Computer-assisted blood flow imaging

Abstract

This specification describes blood flow imaging based on area under a time-contrast effect curve predicted by a computer learning model. Image data, including a plurality of corresponding images capturing at least a portion of one or both of the contrast agent's increasing and decreasing phases in the target cardiovascular system, is input to a machine learning model to predict the area under the time-contrast effect curve of the contrast agent in the target cardiovascular system, where the predicted area under the time-contrast effect curve represents the sum of the time products of contrast agent concentrations in the target cardiovascular system. One example is a computer implementation method for blood flow imaging, comprising: obtaining image data including a plurality of corresponding images capturing at least a portion of one or both of the contrast agent's increasing and decreasing phases in the target cardiovascular system; providing the image data to a machine learning model to predict the area under the time-contrast effect curve of the contrast agent in the target cardiovascular system; selecting a target region in the target cardiovascular system in the image data; and determining blood flow characteristics through the target region based on the area under the time-contrast effect curve. A system for performing the method and a non-temporal computer-readable medium are also described.

Description

[Technical Field]

[0001] Background of the Invention Field of Invention This invention relates to dynamic imaging of flow, and more particularly to the evaluation of blood flow characteristics of a subject based on dynamic imaging of the flow of contrast agent through blood vessels or cardiac structures. [Background technology]

[0002] Explanation of conventional technology Dynamic contrast-enhanced (DCE) computed tomography (CT) is used to assess intravascular blood flow and blood flow pressure, as described, for example, in the international PCT application PCT / CA2019 / 050668 filed on 16 May 2019 by the same applicant (published as international publication 2019 / 218076 on 21 November 2019), which is incorporated herein by reference. In this disclosure, the indicator dilution principle is used for blood flow assessment derived from DCE CT, which represents the average value over many imaging scans of many seconds, often exceeding 10 seconds, and therefore does not have sufficient temporal resolution to achieve 4D flow imaging, i.e., sufficient temporal resolution to track blood flow characteristics in selected image voxels with very fine temporal resolution and to calculate changes in flow velocity at time intervals of less than 1 second, for example.

[0003] Improvements regarding finer temporal resolution are described in the International PCT Application PCT / CA2021 / 051189, filed on 26 August 2021, by the same applicant (published on 3 March 2022 as International Publication No. 2022 / 040806), which is incorporated herein by reference. In this disclosure, the indicator dilution principle and the Reynolds transport theorem are used to substantially improve the temporal resolution of non-invasive blood flow assessment using DCE CT imaging data.

[0004] These two disclosures improve the existing state of CT imaging and introduce novel extraction and implementation of parameters from time-contrast-enhancing effect curves in a target region, including, for example, the area under the curve, the rate of change over time of tracer mass in the target region, or the density of tracers in blood in the target region.

[0005] Because diagnostic imaging is a very active area of ​​clinical workflow, further improvements, such as improved computer efficiency, reduced radiation dose, or improved accuracy, are welcomed by healthcare professionals. [Overview of the Initiative] [Problems that the invention aims to solve]

[0006] Therefore, there is a continuous need for alternative methods and systems for evaluating the blood vessels of a subject based on blood flow imaging. [Means for solving the problem]

[0007] Summary of the Invention In one embodiment, a computer implementation method for blood flow imaging is provided, and this method is Obtain CT or MRI image data including multiple corresponding images that capture at least a portion of one or both of the contrast agent increase and decrease phases in the target cardiovascular system. Extracting the first image feature of the measured time-contrast effect curve from CT or MRI image data. The method involves providing a machine learning model with a first image feature and at least one non-image feature to generate a predicted value for the area under a simulated time-dependent contrast effect curve of a contrast agent in the target cardiovascular system, wherein the predicted value simulates and generates a second set of image acquisition parameters different from a first set of image acquisition parameters used to acquire CT or MRI image data. Convert the predicted area under the simulated time-time contrast effect curve into the sum of the time-time contrast agent concentration products in the target cardiovascular system. Determining the blood flow characteristics of the target cardiovascular system based on the ratio of the mass of the contrast agent in the target cardiovascular system to the sum of the contrast agent concentration-time products in the target cardiovascular system. Includes.

[0008] A further relevant example of the method further includes including baseline data points extracted from CT or MRI image data in the measured time-contrast effect curve, wherein the CT or MRI image data includes at least one image capturing the target cardiovascular system prior to contrast agent entry in order to provide the baseline data points.

[0009] In a further relevant example of the method, a machine learning model is trained on image features of a measured time-contrast effect curve and at least one non-image feature as training inputs, and associated with ground truth values ​​of the expected area under the simulated time-contrast effect curve.

[0010] In a further relevant example of the method, the image feature is at least a partial time-contrast effect curve, and extracting the image feature involves generating at least a partial time-contrast effect curve of a contrast agent based on the image data, the at least partial time-contrast effect curve having at least an upward or downward slope. The at least partial time-contrast effect curve may be any preferred truncated or complete time-contrast effect curve.

[0011] In a further relevant example of the method, the image feature is a measurement of the area under at least a partial time-contrast effect curve, and extracting the image feature includes generating, based on the image data, at least a partial time-contrast effect curve of the contrast agent, the at least partial time-contrast effect curve having at least an upward or downward slope, and calculating the area under the at least partial time-contrast effect curve. The at least partial time-contrast effect curve may be any preferred truncated or complete time-contrast effect curve.

[0012] In a further related example of the method, the method further includes providing a second image feature based on a contrast effect of signal intensity, a thickness of a wall of the cardiovascular system of the subject, a size of the cardiovascular system of the subject, a diameter of the cardiovascular system of the subject, a morphology of the cardiovascular system of the subject, a position of a sampling site in the cardiovascular system of the subject, or a degree of stenosis in the cardiovascular system of the subject.

[0013] In a further related example of the method, the non-image feature is based on age, gender, weight, heart rate, blood pressure, X-ray tube voltage, X-ray tube current, gradient pulse sequence, medical condition, contrast agent injection rate, contrast agent volume, or contrast agent concentration.

[0014] In a further related example of the method, the first set of image acquisition parameters is different from a simulated second set of image acquisition parameters by at least one parameter selected from the group consisting of a scan projection axis, an anatomical position of the scan, a congested or resting state of the subject, a duration of the scan, X-ray tube voltage, X-ray tube current, gradient pulse sequence, contrast agent injection rate, contrast agent volume, or contrast agent concentration.

[0015] In a further related example of the method, determining the blood flow characteristics comprises determining a temporal rate of change of contrast agent mass in the cardiovascular system of the subject based on a temporal rate of change of signal enhancement measured from at least a partial temporal contrast effect curve, a coefficient for converting signal enhancement to contrast agent concentration, a predetermined volume value of the injected contrast agent, and a temporal value of the duration for which the signal intensity is higher than a predetermined baseline threshold level; determining the density of the contrast agent in the cardiovascular system of the subject based on the ratio defined in claim 1, the temporal value of the duration for which the signal intensity is higher than a predetermined baseline threshold level, and the mass of the contrast agent in the cardiovascular system of the subject; determining an area measured from the image data in the cardiovascular system of the subject and determining the absolute flow velocity using the Reynolds transport theorem.

[0016] In a further related example of the method, determining the blood flow characteristics includes determining the flow pressure by applying Bernoulli's equation.

[0017] In another aspect, a computer-implemented method for blood flow imaging based on predicting the area under the time-contrast effect curve is provided, the method comprising: obtaining image data including a plurality of corresponding images capturing at least a portion of one or both of the increasing and decreasing phases of a contrast agent in a subject's cardiovascular system; providing the image data to a machine learning model to predict the area under the time-contrast effect curve of the contrast agent within the subject's cardiovascular system; wherein the predicted area under the time-contrast effect curve represents the sum of the contrast agent concentration-time integrals within the subject's cardiovascular system.

[0018] In a further aspect, a computer-implemented method for blood flow imaging is provided, the method comprising: obtaining image data including a plurality of corresponding images capturing at least a portion of one or both of the increasing and decreasing phases of a contrast agent in a subject's cardiovascular system; providing the image data to a machine learning model to predict the area under the time-contrast effect curve of the contrast agent within the subject's cardiovascular system; selecting a region of interest within the subject's cardiovascular system in the image data; determining blood flow characteristics through the region of interest based on the area under the time-contrast effect curve. comprising.

[0019] In other aspects, a system and a non-transitory computer-readable medium for performing the method are also provided.

