Method and apparatus for determining biomarkers of vascular function using BOLD CMR images - Patent Application 20070123333

Abstract

A method and apparatus for obtaining biomarkers of microvascular or macrovascular function in an individual includes receiving consecutive blood oxygen level dependent (BOLD) cardiac magnetic resonance (CMR) image series spanning multiple cardiac cycles and respiratory states, generating multiple phase-matched single-cycle image series that are temporally aligned with multiple phases forming a phase vector, performing a windowed matrix decomposition (WMD) operation on the images of each phase vector to generate low-rank image components, constructing a composite single-cycle image series using the low-rank image components, and calculating one or more oxygen perfusion biomarkers using the composite image series.

Description

[Technical field]

[0001] The present disclosure relates to the use of BOLD CMR images to determine biomarkers of vascular function. [Background technology]

[0002] Macrovascular and microvascular dysfunction is a hallmark of several diseases, including coronary artery disease (CAD or coronary atherosclerosis), the majority of which have high morbidity and mortality. In general, blood supply and oxygen to the heart are affected with consequences for life span and quality of life. Moreover, in the developing atherosclerotic cascade, deterioration of microvascular function is considered one of the first pathophysiological changes, occurring before detectable morphological abnormalities. Therefore, microvascular function is the target of choice for early detection of atherosclerosis and other diseases affecting the heart, such as diabetes, obesity, hypertension, and hypercholesterolemia.

[0003] Oxygenation-sensitive Cardiac Magnetic Resonance (CMR) using the Blood Oxygen-Level-Dependent (BOLD) effect allows noninvasive monitoring of changes in myocardial tissue oxygenation. Oxygenation-sensitive CMR detects changes in hemoglobin oxygenation by exploiting its change in magnetic properties as it transitions from an oxygenated to a deoxygenated state. Oxygenated hemoglobin (oxyHb) is diamagnetic and exhibits weak stability of the magnetic field surrounding the molecule, whereas deoxygenated hemoglobin (de-oxyHb) is paramagnetic and destabilizes the surrounding magnetic field, thereby leading to a loss of magnetic field homogeneity, known as the BOLD effect. CMR protocols sensitive to the BOLD effect show a reduction in localized oxygenation-sensitive signal intensity (OS-SI or BOLD-SI) in tissues with a relative increase in de-oxyHb, as seen in myocardial ischemia.

[0004] Several oxygenation sensing techniques have been used to detect coronary artery disease using myocardial oxygenation changes in response to vasodilation by drugs such as adenosine or dipyridamole as a marker of myocardial ischemia. While healthy blood vessels dilate, leading to an increase in myocardial signal intensity (SI), myocardium enveloped by stenotic blood vessels shows a blunted increase or decrease in myocardial BOLD-SI in response to vasodilation triggers. However, these drugs have undesirable side effects such as bradycardia, arrhythmias, chest pain, bronchospasm, headache, nausea, and heat waves. Furthermore, infusion of such vasoactive substances requires the availability of venous access and a physician, with the added cost of vasodilators, additional preparation time, and the risk of adverse events related to the infused drug. Summary of the Invention [Problem to be solved by the invention]

[0005] Improvements in the determination of biomarkers related to vascular function are desirable. [Means for solving the problem]

[0006] Embodiments of the present disclosure will now be described, by way of example only, with reference to the accompanying figures. [Brief description of the drawings]

[0007] [Figure 1] 1 is a flow chart illustrating a method for obtaining biomarkers of microvascular or macrovascular function in an individual, according to one embodiment of the present disclosure. [Diagram 2] FIG. 1 is a schematic diagram illustrating an example of period aggregation, phase-wise registration, and artifact suppression and SNR boosting of a series of consecutively acquired BOLD CMR images according to an embodiment of the present disclosure. [Figure 3A] FIG. 2 is a schematic diagram illustrating an example of performing a windowed singular value decomposition operation according to an embodiment of the present disclosure. [Figure 3B]FIG. 2 is a schematic diagram illustrating an example of performing a windowed singular value decomposition operation according to an embodiment of the present disclosure. [Figure 3C] FIG. 2 is a schematic diagram illustrating an example of performing a windowed singular value decomposition operation according to an embodiment of the present disclosure. [Figure 4] FIG. 1 is a schematic diagram illustrating an example of performing singular value decomposition on low-rank components of a windowed singular value operation to generate a synthetic image series, according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] An embodiment of the present disclosure relates to a method for processing a series of consecutively acquired BOLD CMR images, sometimes referred to herein as a CINE BOLD CMR image series. A CINE BOLD CMR image series may be acquired under various periodic or non-periodic breathing conditions, such as normal breathing, controlled breathing, such as suppressed or accelerated breathing, and breath holding. The method for processing a CINE BOLD CMR image series may include a step of temporally registering CINE BOLD CMR images of different cardiac or respiratory cycles, referred to herein as cycle aggregation, and for each phase, images from the cycle are spatially registered, referred to herein as phase-wise registration.

[0009] The method, in some embodiments, includes utilizing a windowed matrix decomposition (WMD) operation followed by a matrix decomposition (MD) operation for phase-wise multi-artifact removal, and generating a composite image series utilizing low-rank components generated from the WMD operation from which one or more biomarkers that may provide functional, tissue, or perfusion information are extracted.

[0010] The MD operation utilized in the WMD operation and the optional subsequent MD operation may each be one of a dynamic mode decomposition (DMD) operation or a singular value decomposition (SVD) operation. In some examples, the WMD operation may utilize one of a DMD or SVD operation, and the subsequent MD operation may utilize the other of a DMD or SVD operation. In some examples, performing an SVD operation on a CINE BOLD CMR image in accordance with the present disclosure may provide more robust noise reduction compared to a DMD operation, and therefore utilizing an SVD operation may be more desirable compared to a DMD operation.

