Brain multi-mode image data processing method and device

Abstract

The invention discloses a brain multi-modal image data processing method and device. The method comprises the steps of obtaining a brain multi-modal image sequence of a target object; fusing the three-dimensional contrast-agent-free angiography with the plaque internal hemorrhage image sequence and the time leap method imaging image sequence to obtain cerebrovascular image data; cerebrovascular image data is used as a tree root node and a trunk node, hemodynamic parameter data in a phase contrast angiography sequence is used as a branch node for representing a blood supply state, and neural structure parameter data in a diffusion magnetic resonance imaging sequence is used as a branch node for representing a neural structure. Physical and chemical parameter data in the liquid attenuation inversion recovery sequence are used as leaf nodes for representing functional metabolism, hemodynamic parameter data in the magnetic resonance brain perfusion imaging sequence are refined to a pixel level and then are used as leaf nodes for representing metabolic activity, and a brain biological characteristic tree of the target object is constructed. According to the method, the brain multi-modal image data can be quantitatively analyzed.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and apparatus for processing multimodal brain image data. Background Technology

[0002] This section is intended to provide background or context for the embodiments of the invention set forth in the claims. The description herein is not an admission that it is prior art simply because it is included in this section.

[0003] In the field of modern medical imaging diagnosis, magnetic resonance imaging (MRI) technology occupies a pivotal position due to its unique advantages. With the exponential development of technology, MRI scanners have achieved leapfrog breakthroughs in performance. At the hardware level, the field strength of superconducting magnets has continuously increased, progressing from low field strengths to high field strengths such as 3.0T and even 7.0T, significantly enhancing the intensity of magnetic resonance signals and laying a solid foundation for improving image quality. The performance of gradient systems has also been continuously optimized; higher gradient switching rates and intensities enable rapid spatial encoding, significantly shortening scan time. In clinical applications, some rapid scanning sequences can compress examination times that originally took tens of minutes to within minutes. In terms of software algorithms, the widespread application of parallel imaging technology has further accelerated data acquisition speed. Combined with advanced image reconstruction algorithms, the spatial resolution of imaging has also achieved a qualitative leap, now achieving sub-millimeter or even micrometer-level high-resolution imaging, clearly presenting the fine structures of human tissues.

[0004] In the subfield of MRI technology, non-contrast-enhanced MRI sequences have been continuously improved in recent years thanks to collaborative innovation in MRI hardware and software. Compared to traditional imaging methods that rely on contrast agents, non-contrast-enhanced MRI technology has gained widespread attention and promotion in clinical practice due to its significant advantages of being non-invasive, radiation-free, and eliminating the potential risks of contrast agent injections for patients. Especially in the examination of special populations, this technology demonstrates irreplaceable application value. By optimizing pulse sequence design and signal acquisition strategies, it can effectively highlight the natural contrast between tissues, achieving good application results in fields such as vascular imaging and neuropathological analysis.

[0005] Functional magnetic resonance imaging (fMRI), as an important branch of MRI technology, has greatly expanded the application boundaries of magnetic resonance technology. This technology can capture changes in the functional activity of organs such as the brain in real time by detecting blood oxygen level-dependent (BOLD) signals in living organisms.

[0006] Meanwhile, the deep integration of artificial intelligence (AI) with the field of magnetic resonance imaging (MRI) is triggering profound changes in the industry. Deep learning-based image reconstruction algorithms can accurately reconstruct high-quality images from limited sampled data, effectively reducing scan time and the impact of patient motion artifacts. AI-assisted image analysis systems can quickly identify lesion features in images, demonstrating extremely high accuracy and efficiency. Intelligent workflow optimization algorithms can automatically plan the optimal scanning scheme based on individual patient characteristics and examination needs, achieving intelligent management of the MRI examination process. The application of these AI technologies is driving the accelerated evolution of the industry in multiple dimensions, including improving the efficiency of MRI, optimizing diagnostic results, and standardizing examination procedures.