Brief Description of the Drawings

[0020] Brief Description of the Drawings [Figure 1] A schematic diagram of a blood flow imaging system is shown. [Figure 2] A flowchart of a blood flow imaging method is shown. [Figure 3]Figure 2 shows a flowchart of the pre-scan preparation steps in the imaging method. [Figure 4] Figure 2 shows a flowchart of the scan data acquisition process in the imaging method described. [Figure 5] Figure 2 shows a flowchart illustrating an example of predicting the area under the time-contrast effect curve (AUC) in the imaging method shown. [Figure 6A] Figure 2 shows a flowchart illustrating an example of determining blood flow characteristics based on predicted AUC in the imaging method shown. [Figure 6B] Figure 2 shows a flowchart of another example of determining blood flow characteristics based on predicted AUC in the imaging method shown. [Figure 7A] This diagram shows a schematic representation of a method for studying fluid movement. [Figure 7B] This diagram shows a schematic representation of a method for studying fluid movement. [Figure 8A] This diagram shows a schematic representation of the analysis of the inspection volume defined by the voxels in the flow imaging data. It also shows a schematic representation of the inspection volume and its inspection surface. [Figure 8B] This shows a schematic diagram of the analysis of the inspection volume defined by the voxels in the flow imaging data. It also shows a schematic diagram of fluid movement relative to the inspection volume. [Figure 9A] This diagram shows a schematic representation of the dynamic image acquisition and reconstruction method for obtaining ΔHU / Δt, which forms the basis for absolute or relative flow velocity evaluation. It illustrates image acquisition and reconstruction at the same cardiac phase (e.g., 75%RR interval or diastole) across multiple time points. [Figure 9B] This diagram shows a schematic representation of the dynamic image acquisition and reconstruction method for obtaining ΔHU / Δt, which forms the basis for absolute or relative flow velocity evaluation. It illustrates image acquisition covering the complete cardiac cycle (systole and diastole). Multiple image sets corresponding to different cardiac phases are reconstructed. [Figure 10] A schematic diagram illustrating an example of a cardiac CT scan workflow is shown. [Figure 11]A schematic diagram shows an example of using machine learning or deep learning to derive blood flow variables from a pseudo-truncate dynamic perfusion scan. [Figure 12] A schematic diagram shows an example of using machine learning or deep learning to derive blood flow variables from truncated rest dynamic perfusion scans. [Figure 13] A schematic diagram shows an example of using machine learning or deep learning to derive blood flow variables from truncated stress dynamic perfusion scans. [Figure 14] A schematic diagram shows an example of generating training image features for training an ML or DL ​​algorithm to learn and predict AUC values ​​for future truncated perfusion image data. Top: Training path. Bottom: Application path. [Figure 15] A schematic diagram shows an example of image feature extraction involving the incorporation of arbitrarily selected non-image features in the coronary arteries for training ML or DL ​​algorithms. [Figure 16] A schematic diagram shows an example of image feature extraction in myocardium, involving the incorporation of arbitrarily selected non-image features, for training ML or DL ​​algorithms. [Figure 17] A simplified diagram is shown to illustrate an example of a deep neural network (convolutional neural network) architecture. In this diagram, only one convolutional filter is shown, and only one feature map and a pooled feature map are obtained from each round of convolution and pooling. The grid of each feature map and the pooled feature map does not represent the actual dimensions of these maps, but is only there to demonstrate the reduced dimensions of these maps after each operation. If the input image has M × M pixels and the convolutional filter has C × C pixels, then the feature map should have (M - C + 1) × (M - C + 1) pixels. [Figure 18] Figure 17 shows an analog example of a deep neural network, in which three convolutional filters are applied. [Figure 19]A schematic diagram of an example deep neural network architecture for processing dynamic perfusion image sets is shown. Only a portion of the architecture is shown for visual simplification. [Figure 20A] The post-stenosis AUC, normalized to the pre-stenosis AUC, is shown as a function of luminal stenosis in five patients with coronary artery disease (CAD). [Figure 20B] This shows the sampling ROI of the pre-stenosis and post-stenosis segments in a stenotic coronary artery. [Figure 21] This shows the stress AUC normalized to rest AUC in the pre- and post-stenotic coronary artery segments of five patients with CAD. [Figure 22] The graph shows coronary time-enhancement effect curves obtained from two different patients with different heart rates. White circles represent the original data, and solid lines represent the corresponding approximation curves. The predicted baseline and peak enhancement effects for each approximation curve are indicated by blue dashed horizontal and vertical lines, respectively. [Figure 23] The time-enhancement effect curves measured from the left anterior descending (LAD) artery (ascending) and myocardium perfused by the LAD (descending) in the same patient are shown. [Figure 24A] Figure 24B shows the increased thickness of the myocardial wall (dashed line) when comparing the maximum vasodilation stress state (Figure 24A) with the resting state (Figure 24B). The right coronary artery (RCA, indicated by the arrow) also appeared to be more dilated under stress. [Figure 24B] Figure 24B shows the increased thickness of the myocardial wall (dashed line) when comparing the maximum vasodilation stress state (Figure 24A) with the resting state (Figure 24B). The right coronary artery (RCA, indicated by the arrow) also appeared to be more dilated under stress. [Figure 24C] The corresponding time-dependent contrast-enhancing effect curves measured from RCA during stress and rest are shown. [Figure 25]The left figure shows the default orientation of two CT slices relative to the coronary artery. The red arrows indicate the direction of blood flow within the artery. The right figure shows that slice reshaping is necessary at one slice location to ensure that the slices are perpendicular to the direction of blood flow before quantitative blood flow evaluation. [Figure 26A] This shows a short-axis view of a contrast-enhanced cardiac image from the same patient. [Figure 26B] This shows an axial view of a contrast-enhanced cardiac image from the same patient. [Figure 26C] This shows a comparison of coronary time-enhancement effect curves sampled at the same spatial location in the left anterior descending artery in both short-axis and axial views. [Figure 27A] Figure 27C shows the stress-dynamic perfusion images, illustrating the time-enhancement effect curves of coronary artery stenosis sampled at the proximal and distal ends of the RCA. [Figure 27B] Figure 27D shows the coronary time-enhancement effect curve sampled at the same location across the entire RCA stenosis in the rest dynamic perfusion image. [Figure 27C] Please note that only the distal sampling location is shown. [Figure 27D] Please note that only the distal sampling location is shown. [Figure 28] This graph shows the AUC of stress time-to-angiography (TAI) versus rest time-to-angiography (TAI) curves, sampled from different right coronary arteries with varying degrees of stenosis (percentage of luminal stenosis). All data points shown in the graph are actual measured AUC values. [Figure 29]The following shows two sections of Python code. The upper section shows that three hidden layers (L1, L2, and L3) have been implemented in the neural network. The activation is not specified in the output layer (the last line of code in the upper section), which means that linear regression was used. By default, a(x) = x, where a is the activation function, x = w1b1 + w2b2 + ..., where w are the weights and b are the parameters. The code shown in the lower section was used to run the neural network layers defined above. [Figure 30A] This shows the predicted stress AUC value (represented as a cross) predicted by a trained machine learning model versus the actual stress AUC value measured from stress dynamic perfusion images. The training (circle) and test (square) AUC datasets used for the machine learning model are shown in the background for comparison. The substantial overlap between the cross and the training / test datasets indicates a good prediction of the stress AUC value made by the model. [Figure 30B] This shows the training and test scores and losses related to model training. [Figure 31] A schematic diagram of a proposed machine learning method for predicting the area under the curve (AUC) of a time-spectrum effect curve during stress (maximum congestion) without actually acquiring the curve is shown. The input AUC can be a coronary artery or myocardial time-spectrum effect curve. Similarly, the output AUC can be a coronary artery or myocardial time-spectrum effect curve, because these curves are closely related to each other. The coronary artery AUC can be used as input to predict the myocardial AUC, and vice versa. These curves can also be used simultaneously as input training data for a machine learning model. [Figure 32]A schematic diagram illustrates how a trained machine learning model can be integrated into existing image processing software to facilitate the prediction of blood flow parameters corresponding to maximum congestion. The asterisk (*) indicates that the algorithm is described in PCT / CA2019 / 050668 filed on 16 May 2019 or PCT / CA2021 / 051189 filed on 26 August 2021 by the same applicant. [Figure 33A] This shows a truncated perfusion dataset generated by removing data points after peak contrast enhancement. [Figure 33B] The AUC (Area of ​​Concentration) calculated for the generated, truncated dataset is shown, and this was used for machine learning. [Figure 34] This shows the AUC of the full time-enhancement effect curve sampled from dynamic perfusion images versus the AUC of the truncated time-enhancement effect curve generated from the full curves of different coronary arteries with varying degrees of stenosis. [Figure 35A] This shows the AUC (indicated by the crosshairs) of the full coronary time-enhancement effect curve predicted by a trained machine learning model, compared to the AUC of the full time-enhancement effect curve measured from a full dynamic perfusion scan. Training (circle) and test (square) datasets are provided as background. The substantial overlap between the crosshairs and the training / test datasets indicates that the trained model made accurate predictions about the full AUC. [Figure 35B] The training and test losses of the machine learning model are shown as a function of epochs (the number of times the entire training dataset has been completely traversed). The flat portion of the plot indicates that the training and test losses are low and close to each other, suggesting that the training data was not overfitted by the model. [Figure 35C] The training loss and test loss, as well as the training score and test score, returned by the model are shown. [Figure 36] The machine learning scheme discussed in Example 6 is shown below. [Figure 37]Figure 36 illustrates how trained machine learning, trained according to the scheme shown, can be used for clinical hemodynamic assessment. The asterisk (*) indicates that the algorithm is described in PCT / CA2019 / 050668 filed on May 16, 2019, or PCT / CA2021 / 051189 filed on August 26, 2021, by the same applicant. [Figure 38A] The measured minor axis AUC values ​​are plotted against the measured axial axis AUC values ​​(circular dots). [Figure 38B] The predicted short-axis AUC values ​​(crosshairs) are shown compared to the measured AUC values ​​used in training (circle) and testing (square). [Figure 38C] This shows the training / test loss and score related to model training. [Figure 39A] Feature-labeled contrast-enhanced cardiac CT images are shown in a 2D view. [Figure 39B] Shows the corresponding 3D view. [Figure 39C] The MATLAB code used to implement the three-dimensional U-NET architecture for segmenting the cardiac CT volumetric image set shown in Figures 39A and 39B is shown. The configuration used to train the U-NET is also shown. [Figure 40] This section demonstrates how the training results from deep learning models can be used as input labels to train the machine learning models described in Examples 5-7. [Modes for carrying out the invention]

[0021] Detailed description of preferred embodiments Referring to the drawings, a system and method for blood flow imaging will be described. The system and method offer advantages compared to current blood flow imaging techniques.

[0022] Figure 1 shows a computer-implemented imaging system 2 incorporating a computed tomography (CT) scanner 4. The CT scanner 4 may be any multi-row or multi-slice CT scanner, typically including a radiation source and radiation detector positioned within a gantry, and an adjustable, often motorized, support or table for maintaining the subject in a desired position (e.g., prone or supine) within an open central chamber formed within the gantry during the scanning procedure. The radiation source generates radiation across one or more predetermined sampling sites, targeting the blood vessels of interest in the subject, in synchronization with a contrast agent (also called a tracer) administered to the subject. The radiation detector often consists of a rotating detector panel that receives radiation across the subject at predetermined sampling sites and provides projection data (also called scan data) over a time range including the increasing and optionally decreasing phases of the contrast agent flowing through the blood vessels of interest.

[0023] The imaging system 2 includes a data acquisition component 6 incorporating a data acquisition scheme or data acquisition computer code for receiving, organizing, and storing projection data from the radiation detector of a CT scanner. The projection data is sent to an image reconstruction component 8 incorporating an image reconstruction computer code. The projection data can then be processed using the image reconstruction computer code to obtain image data containing multiple images of a predetermined sampling site over the increasing and optionally decreasing phases of contrast agent flowing through the target vessel. The image reconstruction computer code can be easily modified to accommodate any available CT imaging technique. The image data can then be processed by an image analysis component 10 incorporating an image analysis computer code for generating a time-contrast effect curve of the contrast agent signal from the image data. The time-contrast effect curve data can then be processed by a blood flow estimation component 12 incorporating a blood flow estimation computer code to determine the blood flow characteristics of the target vessel from the time-contrast effect curve data. The imaging system 2 is controlled by a computer 16, and data and operation commands are communicated via a bus 14. The imaging system 2 may include, as necessary, any additional components for evaluating the target blood vessels, including a multiplexer, a digital-to-analog conversion board, a microcontroller, a physical computer interface device, input / output devices, a display device, and a data storage device. The imaging system 2 may include dedicated controllers for various components of the CT scanner 4, such as a radiation source controller that provides power and timing signals to control the radiation source, a gantry controller that provides power and timing signals to the gantry motors that control the rotation of the gantry and thereby control the rotation of the radiation source and detector, and a table controller that provides power and timing signals to the table motors to control the position of the table, thereby controlling the position of the subject in the gantry by moving the subject along the z-axis through the opening of the gantry that communicates with the internal open chamber of the gantry.Imaging system 2 is shown with a CT scanner as an example for illustrative purposes only, and the system may be modified to include other imaging modalities, such as non-CT X-ray imaging or MRI.

[0024] Figure 2 shows a computer implementation method 20 for blood flow imaging. Method 20 includes pre-scan preparation 30 and positioning of a subject for CT scanning of a desired sampling site. The subject is prepared and placed in the CT scanner, then a contrast agent solution is injected into the subject (40), and CT scanning (50) is synchronized with the injection of the contrast agent solution to acquire projection data (also called scan data) over a time range including the flow of contrast agent through the blood vessels of the sampling site. The projection data is processed to reconstruct image data from the projection data (60). The image data is analyzed to predict the area under the time-contrast effect curve of contrast agent signal parameters, such as contrast agent signal intensity, extracted from the image data (70). Blood flow values ​​are calculated based on the time-contrast effect curve (80).

[0025] Figure 3 shows an example of pre-scan preparation 30 of a subject for CT scanning. Pre-scan preparation 30 includes identifying a target area within the subject (32). For example, the target area may be a part of a blood vessel to be evaluated for blood flow within the vessel. Once the target area is defined, a sampling site for CT scan slices is identified within or near the target area (34). Based on the determined sampling site, the subject is positioned within the CT scanner in a configuration that allows the radiation source of the CT scanner to direct radiation to the sampling site (36). Before scanning, the subject optionally holds their breath (38) and maintains breath-holding throughout the scanning. As a further option, a hyperemia state can be induced in the subject, for example, by administering a vasodilator to the subject.

[0026] Figure 4 shows an example of CT scanning 50 synchronized with contrast agent injection. Synchronized CT scanning 50 includes initiating a dynamic CT scan at a desired time based on the injection of contrast agent. Optionally, the CT scanning may be synchronized with an electrocardiogram (ECG), provided by, for example, prospective ECG-gated contrast-enhanced dynamic CT imaging. When ECG synchronization is performed, retrospective or prospective ECG synchronization may be used. The dynamic CT scan includes acquiring projection data at the sampling site before contrast agent entry to establish a baseline (54), acquiring projection data during the contrast agent increase phase at the sampling site (56), and acquiring projection data during the contrast agent decay phase at the sampling site (58). The increasing phase refers to the period following the initial entry of the contrast agent into the sampling site, during which the mass of the contrast agent at the sampling site increases over time, while the decreasing phase refers to the period before the contrast agent is substantially completely removed from the sampling site, during which the mass of the contrast agent at the sampling site decreases over time. The peak (maximum value) of the contrast agent mass at the sampling site occurs during the transition from the increasing phase to the decreasing phase. The time elapsed from the entry to the removal of the contrast agent at the sampling site can be called the transit time of the contrast agent. The duration of the CT scan is not limited by the requirement to capture the complete transit time of the contrast agent at the sampling site, provided that at least a portion of both the increasing and decreasing phases is captured.

[0027] Figure 5 shows an example of image analysis for predicting the area under the time-contrast effect curve using a machine learning model (70). Predicting the area under the time-contrast effect curve may involve identifying the target voxel or region in multiple corresponding images at the sampling site (72). Contrast agent signal data, e.g., contrast agent signal intensity, is extracted from each of the multiple corresponding images from the area defined by the target voxel or region (74). The area under the time-contrast effect curve is predicted by a computer-learned model 74 based on the contrast agent signal data for the increasing, decreasing, or both increasing and decreasing periods at the sampling site.

[0028] Figure 6A shows an example of estimating blood flow 80 in a target voxel or region based on a time-contrast effect curve. Blood flow estimation can be achieved by determining the flow velocity value based on the time-contrast effect curve. Determining the flow velocity value may include selecting a first time point and a second time point from the time-contrast effect curve (82). The rate of change of contrast effect (ΔHU / Δt) can be determined by subtracting the CT image data at the selected first and second time points (84). The rate of change of tracer mass (dm / dt) is calculated using Equation 17, for example, as shown in PCT / CA2021 / 051189 filed by the same applicant on August 26, 2021, based on the rate of change of contrast effect (ΔHU / Δt) and the fractional volume of tracer solution passing through the target voxel per unit time (85). The tracer density can be determined from the predicted area under the time-contrast effect curve from step 76, for example, using equations 13, 14, and 15 shown in PCT / CA2021 / 051189 filed by the same applicant on August 26, 2021 (86). The flow velocity through the target voxel is determined based on the rate of change of the tracer mass (dm / dt) and tracer density (ρ), for example, using equation 11B derived from the Reynolds transport theorem shown in PCT / CA2021 / 051189 filed by the same applicant on August 26, 2021 (88). The determined flow velocity value can be communicated or displayed to the technician / operator or other end user via any conventional computer or display device.

number

[0029] Equation (11B) shows that if the rate of change of the fluid mass within the examination volume and the density of the fluid are known, the magnitude of the net fluid velocity within a given image voxel (examination volume) can be estimated. Both pieces of information can be obtained from dynamic contrast-enhanced CT imaging. A is an independent variable that can be selected / controlled by the operator because it depends on the area of ​​the selected ROI.

