate better performance.

[0045] Cross-validation was used to establish the optimal hyperparameters for the neural network models. This technique involves partitioning the data into subsets, training the model on one subset (the training set) and validating the model on another subset (the validation set). The process was repeated multiple times, with different partitions each time, to ensure the model’s generalisability and prevent overfitting. The linear regression model predicted resting coronary transit time in seconds from TIMI frame counts with an R2of 0.33, indicating that 33% of the variance in the transit time could be explained by the model. The model had an MAE of 0.32 and an RMSE of 0.39. The MSE was 0.15. The linear regression model used had a coefficient of 1 .16 and an intercept of 0.69. In some examples, the linear regression model also took, as inputs, the vessel studied, the sex of the patient, and the coronary dominance. This example has been tested on fifty complete datasets.

[0046] The first MLPR model was found to predict hyperaemic distal coronary pressure (Pd) in millimeters of mercury from resting proximal aortic pressure (Pu) with an R2of 0.75, indicating that 75% of the variance in Pdcould be explained by the model. The model had an MAE of 4.26 and an RMSE of 5.70, indicating relatively small average errors in prediction. The MSE was 32.48, which is the square of the RMSE.

[0047] The second MLPR model was found to predict hyperaemic transit time in seconds from predicted resting coronary transit time, training data resting transit times, and TIMI frame counts with an R2of 0.45, indicating that 45% of the variance in transit time could be explained by the model. The model had an MAE of 0.09 and an RMSE of 0.12. The MSE was 0.01.

[0048] In addition to the second MLPR model, a polynomial regression model and a random forest regression model were also used to predict hyperaemic transit time. The polynomial regression model had an R2of 0.45 and an MSE of 0.01 , while the random forest regression model had an R2of 0.37 and an MSE of 0.02.

[0049] Two cardiovascular parameters were derived using the combination of the models (‘the combined model’): the index of microvascular resistance (IMR) and the coronary flow reserve (CFR). These parameters were calculated using the predicted values from the previous models. CFR is calculated using the equation CFR =Restina Tmnand IMR is calculated using the equation IMR =

[0050] Hyperaemic Tmn

[0051] Pdx Hyperaemic Tmn.

[0052] The combined model predicted CFR with an R2of 0.68, which indicates that 68% of the variance in CFR could be explained using the model. The model had an MAE of 0.23 and an RMSE of 0.31 , indicating relatively small average errors in the prediction. The MSE was 0.10.

[0053] The combined model predicted IMR with an R2of 0.99. This indicates that 99% of the variance in the IMR could be explained by the model. The model had an MAE of 0.90 and an RMSE of 2.34, the

[0054] MSE was 5.49. Fig. 1 shows a method 100. In a first step, 102, a flow value and blood pressure value are received. Receiving the flow value may include processing an angiogram cine to calculate the flow value, by applying an edge detection algorithm to the angiogram cine. Receiving the flow value may include a user providing the flow value through a data Receiving the blood pressure value may include receiving an input from a user indicative of the blood pressure value, or might include receiving the blood pressure value from (for example) a networked angiography system.

[0055] Next, steps 104 and 106 are performed either in series (in either order) or in parallel. Step 104 includes predicting a first value by providing the flow value as a predictor to a first machine-learning model such as the linear regression model identified above. Step 106 includes predicting a second value by providing the blood pressure value to a second machine-learning model such as the first MLPR model discussed above. After both predictions, the method moves to step 108, where the one or more cardiovascular parameters are estimated using the predicted first and second values. Step 108 can include further sub-steps, such as providing the first value and the flow value to a third machine-learning model such as the second MLPR model discussed above. This results in prediction of a third value, which can be used to calculate clinical parameters such as IMR and CFR respectively.

