Computer-implemented methods, systems and non-transitory computer-readable media for determination of blood flow characteristics of patient

Abstract

To provide favorable systems and methods for determining blood flow characteristics of a patient.SOLUTION: One method includes: receiving, in an electronic storage medium, patient-specific image data of at least a portion of vasculature of the patient having geometric features at one or more points; generating a patient-specific reduced order model from the received image data, the patient-specific reduced order model comprising estimates of impedance values and a simplification of the geometric features at the one or more points of the vasculature of the patient; creating a feature vector comprising the estimates of impedance values and the geometric features for each of the one or more points of the patient-specific reduced order model; and determining blood flow characteristics at the one or more points of the patient-specific reduced order model using a machine learning algorithm trained to predict blood flow characteristics based on the created feature vectors at the one or more points.SELECTED DRAWING: None

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to U.S. Provisional Patent Application No. 62 / 396,965, filed September 20, 2016, the entire disclosure of which is incorporated herein by reference.

[0002] Various embodiments of the present disclosure relate generally to diagnosis and treatment planning of vasculature(s). More specifically, certain embodiments of the present disclosure relate to systems and methods for estimating blood flow characteristics using reduced order models and / or machine learning. [Background technology]

[0003] Blood flow within coronary arteries can provide useful information, including the presence or degree of ischemia, the presence or degree of blood perfusion to the myocardium, and the like. Because direct measurement of blood flow within small arteries can be difficult, blood flow can be simulated by solving the Navier-Stokes equations for patient-specific three-dimensional (3D) geometries derived from medical imaging data, including cardiac computed tomography (CT) scans, magnetic resonance imaging (MRI), ultrasound, and the like. To expedite the solution process, the 3D geometries can be simplified to a one-dimensional framework of centerlines parameterized by area or radius, and blood flow properties (e.g., pressure, flow rate, etc.) can be calculated along these centerlines, for example, by solving the Navier-Stokes equation simplifications. While these techniques can enable significantly faster computation of solutions to the Navier-Stokes equations, they may not be as accurate as solving the Navier-Stokes equations for three-dimensional geometries. Methods involving simplification to 1D geometries are not sufficiently accurate, and there is a need for methods that can provide more precise and accurate calculations of blood flow properties in localized regions of an anatomical model. Such a desired method can significantly improve computation time while maintaining accuracy. There is also a need for a method for utilizing these models to determine optimal geometric parameterizations, which will provide an optimal solution and / or improve knowledge of the geometric characteristics of the patient's anatomy, thereby improving medical images. Summary of the Invention [Means for solving the problem]

[0004] Described below are various embodiments of the present disclosure of systems and methods for estimating blood flow characteristics using reduced order models and / or machine learning.

[0005] One method includes receiving, in an electronic storage medium, patient-specific image data of at least a portion of a patient's vasculature having geometric features at one or more points; generating a patient-specific reduced-order model from the received image data, the patient-specific reduced-order model including estimates of impedance values ​​and simplifications of the geometric features at the one or more points of the patient's vasculature; creating a feature vector for each of the one or more points of the patient-specific reduced-order model including the estimates of the impedance values ​​and the geometric features; and determining blood flow characteristics at the one or more points of the patient-specific reduced-order model using a machine learning algorithm trained to predict blood flow characteristics based on the generated feature vector at the one or more points.

[0006] According to another embodiment, a system for estimating blood flow characteristics using reduced order models and / or machine learning includes a storage device storing instructions for estimating blood flow characteristics using reduced order models and / or machine learning; and a processor configured to: receive, in an electronic storage medium, patient-specific image data of at least a portion of a patient's vasculature having geometric features at one or more points; generate a patient-specific reduced order model from the received image data, the patient-specific reduced order model including estimates of impedance values ​​and simplifications of the geometric features at the one or more points of the patient-specific reduced order model; create, for each of the one or more points of the patient-specific reduced order model, a feature vector including the estimates of the impedance values ​​and the geometric features; and determine the blood flow characteristics at the one or more points of the patient-specific reduced order model using a machine learning algorithm trained to predict blood flow characteristics based on the feature vector created at the one or more points.