[0007] However, despite the remarkable achievements of the aforementioned individual technologies, significant challenges remain in practical clinical and research applications. Currently, medical imaging data exhibits multimodal and large-scale characteristics, encompassing not only MRI data but also imaging information from CT, PET, ultrasound, and other modalities, as well as non-imaging data such as gene sequencing and clinical laboratory tests. However, existing technologies lack a systematic, comprehensive, and effective methodology for the quantitative analysis of multimodal data, hindering the full extraction of the rich information contained within and making it difficult to achieve a more comprehensive and accurate assessment of pathology. Summary of the Invention

[0008] This invention provides a method for processing multimodal brain imaging data, capable of quantitative analysis of multimodal brain imaging data, including:

[0009] Obtain a multimodal brain imaging sequence of the target object, wherein the multimodal brain imaging includes a three-dimensional contrast-free angiography and plaque hemorrhage imaging sequence, a time-of-flight imaging sequence, a magnetic resonance brain perfusion imaging sequence, a diffusion-weighted magnetic resonance imaging data sequence, a fluid attenuation inversion recovery sequence, and a phase-contrast angiography sequence.

[0010] Spatial alignment of brain multimodal image sequences;

[0011] Cerebrovascular imaging data were obtained by fusing spatially aligned 3D contrast-free vascular imaging sequences with intraplaque hemorrhage imaging sequences and spatially aligned time-of-flight imaging sequences.

[0012] Using cerebral vascular imaging data as root and trunk nodes, hemodynamic parameters from spatially aligned phase-contrast vascular imaging sequences as branch nodes representing blood supply status, neural structural parameters from spatially aligned diffusion magnetic resonance imaging sequences as branch nodes representing neural structures, and physicochemical parameters from spatially aligned fluid attenuation inversion recovery sequences as leaf nodes representing functional metabolism, and refining hemodynamic parameters from spatially aligned magnetic resonance brain perfusion imaging sequences to pixel level as leaf nodes representing metabolic activity, a brain biometric tree of the target object is constructed. This brain biometric tree of the target object is used for brain function analysis of the target object.

[0013] This invention also provides a brain multimodal image data processing device, capable of quantitative analysis of brain multimodal image data, including:

[0014] The brain multimodal image sequence acquisition module is used to acquire brain multimodal image sequences of the target object. The brain multimodal images include three-dimensional contrast-free angiography and plaque hemorrhage image sequences, time-of-flight imaging image sequences, magnetic resonance brain perfusion imaging sequences, diffusion nuclear magnetic resonance data sequences, fluid attenuation inversion recovery sequences, and phase-contrast angiography sequences.

[0015] The spatial alignment module is used for spatial alignment of brain multimodal image sequences;

[0016] The fusion module is used to fuse spatially aligned 3D contrast-free vascular imaging sequences with plaque hemorrhage imaging sequences and spatially aligned time-of-flight imaging sequences to obtain cerebral vascular imaging data.

[0017] A brain biometric tree construction module is used to construct a brain biometric tree for a target object. This tree uses cerebral vascular imaging data as root and trunk nodes, hemodynamic parameters from spatially aligned phase-contrast vascular imaging sequences as branch nodes representing blood supply status, neural structural parameters from spatially aligned diffusion magnetic resonance imaging sequences as branch nodes representing neural structures, and physicochemical parameters from spatially aligned fluid attenuation inversion recovery sequences as leaf nodes representing functional metabolism. Furthermore, hemodynamic parameters from spatially aligned magnetic resonance brain perfusion imaging sequences are refined to the pixel level and used as leaf nodes representing metabolic activity. This brain biometric tree is then used to perform brain function analysis on the target object.

[0018] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described brain multimodal image data processing method.

[0019] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described brain multimodal image data processing method.

[0020] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described brain multimodal image data processing method.

[0021] In this embodiment of the invention, a multimodal brain image sequence of the target object is obtained. The multimodal brain images include a three-dimensional contrast-free angiography and plaque hemorrhage image sequence, a time-of-flight imaging image sequence, a magnetic resonance brain perfusion imaging sequence, a diffusion-weighted magnetic resonance imaging (MRI) data sequence, a fluid attenuation inversion recovery (FIR) sequence, and a phase-contrast angiography sequence. The multimodal brain image sequences are spatially aligned. The spatially aligned three-dimensional contrast-free angiography and plaque hemorrhage image sequence and the spatially aligned time-of-flight imaging image sequence are fused to obtain cerebral vascular image data. The cerebral vascular image data are used as the root node and trunk node of a tree structure. Hemodynamic parameters from spatially aligned phase-contrast angiography sequences are used as tree nodes representing blood supply status; neural structural parameters from spatially aligned diffusion magnetic resonance imaging sequences are used as tree nodes representing neural structures; physicochemical parameters from spatially aligned fluid attenuation inversion recovery sequences are used as leaf nodes representing functional metabolism; and hemodynamic parameters from spatially aligned magnetic resonance brain perfusion imaging sequences are refined to pixel level and used as leaf nodes representing metabolic activity. This process constructs a brain biometric tree for the target subject, which is then used for brain function analysis. Through this process, the acquired multimodal brain imaging sequences are given biological meaning, forming a quantified brain biometric tree, which can then be used for brain function evaluation from a naturalistic perspective. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0023] Figure 1 This is a flowchart of the brain multimodal imaging data processing method in an embodiment of the present invention;