[0056] Fig. 2 shows a computer 200 according to the present invention. The computer 200 includes a processor 202, which is connect to each of: volatile storage 204 (itself connected to long-term storage 206); an input / output interface 208; and a network interface 210. The long-term storage 206 contains machine-executable instructions which, when loaded into volatile storage 204 (e.g., RAM) and executed on the processor 202 cause the processor to perform the methods disclosed herein. Specifically, the step of receiving the various input values (flow value, blood pressure value, FFR, and selected coronary artery) can be performed using one or more of the input / output interface 208 and the network interface 210. For example, a user may enter the various values via a keyboard or other human-machine interface and / or may provide various values from a further device (not shown) over a computer network and via the network interface 210.

[0057] In other examples, the computer can include or be connected to (e.g., via the network interface and / or input / output interface) one or more medical devices configured to obtain the flow value and / or the blood pressure value. For example, the computer may include or be connected to a digital angiogram and / or a blood pressure monitor. It may be configured to collect the flow value and blood pressure value autonomously or semi-autonomously (e.g., in collaboration with a clinician who places the catheter and / or blood pressure cuff). The computer may then automatically, after collecting the respective values, apply the trained models to arrive at the one or more cardiovascular parameters of the patient. Fig. 3 shows a system including the computer 200 above. Here, the computer 200 is connected via network 220 (e.g., a wide area network such as the internet) to a further computer 230. In this example, the computer 230 may be a local terminal (e.g., such as used by a clinician) to input the various values. These values are then transmitted over network 220 to the computer 200. The computer 200 then calculates the clinical parameters discussed above, and provides the resulting values back over the network 220 to the computer 230.

[0058] Fig. 4 shows four Bland-Altman plots for the mean hyperaemic distal coronary pressure (Pd) (upper left), mean resting transit time (upper right), mean hyperaemic transit time (lower left), and mean index of microcirculatory resistance (IMR), (lower right). Fig. 5 shows an ROC curve for the IMR classification method as a whole.

[0059] The systems and methods of the above embodiments may be implemented in a computer system (in particular in computer hardware or in computer software) in addition to the structural components and user interactions described.

[0060] The term “computer system” includes the hardware, software and data storage devices for embodying a system or carrying out a method according to the above described embodiments. For example, a computer system may comprise a central processing unit (CPU), input means, output means and data storage. The computer system may have a monitor to provide a visual output display. The data storage may comprise RAM, disk drives or other computer readable media. The computer system may include a plurality of computing devices connected by a network and able to communicate with each other over that network.

[0061] The methods of the above embodiments may be provided as computer programs or as computer program products or computer readable media carrying a computer program which is arranged, when run on a computer, to perform the method(s) described above.

[0062] The term “computer readable media” includes, without limitation, any non-transitory medium or media which can be read and accessed directly by a computer or computer system. The media can include, but are not limited to, magnetic storage media such as floppy discs, hard disc storage media and magnetic tape; optical storage media such as optical discs or CD-ROMs; electrical storage media such as memory, including RAM, ROM and flash memory; and hybrids and combinations of the above such as magnetic / optical storage media.

[0063] While the disclosure has been described in conjunction with the exemplary embodiments described above, many equivalent modifications and variations will be apparent to those skilled in the art when given this disclosure. Accordingly, the exemplary embodiments of the disclosure set forth above are considered to be illustrative and not limiting. Various changes to the described embodiments may be made without departing from the spirit and scope of the disclosure.

[0064] In particular, although the methods of the above embodiments have been described as being implemented on the systems of the embodiments described, the methods and systems of the present disclosure need not be implemented in conjunction with each other, but can be implemented on alternative systems or using alternative methods respectively.

[0065] The features disclosed in the description, or in the following claims, or in the accompanying drawings, expressed in their specific forms or in terms of a means for performing the disclosed function, or a method or process for obtaining the disclosed results, as appropriate, may, separately, or in any combination of such features, be utilised for realising the disclosure in diverse forms thereof.

[0066] While the disclosure has been described in conjunction with the exemplary embodiments described above, many equivalent modifications and variations will be apparent to those skilled in the art when given this disclosure. Accordingly, the exemplary embodiments of the disclosure set forth above are considered to be illustrative and not limiting. Various changes to the described embodiments may be made without departing from the spirit and scope of the disclosure.