[0007] According to another embodiment, in a non-transitory computer-readable medium for use on a computer system containing computer-executable programming instructions for estimating blood flow characteristics using reduced-order models and / or machine learning, the method includes: receiving, in an electronic storage medium, patient-specific image data of at least a portion of a patient's vasculature having geometric features at one or more points; generating a patient-specific reduced-order model from the received image data, the patient-specific reduced-order model including estimates of impedance values ​​and simplifications of the geometric features at the one or more points of the patient-specific reduced-order model; creating, for each of the one or more points of the patient-specific reduced-order model, a feature vector including the estimates of the impedance values ​​and the geometric features; and determining the blood flow characteristics at the one or more points of the patient-specific reduced-order model using a machine learning algorithm trained to predict blood flow characteristics based on the feature vector created at the one or more points. The present specification also provides, for example, the following items: (Item 1) receiving, in an electronic storage medium, the patient-specific image data of at least a portion of the patient's vasculature having geometric characteristics at one or more points; generating a patient-specific reduced-order model from the received image data, the patient-specific reduced-order model including estimates of impedance values ​​and simplifications of the geometric features at the one or more points of the patient's vasculature; creating a feature vector for each of the one or more points of the patient-specific reduced-order model, the feature vector including an estimate of the impedance value and a geometric feature; determining blood flow characteristics at the one or more points of the patient-specific reduced-order model using a machine learning algorithm trained to predict blood flow characteristics based on the created feature vector at the one or more points; a computer-implemented method for determining the blood flow characteristics of the patient, the method comprising: (Item 2) determining blood flow characteristics at the one or more points of the patient-specific reduced-order model using a machine learning algorithm trained to predict blood flow characteristics based on the created feature vector at the one or more points; receiving, for each of a plurality of individuals, an anatomical model of the individual-specific vasculature having known values ​​of blood flow properties at one or more points of the individual-specific anatomical model and having one or more geometric features at the one or more points of the individual-specific anatomical model corresponding to the vasculature of the individual; forming, for each of the plurality of individuals having known values ​​of the blood flow characteristic at the one or more points, a feature vector including (i) information regarding the location of the one or more points, and (ii) geometric features at the one or more points; for each of the plurality of individuals having a known value of the blood flow characteristic at the one or more points, associating the feature vector with the known value of the blood flow characteristic at the one or more points; training the machine learning algorithm using the associated feature vectors to predict values ​​of the blood flow properties at one or more points of the vasculature from a feature vector comprising geometric features at the one or more points; using the trained machine learning algorithm to determine blood flow characteristics at the one or more points of the patient-specific reduced-order model corresponding to the patient's vasculature; and 2. The computer-implemented method of claim 1, comprising: (Item 3) each feature vector further comprising physiological and / or phenotypic parameters of the patient at one or more points of the patient-specific reduced order model; 2. The computer-implemented method of claim 1, wherein the machine learning algorithm is trained to predict values ​​of the blood flow characteristics at one or more points of the patient's vasculature from a feature vector comprising geometric features and physiological and / or phenotypic parameters at one or more points of the patient's vasculature. (Item 4) 4. The computer-implemented method of claim 3, wherein the physiological and / or phenotypic parameters include one or more of systolic and diastolic blood pressure, heart rate, hematocrit, blood pressure, blood viscosity, the patient's age, the patient's sex, the patient's height, the patient's weight, the patient's lifestyle characteristics, and delivered tissue mass. (Item 5) determining blood flow characteristics at the one or more points of the patient-specific reduced-order model using a machine learning algorithm trained to predict blood flow characteristics based on the created feature vector at the one or more points; receiving a patient-specific anatomical model based on the patient-specific image data of the patient's vasculature, the patient-specific anatomical model having geometric characteristics at one or more points of the patient-specific anatomical model that correspond to the patient's vasculature; applying boundary conditions to locations on the patient-specific anatomical model to simulate blood flow through the patient-specific anatomical model; determining values ​​for blood flow characteristics at one or more points of the patient-specific anatomical model from a simulation of blood flow using computational fluid dynamics (CFD); forming a feature vector comprising: (i) information regarding the location of the one or more points of the patient-specific anatomical model; and (ii) geometric features of the received patient-specific anatomical model at the one or more points; Associating the feature vector with the determined value of the blood flow characteristic at the one or more points of the patient-specific anatomical model; training a machine learning algorithm using the associated feature vectors to predict values ​​of the blood flow characteristics at one or more points of the patient's vasculature from a feature vector comprising geometric features at the one or more points; using the trained machine learning algorithm to determine blood flow characteristics at the one or more points of the patient-specific reduced-order model corresponding to the patient's vasculature; and 2. The computer-implemented method of claim 1, comprising: (Item 6) 6. The computer-implemented method of claim 5, wherein applying boundary conditions comprises clipping the patient-specific anatomical model at locations where appropriate boundary conditions can be applied, the locations including inflow boundaries of blood flow, outflow boundaries of blood flow, and vessel walls. (Item 7) 2. The computer-implemented method of claim 1, further comprising: dividing the received patient-specific image data into one or more regions of the patient's vasculature, wherein a reduced-order model is generated for each of the one or more regions of the vasculature. (Item 8) 2. The computer-implemented method of claim 1, wherein the machine learning algorithm comprises one or more of a support vector machine (SVM), a multilayer perceptron (MLP), a multivariate regression (MVR), a neural network, a tree model classifier, and a weighted linear regression or logistic regression. (Item 9) 2. The computer-implemented method of claim 1, wherein the blood flow characteristics include one or more of blood pressure, fractional flow reserve (FFR), blood flow volume or velocity, velocity or pressure field, blood flow force, and organ and / or tissue perfusion characteristics. (Item 10) a data storage device storing instructions for determining a blood flow characteristic of a patient; receiving, in an electronic storage medium, patient-specific image data of at least a portion of the patient's vasculature having geometric characteristics at one or more points; generating a patient-specific reduced-order model from the received image data, the patient-specific reduced-order model including estimates of impedance values ​​and simplifications of geometric features at the one or more points of the patient's vasculature; creating a feature vector for each of the one or more points of the patient-specific reduced-order model, the feature vector including an estimate of the impedance value and a geometric feature; determining blood flow characteristics at the one or more points of the patient-specific reduced-order model using a machine learning algorithm trained to predict blood flow characteristics based on the created feature vector at the one or more points; a processor configured to execute instructions for performing a method comprising: 1. A system for determining blood flow characteristics of a patient, comprising: (Item 11) determining blood flow characteristics at the one or more points of the patient-specific reduced-order model using a machine learning algorithm trained to predict blood flow characteristics based on the created feature vector at the one or more points; receiving, for each of a plurality of individuals, an anatomical model of the individual-specific vasculature having known values ​​of blood flow properties at one or more points of the individual-specific anatomical model and having one or more geometric features at the one or more points of the individual-specific anatomical model corresponding to the vasculature of the individual; forming, for each of the plurality of individuals having known values ​​of the blood flow characteristic at the one or more points, a feature vector including (i) information regarding the location of the one or more points, and (ii) the geometric features at the one or more points; for each of the plurality of individuals having a known value of the blood flow characteristic at the one or more points, associating the feature vector with the known value of the blood flow characteristic at the one or more points; training the machine learning algorithm using the associated feature vectors to predict values ​​of the blood flow properties at one or more points of the vasculature from a feature vector comprising geometric features at the one or more points; using the trained machine learning algorithm to determine blood flow characteristics at the one or more points of the patient-specific reduced-order model corresponding to the patient's vasculature; and Item 11. The system according to item 10, comprising: (Item 12) determining blood flow characteristics at the one or more points of the patient-specific reduced-order model using a machine learning algorithm trained to predict blood flow characteristics based on the created feature vector at the one or more points; receiving a patient-specific anatomical model based on the received patient-specific image data of the patient's vasculature, the patient-specific anatomical model having geometric characteristics at one or more points of the patient-specific anatomical model that correspond to the patient's vasculature; applying boundary conditions to locations on the patient-specific anatomical model to simulate blood flow through the patient-specific anatomical model; determining values ​​for blood flow characteristics at one or more points of the patient-specific anatomical model from a simulation of blood flow through the patient-specific or population-derived anatomical model using computational fluid dynamics (CFD); forming a feature vector comprising: (i) information regarding the location of the one or more points of the patient-specific anatomical model; and (ii) the geometric features at the one or more points of the patient-specific anatomical model; Associating the feature vector with the determined value of the blood flow characteristic at the one or more points of the patient-specific anatomical model; training a machine learning algorithm using the associated feature vectors to predict values ​​of the blood flow characteristics at one or more points of the patient's vasculature from a feature vector comprising geometric features at the one or more points; and using the trained machine learning algorithm to determine blood flow characteristics at the one or more points of the patient-specific reduced-order model corresponding to the patient's vasculature. (Item 13) Item 13. The system of item 12, wherein applying boundary conditions includes clipping the patient-specific anatomical model at locations where appropriate boundary conditions can be applied, the locations including an inflow boundary of blood flow, an outflow boundary of blood flow, and a vessel wall. (Item 14) 11. The system of claim 10, further comprising: dividing the received patient-specific image data into one or more regions of the patient's vasculature, wherein a reduced-order model is generated for each of the one or more regions of the vasculature. (Item 15) Item 11. The system of item 10, wherein the machine learning algorithm comprises one or more of a support vector machine (SVM), a multilayer perceptron (MLP), a multivariate regression (MVR), a neural network, a tree model classifier, and a weighted linear regression or logistic regression. (Item 16) 11. The system of claim 10, wherein the blood flow characteristics include one or more of blood pressure, fractional flow reserve (FFR), blood flow or velocity, velocity or pressure field, blood flow force, and organ and / or tissue perfusion characteristics. (Item 17) 1. A non-transitory computer-readable medium storing instructions that, when executed by a computer, cause the computer to perform a method for determining blood flow characteristics of a patient, the method comprising: receiving, in an electronic storage medium, patient-specific image data of at least a portion of the patient's vasculature having geometric characteristics at one or more points; generating a patient-specific reduced-order model from the received image data, the patient-specific reduced-order model including estimates of impedance values ​​and simplifications of the geometric features at the one or more points of the patient's vasculature; creating a feature vector for each of the one or more points of the patient-specific reduced-order model, the feature vector including an estimate of the impedance value and a geometric feature; determining blood flow characteristics at the one or more points of the patient-specific reduced-order model using a machine learning algorithm trained to predict blood flow characteristics based on the created feature vector at the one or more points; The non-transitory computer-readable recording medium comprising: (Item 18) determining blood flow characteristics at the one or more points of the patient-specific reduced-order model using a machine learning algorithm trained to predict blood flow characteristics based on the created feature vector at the one or more points; receiving, for each of a plurality of individuals, an anatomical model of the individual-specific vasculature having known values ​​of blood flow properties at one or more points of the individual-specific anatomical model, the anatomical model having one or more geometric features at the one or more points of the individual-specific anatomical model corresponding to the vasculature of the individual; forming, for each of the plurality of individuals having known values ​​of the blood flow characteristic at the one or more points, a feature vector including (i) information regarding the location of the one or more points, and (ii) geometric features at the one or more points; for each of the plurality of individuals having a known value of the blood flow characteristic at the one or more points, associating the feature vector with the known value of the blood flow characteristic at the one or more points; training the machine learning algorithm using the associated feature vectors to predict values ​​of the blood flow properties at one or more points of the vasculature from a feature vector comprising geometric features at the one or more points; using the trained machine learning algorithm to determine blood flow characteristics at the one or more points of the patient-specific reduced-order model corresponding to the patient's vasculature; and Item 18. The non-transitory computer-readable medium of item 17, comprising: (Item 19) determining blood flow characteristics at the one or more points of the patient-specific reduced-order model using a machine learning algorithm trained to predict blood flow characteristics based on the created feature vector at the one or more points; receiving a patient-specific anatomical model based on the received patient-specific image data of the patient's vasculature, the patient-specific anatomical model having geometric characteristics at one or more points of the patient-specific anatomical model that correspond to the patient's vasculature; applying boundary conditions to locations on the patient-specific anatomical model to simulate blood flow through the patient-specific anatomical model; determining values ​​for blood flow characteristics at one or more points of the patient-specific anatomical model from a simulation of blood flow through the patient-specific anatomical model using computational fluid dynamics (CFD); forming a feature vector comprising: (i) information regarding the location of the one or more points of the patient-specific anatomical model; and (ii) geometric features of the patient-specific anatomical model at the one or more points; Associating the feature vector with the determined value of the blood flow characteristic at the one or more points of a patient-specific anatomical model; training a machine learning algorithm using the associated feature vectors to predict values ​​of the blood flow characteristics at one or more points of the patient's vasculature from a feature vector comprising geometric features at the one or more points; using the trained machine learning algorithm to determine blood flow characteristics at the one or more points of the patient-specific reduced-order model corresponding to the patient's vasculature; and Item 18. The non-transitory computer-readable medium of item 17, comprising: (Item 20) Item 18. The non-transitory computer-readable medium of item 17, wherein the machine learning algorithm comprises one or more of a support vector machine (SVM), a multilayer perceptron (MLP), a multivariate regression (MVR), a neural network, a tree model classifier, and a weighted linear regression or logistic regression.