[0011] By utilizing MD techniques, the method can generate CINE BOLD CMR image series with reduced or no noticeable artifacts, such as motion artifacts due to individual movement during image acquisition, facilitating that CINE BOLD CMR image series acquired under any respiratory state or combination of respiratory states can be utilized for analysis of myocardial perfusion by oxygen-related CMR signal intensity at each pixel and each time phase of the cardiac or respiratory cycle, as well as its functional and tissue properties.

[0012] MD operations, including SVD and DMD operations, have traditionally been used for image decomposition in video sequences, e.g., decomposing the vector field of weather patterns into low-rank components indicative of larger overall motion and higher-rank components indicative of smaller eddy currents. Traditionally, MD operations such as SVD and DMD have not been applied techniques in medical image analysis.

[0013] Traditionally, cardiac analysis utilizing CINE BOLD CMR images acquired during any type of patient breathing paradigm other than breath-holding has been hindered because the acquired images are typically misaligned with respect to one another and contaminated with severe motion artifacts due to motion caused by the patient's breathing. Such misalignment and motion artifacts create problems for deriving clinical information from such CINE BOLD CMR images.

[0014] The present disclosure provides methods for post-processing CINE BOLD CMR images acquired during any respiratory state and / or acquired with or without contrast or stress agents to measure, derive and visualize multiple clinically relevant biomarkers decomposed into respiratory and cardiac cycles.

[0015] In one embodiment, the present disclosure provides a method for obtaining biomarkers of microvascular or macrovascular function in an individual, the method comprising the steps of: receiving consecutive blood oxygen level dependent (BOLD) cardiac magnetic resonance (CMR) image series spanning a plurality of cardiac cycles; cropping the received BOLD CMR image series into a plurality of single cycle image series, each single cycle image series spanning a single cardiac cycle; phase matching the plurality of single cycle image series to generate a plurality of phase matched single cycle image series that are time-registered at a plurality of phases, each image of the phase matched single cycle image series at a particular phase forming a phase vector; and for each phase vector, applying an overlapping, sliding window method to the images of the phase vector. The method includes performing a windowed matrix decomposition (WMD) operation in a windowed manner to generate, for each window, a low-rank image component containing physiological information salient to the window and a high-rank image component containing sparse information; reconstructing a plurality of noise-reduced phase vectors using the low-rank image components for each phase vector; generating, for each noise-reduced phase vector, a synthetic phase image based on an image of the noise-reduced phase vector; constructing a synthetic single-cycle image series consisting of the generated synthetic phase images of the phase vectors; and calculating one or more oxygen perfusion biomarkers using the synthetic image series.

[0016] In one exemplary embodiment, the WMD operation is a windowed singular value decomposition (WSVD) operation.

[0017] In an exemplary embodiment, the BOLD CMR image series includes a Digital Imaging and Communications in Medicine (DICOM) image that includes timing tags, and cropping the received BOLD CMR image series includes cropping the BOLD CMR image series into a plurality of single-cycle image series using the timing tags.

[0018] In an exemplary embodiment, cropping the received BOLD CMR image series includes cropping the BOLD CMR image series using an image-based technique to determine multiple cardiac cycles.

[0019] In an exemplary embodiment, the image-based technique includes identifying diastolic images in the BOLD CMR image series by comparing the relative size of the left ventricle and sequential image similarity metrics in images of the BOLD CMR image series, and determining each single cycle image series as an image in the BOLD CMR image series between two consecutive identified diastolic images.

[0020] In an exemplary embodiment, the method further includes, for each phase vector, performing a matrix decomposition (MD) operation on the low-rank image components generated by the WMD operation to generate a plurality of ranked eigen modes for each low-rank image component, and reconstructing the plurality of noise-reduced phase vectors includes reconstructing an image of the noise-reduced phase vector using a predetermined number of the lowest-ranked modes among the generated ranked eigenmodes, the top eigenmodes.

[0021] In one exemplary embodiment, the MD operation is a singular value decomposition (SVD) operation.

[0022] In an exemplary embodiment, the method further includes, for each phase vector, spatially registering the image of the phase vector prior to performing the WMD operation.

[0023] In an exemplary embodiment, the step of spatially registering the images of the phase vectors is performed utilizing a non-rigid registration operation.

[0024] In an exemplary embodiment, constructing a synthetic phase image utilizing the image of the noise-reduced phase vectors includes performing one of a two-dimensional (2D) median operation, a 2D mean operation, a principle component analysis operation, a spectral-based operation, or a machine learning based operation on the image of the noise-reduced phase vectors.

[0025] In an exemplary embodiment, phase matching a plurality of single cycle image series to generate a plurality of phase matched single cycle image series includes generating a phase matched matrix, such that a given phase matched single cycle image series is included in a respective row of the phase matched matrix, the columns correspond to respective phases, and a given image is the phase vector of the phase corresponding to that column.

[0026] In an exemplary embodiment, the step of calculating one or more oxygen perfusion biomarkers utilizing the composite image series includes: segmenting each phase of the composite image series to isolate myocardial tissue to generate a segmented image series; and calculating the one or more oxygen perfusion biomarkers utilizing the segmented image series.

[0027] In an exemplary embodiment, the segmenting step is performed by a machine learning system trained to isolate myocardial tissue or by manual myocardial tissue definition.

[0028] In an exemplary embodiment, the one or more oxygen perfusion biomarkers include one or more of total signal intensity over time, oxygenation, deoxygenation, ratio of oxygenation to deoxygenation, oxygenation to deoxygenation differential, oxygenation reaction rate, deoxygenation reaction rate, ratio of oxygenation reaction rate to deoxygenation reaction rate, difference of oxygenation reaction measure to deoxygenation reaction measure, End-Diastolic (ED) to End-Systolic (ES) signal intensity ratio, difference between ED and ES, total oxygen variance, vascular function change, or vascular function change related to respiration.