[0067] For the avoidance of any doubt, any theoretical explanations provided herein are provided for the purposes of improving the understanding of a reader. The inventors do not wish to be bound by any of these theoretical explanations.

[0068] Any section headings used herein are for organizational purposes only and are not to be construed as limiting the subject matter described.

[0069] Throughout this specification, including the claims which follow, unless the context requires otherwise, the word “comprise” and “include”, and variations such as “comprises”, “comprising”, and “including” will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps.

[0070] It must be noted that, as used in the specification and the appended claims, the singular forms “a,” “an,” and “the” include plural referents unless the context clearly dictates otherwise. Ranges may be expressed herein as from “about” one particular value, and / or to “about” another particular value. When such a range is expressed, another embodiment includes from the one particular value and / or to the other particular value. Similarly, when values are expressed as approximations, by the use of the antecedent “about,” it will be understood that the particular value forms another embodiment. The term “about” in relation to a numerical value is optional and means for example + / - 10%.

Claims

CLAIMS1 . A computer-implemented method of estimating one or more cardiovascular parameters of a patient, comprising steps of: receiving: a flow value, which is indicative of the time required for contrast injected into the patient to reach a predefined distal landmark, and a blood pressure value for the patient; predicting a first value by providing the flow value as a predictor to a first machine-learning model; predicting a second value by providing the blood pressure value as a predictor to a second machine-learning model; and estimating the one or more cardiovascular parameters using the predicted first value and the predicted second value.

2. The computer-implemented method of claim 1 , wherein the first value is a resting coronary transit time, Resting Tmn.

3. The computer-implemented method of claim 1 or claim 2, wherein the second value is a hyperaemic distal coronary pressure, Pd.

4. The computer-implemented method of any preceding claim, wherein the first machinelearning model is a linear regression model, trained on data comprising pairs of flow values and resting coronary transit times, so as to predict the first value when provided with the flow value.

5. The computer-implemented method of any preceding claim, wherein the second machinelearning model is a multi-layer perceptron neural network regression model, trained on data comprising: hyperaemic distal coronary pressures and resting proximal aortic pressures, so as to predict the second value when provided with the resting proximal aortic pressure.

6. The computer-implemented method of any preceding claim, further comprising a step of predicting a third value, by providing the flow value and the first value to a third machine-learning model.

7. The computer-implemented method of claim 6, wherein the third value is a hyperaemic transit time, Hyperaemic Tmn.

8. The computer-implemented method of claim 6 or 7, wherein the third machine-learning model is a multi-layer perceptron neural network regression model, trained on data comprising: resting coronary transit times and flow values, so as to predict the third value when provided with the flow value and the first value.

9. The computer-implemented method of any preceding claim, wherein the one or more cardiovascular parameters including an index of microvascular resistance, I MR, and / or a coronary flow reserve, CFR.

10. The computer-implemented method of claim 9 as dependent on any of claims 6 - 8, wherein the index of microvascular resistance is estimated using the second value and the third value.11 . The computer-implemented method of claim 10, wherein the IMR is estimated using the equation IMR = Pdx Hyperaemic Tmn, where Pdis the predicted hyperaemic distal coronary pressure and Hyperaemic Tmnis the predicted hyperaemic transit time.

12. The computer-implemented method of any of claims 6 - 11 , wherein the coronary flow reserve is estimated using the first value and the third value.

13. The computer-implemented method of claim 12, wherein the coronary flow reserve is estimated using the equation CFR =Restma Tmn.Hyperaemic Tmn14. The computer-implemented method of any preceding claim, wherein receiving the flow value includes processing an angiogram cine to calculate the flow value.

15. The computer-implemented method of claim 14, wherein calculating the flow value includes applying an edge detection algorithm to the angiogram cine.

16. A computer, comprising a processor and memory, the memory containing machineexecutable instructions which, when executed on the processor, cause the processor to perform the method of any preceding claim.

17. A non-transitory computer-readable storage medium containing machine executable instructions which, when executed on a processor, cause the processor to perform the method of any of claims 1 - 15.

Images