[0008] Additional objects and advantages of the disclosed embodiments will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the disclosed embodiments. The objects and advantages of the disclosed embodiments will be realized and attained by means of the elements and combinations particularly pointed out in the appended claims.

[0009] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not intended to limit the scope of the disclosed embodiments as claimed.

[0010] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate various exemplary embodiments and, together with the description, serve to explain the principles of the disclosed embodiments. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 is a block diagram of an example system and network 100 for predicting or estimating blood flow characteristics using reduced order models and / or machine learning, in accordance with an example embodiment of the present disclosure. [Figure 2] FIG. 2 is a block diagram of a general method 200 for predicting or estimating blood flow characteristics using reduced order models and / or machine learning, according to an exemplary embodiment of the present disclosure. [Figure 3] FIG. 3 is a block diagram of a general method 300 for generating a reduced-order model from image data and using the reduced-order model to determine impedance values, according to an exemplary embodiment of the present disclosure. [Figure 4A] FIG. 4 is a block diagram of an example method 400A for training and running a machine learning algorithm for estimating blood flow characteristics using a reduced order model, according to an example embodiment of the present disclosure. [Figure 4B] FIG. 4B is a block diagram of an example method 400B for training and running a machine learning algorithm for estimating blood flow characteristics using a reduced order model, according to an example embodiment of the present disclosure. [Figure 4C] FIG. 4C is a block diagram of an example method 400C for training and running a machine learning algorithm for estimating blood flow characteristics using a reduced order model, according to an example embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0012] The steps described in the methods may be performed in any order or in conjunction with any other steps. It is also contemplated that one or more steps may be omitted to perform the methods described herein.

[0013] Reference will now be made in detail to the exemplary embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numbers will be used throughout the drawings to refer to the same or like parts.

[0014] Various embodiments of the present disclosure may provide systems and methods for estimating blood flow characteristics using reduced-order models and / or machine learning. For purposes of this disclosure, blood flow characteristics may include, but are not limited to, blood pressure, fractional flow reserve (FFR), blood flow rate or velocity, velocity or pressure field, blood flow force, and organ and / or tissue perfusion characteristics. At least some embodiments of the present disclosure may provide advantages, for example, through the use of reduced-order models, such as providing faster calculation of blood flow characteristics from image data, as well as ensuring more accurate calculation of blood flow characteristics by, for example, utilizing trained machine learning algorithms. It is contemplated that other models having simplified forms other than reduced-order models may be used instead of or in addition to reduced-order models to achieve these advantages.

[0015] Referring now to the drawings, FIG. 1 illustrates a block diagram of an exemplary system 100 and network for estimating blood flow characteristics using reduced-order models and / or machine learning, according to exemplary embodiments. Specifically, FIG. 1 illustrates multiple physicians 102 and third-party providers 104, any of which may be connected to an electronic network 100, such as the Internet, via one or more computers, servers, and / or handheld mobile devices. The physicians 102 and / or third-party providers 104 may create or otherwise obtain images of one or more patient anatomy. The physicians 102 and / or third-party providers 104 may also obtain any combination of information, including, but not limited to, patient-specific and / or baseline anatomical images, physiological measurements, and / or geometric and / or anatomical features of the patient's vessels of interest, blood flow characteristics, impedance values ​​of the vessels of interest, etc. In some embodiments, the physicians 102 and / or third-party providers 104 may also obtain baseline values ​​for blood flow characteristics as they relate to the reduced-order or lumped parameter models. For example, in the case of a reduced-order model that simplifies blood flow characteristics into a one-dimensional electrical circuit, the physician 102 and / or third-party provider 104 can obtain resistance, capacitance, and / or inductance values ​​from a library or lookup table of parameters based on the blood flow characteristics that can be simplified.

[0016] The physician 102 and / or third-party provider 104 can transmit anatomical images, physiological information, and / or information about the vessel of interest to the server system 106 over the electronic network 100. The server system 106 can include a storage device for storing images and data received from the physician 102 and / or third-party provider 104. The server system 106 can also include a processing device for processing the images and data stored in the storage device.

[0017] FIG. 2 illustrates a method 200 for estimating blood flow characteristics using reduced order models and / or machine learning, according to an exemplary embodiment of the present disclosure.

[0018] In some embodiments, step 202 of method 200 may include receiving a patient-specific anatomical model of the patient's vasculature, vasculature, or blood vessel of interest. In some embodiments, instead of the model, patient-specific image data may be received from the patient's vasculature, vasculature, or blood vessel of interest. The vasculature to which the vessel or vasculature of interest may belong may include a coronary vascular model, a cerebrovascular model, a peripheral vascular model, a hepatic vascular model, a renal vascular model, a visceral vascular model, or any vascular model that includes blood vessels supplying blood that are susceptible to stenotic lesions or plaque formation. In some embodiments, other patient data, such as measured blood flow characteristics and / or properties, of the patient's vasculature, vasculature, or blood vessel of interest may be received. The image data and / or blood flow characteristics and / or properties may be obtained non-invasively and / or invasively from the patient (e.g., via a scanning or medical device) or through a population study (e.g., based on similarity to the patient).

[0019] Step 204 may include cropping the patient-specific anatomical model where appropriate boundary conditions may be applied, such as in a region distal to the location of disease in an artery visible to the imaging device and encompassing one or more vessels identified from the anatomical information (e.g., the anatomical information received in step 202) to capture the vascular stenosis region.

[0020] Step 206 may include applying boundary conditions to the truncated patient-specific anatomical model to estimate blood flow characteristics. The estimated blood flow characteristics can provide approximations that can be used to generate a reduced-order model (e.g., as in 212A). In some embodiments, the applied boundary conditions can be used to ultimately solve for the blood flow characteristics using computational fluid dynamics (CFD) (e.g., 212B). The boundary conditions provide information about hemodynamics at the boundaries of the three-dimensional model, such as inflow boundaries or inlets, outflow boundaries or outlets, vessel wall boundaries, etc. An inflow boundary or inlet can include a boundary through which flow is directed into an anatomical structure of the three-dimensional model, such as the aorta. The inflow boundary can be assigned a predetermined value or field, e.g., for velocity, flow rate, pressure, or other property, such as by coupling a cardiac model and / or a lumped parameter model to the boundary. Flow rate at the aorta can be estimated by cardiac output, measured directly, or derived from the patient's mass using a scaling law. In some embodiments, aortic flow can be estimated by cardiac output using the method described in U.S. Patent No. 9,424,395, filed April 17, 2013 ("Method and system for sensitivity analysis in modeling blood flow characteristics"), which is incorporated herein by reference in its entirety.