[0029] In an exemplary embodiment, receiving successive BOLD CMR image series includes receiving successive BOLD CMR image series acquired utilizing at least two different respiratory paradigms.

[0030] In one exemplary embodiment, the at least two different breathing paradigms are at least two of normal breathing, hyperventilation, or breath holding.

[0031] In one exemplary embodiment, the method includes utilizing the composite image series to calculate at least one functional biomarker, the at least one functional biomarker being at least one of radial strain, circumferential strain, ejection fraction, or systolic wall thickening.

[0032] In another embodiment, the present disclosure provides an apparatus for obtaining biomarkers of microvascular or macrovascular function in an individual, the apparatus comprising: receiving successive blood oxygen level dependent (BOLD) cardiac magnetic resonance (CMR) image series spanning a plurality of cardiac cycles; The method includes: cropping the CMR image series into a plurality of single cycle image series, each single cycle image series spanning a single cardiac cycle; phase matching the plurality of single cycle image series to generate a plurality of phase-matched single cycle image series temporally aligned at a plurality of phases, each image of the phase-matched single cycle image series at a particular phase forming a phase vector; for each phase vector, performing a windowed matrix decomposition (WMD) operation on the image of the phase vector in an overlapping sliding window manner to generate, for each window, a low-rank image component containing physiological information salient to the window and a high-rank image component containing sparse information; reconstructing a plurality of noise-reduced phase vectors using the low-rank image components of each phase vector; generating, for each noise-reduced phase vector, a synthetic phase image based on the image of the noise-reduced phase vector; constructing a synthetic single cycle image series consisting of the generated synthetic phase images of the phase vectors; and calculating one or more oxygen perfusion biomarkers using the synthetic image series.

[0033] In one exemplary embodiment, the WMD operation is a windowed singular value decomposition (WSVD) operation.

[0034] In an exemplary embodiment, the BOLD CMR image series includes a Digital Imaging and Communications in Medicine (DICOM) format including timing tags, and the processor configured to crop the received BOLD CMR image series includes a processor configured to utilize the timing tags to crop the BOLD CMR image series into a plurality of single-cycle image series.

[0035] In an exemplary embodiment, the processor configured to crop the received BOLD CMR image series includes a processor configured to crop the BOLD CMR image series using an image-based technique to determine multiple cardiac cycles.

[0036] In an exemplary embodiment, a processor configured to utilize image-based techniques includes a processor configured to identify diastolic images in the BOLD CMR image series by comparing the relative size of the left ventricle in images of the BOLD CMR image series with a successive image similarity metric, and determine each single cycle image series as an image in the BOLD CMR image series between two successive identified diastolic images.

[0037] In an exemplary embodiment, the processor is further configured to perform a matrix decomposition (MD) operation on the low-rank image components generated by the WMD operation for each phase vector to generate a plurality of ranked eigenmodes for each low-rank image component, and the processor configured to reconstruct the plurality of noise-reduced phase vectors includes a processor configured to reconstruct an image of the noise-reduced phase vector using a predetermined number of the lowest-ranked modes among the generated ranked SVD modes, the top eigenmodes.

[0038] In one exemplary embodiment, the MD operation is a singular value decomposition (SVD) operation.

[0039] In an exemplary embodiment, the processor is further configured to, for each phase vector, spatially align the image of the phase vector prior to performing the WMD operation.

[0040] In an exemplary embodiment, the processor configured to spatially align the images of the phase vectors includes a processor configured to perform a non-rigid registration operation.

[0041] In an exemplary embodiment, the processor configured to construct a synthetic phase image utilizing the image of the noise-reduced phase vector includes a processor configured to perform one of a two-dimensional (2D) median operation, a 2D mean operation, a principal component analysis operation, a spectral-based operation, or a machine learning based operation on the image of the noise-reduced phase vector.

[0042] In an exemplary embodiment, a processor configured to phase match a plurality of single cycle image series to generate a plurality of phase matched single cycle image series includes a processor configured to generate a phase matched matrix, such that the images of a given phase matched single cycle image series are included in respective rows of the phase matched matrix, the columns correspond to respective phases, and a given image is the phase vector of the phase corresponding to that column.

[0043] In an exemplary embodiment, the composite image series is a segmented image series in which each phase of the composite image series is segmented to isolate myocardial tissue, and the processor configured to calculate the one or more oxygen perfusion biomarkers includes a processor configured to calculate the one or more oxygen perfusion biomarkers utilizing the segmented image series.

[0044] In an exemplary embodiment, the segmented image series is generated by a machine learning system trained to isolate myocardial tissue or by manual myocardial tissue definition.

[0045] In an exemplary embodiment, the one or more oxygen perfusion biomarkers include one or more of total signal intensity over time, oxygenation, deoxygenation, ratio of oxygenation to deoxygenation, difference of oxygenation to deoxygenation, oxygenation reaction rate, deoxygenation reaction rate, ratio of oxygenation reaction rate to deoxygenation reaction rate, difference of oxygenation reaction measure to deoxygenation reaction measure, end-diastolic (ED) to end-systolic (ES) signal intensity ratio, difference between ED and ES, total oxygen dispersion, vascular function change, or vascular function change related to respiration.

[0046] In an exemplary embodiment, the processor configured to receive successive BOLD CMR image series includes a processor configured to receive successive BOLD CMR image series acquired utilizing at least two different respiratory paradigms.

[0047] In one exemplary embodiment, wherein the at least two different breathing paradigms are at least two of normal breathing, hyperventilation, or breath holding.