[0024] Figure 2 This is a flowchart illustrating the spatial alignment of brain multimodal image sequences in an embodiment of the present invention;

[0025] Figure 3 This is a flowchart illustrating the registration of multimodal brain images in an embodiment of the present invention;

[0026] Figure 4 This is a flowchart illustrating the process of obtaining cerebral vascular imaging data in an embodiment of the present invention;

[0027] Figure 5 This is a flowchart illustrating the construction of a brain biometric tree of a target object in an embodiment of the present invention;

[0028] Figure 6 This is a flowchart illustrating the verification of the brain biometric tree in an embodiment of the present invention;

[0029] Figure 7 This is a visualization of the brain biometric tree in an embodiment of the present invention;

[0030] Figure 8 This is a flowchart of brain function analysis in an embodiment of the present invention;

[0031] Figure 9 This is a schematic diagram of the brain multimodal imaging data processing device in an embodiment of the present invention;

[0032] Figure 10 This is a schematic diagram of a computer device in an embodiment of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0034] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0035] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0036] Figure 1 The flowchart of the brain multimodal imaging data processing method in this embodiment of the invention includes:

[0037] Step 101: Obtain the brain multimodal imaging sequence of the target object. The brain multimodal imaging includes three-dimensional contrast-free angiography and plaque hemorrhage imaging sequence, time-of-flight imaging sequence, magnetic resonance brain perfusion imaging sequence, diffusion nuclear magnetic resonance data sequence, fluid attenuation inversion recovery sequence, and phase contrast angiography sequence.

[0038] Step 102: Spatial alignment of the brain multimodal image sequence;

[0039] Step 103: The spatially aligned three-dimensional contrast-free vascular imaging sequence and the plaque hemorrhage imaging sequence, as well as the spatially aligned time-of-flight imaging sequence, are fused to obtain cerebral vascular imaging data.

[0040] Step 104: Using cerebral vascular imaging data as the root and trunk nodes, hemodynamic parameter data from spatially aligned phase-contrast vascular imaging sequences as branch nodes representing blood supply status, neural structure parameter data from spatially aligned diffusion magnetic resonance imaging sequences as branch nodes representing neural structures, and physicochemical parameter data from spatially aligned fluid attenuation inversion recovery sequences as leaf nodes representing functional metabolism, and refining hemodynamic parameter data from spatially aligned magnetic resonance brain perfusion imaging sequences to the pixel level as leaf nodes representing metabolic activity, a brain biometric tree of the target object is constructed. The brain biometric tree of the target object is used for brain function analysis of the target object.

[0041] In this embodiment of the invention, the acquired brain multimodal image sequences are given biological meaning, forming a quantified brain biometric tree, which can then be used for brain function evaluation from a natural perspective.

[0042] Each step is described in detail below.

[0043] In step 101, a multimodal brain imaging sequence of the target object is obtained. The multimodal brain imaging includes a three-dimensional contrast-free angiography and plaque hemorrhage imaging sequence, a time-of-flight imaging sequence, a magnetic resonance brain perfusion imaging sequence, a diffusion-weighted magnetic resonance imaging data sequence, a fluid attenuation inversion recovery sequence, and a phase-contrast angiography sequence.

[0044] Three-dimensional contrast-free angiography and intraplaque hemorrhage imaging (3D SNAP-MRA) takes into account the differences in vascular morphology and blood flow time patterns in each individual, thus enabling precise analysis of hemodynamics from the heart to the distal cerebral vessels. This technique can detect abnormalities such as smoothness of the vessel wall, plaque formation, obstruction, integrity of the vessel wall, and hemorrhage. Furthermore, black-blood flow morphology can be reconstructed using low-intensity pulsed light (PISR). Validated through signal derivation, simulation sequence parameter optimization, fluid model experiments, and clinical scanning trials, this technique can capture biophysically significant blood flow velocity information, conforming to the "interval flow velocity" characteristic in physics. It exhibits higher sensitivity and specificity in detecting pathological changes in cerebral blood flow.