[0030] In equation (11B), ρ represents the density of iodine in the mixture of blood and contrast agent solution. To estimate the density of iodine in the examination volume (image voxel), it is necessary to know i) the total mass of contrast agent injected into the patient's body, and ii) the total volume of blood mixed with the contrast agent solution.

[0031] The total mass of contrast agent injected into the patient's body is given by the following formula: Mi=C o ×D×V t (12) Here, Mi is the mass of iodine in milligrams (mg), Co is the original concentration of the iodine-based contrast solution in mg / milliliter (mg / mL), D is the contrast agent dilution factor in the range of 0 to 1, and Vt is the total amount of contrast agent injected in milliliters (mL).

[0032] The volume of blood mixed with the contrast agent solution can be estimated in a two-step process using the area under the time-contrast effect curve.

number

[0033] Volumetric flow rate represents the amount of blood passing through per unit time, therefore, the amount of blood mixed with iodine V i This can be determined by the following formula. V i =Q × T en (14) Here, T en This is the period during which the signal intensity of the examination volume (image voxel) is higher than the baseline level, and can be determined using a graph from the measured time-contrast effect curve. The density of the contrast agent within the examination volume (image voxel) can be estimated using the following formula.

number

[0034] The time rate of change of tracer mass within the examination volume (image voxel) can be determined by the following formula.

number

number

[0035] Figure 6B shows another example (an alternative form to the example shown in Figure 6A) of estimating blood flow in a target vessel based on a time-contrast effect curve (80a). The estimation of blood flow (80a) can be achieved by determining the flow reserve ratio (FFR) value based on the time-contrast effect curve. Determining the FFR value may include input 92 of the predicted area under the time-contrast effect curve from step 76. The flow rate can be determined based on the predicted area under the time-contrast effect curve using, for example, the indicator dilution principle represented by Equation 1 shown in PCT / CA2019 / 050668 filed by the same applicant on May 16, 2019 (94). The flow velocity can be determined based on the flow rate at the sampling site and the calculated cross-sectional area of ​​the vessel lumen using, for example, Equation 2 shown in PCT / CA2019 / 050668 filed by the same applicant on May 16, 2019 (95). The flow pressure 96 can be determined from the flow velocity using, for example, Bernoulli's formula as shown in formula 3A or 3B in PCT / CA2019 / 050668 filed by the same applicant on May 16, 2019. Based on the flow pressure 96 determined from at least two sampling sites, the pressure gradient can be calculated, and the FFR value can be determined based on the calculated pressure gradient and systolic blood pressure value using, for example, formula 11 as shown in PCT / CA2019 / 050668 filed by the same applicant on May 16, 2019. The determined FFR value can be communicated or displayed to a technician / operator or other end user via any conventional computer or display device.

[0036] The blood flow imaging systems and methods have been mathematically validated. The mathematical analyses described in PCT / CA2019 / 050668, filed on 16 May 2019, or PCT / CA2021 / 051189, filed on 26 August 2021, by the same applicant, can all be incorporated into this disclosure and provide numerous examples of deriving blood flow characteristics from dynamic contrast-enhanced dynamic CT imaging sessions.

[0037] Fluid motion can typically be evaluated using two methods. The first method involves monitoring the movement of individual particles in a fluid over time (Figure 7A). The Navier-Stokes equations (derived from Newton's second law) can be used to describe the movement of individual particles in a fluid in any direction. The second method involves monitoring the passage of a small amount of liquid within a fixed region (volume) over time (Figure 7B). The monitored region or reference frame (the rectangle enclosed by the black dashed line in Figure 7B) does not move over time. Although a small portion of the fluid contains many individual particles that cannot be resolved, this second method is less computationally intensive than the first method and, as a result, can provide a good approximation of fluid motion with reasonably good spatial resolution. While we do not wish to be constrained by theory, the second method provides the basis for the analytical imaging methods disclosed herein.

[0038] In CT, an image voxel or block of image voxels is selected as a fixed region to monitor the movement of a fluid (e.g., blood) over time. This reference frame is called the examination volume, and the surfaces on either side of the examination volume (as shown in Figure 8A) are called the examination surfaces. The fluid can enter and exit the examination surfaces in any direction.

[0039] For illustrative purposes, the example shown in Figure 8B illustrates the movement of a small amount of fluid (gray area outlined by a gray dashed line) within a volume of inspection (CV). Here, the CV is shown in a two-dimensional view. At time 0, the fluid in question (gray area marked by a mid-gray stripe) completely fills the CV. At a later time (time 0+Δt), the same fluid in question begins to move out of the CV in the flow direction shown in Figure 8B. The dark gray stripe indicates the portion (mass) of fluid leaving the CV at time 0+Δt, and the empty space within the CV is filled with incoming fluid (marked by a light gray stripe). The mid-gray stripe represents the portion of fluid remaining in the CV at time 0+Δt. If the mass of the incoming fluid filling the empty space in the CV is equal to the mass of the fluid leaving the CV, the flow state is considered steady. If the two masses (the light gray portion and the dark gray portion) are not equal to each other, the flow is considered transient. Note that "unsteady" means that the flow is either turbulent or in a transition between a steady state and a turbulent state. At any given time, the total mass of the fluid in question (the fluid region enclosed by the gray dashed line in Figure 8B) remains constant. That is, at time 0, the mass of the region marked with the mid-gray stripe must be equal to the total mass of the region marked with the mid-gray and dark-gray stripe at time 0+Δt. The movement of the fluid relative to CV can be described using the Reynolds transport theorem.

[0040] Figures 9A and 9B show schematic diagrams of two different dynamic image acquisition and reconstruction techniques that form the basis for absolute or relative flow velocity assessment. Figure 9A shows image acquisition and reconstruction at the same cardiac phase (e.g., 75%RR interval or diastole) across multiple time points. Figure 9B shows image acquisition covering the complete cardiac cycle (systole and diastole). Multiple image sets corresponding to different cardiac phases are reconstructed.

[0041] The blood flow imaging techniques described herein use supervised machine learning (ML) or supervised deep learning (DL) algorithms to simplify the clinical workflow for functional assessment in vascular diseases using CT. While cardiac applications are used for illustrative purposes, blood flow imaging techniques may be applicable to other vascular diseases.

[0042] In a standard cardiac CT scan, coronary CT angiography (CCTA) is acquired first to determine if the patient has occlusive stenosis in the coronary arteries. Prior to the CCTA scan, a bolus tracking (BT) scan is initiated to determine the optimal acquisition time for the CCTA scan. The BT scan is a real-time tracking technique in which the signal intensity or contrast effect in a region of interest (ROI) is monitored at consecutive points in time after intravenous bolus injection of contrast agent solution into the patient. When the signal intensity of the monitored ROI reaches a predetermined threshold, the CCTA scan is automatically performed with a delay time (approximately 5-7 seconds) to ensure that the CCTA scan captures the peak or near-peak contrast effect in the coronary arteries. Alternatively, a non-real-time tracking timing bolus (TB) scan can be used instead of a real-time tracking BT scan to estimate the peak contrast effect duration in the target artery. The TB scan is similar to a dynamic perfusion scan, except that the volume of contrast agent solution and iodine concentration injected for the TB scan are smaller than those typically used in a CCTA scan or perfusion scan. Furthermore, TB scans are established as a separate series from CCTA scans. While BT scans are used to illustrate ML / DL implementation, it will be understood that the same concept can be easily extended by using TB scans instead of BT scans.

[0043] In a typical clinical workflow, if CCTA imaging does not show occlusive stenosis, further functional assessment is not necessary. If the CCTA scan confirms the presence of occlusive stenosis, especially if the degree of stenosis is 40–90% of the lumen diameter, a dynamic perfusion scan can be taken during maximal vasodilation to functionally assess the stenosis (i.e., to determine whether the stenosis is restricting flow). In some situations, for a more accurate functional assessment, a dynamic perfusion scan may also be taken at rest to determine the magnitude of the increase in blood flow from baseline.

[0044] Therefore, as summarized in Figure 10, in an example of the progression of the clinical workflow for cardiac CT, BT and CCTA scans are typically performed during cardiac CT, while perfusion scans are optional (depending on CCTA findings). Both BT and CCTA scans are performed in a resting state.

[0045] Figure 11 shows a schematic diagram of an example of blood flow imaging incorporating a machine learning model. 1. Combine BT scan and CCTA scan as a pseudo-truncate dynamic perfusion scan. 2. Track the temporal changes in contrast enhancement in each coronary artery using a simulated, truncated dynamic perfusion scan. 3. Using ML or DL, predict the area under the time-enhancement effect curve in the coronary arteries throughout the first-pass phase, based on the findings in (2). 4. Using ML or DL, predict the area under the time-enhancement effect curve in the coronary artery during maximal vasodilation (congestion) from the AUC predicted in (3). The AUC predicted in 5.(4) is used in combination with the analysis algorithms disclosed in previous techniques to derive blood flow variables such as the flow reserve ratio (FFR).

[0046] Similar techniques can be applied to other scenarios. More specifically, one example is applying ML or DL ​​to predict the area under the complete coronary time-enhancement curve from a genuine truncated dynamic rest perfusion scan containing five or more time points (Figure 12). Another example is applying ML or DL ​​to predict the area under the complete coronary time-enhancement curve from a genuine truncated dynamic stress perfusion scan containing five or more time points (Figure 13). Furthermore, ML or DL ​​can also be applied to predict the area under the time-enhancement curve at one location of a coronary artery in the short-axis plane from time-enhancement curves sampled at the same location of the coronary artery in the axial plane.

[0047] Both ML (Machine Learning) and DL (Deep Learning) are artificial intelligence techniques that teach computers to learn from training datasets in a way similar to how humans learn from past experience. Furthermore, supervised ML and DL refer to the use of labeled training datasets for computer learning. The difference between ML and DL is that ML algorithms require human intervention (such as manual feature extraction) for computer learning, while DL algorithms can automatically extract relevant features from training datasets without human intervention. The comparison between ML and DL is for convenience in discussing segmentation tasks, as both ML and DL can perform AUC prediction tasks, although DL is advantageous for them.

[0048] DL is a subset of ML, and therefore the term ML encompasses DL. When ML and DL are contrasted, it is intended to contrast non-deep learning with DL methods, and does not imply that ML and DL are mutually exclusive categories of computer learning models.

[0049] Machine learning techniques. In an example of an ML technique, a computer is trained to develop a regression model that predicts the area under the complete coronary time-enhancement effect curve by learning from several labeled images provided to the computer. The training image set is obtained from pseudo- or true truncated dynamic perfusion scans. The training images are assigned labels corresponding to the AUC values ​​of the complete coronary time-enhancement effect curve. The labels (ground truth) are obtained from true, complete dynamic perfusion scans acquired from the same patient. The features that the computer can learn can be image-based, non-image-based, or both (Figure 14).

[0050] Image-based features are features that can be observed or measured from image data and may include, but are not limited to, the following variables: contrast enhancement in the coronary duct lumen (pixel signal intensity), thickness of the myocardial wall surrounding the left ventricle, size of the left atrium, size of the left ventricle, diameter of the coronary arteries, morphology of the coronary arteries, and degree of coronary artery stenosis.

[0051] Non-image-based features are features not measured from image data and may include, but are not limited to, the following variables: patient age, sex, weight, heart rate, blood pressure, X-ray tube voltage, X-ray tube current, symptoms, contrast agent injection rate, iodine concentration in the contrast agent solution, etc.

[0052] Figures 15 and 16 provide schematic diagrams illustrating how different image-based features can be used to train an ML algorithm to make predictions about AUC values.

[0053] In Figure 1, coronary angiography effects at different time points can be combined into a single image-based feature using one of the following methods (Figure 15). 1. The difference in Hounsfield units (HU, units of CT image pixel values) between any two time points in the non-baseline (extension) period. 2. The difference in HU between any two time points in the non-baseline period, divided by the time difference between two selected time points. 3. Area under the truncated time-enhancement effect curve between any two time points in the non-baseline period. 4. One of the following options #1 to #3 involves unit conversion from HU to iodine concentration (milligrams / mL).

[0054] In Figure 2, myocardial wall thickness in different regions can be used individually or collectively as input features for training an ML algorithm (Figure 16). Furthermore, as shown in Figures 6 and 7, image and non-image features can be used in combination as input features for training. The computer determines which of the input features has the highest correlation with the labeled feature (AUC value) and retains it in a regression model for future predictions of the AUC value.

[0055] Deep learning techniques. Unlike ML techniques, DL techniques rely on deep artificial neural networks and can recognize relevant image features on their own.

[0056] As shown in Figure 17, a basic artificial neural network consists of three layers: an input layer, a hidden layer, and an output layer. A "deep" artificial neural network is a network with multiple hidden layers. Many hidden layers are needed to accomplish more advanced tasks, which is relevant to our applications. The input layer is where the input values ​​are accepted by the network, and the output layer is where the final prediction is made. In between, the hidden layers take the input values ​​and perform several calculations based on the input variables. The results generated by the hidden layers are propagated to the output layer where the final prediction is made (i.e., regression in the current example).

[0057] In one example, the hidden layer consists of multiple convolutional layers, multiple pooling layers, planarization layers, and multiple dense layers (or fully connected layers).