[0021] For example, the net cardiac output (Q) is [ka] (Cardiac output) can be calculated from body surface area (BSA). Body surface area (BSA) is [ka] It can be calculated from height (h) and weight (w) as follows: Coronary blood flow (q cor) is the myocardial mass (m myo )from [ka] where: [ka] is the expansion factor. Therefore, the flow in the artery is [ka] It could be.

[0022] Similarly, alternatively, or additionally, step 210 may include determining blood flow characteristics for the truncated patient-specific anatomical model using CFD.

[0023] Accordingly, step 210 may include partitioning the truncated patient-specific anatomical model into one or more regions. This partitioning can be performed based on estimates of blood flow characteristics from applied boundary conditions (e.g., from step 206). Alternatively or additionally, step 210 may include partitioning the truncated patient-specific anatomical model into one or more regions based on blood flow characteristics determined using CFD (e.g., from step 212B described herein). In some embodiments, measurements of blood flow characteristics of at least a portion of a patient's vasculature, vasculature, or vessel of interest may also be received along with the patient-specific anatomical model of the vasculature, vasculature, or vessel of interest (e.g., as in step 202). In such embodiments, the truncated patient-specific anatomical model may be partitioned into one or more regions based on the measured blood flow characteristics. The model may be partitioned into different regions based on the flow characteristics, for example, (i) ostial bifurcation, (ii) non-ostial bifurcation, (iii) stenotic region, (iv) post-stenotic dilation region, (v) healthy region, etc. Each of these regions may be further divided into sub-regions having a predetermined length. In some embodiments, steps 204 and 210 may be combined (e.g., clipping at the region boundary), or one of the two steps may be skipped (e.g., using only a single region in the method).

[0024] Step 212A may include generating a reduced-order model for each of the one or more regions. The reduced-order model may parameterize the geometry using a set of radii along a centerline and may solve simplified Navier-Stokes equations by assuming a blood flow profile. Because the reduced-order model or lumped parameter model may inherently lack the geometric features of a more complex (e.g., 3D) model, the reduced-order model or lumped parameter model may represent a simplification of the geometric features of the more complex model ("simplified geometric features"), e.g., the geometric features are described in terms of one dimension. For example, the three-dimensional geometric features of perivascular stenosis may be simplified geometric features of vessel diameter reduction on the reduced-order model and / or lumped parameter model. Thus, the 3D geometry may be represented by a set of radii along the centerline of the reduced-order model and / or lumped parameter model. However, it is contemplated that the geometric features may be quantified and / or discretized from the original 3D anatomical structure or model, e.g., for purposes of creating a feature vector in a machine learning algorithm. Examples of geometric characteristics may include, but are not limited to, distance from the nearest bifurcation, distance from the ostium, minimum upstream diameter, and the like.

[0025] Step 214A may include estimating impedance values ​​for one or more points within each reduced-order model (or one or more reduced-order models from step 212). The impedance values ​​may include estimates of blood flow characteristics based on prior reduced-order models and / or lumped parameter models. For example, the blood flow characteristics may be estimated from boundary conditions or measured from the patient. In some embodiments, these reduced-order models may be based on geometric features of the region. In one embodiment, these reduced-order models may be further based on blood flow properties (e.g., viscosity, concentration, flow rate, etc.) and / or blood flow characteristics. In such an embodiment, measurements of blood flow characteristics for at least a portion of the patient's vasculature, vasculature, or vessel of interest may also be received (e.g., as in step 202) along with the patient-specific anatomical model of the vasculature, vasculature, or vessel of interest. Similarly, the patient's physiological and / or phenotypic parameters may also affect the estimation of the impedance values, and these parameters may also be received (e.g., in step 202). In some embodiments, the reduced-order model and / or lumped parameter model can simplify the anatomical model into a one-dimensional electrical circuit that represents the path of blood flow through the anatomical model. In such embodiments, impedance can be represented by a resistance (and other electrical characteristics) on the electrical circuit.

[0026] Steps 212A and 214A above may describe methods for estimating and / or determining impedance values ​​via reduced order models and / or lumped parameter models, while steps 212B and 214B may describe methods for estimating and / or determining impedance values ​​via CFD analysis.

[0027] For example, step 212B may include solving for blood flow characteristics to use CFD analysis or simulation. Step 212B may utilize boundary conditions applied (e.g., from 206) in the CFD analysis or simulation.

[0028] Additionally, step 212B may include calculating blood flow characteristics for each region of the entire system (e.g., represented by a patient-specific anatomical model of the patient's vasculature, vascular structure, or vessel of interest), a truncated patient-specific anatomical model, or one or more regions of the patient-specific anatomical model (divided in step 210). For example, step 212B may include solving equations governing blood flow for velocity and pressure using the boundary conditions applied in step 206. In one embodiment, step 212B may include calculating a blood velocity or flow field for one or more points or regions of the anatomical model using the assigned boundary conditions. This velocity or flow field may be the same field calculated by solving the blood flow equations using the physiological conditions and / or boundary conditions provided above. Step 212B may further include solving a scalar advection-diffusion equation governing blood flow at one or more locations of the patient-specific anatomical model.

[0029] Step 214B may include determining impedance values ​​for one or more points for each region of the one or more regions using the blood flow characteristics calculated using CFD. In some embodiments, the impedance values ​​may be approximations and / or simplifications from the blood flow characteristics solved using CFD (as opposed to estimates from reduced order and / or lumped parameter models in step 212A).

[0030] For steps 214A and / or 214B, the impedance values ​​may include, for example, resistance values, capacitance values, and / or inductance values.

[0031] Resistance can be constant, linear, or nonlinear, depending on, for example, the estimated flow rate through the corresponding segment of a blood vessel. For more complex geometries, such as stenoses, resistance can vary with flow rate. Resistance for various geometries can be determined based on computational analysis (e.g., finite difference, finite volume, spectral, lattice Boltzmann, particle-based, level set, isogeometric, or finite element methods, or other computational fluid dynamics (CFD) analytical methods), and multiple solutions from computational analyses performed under different flow and pressure conditions can be used to derive patient-specific, vessel-specific, and / or lesion-specific resistances. The results can be used to determine resistance for various types of features and geometries of any segment that can be modeled. As a result, deriving patient-specific, vessel-specific, and / or lesion-specific resistances as described above can enable a computer system to recognize and evaluate more complex geometries, such as asymmetric stenoses, multiple lesions, lesions at bifurcations and bifurcations, and tortuous vessels.

[0032] Capacitors can also be included as impedances and / or features in reduced-order or lumped parameter models. The capacitance can be determined, for example, based on the elasticity of the vessel wall of the corresponding segment. Inductors may also be included, and the inductance can be determined, for example, based on inertial effects associated with the acceleration or deceleration of the blood volume flowing through the corresponding segment.

[0033] Individual values ​​for resistance, capacitance, inductance, and other variables associated with other electrical components used in reduced-order and / or lumped-parameter models can be derived based on data from many patients, with similar vessel geometries having similar values. Thus, empirical models can be developed from large populations of patient-specific data to create libraries of values ​​corresponding to specific geometric features, which can be applied to similar patients in future analyses. Geometry can be matched between two different vessel segments, automatically selecting values ​​for the patient's segment or vessel of interest from a previous simulation.