[0045] Time-of-flight imaging (3D TOF-MRA) is a classic magnetic resonance angiography technique widely used in the diagnosis of cerebral vascular stenosis, aneurysms, and vascular malformations. This technique utilizes the "inflow enhancement effect" of blood flow for vascular imaging. By employing a shorter repetition time (TR) and a larger flip angle, it effectively suppresses background tissue signals, making blood flow signals stand out more. Because the background appears black and the blood vessels appear white, it is also known as "bright blood" in magnetic resonance imaging. Its advantages include good background tissue signal suppression, fast scanning speed, and favorable visualization of arterial blood. However, TOF angiography also has some limitations, such as poor visualization of slow blood flow. In addition, this technique only emphasizes vascular morphology and fails to provide information on intravascular blood supply function, such as the luminal wall, plaque, and flow velocity, which are crucial for determining the natural state of blood supply to brain tissue.

[0046] Magnetic resonance brain perfusion imaging (MRI) is a non-invasive method for assessing cerebral hemodynamics. This technique reflects brain blood flow by detecting changes in the ratio of oxyhemoglobin to deoxyhemoglobin in the blood, thereby providing insight into the brain's functional state. There are two main methods of MRI: static MRI without exogenous contrast agents and dynamic MRI with contrast agents. This invention employs the static method, abbreviated as pCASL, which assesses changes in cerebral blood flow by comparing MRI images at different time points.

[0047] Diffusion magnetic resonance imaging (DMRI) sequences, and neural diffusion tensor imaging (DTI) sequences, are used to obtain not only quantitative values ​​of neural DTI sequences but also information about the density and directional dispersion of neurites (including axons and dendrites), as well as important information about the microstructure of brain white matter and other neural tissues. By analyzing parameters such as fiber bundle anisotropy (FA), diffusion coefficient (ADC), intrasynaptic fraction (NDI), directional dispersion index (ODI), and free water fraction (FFW), tissue characteristics can be better understood.

[0048] Fluid-attenuated inversion recovery (FLAIR) sequences are an advanced brain imaging technique. They effectively suppress cerebrospinal fluid signals, making small lesions or other abnormalities near the ventricles clearly visible. The main advantage of this sequence lies in its ability to suppress free water in T2-weighted images, which greatly improves the ability to identify intracranial lesions, especially in diagnosing cerebral infarction, ischemic lesions, and edema. Magnetic susceptibility-weighted imaging (QSM) sequences are a subset of FLAIR sequences, allowing for the visualization of analytical data regarding outcomes such as brain functional status.

[0049] Phase-contrast angiography (PCA) sequences utilize the phase changes of protons in blood flow to distinguish blood vessels from surrounding tissues. By applying a gradient magnetic field, blood flow velocity and direction are measured to generate vascular images. Primarily used for blood flow analysis in cerebral or cardiovascular systems, such as assessing the hemodynamic status of aneurysms, vascular stenosis, or arteriovenous malformations, it provides quantitative blood flow velocity data.

[0050] In step 102, the brain multimodal image sequence is spatially aligned;

[0051] This invention proposes a spatiotemporal alignment algorithm based on a biomechanical deformation model. Targeting physiological motion characteristics such as cerebral vascular pulsation and cerebrospinal fluid flow, it unifies the spatiotemporal coordinates of different sequences (e.g., dynamic perfusion sequences and static structural sequences) into a physiologically gated coordinate system through a phase correction module. This addresses the inadequacy of traditional rigid registration in adapting to minute movements of the vascular wall. It can achieve sub-voxel level spatial alignment accuracy (≤0.3mm) and simultaneously eliminate artifacts such as respiration and heartbeat through dynamic time-series modeling, making it particularly suitable for image registration in elderly patients or those with tortuous blood vessels.

[0052] Figure 2 This is a flowchart illustrating spatial alignment of brain multimodal image sequences in an embodiment of the present invention. In one embodiment, spatial alignment of brain multimodal image sequences includes:

[0053] Step 201: Unify the brain multimodal images into a preset standardized space;

[0054] Specifically, rigid body registration can be performed on brain multimodal image sequences (such as 3D SNAP-MRA sequences, 3D TOF-MRA sequences, DTI, etc.) to unify different sequence images into a standard space (such as the MNI template) and eliminate global offset caused by individual anatomical differences.