[0048] In an exemplary embodiment, the processor is further configured to utilize the composite image series to calculate at least one functional biomarker, the at least one functional biomarker being at least one of radial strain, circumferential strain, ejection fraction, or systolic wall thickening.

[0049] For brevity and clarity of description, reference numbers may be repeated among the figures to indicate corresponding or similar elements. Numerous details are set forth to provide an understanding of the embodiments described herein. The embodiments may be practiced without these details. In other instances, well-known methods, procedures, and components have not been described in detail to avoid obscuring the described embodiments.

[0050] Referring to FIG. 1, a flow chart illustrating a method for obtaining biomarkers of microvascular or macrovascular function in an individual is shown. The illustrated method may be implemented by software executed, for example, by one or more processors. The one or more processors may be included in a computer configured to process medical images, including CMR images, or a computing system comprising two or more computers. Coding software to implement the disclosed method is within the scope of one of ordinary skill in the art given the present specification. The method may include additional processes, or fewer processes than shown and / or described, and may be performed in a different order. Computer readable code executable by at least one processor to perform the process may be stored in a computer readable storage medium, such as a non-transitory computer readable medium.

[0051] At 102, a CINE BOLD CMR image series is received. The received CINE BOLD CMR may be acquired over multiple cardiac cycles and during any breathing paradigm performed by the individual, including normal breathing, breath holding, and hyperventilation. In some examples, a portion of the received CINE BOLD CMR image series may be acquired while one breathing paradigm is performed by the individual, and another portion of the received CINE BOLD CMR image series is acquired while a different breathing paradigm is performed.

[0052] In some examples, in addition to being acquired while the individual is performing one or more breathing paradigms, the CINE BOLD CMR image series may be acquired while one or more additional external stimuli are administered to the individual. For example, the external stimuli may include one or more of a contrast enhancing agent or a vasoactive or pharmacological stress agent being administered to the individual prior to acquiring the CINE BOLD CMR image series. Alternatively or additionally, the additional external stimulus may be induced physical activity.

[0053] In some examples, the received CINE BOLD CMR image series may be in Digital Imaging and Communications in Medicine (DICOM) file format. The DICOM format CINE BOLD CMR image series may include metadata related to the received CINE BOLD CMR image series. In one example, the metadata includes timing tags that may be utilized to identify which cardiac cycle each image in the CINE BOLD CMR image series relates to. Alternatively or additionally, the metadata may include information regarding one or more breathing paradigms the individual is instructed to perform or what external stimuli, if any, were given to the individual during acquisition of the received CINE BOLD CMR image series.

[0054] At 104, the received CINE BOLD CMR image series is cropped into a plurality of single cycle image series, each single cycle image series spanning a single cardiac cycle in the individual. In some examples, the received CINE BOLD CMR image series may be cropped such that each single cycle image series spans the interval between two successive R waves in an electrocardiogram, sometimes referred to as a single "RR cycle." In some examples, the cropping performed at 104 is performed utilizing cycle ID numbers included in timing tag metadata, if such timing tags exist, for each image in the DICOM formatted CINE BOLD CMR image series.

[0055] In examples where timing tag metadata is not included in the received CINE BOLD CMR image series, the cropping at 104 may be performed utilizing an image-based technique. In one example, the image-based technique may include identifying diastolic images in the received CINE BOLD CMR image series, for example, by comparing the relative size of the left ventricle using a successive image similarity metric. In this example, images from one identified diastolic image to the next subsequent identified diastolic image are cropped as a single cycle image series.

[0056] At 106, the single period image series from 104 is phase matched. Phase matching may be performed, for example, by time-registering each image in the single period image series to generate a plurality of time-registered phase-matched single period image series.

[0057] In one example, multiple phase-matched single-cycle image series are arranged in a matrix array, with the single-cycle image series forming rows of the matrix and each column of the matrix being a particular phase within a single cardiac cycle. The time-registered images of the single-cycle image series at a particular phase form what is referred to herein as a "phase vector."

[0058] An example of such a matrix arrangement is shown in Figure 2. The example matrix 200 shown in Figure 2 is composed of four single cycle image series 202-208. Each single cycle image series 202-208 includes seven images that are temporally aligned with the images of the other single cycle image series 202-208 to form seven phase vectors 210-222. Although the example matrix 200 shown in Figure 2 includes four single cycle image series each including seven phases, in general any number of single cardiac cycle image series and any number of phases may be utilized.

[0059] 1, optionally, the images forming each phase vector are spatially registered with other images of that phase vector at 108. In one example, spatially registering the images of each phase vector utilizes a non-rigid registration, where a geometric registration is determined that moves each point in one image to align with a related point in another image. In other examples, other suitable methods for implementing the optional spatial registration of the images of each phase vector.

[0060] 2, performing spatial registration of each image of phase vectors 210-222 is represented by 224. The spatial registration is performed before performing artifact suppression and signal-to-noise ratio (SNR) boosting, represented by 226. The artifact suppression and SNR boosting 226 includes performing a windowed matrix decomposition (WMD) operation, and in some cases may include a subsequent matrix decomposition (MD) operation, as described in more detail below.

[0061] At 110, a WMD operation is performed on the images of the phase vector in an overlapping sliding window manner. The WMD operation involves performing an MD operation on a predefined number of images of the phase vector, called a "window," to generate low-rank image components in the window that contain salient physiological information of the image, and high-rank image components in the window that contain sparse information of the image. The MD operation is then performed on another window that contains a predefined number of images of the phase vector, with at least a portion of the images in the new window included in the previous window, to generate low-rank image components and high-rank image components in the new window. This process is repeated until the window reaches the end of the phase vector. The size of the window utilized may be empirically determined and may depend, for example, on the data on which the WMD operation is being performed.