[0034] Step 216 may include determining, for one or more points within each region, an error (e.g., difference) in the impedance values ​​of the reduced-order model from impedance values ​​determined using the blood flow characteristics calculated using CFD. Step 218 may include using the error in the impedance values ​​to train a machine learning algorithm to update the reduced-order model with an appropriate set of features. The machine learning regressor may be trained on the errors calculated in step 216. Thus, after solving the reduced-order model and calculating the errors with respect to the blood flow characteristics calculated using CFD, step 218 may include defining a set of features (e.g., geometric, clinical, flow-related, etc.) and mapping the features to the errors to estimate a better blood flow solution. The set of features may include geometric features (e.g., degree of stenosis, distance from ostium, distance from bifurcation, worst upstream stenosis, etc.), flow-related features (e.g., downstream boundary conditions), and / or features calculated directly from the reduced-order model (e.g., resistance). Different candidate machine learning algorithms or regressors (e.g., random decision forests, neural networks, multilayer perceptrons, etc.) may be utilized. The trained machine learning algorithm may be used to update the reduced-order model(s), for example, by determining new (or revised) impedance values. Step 220 may include determining blood flow characteristics using the updated reduced-order model. For example, the learned impedance may be used to estimate flow and pressure. Method 400A of FIG. 4A may describe the training of such a machine learning algorithm in further detail.

[0035] Alternatively, a regressor can be trained to predict an idealized shape that matches the CFD or measurements. This idealized geometry determined using a machine learning algorithm can be used as input to the system (e.g., optimizing simplified geometric features of a reduced-order model and / or lumped parameter model) or as part of a system that enables fast computation of the Navier-Stokes equations. For example, a regressor can be used to determine optimal parameterization and / or simplification of geometric features of a reduced-order model, allowing the reduced-order model to more accurately compute blood flow characteristics. Method 400B or FIG. 4B can describe the training of such a machine learning algorithm in further detail.

[0036] In some embodiments, the results of method 200 can be output to an electronic storage medium or a display. The results can include blood flow characteristics. The results can be visualized using a color map.

[0037] FIG. 3 is a block diagram of a general method 300 for generating a reduced order model from image data and determining impedance values ​​using the reduced order model, according to an exemplary embodiment of the present disclosure.

[0038] Step 302 of method 300 may include receiving image data (or anatomical images and / or information) encompassing a vessel or vascular region of interest. In such embodiments, the image data, anatomical images, and / or information may be stored on an electronic storage medium. The vessel of interest may include, for example, various vessels of the coronary vasculature. In other embodiments, vessels of other vasculature systems may also be captured, including, but not limited to, the coronary vasculature, cerebral vasculature, peripheral vasculature, hepatic vasculature, renal vasculature, visceral vasculature, or any vasculature having blood-supplying vessels susceptible to stenotic lesions or plaque formation. The anatomical images and / or information may be extracted from images and / or image data generated from a scanning imaging device (e.g., magnetic resonance (MR) imaging, computed tomography (CT) imaging, positron emission tomography (PET) imaging, x-ray imaging, etc.) and / or received from an electronic storage device (e.g., a hard drive).

[0039] Step 308 may include partitioning the model into one or more regions. In some embodiments, the partitioning may be based on estimates of anatomical and / or blood flow characteristics. For example, these anatomical and / or blood flow characteristics may include, but are not limited to, ostial bifurcation 310A, non-ostial bifurcation 310B, stenotic region(s) 310C, dilated region (e.g., post-stenosis) 310D, and healthy region(s) 310E. Other region partitioning schemes based on different region data characteristics may also be implemented. In some embodiments, estimates of blood flow characteristics may be obtained via steps 304A and 306, and / or 304B.

[0040] For example, step 304A may include generating a three-dimensional (3D) anatomical model encompassing the vessel of interest using the received image data. In other embodiments, a 2D anatomical model may be generated, or an anatomical model having a temporal dimension may be generated. The anatomical model may be generated from received or stored anatomical images and / or information encompassing the patient's vessel of interest (e.g., from step 302). In such embodiments, building the 3D anatomical model may include segmentation or related methods. Segmentation may occur, for example, by placing seeds based on extracted centerlines and using intensity values ​​from the image data to form one or more segmentation models (e.g., "threshold-based segmentation," as described in U.S. Patent No. 8,315,812, filed January 25, 2011, which is incorporated herein in its entirety). Segmentation can also occur by using intensity values ​​of the image data to find the edge (e.g., of a lumen), placing seeds, and expanding the seeds until the edge is reached (e.g., "threshold-based segmentation" as described in U.S. Patent No. 8,315,812, filed January 25, 2011, which is incorporated herein in its entirety). In some embodiments, a marching cubes algorithm can also be used for segmentation.

[0041] Step 306 may include determining blood flow characteristics of the vessel of interest using the 3D anatomical model.

[0042] Additionally or alternatively, the segmentation in step 308 may result from an estimate of the received blood flow characteristics. For example, step 304B may include receiving an estimate of the blood flow characteristics of the vessel of interest from, for example, population-derived data, patient studies, or measurements. Step 312 may include deriving a simplified geometry from the segmented image data of the region(s). In one embodiment, the geometry can be derived from the original 3D geometry of the image data received in step 302 or from the 3D model generated in step 304A. The geometry may be defined by centerline points and associated radii. In one embodiment, an optimal 1D geometry can be derived or learned, which will result in optimal performance of the 1D blood flow simulation compared to the 3D blood flow simulation.

[0043] Once the geometry is defined, each centerline point [ka] Impedance values ​​at can be calculated based on the defined geometric shape. Thus, step 314 may include determining one or more impedance values ​​for blood flow through the region(s). Impedance may include, for example, blood flow characteristics (e.g., pressure, flow rate, etc.) or similar representations or simplifications thereof (e.g., resistance, capacitance, inductance, etc.). Thus, step 314 may include determining one or more impedance values ​​for blood flow through the region(s). For example, hydrodynamic impedance may be estimated from data. In some cases, such data may be estimated from a 3D blood flow simulation (e.g., as in step 306 or steps 210B, 212B, and 214B of FIG. 2), or the data may be data from or derived from measurements of flow and pressure (e.g., as in step 304B). Impedance may include resistance to flow (ratio of blood pressure to flow velocity) 314A, the ability of the coronary artery to pulsate (from arterial elasticity) 314B, etc.

[0044] Each centerline point x i The pressure at, starting from a given aortic pressure,

number

[0045]

number

[0046]

number

[0047]

number

[0048]

number

[0049]

number

[0050]

number

[0051]

number

[0052]

number

[0053] [ka] can be determined in a variety of ways, for example: [ka] may be determined using any empirical model or selected as a predetermined value (e.g., zero) learned using machine learning.

[0054] Example errors in this reduced-order model (compared to a full 3D CFD calculation) can be classified into two main categories: (1) invalid assumptions in the reduced-order resistance model, and (2) geometric errors introduced by the geometry simplification process. The primary contribution to the first type of error can be attributed to the laminar flow assumptions underlying the accuracy of both resistance models. While the assumptions may work in healthy arterial regions away from the bifurcation location, highly non-laminar flow may exist both at the bifurcation location and within the dilated region after the stenosis. The second category of ROM error can depend on the method used to create the simplified geometry, e.g., creating centerline points and defining a radius at each point.

[0055] Overall, the system can be expressed as:

[0056]

number

[0057] During the ceremony, [ka] can be calculated from the data, which may depend on the flow rate. In some embodiments, the resistance difference can be approximated using a machine learning approximation of, for example, the following equation:

[0058]

number

[0059] In step 316, one or more impedance values ​​determined for blood flow through the region(s) may be integrated and / or incorporated into a reduced order model.

[0060] The integrated reduced-order model may be used to estimate and / or determine blood flow characteristics described herein (e.g., as in method 400C of FIG. 4C ). The integrated reduced-order model may also be used to plan treatment of a lesion, for example, by modifying impedance values ​​or geometric features (simplified from the 3D image data) and / or by simulating blood flow. In some embodiments, the integrated reduced-order model may be output to an electronic storage medium and / or a display on the server system 106.

[0061] 4A-4C are block diagrams of example methods 400A-400C for training and applying a machine learning algorithm for blood flow characteristic estimation using a reduced-order model, according to an example embodiment of the present disclosure. Furthermore, FIG. 4A illustrates an example method 400A for training a machine learning algorithm to predict blood flow characteristics of a model (e.g., a reduced-order model or a lumped parameter model) derived from image data. FIG. 4B illustrates an example method 400B for training a machine learning algorithm to predict geometric features of a model derived from image data (e.g., simplified geometric features of a reduced-order model or a lumped parameter model) using the blood flow characteristics. FIG. 4C illustrates an example method 400C for applying the trained machine learning algorithm to determine blood flow characteristics (e.g., more accurately) using a reduced-order model or a lumped parameter model, or to determine geometric features of a model.