[0058] A CT image has 512 × 512 = 262,144 pixels. For any grayscale image like a CT image, every pixel in the image has a value in the range of 0 to 255. Traditionally, darker pixels have lower values, and brighter pixels have higher values. CT images are not passed directly to the dense layer. Instead, a mathematical operation called convolution is used to obtain only the desired information from the image as input for the dense layer. During the convolution process, multiple filters are overlaid on the two-dimensional CT image, slicing the image (from top left to bottom right). Each filter is a mathematical matrix with small dimensions, such as 3 × 3 or 5 × 5 pixels. The image information extracted from the convolution by each filter is passed through an activation function, and the result is stored in a new matrix with reduced dimensions (also called a feature map). Then, a pooling filter (normally with dimensions of 2 × 2 pixels) is applied to the feature map to reduce the dimensions of the feature map while retaining the most identified information. As a result, a pooled feature map is produced. This image processing sequence (convolution -> activation -> pooling) can be repeated several times, and the dimensions of the features and the pooled feature maps are reduced after each iteration. After the last iteration, the pooled maps are transformed into one-dimensional vectors (flattening layer) to be used as input for the dense layer.

[0059] Each convolutional filter is designed for specific pattern recognition, and therefore more filters are needed for more complex image patterns. The number of feature maps generated after convolution is equal to the number of convolutional filters applied (Figure 18).

[0060] Each dense layer has many interconnected nodes called neurons. In each neuron, each input variable is multiplied by a weight (a bias may also be added) before undergoing a nonlinear transformation by passing through an activation function. The purpose of this calculation is to determine whether a particular input variable should be passed to the next layer.

[0061] A dynamic image set can be processed by a DL algorithm as follows: First, multiple inputs are generated by a convolutional stack for each image corresponding to a specific time point. Next, concatenation is applied to the flattening layer corresponding to each time point to collect temporal information from all images in the dynamic series. Finally, the concatenated flattening layer provides the input to the first dense layer (Figure 19).

[0062] In summary, both ML and DL methods can predict the AUC of a truncated time-enhancement effect curve from image features shown in dynamic perfusion images (non-image features may also be used), but the input features may differ. When a trained ML algorithm is used, the user needs to provide input features using image processing tools within the software. For example, a sampling ROI is placed in a pseudo- or true dynamic image set to obtain a truncated time-enhancement effect curve, or a measurement tool is placed across the myocardium to measure wall thickness. The ML algorithm then acquires the input features and predicts the AUC value, from which blood flow variables such as FFR can be calculated in the analysis steps disclosed in our prior art. On the other hand, when a trained DL algorithm is used, the user only needs to load a dynamic image set, and the algorithm automatically identifies the relevant input features and predicts the AUC value, and then the blood flow variables.

[0063] The blood flow imaging systems and methods have been validated through experimental testing. The experimental results demonstrate the ability of the blood flow imaging systems and methods to determine one or more of several blood flow characteristics. The following experimental examples are for illustrative purposes only and are not intended to limit the scope of the research. [Examples]

[0064] Experimental examples: Experimental example 1 (Relationship between AUC and the degree of lumen stenosis). As described in PCT / CA2019 / 050668 filed by the same applicant on 16 May 2019 or PCT / CA2021 / 051189 filed by the same applicant on 26 August 2021, the vascular volumetric flow rate F (indicated as Q in Equation 1) can be estimated by the indicator dilution principle.

number

[0065] In stenotic coronary arteries, the magnitude of the increase in volumetric flow from baseline (resting state) is attenuated in the post-stenotic coronary segment when the stenosis exceeds a certain degree (approximately 40% lumen stenosis), and this is reflected in the larger ratio of stress AUC to rest AUC in this segment compared to the pre-stenotic segment (Figure 21). The graphs shown in Figures 20 and 21 summarize that the AUC of the coronary time-angiography effect curve is predictable when other information such as the degree of coronary artery stenosis is known, which supports the concept of learning and predicting AUC values ​​using ML or DL ​​algorithms.

[0066] Experimental Examples: Experimental Example 2 (Relationship between time-based contrast effect curve and heart rate). The symmetry of the coronary time-enhancement effect curve after bolus intravenous injection of contrast agent is relatively unaffected by the patient's heart rate, although the width of the curve may be partially dependent on the patient's heart rate (Figure 22). This suggests that non-image features such as the patient's heart rate can also be useful input variables for ML or DL ​​algorithms to make predictions about AUC.

[0067] Experimental example: Experimental example 3 (Shape of time-enhancement effect curves at different vessel sizes). Figure 23 shows that the approximate time-to-intervention curve obtained from the myocardium (after removing contrast agent retention and recirculation effects) exhibits similar symmetry to the approximate time-to-intervention curve obtained from the supply epicardial coronary artery. This finding also suggests that the proposed ML and DL methods can be applied to estimate the first-pass time-to-intervention curve at the microvessel level in pseudo- or true truncated dynamic perfusion image sets.

[0068] Experimental Examples: Experimental Example 4 (Relationship between myocardial wall thickness, coronary artery diameter, and AUC). As discussed in Experimental Example 1, AUC differs between resting and stressed (congestive) states. During maximal congestion resulting from intravenous adenosine administration, the patient's myocardial contractility and coronary artery diameter can increase substantially compared to baseline levels. Therefore, imaging features such as myocardial wall thickness and coronary artery diameter (or radius or circumference) can be used as input variables for ML or DL ​​algorithms to predict AUC (Figure 24).

[0069] Experimental Examples: Experimental Example 4 (Time-Anti-Enhancement Curves in Different Tomographic Views). As described in PCT / CA2019 / 050668 filed by the same applicant on 16 May 2019 or PCT / CA2021 / 051189 filed by the same applicant on 26 August 2021, the pressure gradient between two sampling slices A and B in a blood vessel can be estimated using Bernoulli's equation.

number

[0070] The axial view is the default tomographic plane for CT image reconstruction. As mentioned above, quantitative blood flow assessment requires short-axis reshaping to obtain time-enhancement curves in slices perpendicular to the direction of blood flow. However, reformatting a fully dynamic (4D) image set can be very time-consuming. Using the proposed ML or DL ​​method, the short-axis time-enhancement curve can be predicted based on time-enhancement curves sampled from the axial plane without requiring image reshaping. Figure 26 compares the short-axis time-enhancement curve of one coronary artery with the axial time-enhancement curve. It can be seen that the short-axis curve is slightly more variable than the axial curve. However, the details of the curve shape are irrelevant to FFR calculation, and only the area under the curve matters. This reduces the complexity of ML and DL learning, making the proposed method feasible for clinical application.

[0071] Experimental Examples: Experiment Example 5 (Training a machine learning model to predict the AUC of the time-contrast effect curve during maximum congestion (stress) from the AUC of the time-contrast effect curve sampled during rest). This example demonstrates a machine learning method to predict the AUC of the time-enhancement effect curve during maximal hyperemia from the AUC of the time-enhancement effect curve sampled during the resting state. The proposed method works for both coronary and myocardial time-enhancement effect curves as well as for other cardiovascular subjects, and here an example using the coronary time-enhancement effect curve is provided. In this method, cardiac CT studies with dynamic rest and stress perfusion scans performed on the same patient on the same day were used to train a linear regression model in machine learning. The machine learning implementation was realized using Python and TensorFlow. Figure 27 shows the coronary time-enhancement effect curves sampled over the entire RCA (right coronary artery) stenosis of a patient during rest and maximal hyperemia (stress). The sampled coronary time-enhancement effect curves were denoised with a modified gamma variable function, and the AUC of the denoised (approximate) curve corresponding to the resting state was used as the input (independent) variable to the machine learning regression. Other input (independent) variables include non-image data such as patient age, heart rate, and blood pressure measured during the resting state.

[0072] Figure 28 shows a scatter plot of the AUC of measured stress-time angiography against the AUC of corresponding measured rest-time angiography curves sampled from different right coronary arteries with different degrees of stenosis. Although there is some overlap between different groups of AUC data, it can be seen that the AUC of rest-time and stress-time angiography curves tends to increase with the degree of stenosis in the target arteries.

[0073] The AUC data shown in Figure 28 was split into a training dataset and a test dataset for machine learning (approximately 80% of the data was used for training). The computer code shown in Figure 29 shows that three hidden layers were implemented in a neural network, and linear regression was used for this task.

[0074] The training and test results are shown in Figure 30. The crosshairs in the graph represent the AUC of the stress coronary artery TIMEAGI effect curve predicted by the trained neural network, compared to the actual AUC of the stress coronary artery TIMEAGI effect curve. The training and test scores were 0.904 and 0.837, respectively (the highest score is 1). The results suggest that the trained machine learning model can predict stress AUC with high accuracy from measured rest AUC and other non-image data measured at rest (e.g., heart rate, blood pressure). The accuracy of the machine learning model can be further improved by training with more image and non-image data and optimizing some of the training settings.

[0075] The flowchart in Figure 31 illustrates the machine learning scheme used in this example. The flowchart in Figure 32 shows how the trained machine learning model can be used for clinical hemodynamic assessment.

[0076] Experimental Examples: Experiment Example 6 (Training a machine learning model to predict the AUC of the full time-enhancement effect curve from the AUC of the truncated time-enhancement effect curve). This example demonstrates a machine learning method to predict the AUC of a complete coronary TIMEA (TIMEA) curve from a truncated TIMEA curve sampled from a simulated truncated resting perfusion scan. As previously mentioned, the complete curve is defined as covering the baseline as well as the entire uphill and downhill slopes. Before training, a pseudo-truncate perfusion scan was first simulated by removing the post-APA image of the peak in the dynamic perfusion image (Figure 33a). The AUC of the truncated curve was then calculated (Figure 33b). The AUC of the complete (untruncate) curve was used as a pair of input variables to train a machine learning linear regression model to understand the relationship between the two AUC values ​​and make corrected predictions for first-time cases.

[0077] The curve shown in Figure 33 is a simulation of a pseudo-truncate time-enhancement effect curve synthesized by combining image data from a bolus tracking (BT) scan with image data from a subsequent diagnostic CT angiography (CTA) scan. The simulation is based on a realistic clinical scenario in which the diagnostic CTA scan is triggered approximately 4–7 seconds after the trigger threshold of the monitor region is reached (in this simulation, the trigger threshold of the selected monitor region is set to 50 HU, and the trigger interval is specific to each scanner and is a function of the time it takes to reconfigure the scanner to the diagnostic CTA scan setting for the BT scan setting). Therefore, the truncated time-enhancement effect curve has a larger time interval between the third and fourth data points. The proposed machine learning method works for all types of scans, including test bolus (TB) scans, bolus tracking (BT) scans, and diagnostic CTA scans, in addition to perfusion scans. Although the contrast agent volume and X-ray energy used for each scan may differ (Table 1), the AUC calculation remains unaffected if the conversion from image pixel contrast effect to tracer (iodine) concentration can be performed using a conversion factor specific to the X-ray energy used for the scan.

[0078] [Table 1]

[0079] Figure 34 shows a scatter plot of the AUC of the measured full time-enhancement effect curve against the AUC of the relevant measured truncated time-enhancement effect curve, sampled from different coronary arteries with different degrees of stenosis. The AUC data was split into training and test datasets for machine learning (approximately 80% was used for training). As in Experimental Example 5, linear regression analysis was used for machine learning for this task. Preliminary training results (Figure 35) showed high and low training / test scores, suggesting that the trained machine learning regression model could be used to predict the AUC of the full coronary time-enhancement effect curve, even if the time-enhancement effect curve was not obtained to fully cover the entire uphill and downhill slopes.

[0080] The flowchart in Figure 36 illustrates the machine learning scheme discussed in this example. The flowchart in Figure 37 shows how trained machine learning can be used in clinical hemodynamic assessment.

[0081] Experimental Examples: Experiment Example 7 (Train a machine learning model to predict the AUC of the time-enhancement effect curve in a short-axis tomography view from the AUC of the time-enhancement effect curve in a default axial tomography view). As shown in Figure 26, coronary time-enhancement curves can be sampled from dynamic cardiac perfusion images in different tomographic directions. In CT, the default image view is axial, but a complete cross-sectional view of the ductus arteriosus lumen may not be visible in a given axial slice (i.e., the axial slice may have a non-orthogonal orientation to the central longitudinal axis of the vessel or the direction of blood flow). Currently, reformatting axial images to the short axis (i.e., slices with an orientation orthogonal to the central longitudinal axis of the vessel or the direction of blood flow) requires specialized software for volumetric image reshaping. Figure 38 shows preliminary results predicting the AUC of the coronary time-enhancement curve in a short-axis tomographic slice from the AUC of the coronary time-enhancement curve sampled in the corresponding axial slice, using a machine learning technique similar to those described in Experimental Examples 5 and 6.

[0082] Experimental Examples: Experiment Example 8 (Training a deep neural network to segment coronary arteries and myocardium in contrast-enhanced cardiac CT images). In contrast to the methods discussed in Experimental Examples 5 and 6, this example demonstrates a deep learning technique for segmenting coronary arteries and myocardium in contrast-enhanced cardiac CT images. Such segmentation facilitates the extraction of relevant image features to generate input variables for subsequent machine learning. In other words, this technique is a combination of deep learning and machine learning. Figures 39A and 39B show the labeling of different image features, including coronary arteries, ascending aorta, and left ventricular myocardium, with different colors using MATLAB's Medical Image Labeler. Using these image labels, a deep neural network called U-NET was trained to segment these features in a first-hand contrast-enhanced cardiac CT image. The computer code shown in Figure 39C demonstrates a three-dimensional implementation of U-NET using MATLAB. U-NET is based on the convolutional neural network (CNN) described in the previous section, but with several modifications that allow for more accurate segmentation than CNN. The main difference between the two neural networks is that in CNN, the image is downsampled and transformed into a vector used for object classification. Conversely, in U-NET, the image is downsampled to the pixel level and then upsampled. This additional sampling step allows U-NET to achieve more accurate detection of structural details within the image. In other words, while CNN methods can provide information about what an object in an image is, U-NET methods can provide this information plus detailed spatial information about the object (the exact location of the object within the image).