[0062] The WMD operation performed at 110 is one of a windowed dynamic mode decomposition (WDMD) operation or a windowed singular value decomposition (WSVD) operation. In some examples, performing a WSVD operation on a CINE BOLD CMR image in accordance with the present disclosure may provide more robust noise reduction compared to a WDMD operation, and thus utilizing a WSVD operation may be more desirable compared to a WDMD operation.

[0063] 3A-3C, an example of performing a WMD operation on a phase vector 302 is shown. In the illustrated example, the phase vector 302 includes four images 304a-304d, and may be, for example, one of the phase vectors 210-222 of the example matrix 202 shown in FIG. 2. Additionally, in the illustrated example, the WMD operation is a WSVD operation utilizing a window 306 that includes two of the images 304a-304d, although windows of other sizes may be utilized. In other examples, a WDMD operation may be utilized in place of the illustrated WSVD operation in the examples shown in FIGS. 3A-3C.

[0064] In Fig. 3A, an SVD operation 308 is performed on the images 304a and 304b contained in the window 306a to generate a first low-ranked image component C11 and a first high-ranked image component C21. Next, an SVD operation 308 is performed on the images 304b and 304c contained in the window 306b to generate a second low-ranked image component, C12, and a second high-ranked image component C22. Furthermore, an SVD operation 308 is performed on the images 304c and 304d contained in the window 306c to generate a third low-ranked image component, C13, and a third high-ranked image component C23. In the example shown in Figs. 3A to 3C, three SVD operations are performed to generate the three low-ranked image components C11 to C13, although in general the number of SVD operations performed depends on the size of the window 306 and the number of images contained in the phase vector 302.

[0065] Referring again to FIG. 1, an optional MD operation is performed on the low-rank image components generated in the WMD operation for each of the phase vectors at 112. The optional MD operation generates a number of ranked eigenmodes for each of the low-rank image components of the phase vectors. The MD operation performed on the low-rank image components may be the same MD operation performed on the window of the image of the phase vectors. For example, the MD performed at 112 may be the SVD operation 308 described above with respect to FIGS. 3A-3C, or may be a DMD operation.

[0066] In some examples, the WMD operation performed at 110 may utilize one of a DMD or an SVD operation, and the MD operation subsequently performed at 112 may utilize the other of a DMD or SVD operation. However, in some examples, utilizing an SVD operation on the CINE BOLD CMR image for both the WMD operation performed at 110 and the MD operation performed at 112 may provide more robust noise reduction compared to utilizing a DMD operation for either or both of the WMD operation at 110 and the MD operation at 112.

[0067] At 114, the low-ranked image components are utilized to reconstruct a plurality of noise-reduced phase vectors. If the optional MD operation is performed at 112, the low-ranked image components utilized to reconstruct the plurality of noise-reduced phase vectors may be a subset of the lowest-ranked components of the ranked eigenmodes generated by the MD operation, as described in more detail below with respect to the example shown in FIG. 4. Similar to the low-ranked and high-ranked image components generated by the WMD operation performed at 110, the lowest-ranked modes of the ranked eigenmodes generated by the MD operation performed at 112 will contain the majority of the physiologically relevant information, and the higher ranked modes will contain noise and artifact information.

[0068] In examples where the optional MD operation is not performed, reconstructing the noise-reduced phase vector at 114 may be performed utilizing the low-rank image components generated by the WMD operation performed at 110 .

[0069] In one example, reconstruction of the multiple noise reduced phase vectors may be performed utilizing the following recombination equation:

[0070]

number

[0071] where Ψ are the eigenmodes, ω are the eigenvectors, Σ are the singular values, Λ are the right singular vectors, and 1...r denote that these values ​​are taken for the top r eigenmodes. All these parameters are derived when performing the low-rank decomposition.

[0072] 4, an example is shown for performing an optional MD operation at 112 to generate ranked eigenmodes and reconstructing noise-reduced phase vectors from the ranked eigenmodes at 114. In the example shown in FIG 4, the MD operation is an SVD operation, however, in other examples, a DMD operation may be used in place of the SVD operation.

[0073] Low-rank image components 400 generated for a particular phase vector using a WSVD operation are provided as input to an SVD operation 402. The low-rank components may be generated similarly to the example described above with respect to Figures 3A-3C. In the example shown in Figure 4, n low-rank image components, C11 to C12, generated by the WSVD operation performed at each phase vector are provided as input to an SVD operation 402. n There is.

[0074] The number of ranks utilized may be empirically determined based on the data on which the SVD operation is performed. The number of ranks utilized may also depend on whether or not spatial alignment is performed at 108. For example, if spatial alignment is not performed at 108, fewer ranks may be desirable compared to when spatial alignment is performed in order to remove more dynamic modes so that more spatial deformation modes are eliminated by the SVD operation.

[0075] An SVD operation 402 is performed on each of the low components 400, resulting in K ranked SVD modes for each of the low-rank components 400. The generated ranked SVD modes for all of the low-rank components 400 are illustrated by a matrix 404 shown in FIG. 4, where each column of the matrix 404 represents a low-rank image component, C11 through C12. n For example, E11 through EK1 are ranked eigencomponents generated on the low-rank image component C11, E12 through EK2 are ranked SVD components generated on the low-rank image component C12, etc.

[0076] Once the ranked eigenmodes for all of the low-rank image components 400 are generated, a predetermined number, r, of the K ranked eigenmodes are utilized to reconstruct 406 a noise-reduced phase vector 408, and other higher ranked eigenmodes are discarded.

[0077] The number of lowest ranked eigenmodes, r, may be empirically determined such that at least a minimum amount of dynamic information in the BOLD CMR image series is preserved, including the actual BOLD signal from the BOLD CMR images, which is the signal change caused by the underlying physiological process. The empirical method utilized to determine the number of lowest ranked eigenmodes, r, may utilize, for example, a Pareto analysis.