[0062] Referring now to FIG. 4A , which discloses an exemplary method for training a machine learning algorithm to predict blood flow characteristics, step 402 may include receiving image data of at least one vasculature of interest of a patient in an electronic storage medium. The received image data may be used, for example, to generate a patient-specific 3D anatomical model of at least the vasculature of interest (e.g., as in 404C). Step 406C may include applying boundary conditions to the generated 3D anatomical model. Furthermore, step 408C may include determining blood flow characteristics at one or more points of the patient-specific 3D anatomical model using computational fluid dynamics (CFD) (e.g., using the Navier-Stokes equations). In some embodiments, steps 406C and 408C may use the methods described in steps 206, 210B, and 212B of method 200, as shown in FIG. 2 .

[0063] The received image data can also be used to determine and / or receive a corresponding population-derived 3D anatomical model of at least the vasculature of interest (e.g., as in step 404B). In such embodiments, step 408B can include receiving blood flow characteristics at one or more points of the population-derived 3D anatomical model.

[0064] In some embodiments, independently or using the received image data, step 404A may include receiving, from each of the plurality of individuals, an individual-specific 3D anatomical model of at least the vasculature of interest. In such embodiments, step 408A may include receiving, from each of the plurality of individuals, blood flow characteristics at one or more points of the individual-specific 3D anatomical model.

[0065] It is contemplated that in some embodiments, physiological and / or phenotypic parameters for the anatomical model may also be received, for example, from the patient in step 404C, from the population-derived data from 404B, or from each of the multiple individuals in 404A. These physiological and / or phenotypic parameters may be included in the features used to form the feature vector in step 410A described herein.

[0066] For steps 404A and 404B, the type of anatomical model selected or the individual of the received model may be guided by the original received image data (e.g., from step 402) or the patient. Thus, steps 404A-404C are examples of steps for receiving a 3D anatomical model for developing a feature vector as part of a domain for a training data set for a machine learning algorithm. Steps 408A-408C can be examples of steps for receiving blood flow characteristics for one or more points of the 3D anatomical model from steps 404A-404C, where the blood flow characteristics determined in steps 408A-408C may serve as the scope of the training data set for the machine learning algorithm.

[0067] Step 410A may include creating a feature vector including one or more features at one or more points of the patient-specific 3D anatomical model (e.g., following steps 404C, 406C, and / or 408C). In some embodiments, a feature vector can be formed for one or more points of the population-derived 3D anatomical model (e.g., following steps 404B and / or 408B). Alternatively or additionally, a feature vector may be formed for one or more points of each or some of the individual-specific 3D anatomical models (e.g., following steps 404A and / or 408A). A feature vector may be formed for multiple points within the 3D anatomical model for which blood flow characteristics have been received or determined. Features in the feature vector may include a numerical description of the patient-specific geometry at the time and estimates of physiological or phenotypic parameters of the patient or individual for whom the anatomical model is received. Physiological and / or phenotypic parameters can include, but are not limited to, for example, (i) person characteristics such as patient age, sex, height, and weight; (ii) disease characteristics such as presence or absence of diabetes or myocardial infarction, malignant and rheumatic status, and peripheral vascular status; (iii) lifestyle characteristics such as current medication / drug status and smoker / non-smoker; (iv) hemodynamic forces such as axial plaque stress and wall shear stress; (v) systolic and diastolic blood pressure; and (vi) blood properties including plasma, red blood cells (erythrocytes), hematocrit, white blood cells (leukocytes), and platelets (thrombocytes), viscosity, and yield stress. A feature vector can include both global and regional physiological or phenotypic parameters, where for global parameters, all points have the same numerical value, and for regional parameters, the value(s) can vary at different points within the feature vector. The server system 106 can then associate this feature vector with the received or simulated value of the blood flow characteristic at this point.Thus, step 412A may include associating feature vectors with blood flow characteristics at one or more points of each of the patient-specific 3D anatomical model, the population-derived 3D anatomical model, and / or the individual-specific 3D anatomical model.

[0068] Step 414A may include training a machine learning algorithm to predict blood flow characteristics at one or more points of a model derived from image data from feature vector(s) at one or more points. The training may use the relevant features from step 412A, for example, to determine relevance and / or feature weights. Examples of machine learning algorithms that can perform this task include support vector machines (SVMs), neural networks, multilayer perceptrons (MLPs), multivariate regression (MVR) (e.g., weighted linear or logistic regression), and / or other supervised machine learning techniques known to those skilled in the art. The server system 106 may then store the results of the machine learning algorithm (e.g., feature weights) in a digital representation (e.g., in the memory of a computing device such as a computer, laptop, DSP, server, or in digital storage (e.g., a hard drive, network drive)). The stored feature weights may define the degree to which features (e.g., geometric descriptions, boundary conditions, physiological and / or phenotypic parameters, anatomical features, etc.) can predict blood flow and / or blood pressure at one or more points of the model or the system represented by the model.

[0069] Step 416A may include outputting the trained machine learning algorithm (e.g., to an electronic storage medium). The trained machine learning algorithm can be used in method 400C of FIG. 4C to determine blood flow characteristics from a model having a simplified geometry (e.g., a reduced-order model or a lumped parameter model), for example.

[0070] 4B illustrates an exemplary method 400B for training a machine learning algorithm to predict geometric features from a feature vector containing blood flow characteristics at one or more points of a model derived from image data. Steps 402, 408A-C, 406, and / or 408A-C of method 400A illustrated in FIG. 4 can be performed before the steps of method 400B illustrated in FIG. 4B.

[0071] Step 410B may include creating a feature vector including blood flow characteristics at one or more points of the patient-specific 3D anatomical model (e.g., following steps 404C, 406C, and / or 408C). In some embodiments, a feature vector can be formed for one or more points of the population-derived 3D anatomical model (e.g., following steps 404B and / or 408B). Alternatively or additionally, a feature vector can be formed for one or more points of each or some of the individual-specific 3D anatomical models (e.g., following steps 404A and / or 408A). A feature vector can be formed for points in the 3D anatomical model with known geometric features. These geometric features may include a numerical description of the patient-specific geometry at that point. The feature vector can include both global and regional physiological or phenotypic parameters, where for global parameters, all points have the same numerical value and for regional parameters, the value(s) may vary at different points within the feature vector. Server system 106 may then associate this feature vector with the value of the received or simulated blood flow characteristics at this point. Thus, step 412B may include associating a feature vector containing blood flow characteristics at one or more points of the patient-specific 3D anatomical model and / or each of the individual-specific 3D anatomical models with geometric features at the one or more points.

[0072] Step 414B may include training a machine learning algorithm to predict geometric features at one or more points of a model derived from the image data from a feature vector(s) containing blood flow characteristics at one or more points. The training may use the relevant features from step 412B, for example, to determine relevance and / or feature weights. Examples of machine learning algorithms that can perform this task include support vector machines (SVMs), neural networks, multilayer perceptrons (MLPs), multivariate regression (MVR) (e.g., weighted linear or logistic regression), and / or other supervised machine learning techniques known to those skilled in the art. The server system 106 may then store the results of the machine learning algorithm (e.g., feature weights) in a digital representation (e.g., in the memory of a computing device such as a computer, laptop, DSP, server, or in digital storage (e.g., a hard drive, network drive)). The stored feature weights may define the degree to which the blood flow characteristics predict geometric features at one or more points of the model or the system represented by the model.

[0073] Step 416B may include outputting the trained machine learning algorithm (e.g., to an electronic storage medium). The trained machine learning algorithm can be used in method 400C of FIG. 4C to, for example, update, further refine, and / or generate geometric features of a model derived from the image data. This model can be a 2D or 3D anatomical model, or can have a simplified geometry (e.g., a reduced-order model or a lumped parameter model).