[0083] The segmentation task requires excellent spatial resolution to be accomplished because the target cardiovascular system (such as the coronary arteries) may not remain in exactly the same position during the time series of scan images (such as those acquired in dynamic perfusion scans) due to residual respiration and cardiac movement of the patient during image acquisition. Therefore, deep neural networks (i.e., U-NETs) that can better preserve spatial information of images are more suitable for our task than deep neural networks (i.e., CNNs) that have a lower ability to preserve spatial information of images. Examples of medical image segmentation using U-NET and CNN have already been described (Ronneberger, O., Fischer, P., Brox, T. (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation. In: Navab, N., Hornegger, J., Wells, W., Frangi, A. (eds) Medical Image Computing and Computer-Assisted Intervention - MICCAI 2015. MICCAI 2015. Lecture Notes in Computer Science, vol 9351. Springer, Cham) (Krizhevsky A, Sutskever I, Hinton GE (2012) ImageNet classification with deep convolutional neural networks. Adv Neural Inf Process Syst 25).

[0084] Due to the large number of CT images in a dynamic perfusion scan covering the entire heart (e.g., a 14cm axial scan range with a 1mm image slice thickness) and the large number of pixels in each CT image (512×512), image segmentation using U-NET requires a high-performance computer to achieve. For example, the minimum RAM requirement is typically 64GB, and the graphics card could be an NVIDIA RTX 3060 Ti (8GB, GDDR6). However, to ensure efficient deep learning, the preferred hardware requirements are higher than usual. For example, an NVIDIA RTX A5000 graphics card (24GB, GDDR6) is used by some researchers for deep learning. Therefore, a standard laptop or desktop computer built for light loads may not be suitable for this task.

[0085] Approximately 60 and 20 contrast-enhanced cardiac CT image sets were used for training and testing U-NET, respectively. Each CT image set had labels for the coronary arteries (i.e., RCA, LAD, LCX, and LM) and background (any features unrelated to the coronary arteries). Initially, the focus was on training U-NET to segment the left main artery (LM) and left circumflex artery (LCX), and the test accuracy for segmenting the LM and LCX was 70.5%. Higher test accuracy can be expected by using more complex image sets for training and adjusting training settings such as the learning rate. Figure 40 shows how the training results from the deep learning model can be used as input labels to train the machine learning models described in Experimental Examples 5-7.

[0086] As demonstrated above, the use of machine learning models can significantly streamline clinical workflows related to hemoflow imaging by employing hybrid computer learning and analysis techniques. As described in PCT / CA2019 / 050668 filed May 16, 2019, or PCT / CA2021 / 051189 filed August 26, 2021, by the same applicant, hemoflow variables depend on many other physiological and morphological variables. If a computer needs to learn the relationships between individual hemoflow variables and all of their dependent variables, it requires a very large amount of training / evaluation data, image feature extraction and labeling, and computation time. The methods described herein restrict computer learning to relevant central parameters, such as the area under time-enhancement effect curve (AUC), for which an analysis step is required to derive all blood flow characteristics described in PCT / CA2019 / 050668 filed on 16 May 2019 or PCT / CA2021 / 051189 filed on 26 August 2021 by the same applicant.

[0087] The currently disclosed hybrid computer learning analysis blood flow imaging techniques have many advantageous features, including, for example, predicting the time-congestive enhancement curve from the rest-time enhancement curve, predicting the time-congestive enhancement curve at the microvessel level from the large vessel rest-time enhancement curve, or predicting the time-congestive enhancement curve at one location of a vessel in one direction from time-congestive enhancement curves sampled at the same location of a vessel in another direction.

[0088] The currently disclosed hybrid computer-learned analysis hemodynamic imaging techniques can simplify clinical workflows while providing reliable functional assessments of vascular diseases using CT, through the following means. 1. To derive blood flow characteristics without performing a true dynamic perfusion scan or a short or truncated dynamic perfusion scan. 2. To determine blood flow characteristics without administering vasodilators. 3. To derive blood flow characteristics from a dynamic (four-dimensional) image set without reformatting it from one tomographic view to another.

[0089] Several exemplary variations of methods or systems for blood flow imaging have been described above. Further variations and modifications are described below. In addition, guiding relationships for constructing these variations and modifications are also described below. Further variations and modifications are conceived and will be recognized by those skilled in the art. It should be understood that the guiding relationships and exemplary variations or modifications are provided for the purpose of improving the understanding of those skilled in the art and are not intended to be limiting.

[0090] For example, the blood flow imaging method 20 shown in Figure 2 is merely illustrative, and one or more steps shown in Figure 2 may be replaced or omitted as needed for a particular implementation, so it should not be considered an exhaustive blood flow imaging method. For example, in a particular implementation, the CT scanning of the subject may be geographically or temporally staggered from the image reconstruction. An example of a possible deformed morphological blood flow imaging method includes both projection data from a CT scan and image reconstruction performed in a preceding step, and the reconstructed image is stored for later analysis or analysis by a third party. The deformed morphological blood flow imaging method may begin by acquiring the stored image data. Contrast agent signal data may be extracted from the stored image data, and it is optional that the target vessel in the image data is not explicitly identified. A time-contrast effect curve is generated based on the contrast agent signal data, and the time-contrast effect curve has an upward slope plotted from data points obtained during the increasing phase of the contrast agent signal data and a downward slope plotted from data points obtained during the decay phase of the contrast agent signal data. Therefore, the flow velocity value is determined according to the same method steps as shown in Figure 6A or Figure 6B.

[0091] As another example, blood flow imaging methods and systems are not limited to computed tomography (CT) scanning and can be readily adapted to other imaging modalities, including MRI and other X-ray imaging techniques (i.e., X-ray imaging techniques other than CT imaging), including MRI and, for example, fluoroscopy, that have sufficient spatial resolution to image blood vessels and show a proportional increase in signal intensity within an ROI as a function of the mass of contrast agent present in the ROI (more contrast agent or tracer results in a higher signal at the ROI). X-ray-based scanning is a form of medical imaging that involves the transmission of high-frequency electromagnetic signals that attenuate as they pass through the body of the subject, with the remaining signals being captured by a detector for subsequent analysis. The experimental example data were acquired with a single-energy CT (SECT) scanner. Most clinical CT scanners use single-energy acquisition. However, dual-energy CT (DECT) scanners may also be available. Dual-energy CT refers to the use of two X-ray energy sources, rather than a single X-ray energy source, for scanning the object. Existing literature has shown that dual-energy CT can perform dynamic CT acquisition as well as single-energy CT. From an image processing perspective, without any modification, the methods described herein or in PCT / CA2019 / 050668 filed on May 16, 2019, or PCT / CA2021 / 051189 filed on August 26, 2021, by the same applicant, such as methods based on the Reynolds transport theorem, the indicator dilution principle, or Bernoulli's formula, can be applied to both SECT and DECT.

[0092] Magnetic resonance imaging (MRI) is an alternative to X-ray-based scanning. It has well-known medical imaging applications, including imaging for diagnosing diseases of soft tissues such as the brain, lungs, liver, muscles, and heart. MRI scans involve applying a magnetic field to the patient and transmitting radio frequency pulses. The resonance energy is emitted by the patient and picked up by a receiver / detector that acquires scan data for subsequent analysis. To improve image clarity, contrast agents must be administered orally or intravenously to the patient for both X-ray and MRI scans. Contrast agents for X-ray imaging techniques include, for example, iodine-based contrast agents. Contrast agents for MRI imaging techniques include, for example, gadolinium-based contrast agents. While scan data acquired from X-ray-based scanner devices / systems is often referred to synonymously as scan data or projection data, scan data acquired from MRI scanner devices / systems is usually referred to simply as scan data. Therefore, the term scan data is understood to encompass the term projection data.

[0093] The methods described herein, comprising PCT / CA2019 / 050668 filed on May 16, 2019, by the same applicant, or PCT / CA2021 / 051189 filed on August 26, 2021, have been demonstrated with dynamic contrast-enhanced CT imaging data obtained after intravenous bolus injection of iodine-based contrast agents and are also applicable to dynamic MRI imaging data obtained after intravenous bolus injection of gadolinium (Gd)-based contrast agents. In preclinical trials of PCT application PCT / CA2019 / 050668 (filed on May 16, 2019) by the same applicant, the inventors demonstrated that a time-contrast effect curve of the target region can be obtained from dynamic contrast-enhanced MRI imaging in a manner similar to that of dynamic contrast-enhanced CT imaging. Furthermore, the temporal change in signal intensity over time (e.g., T1 relaxation time) is caused by the migration of the Gd contrast agent in the target region, and the magnitude of the signal change is closely related to the concentration of Gd molecules (tracers). When using low-density Gd contrast agents, the change in MRI signal intensity in the target region and the contrast agent concentration show a relatively linear relationship. Moreover, with respect to PCT / CA2021 / 051189 filed by the same applicant on August 26, 2021, this linear relationship makes it easy to estimate the rate of change of tracer mass over time (dm / dt in Equation 11 of PCT / CA2021 / 051189 filed by the same applicant on August 26, 2021) using the RTT method to derive the flow velocity.

[0094] The contrast agents (also called tracers) for various imaging modalities are well-established in the current literature and remain an active area for the development of new alternatives. Blood flow imaging methods and systems can accommodate any suitable combination of contrast agent and imaging modality, provided that the imaging modality provides sufficient temporal and spatial resolution to image the target cardiovascular system, such as the target blood vessel or a portion of the target blood vessel, or the target cardiac chamber or a portion of the lumen of the target cardiac chamber.

[0095] Blood flow imaging methods and systems may include the selection of target voxels or pixels in acquired image data (i.e., acquired pixel data) or image data output by a computer learning model, and the analysis of pixel data in the selected voxels or pixels. While voxels provide accuracy for volumetric imaging, voxel-based evaluation can be disadvantageous with large datasets that are difficult to manage given the bandwidth of typical computers. The systems and methods described herein provide efficient handling of large data files using commonly available computers, enabling fine temporal resolution with interactive visualization and near real-time evaluation.

[0096] A voxel is the smallest 3D element of volume and is typically represented as a cube or box with dimensions of height, width, and depth (or 3D Cartesian coordinates x, y, and z). Just as a 2D image is composed of multiple pixels (represented as squares with height and width or x and y dimensions), and smaller pixels result in better image quality, the same concept applies to 3D data volume. In data acquisition, each 3D voxel represents a specific X-ray absorption. A voxel described as isotropic means that all dimensions of that isotropic voxel are the same, and is typically a perfect cube with uniform resolution in all directions. In contrast, voxels described as anisotropic or anisotropic mean that anisotropic voxels are not perfect cubes, either because all dimensions of the voxel are not the same (i.e., at least one dimension of the anisotropic voxel differs from the others), or because the anisotropic voxel contains partial voxel units (usually multiple voxel units). The systems and methods described herein provide efficient manipulation of image data, enabling operability with a choice of either or both isotropic or anisotropic target voxels.

[0097] The terms ROI and target voxel are related because in reconstructed 3D image data, the ROI encompasses either a target voxel or a block of adjacent target voxels. In reconstructed 2D image data, the terms ROI and target pixel are also related because the ROI encompasses either a target pixel or a block of adjacent target pixels. The terms voxel and pixel are related because a voxel is the 3D version of a pixel. Voxel size is related to both pixel size and slice thickness. Pixel size depends on both the field of view and the image matrix.

[0098] A selected isotropic target voxel may be a single isotropic voxel or a contiguous block of neighboring or adjacent voxels whose block is isotropic. A selected anisotropic target voxel typically contains two or more voxel units, but may approach the volume of a single voxel unit, or may be a contiguous block of neighboring or adjacent voxels whose block is anisotropic. An anisotropic block of voxels may include a portion of the voxel at its boundary, as would be expected if the target voxel were a non-square shape such as a circle or triangle. Therefore, a block of target voxels or target pixels can correspond to ROIs of various shapes (including circles, triangles, or irregular shapes), and does not need to be limited to a complete voxel or pixel unit; an ROI can define a block of neighboring voxels or a pixel having a partial voxel at the boundary of the ROI.

[0099] The elapsed time of the imaging scan procedure corresponds to the duration of scan data acquisition and may be modified as necessary to acquire sufficient data to estimate or predict the area under the time-contrast effect curve, provided that the imaging scan captures at least a portion of the contrast agent's increasing or decreasing phase at the sampling site. Generally, imaging scans of more than 5 seconds are required to capture a portion of both the increasing and decreasing phases. In specific cases, imaging scans may be configured to capture scan data exceeding 6 seconds, 7 seconds, 8 seconds, 9 seconds, or 10 seconds. There are no upper time limits or constraints on contrast agent transit time, but in most cases, the imaging scan will not significantly exceed the expected transit time of the contrast agent at the sampling site.