[0078] The minimum amount of dynamic information may depend, for example, on whether the phase vector images are properly spatially aligned. For example, if the phase vector images are spatially aligned at 108, the number, r, may be determined such that, for example, 30% of the dynamic information is retained. In this case, it is assumed that the physical motion of the individual during acquisition of the CINE BOLD CMR images has been removed through the spatial alignment process.

[0079] However, when spatial registration of the phase vector images is not performed, the actual BOLD signal is mixed with the effects of the individual's physical motion during the acquisition of the CINE BOLD CMR images, and since a larger amount of dynamic information must be preserved when spatial registration is not performed, the number, r, may be determined such that 80% of the dynamic information is preserved.

[0080] After the noise-reduced phase vector is reconstructed, a synthetic phase image is generated at 116. The synthetic phase image may be generated such that the SNR is increased. In one example, the SNR may be increased by generating the synthetic phase image as a two-dimensional (2D) median of images of the noise-reduced phase vector. In other examples, different techniques may be utilized to generate the synthetic phase image, including, for example, 2D average, principal component analysis, spectral-based methods, and machine learning-based operations.

[0081] The composite phase images associated with each phase vector are then combined together to generate a series of composite single cycle images at 116 . At 118, the composite single cycle image series is utilized to calculate one or more oxygen perfusion biomarkers. Calculating the oxygen perfusion biomarkers may be performed, for example, by segmenting each phase of the composite single cycle image series to isolate myocardial tissue, which generates a segmented image series, and calculating the one or more oxygen perfusion biomarkers utilizing the segmented image series. Segmenting the images of the composite single cycle image series may be performed by a machine learning system trained to isolate myocardial tissue, or by manual myocardial tissue definition.

[0082] The biomarker oxygen perfusion calculated at 118 is a function of total signal intensity over time, which is the sum of all signal intensity values ​​over time, oxygenation, which is the sum of increasing signal intensity values ​​over time, deoxygenation, which is the sum of decreasing signal intensity values ​​over time, cumulative oxygen load divided by disappearance of oxygen flush, and oxygen perfusion over time, which is the sum of increasing signal intensity values ​​over time. the oxygenation to deoxygenation ratio, which is the ratio of the oxygenation to deoxygenation ratio (the oxygenation to deoxygenation ratio), the difference between the oxygen load cumulative and the oxygen flush disappearance, the oxygenation to deoxygenation difference, which is the difference between the oxygen load cumulative and the oxygen flush disappearance, the oxygenation response rate, which may be the first derivative, i.e., rate, and / or the second derivative (i.e., acceleration) of the increasing signal intensity value over time, the deoxygenation response measure, which may be the first derivative, i.e., rate, and / or the second derivative (i.e., acceleration) of the decreasing signal intensity value over time, the ratio of the oxygenation response measure to the deoxygenation response measure, the difference of the oxygenation response measure to the deoxygenation response measure, the ratio of the signal intensity at the ED to the signal intensity at the ES, the end-diastolic (ED) to end-systolic (ES) signal intensity ratio, which is the difference between the signal intensity at the ED and the signal intensity at the ES, the difference between the ED and ES, the total oxygen variance, which is the statistical variance of the signal intensity over time, the vascular function change, or the vascular function change related to respiration. These biomarkers can only be determined from images acquired during active breathing. These biomarkers cannot be obtained using previous image processing techniques due to intrinsic noise in the image data acquired during active breathing.

[0083] In addition to the oxygen perfusion biomarkers calculated at 118, one or more functional biomarkers may be calculated utilizing the composite single cycle image series. Functional biomarkers calculated from the composite single cycle image series may include, for example, one or more of radial strain, which is myocardial tissue displacement relative to the ventricular chamber centroid, circumferential strain, which is myocardial tissue displacement relative to the radial vector normal to the ventricular chamber centroid, ejection fraction, which is the ratio of minimum ventricular chamber area divided by maximum ventricular chamber area, or systolic wall thickening, which is the difference between ED and ES wall thickness.

[0084] The embodiments of the present disclosure provide processing of CINE BOLD CMR images to remove artifacts and noise and to enhance SNR, so that CINE BOLD CMR images acquired under any respiratory state or combination of respiratory states can be utilized for the analysis of myocardial perfusion. WMD operations, and in some embodiments MD operations following the WMD operations, are utilized to remove artifacts from the CINE BOLD CMR images, for example, caused by physical movement of the individual during image acquisition. Additionally, a synthetic single cycle image series is constructed from low-rank image components generated from the WMD operations, and in some embodiments further MD operations, to enhance SNR. In some embodiments, artifact suppression and SNR enhancement are enhanced by spatially registering the images of each phase vector prior to the WMD operations. By removing noise caused by, for example, physical movement of the patient due to breathing during image acquisition, the above biomarkers can be obtained that cannot be obtained using previous techniques.

[0085] In the preceding description, for purposes of explanation, numerous details are set forth in order to achieve a thorough understanding of the embodiments. However, it will be apparent to one skilled in the art that these specific details are not required. In other instances, well-known electrical structures and circuits are shown in block diagram form in order not to obscure the understanding. For example, specific details are not provided as to whether the embodiments described herein are implemented as a software routine, a hardware circuit, firmware, or a combination thereof.

[0086] The embodiments of the present disclosure can be represented as a computer program product stored on a machine-readable medium (also referred to as a computer-readable medium having computer-readable program code embodied therein, a processor-readable medium, or a computer usable medium). The machine-readable medium can be any suitable tangible, non-transitory medium, including magnetic, optical, or electrical storage media, including diskettes, compact disk read-only memories (CD-ROMs), memory devices (volatile or non-volatile), or similar storage mechanisms. The machine-readable medium can include various sets of instructions, code sequences, configuration information, or other data that, when executed, cause a processor to perform steps of a method according to the embodiments of the present disclosure. Those skilled in the art will appreciate that other instructions and operations necessary to implement the described implementations may also be stored on the machine-readable medium. The instructions stored on the machine-readable medium are executable by a processor or other suitable processing device and can interface with circuitry to perform the tasks described.