[0074] FIG. 4C shows an example method 400C of applying a trained machine learning algorithm to determine (e.g., more accurately) blood flow characteristics using a reduced-order or lumped parameter model, or to determine geometric features of the model.

[0075] Step 402 may include receiving image data of at least one vasculature of interest of a patient in an electronic storage medium (e.g., as in methods 400A and 400B). The image data may be received from one or more imaging modalities (e.g., computed tomography, angiography, magnetic resonance, x-ray, etc.), an electronic storage medium, a third party device, or via the cloud.

[0076] Having completed the steps for training machine learning algorithms in methods 400A and / or 400B, or having received trained machine learning algorithms as described in these methods, steps 418 through 430 describe the application of these trained machine learning algorithms, for example, to determine blood flow characteristics and / or geometric features.

[0077] For example, step 418 may include segmenting the image data (received in step 402) into one or more region(s). Step 418 may be similar to step 204 or 210A of method 200 shown in Figure 2, or step 308 of Figure 3. In some embodiments, this segmentation may be based on estimated blood flow characteristics, or may be performed manually or automatically to identify voxels belonging to various regions (e.g., the aorta to the lumen of the coronary arteries).

[0078] Step 420 may include deriving a simplified geometric shape for each region. Step 420 may be similar to step 312 of method 300, as shown in FIG.

[0079] Step 422 may include generating reduced-order and / or lumped parameter model(s) for each (or one or more) region(s) using the simplified geometry. Method 300 of FIG. 3 broadly describes at least some embodiments of this generation step.

[0080] Step 426 may include creating a feature vector including one or more features at one or more points of the reduced-order model and / or lumped parameter model. The features may reflect or be similar to the features used in the training phase described in method 400A. In some embodiments, these features may include, but are not limited to, a local radius or diameter, a local indicator of stenosis severity (e.g., percentage), a minimum upstream diameter, a minimum upstream stenosis severity indicator, a minimum downstream diameter, a minimum downstream stenosis severity indicator, a distance to the nearest bifurcation, a diameter of the nearest upstream bifurcation, a distance to the ostium, an average downstream outlet diameter, a minimum downstream outlet diameter, a minimum, maximum, mean, or median downstream resistance (e.g., or a boundary condition), aortic pressure, and the physiological and / or phenotypic parameters described above.

[0081] In some embodiments, the feature vector may include as features blood flow characteristics at one or more points estimated using, for example, a reduced order model and / or lumped parameter model(s). In such embodiments, step 424 may include estimating blood flow characteristics at one or more points of the region(s) using a reduced order model and / or lumped parameter model(s).

[0082] Step 428 may include using a trained machine learning algorithm (e.g., from method 400A of FIG. 4A ) to determine blood flow characteristics at one or more points of the reduced-order model and / or lumped parameter model. The blood flow characteristics can be determined individually for one or more points in each region, or can be solved for the entire vasculature or vascular system. In some embodiments, the blood flow characteristics can be displayed, for example, on an anatomical model or used as part of a blood flow simulation. In further embodiments, the blood flow characteristics can be used to determine important indicators of perfusion and / or tissue viability (e.g., a myocardial perfusion risk index). Similarly, if the vasculature has one or more lesions or stenotic regions, the blood flow characteristics can be used to determine an indicator of the severity of the lesions or stenotic regions (e.g., a plaque vulnerability index).

[0083] Additionally or alternatively, a trained machine learning algorithm (e.g., from method 400B of FIG. 4B ) can be used to update the simplified geometry of the reduced-order model and / or lumped parameter model, as shown in step 430. In some embodiments, step 430 can be used to determine geometric features for generating an unsimplified model (e.g., a 2D or 3D anatomical model). In some embodiments, the updated model(s) or the determined geometric features can be output to an electronic storage medium or a display.

[0084] For example, in forming a feature vector in method 400A of FIG. 4A and method 400C of FIG. 4B, the feature vector may include, but is not limited to, (vii) vessel shape characteristics (such as aortic inlet and outlet cross-sectional areas, aortic surface area and volume, minimum, maximum, and average cross-sectional areas), (viii) coronary branch shape characteristics, and (ix) one or more feature sets.

[0085] In one embodiment, the coronary artery bifurcation shape features include: (i) the volume of the aorta upstream / downstream of the coronary artery bifurcation point; (ii) the cross-sectional area of ​​the coronary artery / aortic bifurcation point, i.e., the entrance to the coronary artery bifurcation; (iii) the total number of vessel bifurcations and the number of upstream / downstream vessel bifurcations; (iv) the average, minimum, and maximum upstream / downstream cross-sectional area; (v) the distance (along the vessel centerline) to the centerline point of the minimum and maximum upstream / downstream cross-sectional area; (vi) the cross-section of the nearest upstream / downstream vessel bifurcation and the distance (along the vessel centerline) to the nearest upstream / downstream vessel bifurcation; (vii) cross-sectional area of ​​the nearest coronary artery outlet and aortic inlet / outlet and distance (along the vessel centerline) to the nearest coronary artery outlet and aortic inlet / outlet, (viii) cross-sectional area of ​​the downstream coronary artery outlet with the smallest / largest cross-sectional area and distance (along the vessel centerline) to the downstream coronary artery outlet with the smallest / largest cross-sectional area, (ix) upstream / downstream volume of the coronary artery vessels, and (x) upstream / downstream volume fraction of the coronary artery vessels with a cross-sectional area below a user-specified tolerance.

[0086] In one embodiment, the first feature set may define cross-sectional area features, including the lumen cross-sectional area along the coronary artery centerline, the exponentiated lumen cross-sectional area, the ratio of the lumen cross-sectional area to the major ostium (LM, RCA), the exponentiated ratio of the lumen cross-sectional area to the major ostium, the degree of taper of the lumen cross-sectional area along the centerline, the location of the stenotic lesion, the length of the stenotic lesion, the location and number of lesions corresponding to 50%, 75%, and 90% area reduction, the distance from the stenotic lesion to the main ostium, and / or the irregularity (or circularity) of the cross-sectional lumen boundary.

[0087] In one embodiment, the lumen cross-sectional area along the coronary artery centerline can be calculated by extracting the centerline from the constructed geometry, smoothing the centerline if necessary, calculating the cross-sectional area at each centerline point, and mapping it to the corresponding surface and volume mesh points. In one embodiment, the exponentiated lumen cross-sectional area can be determined from various sources of scaling laws. In one embodiment, the ratio of the lumen cross-sectional area to the major ostium (LM, RCA) can be calculated by measuring the cross-sectional area at the LM ostium, normalizing the cross-sectional area of ​​the left coronary artery by the LM ostium area, measuring the cross-sectional area at the RCA ostium, and normalizing the cross-sectional area of ​​the right coronary artery by the RCA ostium area. In one embodiment, the exponentiated lumen cross-sectional area ratio to the major ostium can be determined from various sources of scaling laws. In one embodiment, the degree of taper of the lumen cross-sectional area along the centerline can be calculated by sampling centerline points within a fixed interval (e.g., twice the vessel diameter) and calculating the slope of the cross-sectional area of ​​the linear fit. In one embodiment, the location of the stenotic lesion can be calculated by finding the minimum of the cross-sectional area curve, finding the location where the first derivative of the area curve is zero and the second derivative is positive, and calculating the distance from the main ostium (parametric arc length of the centerline). In one embodiment, the length of the stenotic lesion can be calculated by calculating the proximal and distal locations from the stenotic lesion where the cross-sectional area is restored.