[0100] The number of images (also called frames or individual scans) analyzed to predict the area under the time-contrast effect curve may be modified as needed, provided that the number of images cumulatively captures at least a portion of the contrast agent's increasing or decreasing phase at the sampling site, in order to obtain sufficient data to estimate the shape of the time-contrast effect curve. Generally, more than six imaging scans are required to capture both the increasing and decreasing phases. In specific examples, imaging scans may be configured to capture seven or more images, nine or more images, eleven or more images, thirteen or more images, fifteen or more images, seventeen or more images, nineteen or more images, or twenty-one or more images. Furthermore, imaging scans configured to capture at least ten images have been observed to improve consistency between peak value determination and curve shape. While it is not necessary to extract signal intensity values ​​from all of at least ten images, at least ten images often provide a sufficiently large set of images to select an appropriate subset of time-distributed images (usually five or more images) that yields consistency in the estimation of the curve shape.

[0101] Blood flow imaging methods and systems are considered dynamic because they analyze multiple images, distinguishing them from static techniques that evaluate a single image. Most commercially available CT angiography techniques are static. Furthermore, commercially available CT angiography techniques that are minimally dynamic (evaluating 2-3 images) do not recognize or consider the benefits of acquiring scan data from both the increasing and decreasing phases of contrast agent passage, generating a time-contrast effect curve, or predicting the area under the time-contrast effect curve using a computer learning model. In addition, CT angiography that acquires two or three images in slightly different timeframes to compensate for motion or for the physician to select the best image least affected by motion may also be considered a static technique.

[0102] Multiple images used to predict the area under a time-contrast effect curve, for example, at least five images, are considered to be multiple corresponding images, and the image correspondence refers to a chronological sequence of multiple images located within the same sampling site or slice or a group of adjacent sampling sites or slices. Thus, the image correspondence is spatially restricted to a single sampling site or slice or a group of adjacent sampling sites or slices (or a single ROI or a group of adjacent ROIs), and the image correspondence does not include sampling sites or slices spatially separated to be upstream versus downstream of the source of the blood flow anomaly. For example, if the determination of blood flow characteristics involves comparing corresponding values ​​calculated from first and second time-contrast effect curves, the first time-contrast effect curve may be from a first set of multiple corresponding images from a first sampling site or slice located upstream of the suspected source of the blood flow anomaly, and the second time-contrast effect curve may be from a second set of multiple corresponding images from a second sampling site or slice located downstream of the suspected source of the blood flow anomaly. In this example, the first and second sampling sites are spatially separated by the suspected source of intervening blood flow abnormalities, so the first set of corresponding images is not mixed with the second set of corresponding images.

[0103] Each set or group of corresponding images is optionally time-ordered or time-decomposed to benefit input to a computer-learned model for predicting the area under the time-contrast effect curve. The time-contrast effect curve may have an upward slope, a peak, and a downward slope. Time ordering benefits computer learning and computer prediction of the area under the time-contrast effect curve. Time ordering provides an upward slope of the time-contrast effect curve interpolated from time-specific contrast agent signal data points acquired during the increasing phase of contrast agent passage and a downward slope of the time-contrast effect curve interpolated from time-specific contrast agent signal data points acquired during the decay phase of contrast agent passage. Thus, the acquisition of scan data and reconstruction of image data are performed with reference to a time-ordering scheme so that each set of corresponding images obtained from the image data can be ordered in time. The time-ordering scheme may be any convenient scheme including a timestamp with a real-time identifier, relative time identifiers such as the elapsed time since bolus injection, or any customized time identifiers that can be used to identify the absolute or relative time of each image and to time-decompose ordering of the set of corresponding images. Established protocols for the time interval between contrast agent administration and image acquisition can be adopted when designing a time-sequencing scheme. Furthermore, established timing techniques, such as bolus tracking, can be employed to optimize the timing of scan acquisition and the time-sequencing of image data.

[0104] A time-contrast effect curve is a plot of contrast agent signal intensity against time obtained from scan data of contrast agent passage in a single sampling site or a group of adjacent sampling sites. Among other variations, the time-contrast effect curve is sometimes called a time-density curve, signal intensity-time curve, time-dependent signal intensity, or time-intensity curve. Machine learning models do not need to plot time-contrast effect curves (the plots shown in the figure are for illustrative purposes to facilitate comparison with the techniques described in PCT / CA2019 / 050668 filed on May 16, 2019, or PCT / CA2021 / 051189 filed on August 26, 2021, by the same applicant), and can predict the area under the time-contrast effect curve directly from the input image by directly evaluating the contrast effect in the input image or a selected portion thereof. In the time-contrast effect curve, the term "contrast effect" refers to the increase in measured contrast signal intensity relative to a baseline or reference value, such as signal intensity measured at the minimum contrast level, signal intensity measured at the residual contrast level, or signal intensity measured in the absence of contrast. Qualitative terms describing contrast agent passage, such as pre-entry, entry, wash-in, increase phase, decay phase, wash-out, removal, and post-removal, refer to the bolus injection event or, more generally, the contrast agent administration event. Therefore, each of these terms, except pre-entry, describes a portion of the contrast agent passage that occurs after the relevant injection or administration event. The term pre-entry may correspond to a time range that begins earlier than the injection or administration event.

[0105] The blood flow imaging methods and systems described herein enable the determination of blood flow characteristics. Blood flow characteristics can be any indicators used to evaluate blood flow in a subject area of ​​a subject. Blood flow characteristics include those described in PCT / CA2019 / 050668 filed on 16 May 2019 or PCT / CA2021 / 051189 filed on 26 August 2021 by the same applicant, such as flow rate, flow velocity, flow acceleration, flow pressure, and cardiac-induced pulsation reconstruction. Cardiac-induced pulsation refers to the temporal fluctuations in flow rate / flow velocity caused by the contraction and relaxation of the heart (leading to the forward discharge and backward attraction of blood, respectively). Flow rate, flow velocity, and flow acceleration are indicators of blood flow. Blood flow imaging techniques, as described in the international PCT application PCT / CA2019 / 050668 filed by the same applicant on 16 May 2019, may include other blood flow assessment techniques as needed, such as blood flow assessment or blood pressure assessment (using Bernoulli's formula), which also describe the flow reserve ratio (FFR) and shear stress as quantifiable blood flow characteristics, and further examples of blood flow characteristics described include the area under the curve, the rate of change of the area under the curve, the peak (maximum value) of the curve and blood volume.

[0106] Blood flow characteristics can be determined from raw signal intensity measurements or contrast enhancement measurements. In CT, the measured signal intensity can be expressed as a CT value, while in contrast enhancement, normalization relative to a reference value or subtraction of the signal intensity is estimated.

[0107] Determining blood flow characteristics may require, at a minimum, computer-learned model predictions of the area under the time-contrast effect curve, and optionally include computer predictions of other parameters such as the time rate of change of parameters including, for example, the time rate of change of signal intensity, the time rate of change of contrast effect, the time rate of change of tracer mass, the time rate of change of flow velocity, and the time rate of change of flow pressure. Various time rate parameters are associated as described in the mathematical derivations provided in PCT / CA2019 / 050668 filed on 16 May 2019 or PCT / CA2021 / 051189 filed on 26 August 2021 by the same applicant.

[0108] The evaluation of blood flow and the determination of blood flow characteristics can provide diagnostic results. For example, predicting the area under the time-enhancement curve at first and second sampling sites (or first and second ROIs within the same sampling slice) yields the area under the first time-enhancement curve and the area under the second time-enhancement curve, and estimating blood flow characteristics involves a determination that includes corresponding values ​​calculated from the first and second areas under the time-enhancement curve. As another example, relative or absolute flow velocity can be determined at one or more ROIs. The blood flow characteristic values ​​themselves can provide diagnostic results. In a further example, corresponding values ​​calculated from the first and second time-enhancement curves or the first and second flow velocities are compared, and the difference in corresponding values ​​exceeding a predetermined threshold indicates a diagnostic result. Thresholds and corresponding diagnostic results can be adopted from relevant literature and medical guidelines. Furthermore, by repeatedly using blood flow imaging methods and systems, various correlations between indices, thresholds, and diagnostic results can be developed.

[0109] A Region of Interest (ROI) is a region on a digital image that surrounds or includes a desired anatomical location, such as a target blood vessel, a portion of the lumen of a target blood vessel, a target cardiac chamber, or any other target cardiovascular system. The terms ROI and target voxel or target pixel are used to define a region in which the ROI contains one or more voxels (in 3D imaging) or one or more pixels (in 2D imaging). While both voxel and pixel are related in that they depend on pixel data, a voxel is the 3D version of a pixel, an accumulation of pixel data from multiple slices in a 3D image.

[0110] The image processing system enables the extraction of pixel data from ROIs on the image, including, for example, average parameter values ​​calculated for all pixels within the ROI. The sampling site is the location of one or more imaging slices selected to evaluate a desired anatomical location, such as a blood vessel of interest, or a cardiac chamber of interest, or any other cardiovascular system of interest. In some cases, analysis of the time-enhancement curve from a single ROI may be sufficient to determine blood flow characteristics or indicators. In other cases, multiple ROIs at a single sampling site, or multiple ROIs at multiple sampling sites, or multiple imaging slices can be analyzed to obtain multiple corresponding image sets and generate multiple corresponding time-enhancement curves, and any number of multiple corresponding time-enhancement curves can be compared to determine blood flow characteristics or indicators. Conventional scanners can capture 3D image data of all or part of multiple vascular structures, such as all or part of a blood vessel of interest or the cardiovascular system of other interests, and possibly multiple blood vessels of interest. Furthermore, since scans can be subdivided into multiple slices as needed, interlogging multiple sites or slices in the ROI, near the ROI, upstream of the ROI, downstream of any ROI, or any combination thereof is possible and convenient. In multislice or multisite imaging modalities, simultaneous tomographic slices or sampling sites can be extracted with each scan. Therefore, blood flow imaging methods do not need to be limited to the analysis of one or two time-dependent contrast effect curves for scanning contrast agent passage (from entry to exit) in the target vessel, and a single scanning procedure with a single bolus injection of contrast agent can support multiple slices or sampling sites subdivided as needed from the scan data.

[0111] Motion correction or compensation processing of reconstructed image data can be used if the ROI would benefit from being adjusted to adapt to the movement of the vascular wall during the cardiac cycle. Rule-based or machine learning-based motion correction or compensation models are available and can be used as needed in specific implementations.

[0112] The cardiovascular system of interest (also called the vascular structure of interest) may be any blood flow pathway or lumen of the cardiovascular system (also called the circulatory system) and may include any vessel of interest (e.g., systemic arteries, peripheral arteries, coronary arteries, pulmonary arteries, carotid arteries, systemic veins, peripheral veins, coronary veins, pulmonary veins) or any cardiac chamber or opening of interest of interest that can be imaged by contrast imaging techniques. The cardiovascular system of interest typically has a diameter of at least about 0.1 mm, e.g., greater than 0.2 mm or greater than 0.3 mm. The cardiovascular system of interest, such as a vessel of interest or a designated portion of a vessel of interest, can be identified and used as a target for contrast-enhanced hemodynamic imaging to determine the diagnosis of a cardiovascular disorder or vascular disorder, or to determine predisposition to such disorder. The vessel of interest may be within any anatomical region or any organ (e.g., brain, lungs, heart, liver, kidneys, etc.) of an animal body (e.g., human body).

[0113] Blood flow imaging methods are not limited to scan data acquired while the subject is in a hyperemic state (also known as hypertensive stress or vasodilatory stress); time-enhancement effect curves generated from scan data acquired while the subject is in a non-hyperemic state (also known as a resting state) can also yield useful results. Inducing hyperemia is a well-known medical protocol in blood flow assessment and often involves the administration of vasodilators such as adenosine, sodium nitroprusside, dipyridamole, regadenoson, or nitroglycerin. The mode of administration of vasodilators varies depending on the imaging protocol and may include intravenous or intracoronary injection.

[0114] To determine the presence of cardiovascular disorders in the target cardiovascular system, such as vascular disorders in the target vessel, blood flow characteristics are analyzed based on the area under at least one time-enhancement curve, including, for example, the area under a single time-enhancement curve generated from pixel data of ROI in a scan of a single sampling site, or, as another example, the areas under multiple time-enhancement curves generated from corresponding multiple sampling sites. In the case of stenosis, comparing two sampling sites is useful for comparing the blood flow characteristics determined at the sampling site upstream of the stenosis with the blood flow characteristics determined at the sampling site downstream of the stenosis. More generally, once the target vessel is identified, multiple sampling sites can be specified in or near the target vessel. A time-enhancement curve is generated for each of the multiple sampling sites. Desired blood flow characteristics based on each time-enhancement curve are determined for each of the multiple sampling sites. The determined blood flow characteristics for each of the multiple sampling sites are compared to determine vascular disorders. Depending on the specific implementation, determining blood flow characteristics at one or more sampling sites, or determining the presence or absence of vascular disorders based on a comparison of blood flow characteristics at multiple sampling sites, can provide a diagnostic result.