[0087] The above-described embodiments are intended to be examples only. Alterations, modifications and variations may be effected to the particular embodiments by those of skill in the art without departing from the scope, which is defined solely by the claims appended hereto. [Explanation of symbols]

[0088] 200 Queue 202,204,206,208 Image series 210,212,214,216,218,220,222 phasor 224 Non-rigid Registration 226 Artifact Suppression and SNR Boost 302 Phasor 304 images 306 Windows 308 SVD calculation 400 low-rank image components 402 SVD calculation 404 Queue 406 Reconstruction 408 Noise Reduction Phasor

Claims

1. 1. A method for obtaining a biomarker of microvascular or macrovascular function in an individual, comprising: receiving a series of consecutive blood oxygen level dependent (BOLD) cardiac magnetic resonance (CMR) images spanning a plurality of cardiac cycles; cropping the received BOLD CMR image series into a plurality of single cycle image series, each single cycle image series spanning a single cardiac cycle; phase matching the plurality of single period image series to generate a plurality of phase matched single period image series that are time-registered at a plurality of phases, each image of the phase matched single period image series at a particular phase forming a phase vector; For each phase vector, performing a Windowed Matrix Decomposition (WMD) operation on the image of the phase vector with overlapping sliding windows to generate, for each window, a low-rank image component that contains physiological information salient to the window and a high-rank image component that contains sparse information; reconstructing a plurality of noise-reduced phase vectors using the low-rank image components of each phase vector; generating, for each noise reduced phase vector, a synthetic phase image based on an image of said noise reduced phase vector; constructing a synthetic single cycle image series consisting of the generated synthetic phase images of the phase vectors; utilizing said composite single cycle image series to calculate one or more oxygen perfusion biomarkers; A method comprising:

2. The method of claim 1 , wherein the WMD operation is a windowed singular value decomposition (WSVD) operation.

3. the BOLD CMR image series comprises a Digital Imaging and Communications in Medicine (DICOM) image including timing tags; The method of claim 1 , wherein cropping the received BOLD CMR image series comprises cropping the BOLD CMR image series into the plurality of single-cycle image series using the timing tags.

4. The method of claim 1 , wherein cropping the received BOLD CMR image series comprises cropping the BOLD CMR image series using an image-based technique to determine the multiple cardiac cycles.

5. The image-based technique comprises: identifying a diastolic image in the BOLD CMR image series by comparing the relative size of the left ventricle in images of the BOLD CMR image series with a successive image similarity metric; determining each single cycle image series as an image of said BOLD CMR image series between two consecutive identified diastolic images; 5. The method of claim 4, comprising:

6. and performing a matrix decomposition (MD) operation on the low-rank image components generated by the WMD operation for each phase vector to generate a number of ranked eigenmodes for each low-rank image component; 2. The method of claim 1 , wherein reconstructing a plurality of noise-reduced phase vectors comprises reconstructing the image of the noise-reduced phase vectors using a predetermined number of lowest-ranked modes from the generated ranked eigenmodes.

7. The method of claim 6 , wherein the MD operation is a singular value decomposition (SVD) operation.

8. The method of claim 1 , further comprising, for each phase vector, spatially registering the images of the phase vector prior to performing the WMD operation.

9. The method of claim 8 , wherein the step of spatially registering the images of the phase vectors is performed utilizing a non-rigid registration operation.

10. The step of generating a synthetic phase image based on the image of the noise reduced phase vector comprises: - Two-dimensional (2D) median calculation, ・2D average calculation, ・Principal component analysis calculation, Spectral-based operations, or - Machine learning based calculations The method of claim 1 , further comprising the step of:

11. 2. The method of claim 1 , wherein phase matching the plurality of single period image series to generate a plurality of phase matched single period image series comprises generating a phase matched matrix, wherein an image of a given phase matched single period image series is included in a respective row of the phase matched matrix, a column of the phase matched matrix corresponds to a respective phase, and a given image is a phase vector of the phase corresponding to the column.

12. Utilizing the composite single cycle image series to calculate one or more oxygen perfusion biomarkers, segmenting each phase of the composite single cycle image series to isolate myocardial tissue to generate a segmented image series; utilizing the segmented image series to calculate the one or more oxygen perfusion biomarkers; 2. The method of claim 1, comprising:

13. 13. The method of claim 12, wherein the segmenting step is performed by a machine learning system trained to isolate myocardial tissue or by manual myocardial tissue definition.

14. the one or more oxygen perfusion biomarkers being - total signal strength over time, - oxygenation, Deoxygenation, - the ratio of oxygenation to deoxygenation; - the difference between oxygenation and deoxygenation, -oxygenation reaction rate, - Deoxygenation reaction rate, the ratio of the oxygenation reaction rate to the deoxygenation reaction rate; - the difference between the oxygenation response measure and the deoxygenation response measure, - end-diastolic (ED) to end-systolic (ES) signal intensity ratio, ・The difference between ED and ES, Total oxygen dispersion, - Altered vascular function, or - Changes in vascular function related to breathing The method of claim 1, comprising one or more of:

15. The method of claim 1 , wherein receiving successive BOLD CMR image series comprises receiving successive BOLD CMR image series acquired utilizing at least two different respiratory paradigms.

16. 16. The method of claim 15, wherein the at least two different breathing paradigms are at least two of normal breathing, hyperventilation, or breath holding.