[0088] In one embodiment, the set of other features may include, for example, intensity features defining the intensity change along the centerline (the gradient of the linearly fitted intensity change). In one embodiment, the set of other features may include, for example, surface features defining the three-dimensional surface curvature of the geometric shape (Gaussian, maximum, minimum, mean). In one embodiment, the set of other features may include, for example, volumetric features defining the ratio of total coronary artery volume to myocardial volume. In one embodiment, the set of other features may include, for example, centerline features defining the curvature (bending) of the coronary artery centerline, obtained, for example, by calculating the Freinet curvature:

[0089]

number

[0090] Alternatively, it can be obtained by calculating the inverse of the radius of the circumscribed circle along the centerline point. The curvature (bending) of the coronary artery centerline can also be calculated based on the torsion (non-planarity) of the coronary artery centerline, for example, by calculating the Freinet curvature:

[0091]

number

[0092] In one embodiment, another feature set can include SYNTAX scoring features, including, for example, the presence of aortic ostial lesions, detection of lesions located at the origin of the coronary arteries from the aorta, and / or dominance (left or right).

[0093] In one embodiment, another feature set may be, for example, the Hagen-Poiseuille flow assumption. [ka] For example, in one embodiment, the server system 106 may include a simplified physical feature including a fractional flow reserve value derived from [ka] Using [ka] The cross-sectional area of ​​the origin of the coronary artery (LM ostium or RCA ostium) can be calculated from [ka] of coronary vessels [ka] can be calculated, [ka] The coronary blood flow rate at each segment of the vessel can be determined using the resistance boundary condition under [ka] The resistance at

[0094]

number

[0095] where, nominal value

number

[0096] The server system 106 [ka] and the FFR at each sampled position can be estimated as [ka] The location of the minimum cross-sectional area or the interval smaller than the vessel radius can be used as the sampling location. [ka] FFR can be interpolated along the centerline using and the FFR values ​​can be projected onto 3D surface mesh nodes, as needed for training. [ka] We can vary , and obtain a new set of FFR estimates, by perturbing the parameters using the feature set defined above, where [ka] can be a function of diseased length, degree of stenosis, and taper ratio to account for tapered vessels; [ka] teeth, [ka] The FFR can be determined by summing the distributed flows at each outlet based on the same scaling law as in (1), but new scaling laws and hyperemia assumptions can be adopted, and this feature vector can be associated with the measured or simulated FFR at that point.

[0097] In some embodiments, it is contemplated that a trained machine learning algorithm that predicts blood flow characteristics or geometric features from feature vectors can simply be received, for example, on an electronic storage medium and can be readily implemented in method 400C of Figure 4C. In such embodiments, training of such a machine learning algorithm by server system 106 may be unnecessary.

[0098] It is further contemplated that in various embodiments, other models than 3D anatomical models may be used in any of the steps presented herein, for example, 2D models may be used and / or a time component may be added to the model.

[0099] Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.

Claims

1. 1. A computer-implemented method for determining blood flow characteristics of a patient, the method comprising: obtaining a reduced-order model representing at least a portion of a patient-specific anatomical model of at least a portion of a patient's vasculature, the obtained reduced-order model including one or more points having parameter values ​​based on first estimates of blood flow characteristics at one or more locations of a first set of the portion of the patient-specific anatomical model; updating the obtained reduced-order model by using a machine learning algorithm for estimating blood flow characteristics, the machine learning algorithm being trained based on an error determined between at least one parameter value of at least one training reduced-order model different from the obtained reduced-order model and at least one corresponding parameter value determined by computational fluid dynamics, so as to reduce an error in the parameter value of at least one point of the one or more points of the obtained reduced-order model; determining second estimates of the blood flow characteristics at one or more locations of a second set of the portion of the patient-specific anatomical model using the updated reduced-order model; and wherein the second set is different from the first set.

2. A computer-implemented method as described in claim 1, wherein the obtained reduced-order model represents the path of blood flow through the portion of the patient-specific anatomical model as an electrical circuit.

3. A computer-implemented method as described in claim 2, wherein the parameter value is expressed as a resistance in the electrical circuit.

4. The method further comprising obtaining, for each of the one or more points, an estimate of flow rate through a corresponding segment of the portion of the patient's vasculature; The computer-implemented method of claim 3 , wherein the resistance for each point is selectively constant, linear, or non-linear based on the estimate of flow rate for the point.

5. A computer-implemented method as described in claim 3, wherein the obtained reduced-order model is configured so that the resistance at each point varies with the flow rate through the corresponding segment of the portion of the patient's vascular structure.

6. A computer-implemented method as described in claim 1, wherein the parameter values ​​at each point correspond to the geometric shape of the portion of the patient's vascular structure or of the patient-specific model at that point.

7. 2. The computer-implemented method of claim 1, wherein the blood flow characteristics include one or more of blood pressure, fractional flow reserve (FFR), blood flow volume or velocity, velocity or pressure field, blood flow force, and organ and / or tissue perfusion characteristics.

8. 1. A system for determining blood flow characteristics of a patient, the system comprising: at least one memory, the memory storing instructions for determining the blood flow characteristics; the memory storing a reduced-order model representing at least a portion of a patient-specific anatomical model of at least a portion of the patient's vasculature, the stored reduced-order model including one or more points having parameter values ​​based on first estimates of blood flow characteristics at one or more locations of a first set of the portion of the patient-specific anatomical model; the memory storing a machine learning algorithm for estimating blood flow characteristics, the machine learning algorithm being trained based on an error determined between at least one parameter value of at least one training reduced-order model different from the stored reduced-order model and at least one corresponding parameter value determined by computational fluid dynamics, so as to reduce an error in the parameter value of at least one point of the stored reduced-order model; at least one processor operatively connected to said memory; Equipped with the at least one processor is configured to perform actions by executing the instructions; The operation is updating the stored reduced-order model by using the machine learning algorithm, wherein the updating comprises reducing an error in the parameter value of at least one of the one or more points of the stored reduced-order model; determining second estimates of the blood flow characteristics at one or more locations of a second set of the portion of the patient-specific anatomical model using the updated reduced-order model; and wherein the second set is different from the first set.

9. The system described in claim 8, wherein the stored reduced-order model represents the path of blood flow through the portion of the patient-specific anatomical model as an electrical circuit.

10. The system of claim 9, wherein the parameter value is expressed as a resistance in the electrical circuit.

11. The operation is obtaining, for each of the one or more points, an estimate of flow through a corresponding segment of the portion of the patient's vasculature; The system of claim 10 , wherein the resistance for each point is selectively constant, linear, or non-linear based on the estimate of flow rate for the point.

12. The system described in claim 10, wherein the stored reduced order model is configured so that the resistance at each point varies with the flow rate through the corresponding segment of the portion of the patient's vascular structure.

13. The system described in claim 8, wherein the parameter values ​​at each point correspond to the geometric shape of the portion of the patient's vascular structure or of the patient-specific model at that point.

14. 9. The system of claim 8, wherein the blood flow characteristics include one or more of blood pressure, fractional flow reserve (FFR), blood flow or velocity, velocity or pressure field, blood flow forces, and organ and / or tissue perfusion characteristics.

15. A non-transitory computer-readable medium comprising instructions for determining blood flow characteristics of a patient, the instructions being executable by one or more processors to perform operations, the operations comprising: obtaining a reduced-order model representing at least a portion of a patient-specific anatomical model of at least a portion of a patient's vasculature, the obtained reduced-order model including one or more points having parameter values ​​based on first estimates of blood flow characteristics at one or more locations of a first set of the portion of the patient-specific anatomical model; updating the obtained reduced-order model by using a machine learning algorithm for estimating blood flow characteristics, the machine learning algorithm being trained based on an error determined between at least one parameter value of at least one training reduced-order model different from the obtained reduced-order model and at least one corresponding parameter value determined by computational fluid dynamics, so as to reduce an error in the parameter value of at least one point of the one or more points of the obtained reduced-order model; determining second estimates of the blood flow characteristics at one or more locations of a second set of the portion of the patient-specific anatomical model using the updated reduced-order model; and wherein the second set is different from the first set.

16. The obtained reduced-order model represents a path of blood flow through the portion of the patient-specific anatomical model as an electrical circuit; 16. The non-transitory computer-readable medium of claim 15, wherein the parameter value is represented as a resistance in the electrical circuit.

Images