[0115] Cardiovascular disorders or vascular disorders (also known as vascular diseases) evaluated by the methods or systems described herein may be any unhealthy blood flow abnormalities that impair the health of the subject, including any functionally significant blood flow limitation or impaired blood flow in the heart or non-cardiovascular systems, or unhealthy blood flow abnormalities that show symptoms of cardiac chamber abnormalities, such as cardiac valve abnormalities (e.g., aortic valve disease), heart failure, atherosclerosis (e.g., plaque formation), peripheral artery diseases including carotid artery disease and renal artery disease, aneurysms, Raynaud's phenomenon (Raynaud's disease or Raynaud's syndrome), Buerger's disease, peripheral venous diseases and varicose veins, thrombosis and embolism (e.g., thrombosis in veins), blood coagulation disorders, ischemia, angina pectoris, heart attack, stroke and lymphedema.

[0116] Blood flow imaging methods and systems can be used to assess suspected cardiovascular or circulatory disorders by, for example, providing determination of blood flow characteristics in target vessels identified in previous screenings as potentially being sources of unhealthy circulatory abnormalities. Furthermore, partly due to the acquisition of scan data capturing multiple vessels and the reduction of scan data processing time, blood flow imaging methods and systems can be used, as a first example, to proactively assess blood flow in specific vessels or groups of vessels (e.g., pulmonary artery blood flow assessment) and can be implemented as a screening tool that serves as an initial indicator for identifying the causes of unhealthy circulatory abnormalities such as functionally significant stenosis.

[0117] In blood flow imaging methods, the subject or patient does not need to hold their breath during the scanning procedure. In some cases, breath-holding is optional. In other cases, motion correction or motion compensation processing of image data can be used on scan data acquired without breath-holding of the subject or patient. If necessary, motion correction or motion compensation processing of image data can be used on scan data acquired with breath-holding if the ROI benefits from adjusting for the movement of the vessel wall during the cardiac cycle. Rule-based or machine learning-based motion correction or correction models may be used as needed in specific implementations.

[0118] Embodiments or parts thereof disclosed herein can be implemented by programming one or more computer systems or devices using computer-executable instructions embodied in a non-temporary computer-readable medium. When executed by a processor, these instructions operate to cause these computer systems and devices to perform one or more functions specific to the embodiments disclosed herein. The programming techniques, computer languages, devices and computer-readable media necessary to achieve this are known in the art.

[0119] In one example, a non-temporary computer-readable medium embodying a computer program for blood flow imaging may include computer program code for obtaining image data comprising multiple corresponding images capturing at least a portion of one or both of the contrast agent's increasing and decreasing phases in the target cardiovascular system; computer program code for providing image data or image features to a machine learning model to predict the area under the time-contrast effect curve of the contrast agent in the target cardiovascular system; computer program code for selecting a target region within the target cardiovascular system in the image data; and computer program code for determining blood flow characteristics through the target region based on the area under the time-contrast effect curve. In another related example, the image data includes at least one image capturing the target cardiovascular system before the entry of the contrast agent. In yet another related example, the computer-readable medium further includes computer program code for obtaining scan data of the target cardiovascular system from an X-ray-based scan or an MRI scan, and for reconstructing image data based on the scan data.

[0120] Computer-readable media are data storage devices that can store data that can then be read by a computer system. Examples of computer-readable media include read-only memory, random-access memory, CD-ROMs, magnetic tapes, and optical data storage devices. Computer-readable media can be geographically localized or distributed across networked computer systems so that computer-readable code is stored and executed in a distributed manner.

[0121] A computer implementation of a system or method typically includes memory, interfaces, and a processor. The type and configuration of memory, interfaces, and processors may vary depending on the implementation. For example, an interface may include a software interface that communicates with an end-user computing device via an internet connection. An interface may also include a physical electronic device configured to receive requests or queries from a device that transmits digital and / or analog information. In another example, an interface may include a physical electronic device configured to receive signals and / or data related to a blood flow imaging method and system from, for example, an imaging scanner or image processing device.

[0122] The technical fields of medical imaging, particularly CT or MRI imaging as described herein, have come into existence as a result of commercially available computers and have required significant advances in semiconductor technology to enable their implementation in clinics. While the introduction of CT to clinics began in the 1970s, the introduction of MRI medical imaging began in the 1980s.

[0123] Therefore, CT or MRI imaging is closely linked to computers and computer functions. CT or MRI images are communicated and processed using dedicated medical imaging software, which processes and extracts data points to generate temporal contrast-enhancing curves. For example, CT or MRI image data may have a range (continuous, non-discretized) larger than 1000 pixels / voxel values, making it impossible for human experts to accurately quantify the contrast effect. Furthermore, CT or MRI image data is stored and processed in DICOM format, which requires dedicated software to evaluate contrast effect and temporal information to accurately generate temporal contrast-enhancing curves. Neither the temporal component nor the contrast effect component of the temporal contrast-enhancing curve can be evaluated and quantified without dedicated software.

[0124] For example, any suitable processor type can be used depending on the specific implementation, including a microprocessor, programmable logic controller, or field-programmable logic array. Furthermore, any conventional computer architecture, including, for example, memory, mass storage, processor (CPU), graphics processing unit (GPU), read-only memory (ROM), and random access memory (RAM), commonly connected to the system bus of data processing equipment, can be used for computer implementations of this system or method. Memory may be implemented as ROM, RAM, a combination thereof, or simply as a general memory unit. Software modules in the form of routines and / or subroutines for performing functions of the system or method may be stored in memory and then retrieved and processed via the processor to perform specific tasks or functions. Similarly, one or more method steps can be encoded as program components, stored in memory as executable instructions, and then retrieved and processed via the processor. User input devices, such as a keyboard, mouse, or other pointing device, may be connected to a PCI (Peripheral Component Interconnect) bus. If necessary, the software may provide an environment representing programs, files, choices, etc., through icons, menus, and dialog boxes graphically displayed on a computer monitor screen. For example, any number of blood flow images and blood flow characteristics, including time-contrast effect curves, may be displayed.

[0125] Computer implementations of a system or method may accommodate any type of end-user computing device, including computing devices that communicate via a network connection. The computing device may display graphical interface elements for performing various functions of the system or method, including, for example, the display of determined blood flow characteristics for the cardiovascular system of interest. For example, the computing device may be a server, desktop, laptop, notebook, tablet, personal digital assistant (PDA), PDA phone, or smartphone. The computing device may be implemented using any suitable combination of hardware and / or software configured for wired and / or wireless communication. For example, if remote control of the system is desired, communication can be performed via a network.

[0126] Where network connectivity is desired, the system or method may accommodate any type of network. The network may be a single network or a combination of multiple networks. For example, the network may include the Internet and / or one or more intranets, terrestrial line networks, wireless networks and / or other suitable types of communication networks. In another example, the network may include a wireless telecommunications network adapted to communicate with other communication networks such as the Internet (e.g., a mobile phone network). For example, the network may include a computer network utilizing the TCP / IP protocol (including protocols based on the TCP / IP protocol such as HTTP, HTTPS, or FTP).

[0127] The embodiments described herein are illustrative without intentional loss of generality. Further variations, modifications, and combinations thereof are conceived and will be recognized by those skilled in the art. Accordingly, the above detailed description is not intended to limit the scope, applicability, or configuration of the claimed subject matter.

Claims

1. A computer implementation method for blood flow imaging, To obtain CT or MRI image data including multiple corresponding images that capture at least part of one or both of the contrast agent increase and decrease phases in the target cardiovascular system, Extracting a first image feature of the measured time-contrast enhancement curve from the CT or MRI image data, Providing the first image feature and at least one non-image feature to a machine learning model to generate a predicted value of the area under the simulated time-induced contrast effect curve of the contrast agent in the target cardiovascular system, wherein the predicted value simulates and generates a second set of image acquisition parameters different from a first set of image acquisition parameters used to acquire the CT or MRI image data. Convert the predicted value of the area under the simulated time-time contrast effect curve into the sum of the time-time contrast agent concentration products in the target cardiovascular system. The blood flow characteristics in the target cardiovascular system are determined based on the ratio of the mass of the contrast agent in the target cardiovascular system to the sum of the contrast agent concentration-time products in the target cardiovascular system. Computer implementation methods including

2. The method according to claim 1, further comprising including baseline data points extracted from the CT or MRI image data in the measured time-time contrast effect curve, and removing noise from the measured time-time contrast effect curve, wherein the CT or MRI image data includes at least one image capturing the target cardiovascular system before the entry of the contrast agent in order to provide the baseline data points.

3. The method according to claim 1, wherein the machine learning model is trained with training inputs of the image features of the measured time-contrast effect curve and at least one non-image feature, and is associated with the ground truth value of the expected area under the simulated time-contrast effect curve.

4. The method according to claim 1, wherein the image feature is at least a partial time-contrast effect curve, and extracting the image feature includes generating the at least partial time-contrast effect curve of the contrast agent based on the image data, wherein the at least partial time-contrast effect curve has at least an upward or downward slope.

5. The method according to claim 1, wherein the image feature is a measurement of the area under at least a partial time-contrast effect curve, and extracting the image feature includes generating the at least partial time-contrast effect curve of the contrast agent based on the image data, wherein the at least partial time-contrast effect curve has at least an upward or downward slope, and calculating the area under the at least partial time-contrast effect curve.

6. The method according to claim 1, further comprising providing a second image feature based on the contrast effect of signal intensity, the thickness of the wall of the target cardiovascular system, the size of the target cardiovascular system, the diameter of the target cardiovascular system, the morphology of the target cardiovascular system, the location of the sampling site in the target cardiovascular system, or the degree of stenosis in the target cardiovascular system.

7. The method according to claim 1, wherein the non-image features are based on age, sex, weight, heart rate, blood pressure, X-ray tube voltage, X-ray tube current, gradient pulse sequence, medical condition, contrast agent injection rate, contrast agent volume, or contrast agent concentration.

8. The method according to claim 1, wherein the first set of image acquisition parameters differs from a simulated second set of image acquisition parameters by at least one parameter selected from the group consisting of the scan axis direction relative to the longitudinal axis of the target cardiovascular system, the anatomical position of the scan, the hyperemia or resting state of the subject, the duration of the scan, the X-ray tube voltage, the X-ray tube current, the gradient pulse sequence, the contrast agent injection rate, the contrast agent volume, or the contrast agent concentration.

9. Determining the aforementioned blood flow characteristics means The time rate of change of contrast agent mass in the target cardiovascular system is determined based on the time rate of change of signal enhancement measured from at least a partial time-contrast effect curve, a coefficient for converting signal enhancement to contrast agent concentration, a predetermined volume value of the injected contrast agent, and the time value of the duration for which the signal intensity is higher than a predetermined baseline threshold level. Determining the density of the contrast agent in the target cardiovascular system based on the ratio defined in claim 1, the time value of the duration for which the signal intensity is higher than a predetermined baseline threshold level, and the mass of the contrast agent in the target cardiovascular system, To determine the area measured from the image data of the cardiovascular system of the subject. The method according to claim 1, comprising determining the absolute velocity using the Reynolds transport theorem.

10. Determining the aforementioned blood flow characteristics means [Math 1] The method according to any one of claims 1 to 13, comprising determining the flow pressure by applying Bernoulli's formula, which is expressed as:

11. A system for blood flow imaging, A memory for storing CT or MRI image data, which includes multiple corresponding images capturing at least part of one or both of the contrast agent increase and decrease phases in the target cardiovascular system, A CT or MRI image processing component for extracting a first image feature of the time-contrast effect curve measured from the CT or MRI image data, A machine learning model for generating a predicted value of the area under a simulated time-induced contrast effect curve of the contrast agent in the target cardiovascular system by inputting the first image feature and at least one non-image feature into the machine learning model, wherein the predicted value simulates a second set of image acquisition parameters different from a first set of image acquisition parameters used to acquire the CT or MRI image data, A processor that communicates with the CT or MRI image processing component and the machine learning model, converts the predicted value of the area under the simulated time-time contrast effect curve into the sum of the time-time contrast agent concentration products in the target cardiovascular system, and executes instructions to determine the blood flow characteristics in the target cardiovascular system based on the ratio of the mass of the contrast agent in the target cardiovascular system to the sum of the time-time contrast agent concentration products in the target cardiovascular system. A system that includes this.

12. The system according to claim 11, further comprising: including baseline data points extracted from the CT or MRI image data in the measured time-time contrast effect curve; and the processor executing instructions to remove noise from the measured time-time contrast effect curve, wherein the CT or MRI image data includes at least one image capturing the target cardiovascular system before the entry of the contrast agent in order to provide the baseline data points.

13. The system according to claim 11, wherein the machine learning model is trained with training inputs of the image features of the measured time-contrast effect curve and at least one non-image feature, and is associated with the ground truth value of the expected area under the simulated time-contrast effect curve.

14. The system according to claim 11, wherein the image feature is at least a partial time-contrast effect curve, and extracting the image feature includes generating the at least partial time-contrast effect curve of the contrast agent based on the image data, the at least partial time-contrast effect curve having at least an upward or downward slope.

15. The system according to claim 11, wherein the image feature is a measurement of the area under at least a partial time-contrast effect curve, and extracting the image feature includes generating the at least partial time-contrast effect curve of the contrast agent based on the image data, wherein the at least partial time-contrast effect curve has at least an upward or downward slope, and calculating the area under the at least partial time-contrast effect curve.