17. Utilizing the composite single cycle image series to calculate at least one functional biomarker, wherein the at least one functional biomarker is: - Radial strain, - Circumferential strain, Ejection fraction, or ・Systolic wall thickening The method of claim 1 , further comprising at least one of the steps:

18. 1. An apparatus for obtaining a biomarker of microvascular or macrovascular function in an individual, comprising: receiving a series of consecutive blood oxygen level dependent (BOLD) cardiac magnetic resonance (CMR) images spanning multiple cardiac cycles; cropping the received BOLD CMR image series into a plurality of single cycle image series, each single cycle image series spanning a single cardiac cycle; phase matching the plurality of single period image series to generate a plurality of phase matched single period image series that are time-registered at a plurality of phases, each image of the phase matched single period image series at a particular phase forming a phase vector; For each phase vector, performing a windowed matrix decomposition (WMD) operation on the image of the phase vector in an overlapping sliding window manner to generate, for each window, a low-rank image component that contains physiological information salient to the window and a high-rank image component that contains sparse information; reconstructing a plurality of noise-reduced phase vectors using the low-rank image components of each phase vector; generating, for each noise reduced phase vector, a synthetic phase image based on an image of the noise reduced phase vector; constructing a synthetic single cycle image series consisting of the generated synthetic phase images of the phase vectors; Utilizing said composite single cycle image series to calculate one or more oxygen perfusion biomarkers.

16. An apparatus comprising: a processor configured to:

19. The apparatus of claim 18 , wherein the WMD operation is a windowed singular value decomposition (WSVD) operation.

20. the BOLD CMR image series comprises a Digital Imaging and Communications in Medicine (DICOM) image including timing tags; 20. The apparatus of claim 18, wherein the processor configured to crop the received BOLD CMR image series comprises the processor configured to utilize the timing tags to crop the BOLD CMR image series into the multiple single cycle image series.

21. 20. The apparatus of claim 18, wherein the processor configured to crop the received BOLD CMR image series comprises the processor configured to crop the BOLD CMR image series using an image-based technique to determine the multiple cardiac cycles.

22. The processor configured to utilize the image-based technique, identifying a diastolic image of the BOLD CMR image series by comparing the relative size of the left ventricle in images of the BOLD CMR image series to a successive image similarity metric; determining each single cycle image series as an image of said BOLD CMR image series between two consecutive identified diastolic images; 22. The apparatus of claim 21, comprising the processor configured to:

23. the processor is further configured to perform, for each phase vector, a matrix decomposition (MD) operation on the low-rank image components generated by the WMD operation to generate, for each low-rank image component, a number of ranked eigenmodes; 20. The apparatus of claim 18, wherein the processor configured to reconstruct a plurality of noise-reduced phase vectors comprises the processor configured to reconstruct the image of the noise-reduced phase vector using a predetermined number of lowest-ranked modes of the generated ranked eigenmodes.

24. 24. The apparatus of claim 23, wherein the MD operation is a singular value decomposition (SVD) operation.

25. 20. The apparatus of claim 18, wherein the processor is further configured to, for each phase vector, spatially align the images of the phase vector prior to performing the WMD operation.

26. 26. The apparatus of claim 25, wherein the processor configured to spatially align the images of the phase vectors comprises the processor configured to perform a non-rigid registration operation.

27. The processor configured to generate a synthetic phase image based on an image of the noise reduced phase vector includes: - Two-dimensional (2D) median calculation, ・2D average calculation, ・Principal component analysis calculation, Spectral-based operations, or - Machine learning based calculations 20. The apparatus of claim 18, further comprising the processor configured to:

28. 20. The apparatus of claim 18, wherein the processor configured to phase match the plurality of single cycle image series to generate a plurality of phase matched single cycle image series comprises the processor configured to generate a phase matched matrix, wherein an image of a given phase matched single cycle image series is included in a respective row of the phase matched matrix, a column of the phase matched matrix corresponds to a respective phase, and a given image is a phase vector of the phase corresponding to the column.

29. the composite single cycle image series is a segmented image series, wherein each phase of the composite single cycle image series is segmented to isolate myocardial tissue; 20. The apparatus of claim 18, wherein the processor configured to calculate the one or more oxygen perfusion biomarkers comprises the processor configured to calculate the one or more oxygen perfusion biomarkers utilizing the segmented image series.

30. 30. The apparatus of claim 29, wherein the segmented image series is generated by a machine learning system trained to isolate myocardial tissue or by manual myocardial tissue definition.

31. the one or more oxygen perfusion biomarkers being - total signal strength over time, - oxygenation, Deoxygenation, - the ratio of oxygenation to deoxygenation; - the difference between oxygenation and deoxygenation, -oxygenation reaction rate, - Deoxygenation reaction rate, the ratio of the oxygenation reaction rate to the deoxygenation reaction rate; - the difference between the oxygenation response measure and the deoxygenation response measure, - end-diastolic (ED) to end-systolic (ES) signal intensity ratio, ・The difference between ED and ES, Total oxygen dispersion, - Altered vascular function, or - Changes in vascular function related to breathing 20. The apparatus of claim 18, comprising one or more of:

32. 20. The apparatus of claim 18, wherein the processor configured to receive successive BOLD CMR image series comprises the processor configured to receive successive BOLD CMR image series acquired utilizing at least two different respiratory paradigms.

33. 33. The apparatus of claim 32, wherein the at least two different breathing paradigms are at least two of normal breathing, hyperventilation, or breath holding.

34. and wherein the processor utilizes the composite single cycle image series to calculate at least one functional biomarker, the at least one functional biomarker being: - Radial strain, - Circumferential strain, Ejection fraction, or ・Systolic wall thickening The apparatus of claim 18 , further configured to:

Images