16. The system according to claim 11, further comprising providing a second image feature based on the contrast effect of signal intensity, the thickness of the wall of the target cardiovascular system, the size of the target cardiovascular system, the diameter of the target cardiovascular system, the morphology of the target cardiovascular system, the location of the sampling site in the target cardiovascular system, or the degree of stenosis in the target cardiovascular system.

17. The system according to claim 11, wherein the non-image features are based on age, sex, weight, heart rate, blood pressure, X-ray tube voltage, X-ray tube current, gradient pulse sequence, medical condition, contrast agent injection rate, contrast agent volume, or contrast agent concentration.

18. The system according to claim 11, wherein the first set of image acquisition parameters differs from a simulated second set of image acquisition parameters by at least one parameter selected from the group consisting of the scan axis direction relative to the longitudinal axis of the target cardiovascular system, the anatomical position of the scan, the hyperemia or resting state of the subject, the duration of the scan, the X-ray tube voltage, the X-ray tube current, the gradient pulse sequence, the contrast agent injection rate, the contrast agent volume, or the contrast agent concentration.

19. The processor executes instructions for determining the blood flow characteristics, and the determination is The time rate of change of contrast agent mass in the target cardiovascular system is determined based on the time rate of change of signal enhancement measured from at least a partial time-contrast effect curve, a coefficient for converting signal enhancement to contrast agent concentration, a predetermined volume value of the injected contrast agent, and the time value of the duration for which the signal intensity is higher than a predetermined baseline threshold level. Determining the density of the contrast agent in the target cardiovascular system based on the ratio defined in claim 11, the time value of the duration for which the signal intensity is higher than a predetermined baseline threshold level, and the mass of the contrast agent in the target cardiovascular system, To determine the area measured from the image data of the cardiovascular system of the subject. The system according to claim 11, comprising determining the absolute velocity using the Reynolds transport theorem.

20. The processor executes instructions for determining the blood flow characteristics, and the determination is [Math 2] The system according to claim 11, comprising determining the flow pressure by applying Bernoulli's equation, which is expressed as:

21. A computer implementation method for blood flow imaging, To obtain CT or MRI image data including multiple corresponding images that capture at least part of one or both of the contrast agent increase and decrease phases in the target cardiovascular system, Extracting at least one image feature from the CT or MRI image data, The method involves providing a machine learning model with the at least one image feature and the at least one associated non-image feature to generate a predicted value of the area under the time-enhancement curve of the contrast agent within the target cardiovascular system, wherein the machine learning model is trained on the training input of the at least one image feature together with the at least one non-image feature, and generates the area under the time-enhancement curve as ground truth. Convert the predicted value of the area under the time-dependent contrast effect curve into the time-dependent rate of change of the contrast agent concentration in the target cardiovascular system. The blood flow characteristics in the target cardiovascular system are determined based on the ratio of the mass of the contrast agent in the target cardiovascular system to the rate of change of the contrast agent concentration in the target cardiovascular system over time. Computer implementation methods including

22. The method according to claim 21, further comprising extracting baseline data points from the CT or MRI image data before extracting the at least one image feature, wherein the CT or MRI image data includes at least one image capturing the cardiovascular system of the subject before the entry of the contrast agent in order to provide the baseline data points.

23. The method according to claim 21, wherein the machine learning model is trained with training inputs of at least one image feature extracted from a time-contrast effect curve generated from the image data and at least one non-image feature including at least heart rate or blood pressure, and is associated with ground truth values ​​of the expected area under the simulated time-contrast effect curve.

24. The method according to claim 21, wherein the image feature is at least a partial time-contrast effect curve, and extracting the image feature includes generating the at least partial time-contrast effect curve of the contrast agent based on the image data, wherein the at least partial time-contrast effect curve has at least an upward or downward slope.

25. The method according to claim 21, wherein the image feature is a measurement of the area under at least a partial time-contrast effect curve, and extracting the image feature includes generating the at least partial time-contrast effect curve of the contrast agent based on the image data, wherein the at least partial time-contrast effect curve has at least an upward or downward slope, and calculating the area under the at least partial time-contrast effect curve.

26. The method according to claim 21, further comprising providing a second image feature based on the contrast effect of signal intensity, the thickness of the wall of the target cardiovascular system, the size of the target cardiovascular system, the diameter of the target cardiovascular system, the morphology of the target cardiovascular system, the location of the sampling site in the target cardiovascular system, or the degree of stenosis in the target cardiovascular system.

27. The method according to claim 21, wherein the at least one non-image feature includes a first non-image feature which is heart rate or blood pressure, and a second non-image feature which is based on age, sex, weight, heart rate, blood pressure, X-ray tube voltage, X-ray tube current, gradient pulse sequence, medical condition, contrast agent injection rate, contrast agent volume, or contrast agent concentration.

28. The method according to claim 21, wherein the predicted values ​​simulate a second set of image acquisition parameters different from a first set of image acquisition parameters used to acquire the CT or MRI image data, using at least one parameter selected from the group consisting of the scan axis direction relative to the longitudinal axis of the target cardiovascular system, the anatomical position of the scan, the hyperemia or resting state of the subject, the duration of the scan, the X-ray tube voltage, the X-ray tube current, the gradient pulse sequence, the contrast agent injection rate, the contrast agent volume, or the contrast agent concentration.

29. The method according to claim 21, wherein the scan for capturing the image data is a perfusion scan, a bolus tracking (BT) scan, a test bolus (TB) scan, a diagnostic angiography scan, or a combination thereof.

30. The method according to claim 21, wherein determining the blood flow characteristics includes determining the absolute flow velocity using the Reynolds transport theorem, determining the flow pressure by applying Bernoulli's equation, or determining the flow rate by applying the indicator dilution principle.

31. A computer implementation method for blood flow imaging, To obtain image data including multiple corresponding images that capture at least a portion of either or both of the contrast agent increase and decrease phases in the target cardiovascular system, The image data is provided to a machine learning model to predict the area under the time-induced contrast effect curve of the contrast agent within the target cardiovascular system. Selecting the target region within the target cardiovascular system in the aforementioned image data, Based on the area under the time-based contrast-enhancing effect curve, the blood flow characteristics through the target region are determined. Computer implementation methods including

32. The method according to claim 31, wherein the image data includes an increasing phase or a decreasing phase.

33. The method according to claim 32, wherein the image data includes both an increasing phase and a decreasing phase.

34. The method according to claim 33, wherein the image data includes at least one image capturing the target cardiovascular system before the entry of the contrast agent.

35. The method according to any one of claims 31 to 34, wherein the image data is a pseudo-truncate dynamic perfusion scan.

36. The method according to claim 35, wherein the pseudo-truncate dynamic perfusion scan is a collection of bolus tracking (BT) scans that capture images at multiple time points and diagnostic angiography scans that capture images at a single time point.

37. The method according to any one of claims 31 to 34, wherein the image data is a genuine dynamic perfusion scan.

38. The method according to claim 37, wherein the authentic dynamic perfusion scan is an authentic truncated dynamic perfusion scan such that the scan duration covers less than the full increase and full decay phases of the first pass circulation of the contrast agent in the cardiovascular system of the subject, and the scan is a test bolus (TB) scan or a diagnostic angiography scan.

39. The method according to claim 37, wherein the true dynamic perfusion scan is a true full dynamic perfusion scan such that the scan duration covers the full increase and full decay phases of the first pass circulation of the contrast agent in the cardiovascular system of the subject, and the scan is a test bolus (TB) scan or a diagnostic angiography scan.

40. The method according to any one of claims 31 to 39, wherein the machine learning model is trained by extracting image features from the image data.

41. The method according to claim 40, wherein the image features are based on the contrast effect of signal intensity, the thickness of the wall of the target cardiovascular system, the size of the target cardiovascular system, the diameter of the target cardiovascular system, the morphology of the target cardiovascular system, the location of the sampling site in the target cardiovascular system, or the degree of stenosis in the target cardiovascular system.

42. The method according to any one of claims 31 to 39, wherein the machine learning model is trained by associating non-image features with the image data.

43. The method according to claim 42, wherein the non-image features are based on age, sex, weight, heart rate, blood pressure, X-ray tube voltage, X-ray tube current, gradient pulse sequence, medical condition, contrast agent injection rate, contrast agent volume, or contrast agent concentration.

44. The method according to any one of claims 31 to 43, wherein determining the blood flow characteristics includes determining the absolute flow velocity using the Reynolds transport theorem.

45. The method according to any one of claims 31 to 43, wherein determining the blood flow characteristics includes determining the flow pressure by applying Bernoulli's equation.

46. The method according to any one of claims 31 to 43, wherein determining the blood flow characteristics includes determining the flow rate by applying the indicator dilution principle.

47. The method according to any one of claims 31 to 46, further comprising obtaining scan data of the cardiovascular system of the subject from an X-ray-based scan or an MRI scan, and reconstructing image data based on the scan data.

48. The method according to claim 47, wherein the scan data is obtained from a CT scan.

49. The method according to claim 47, wherein the scan data is obtained from an MRI scan.

50. The method according to claim 47, wherein the scan data is obtained from a scan having an elapsed time of more than 5 seconds.

51. The contrast agent is administered to the subject, The subject is scanned to obtain the scan data. The method according to any one of claims 47 to 50, further comprising capturing at least a portion of the increasing or decreasing phase of the contrast agent in the cardiovascular system of the subject.

52. A computer implementation method for blood flow imaging based on predicting the area under the time-induced contrast effect curve, To obtain image data including multiple corresponding images that capture at least a portion of either or both of the contrast agent increase and decrease phases in the target cardiovascular system, The image data is provided to a machine learning model to predict the area under the time-induced contrast effect curve of the contrast agent within the target cardiovascular system. A computer implementation method comprising the above, wherein the predicted area under the time-time contrast effect curve represents the sum of the time products of contrast agent concentrations within the target cardiovascular system.

53. The method according to claim 52, wherein the image data includes an increasing phase or a decreasing phase.

54. The method according to claim 53, wherein the image data includes both an increasing phase and a decreasing phase.

55. The method according to claim 54, wherein the image data includes at least one image capturing the target cardiovascular system before the entry of the contrast agent.

56. The method according to any one of claims 52 to 55, wherein the image data is a pseudo-truncate dynamic perfusion scan.

57. The method according to claim 56, wherein the pseudo-truncate dynamic perfusion scan is a collection of bolus tracking (BT) scans that capture images at multiple time points and diagnostic angiography scans that capture images at a single time point.

58. The method according to any one of claims 52 to 55, wherein the image data is a genuine dynamic perfusion scan.

59. The method according to claim 58, wherein the authentic dynamic perfusion scan is an authentic truncated dynamic perfusion scan such that the scan duration covers less than the full increase and full decay phases of the first pass circulation of the contrast agent in the cardiovascular system of interest.

60. The method according to claim 58, wherein the true dynamic perfusion scan is a true full dynamic perfusion scan such that the scan duration covers the full increase and full decay phases of the first pass circulation of the contrast agent within the cardiovascular system of the subject.

61. The method according to any one of claims 52 to 60, wherein the machine learning model is trained by extracting image features from the image data.

62. The method according to claim 61, wherein the image features are based on the contrast effect of signal intensity, the thickness of the wall of the target cardiovascular system, the size of the target cardiovascular system, the diameter of the target cardiovascular system, the morphology of the target cardiovascular system, or the degree of stenosis in the target cardiovascular system.

63. The method according to any one of claims 52 to 60, wherein the machine learning model is trained by associating non-image features with the image data.

64. The method according to claim 63, wherein the non-image features are based on age, sex, weight, heart rate, blood pressure, X-ray tube voltage, X-ray tube current, gradient pulse sequence, medical condition, contrast agent injection rate, or contrast agent concentration.

65. The method according to any one of claims 52 to 64, further comprising determining blood flow characteristics by determining the absolute flow velocity based on the predicted area under the time-contrast effect curve and the Reynolds transport theorem.

66. The method according to any one of claims 52 to 64, further comprising determining blood flow characteristics by determining the flow pressure based on the predicted area under the time-contrast effect curve and Bernoulli's equation.

67. The method according to any one of claims 52 to 66, wherein the plurality of corresponding images are six or more images.

68. The method according to any one of claims 52 to 67, further comprising obtaining scan data of the cardiovascular system of the subject from an X-ray-based scan or an MRI scan, and reconstructing image data based on the scan data.

69. The method according to claim 68, wherein the scan data is obtained from a CT scan.

70. The method according to claim 68, wherein the scan data is obtained from an MRI scan.

71. The method according to claim 68, wherein the scan data is obtained from a scan having an elapsed time of more than 5 seconds.

72. The contrast agent is administered to the subject, The subject is scanned to obtain the scan data. The method according to any one of claims 68 to 71, further comprising the scan data capturing at least a portion of the increasing or decreasing phase of the contrast agent in the cardiovascular system of the subject.

73. The method according to any one of claims 1 to 10 or 21 to 72, further comprising determining the clinical state of the cardiovascular system of the subject based on the area or blood flow characteristics under the time-time contrast effect curve.

74. A system for blood flow imaging, A memory for storing image data, including multiple corresponding images that capture at least part of one or both of the contrast agent increase and decrease phases in the target cardiovascular system. A processor configured to perform the steps of the method according to any one of claims 1 to 10 or 21 to 73. A system that includes this.

75. A computer-readable medium for embodying a computer-readable code for performing the method according to any one of claims 1 to 10 or 21 to 73.

Images