[0055] Step 202: Extract biomechanically relevant anatomical features from three-dimensional contrast-free angiography and plaque hemorrhage imaging sequences, diffusion magnetic resonance imaging sequences, and fluid attenuation inversion recovery sequences;

[0056] Anatomical features, for example:

[0057] Vessel wall elasticity parameters: Plaque location, vessel wall thickness and hemodynamic data were analyzed using 3D SNAP-MRA sequence to deduce the Young's modulus of the vessel wall.

[0058] White matter fiber bundle pathways: The mechanical conduction pathways of white matter fibers were constructed using the FA values ​​of the neural DTI sequence and the fiber bundle tracing results.

[0059] Deformation-sensitive areas of brain tissue: Edema or infarction areas are identified using FLAIR sequences and marked as mechanically deformation-sensitive points.

[0060] Step 203: Extract dynamic parameters from the magnetic resonance brain perfusion imaging sequence and the phase contrast angiography sequence according to time frames to form spatiotemporal features;

[0061] Specifically, dynamic parameters such as blood flow velocity and perfusion volume are extracted from dynamic sequences (such as pCASL perfusion imaging and PCA blood flow sequences) according to time frames to form spatiotemporal features.

[0062] Steps 201-203 can unify the physical spatial reference of multimodal images and extract biomechanical features.

[0063] Step 204: The brain multimodal image in the preset standardized space is segmented into multiple regions. Based on the anatomical and spatiotemporal characteristics of each region, a finite element model based on tetrahedral mesh is constructed as a biomechanical deformation model.

[0064] Several of these areas include blood vessels, brain tissue, and cerebrospinal fluid.

[0065] Step 205, define biomechanical constraints;

[0066] Biomechanical constraints can be defined as follows:

[0067] Hemodynamic loading: Pulsed blood flow pressure is applied at the inlet of the vascular model based on blood flow velocity data from the PCA sequence.

[0068] Cerebrospinal fluid dynamic constraints: fluid-structure coupling (FSI) boundaries of the ventricular walls are defined by the distribution of cerebrospinal fluid signals in FLAIR sequences.

[0069] Step 206: Under biomechanical constraints, predict the spatiotemporal deformation field using a biomechanical deformation model;

[0070] Specifically, biomechanical equations in biomechanical deformation models can be solved using physics engines (such as ABAQUS and COMSOL) to calculate tissue deformation fields at different time points:

[0071] Φ(x,t)=argmin{E elastic +E fluid +E data}

[0072] Among them, E elastic E represents the elastic potential energy of the tissue (derived from a biomechanical deformation model). fluid E represents the coupling energy between blood flow and tissue (blood flow data from PCA / TOF-MRA). data This provides image feature matching capabilities (such as the consistency constraint between the orientation of DTI fiber bundles and the direction of deformation).

[0073] Step 207: Based on the predicted spatiotemporal deformation field, register the brain multimodal images in the preset standardized space to obtain a spatially aligned brain multimodal image sequence.

[0074] Figure 3 This is a flowchart illustrating the registration of brain multimodal images in an embodiment of the present invention. In one embodiment, based on a predicted spatiotemporal deformation field, brain multimodal images in a preset standardized space are registered to obtain a spatially aligned brain multimodal image sequence, including:

[0075] Step 301: Decompose the predicted spatiotemporal deformation field into global rigid motion and local nonlinear deformation;

[0076] Among them, global rigid motion, such as head displacement, and local nonlinear deformation, such as micro-movements of surrounding tissues caused by vascular pulsation, are decomposed into the following formulas:

[0077]

[0078] Among them, the global rigid motion Φ rigid (t) is used to describe the overall translation / rotation of the head, initialized by physiological gating signals (such as respiration, electrocardiogram).

[0079] Local nonlinear deformation Spatiotemporal deformation field Φ for prediction based on biomechanical model pre (x,t) is used as a priori to optimize details through image features.

[0080] Step 302, construct the energy function;

[0081] The energy function can be expressed as:

[0082]

[0083] Image similarity term L img :

[0084] Normalized cross-correlation (NCC) is applied to structural sequences (such as 3D TOF-MRA sequences and FLAIR sequences) to ensure alignment of anatomical structures;

[0085] For functional sequences (such as neural DTI sequences and pCASL sequences), feature point distance loss is used to constrain the fiber bundle orientation to correspond spatially with the perfusion region.

[0086] Biomechanical consistency term L bm It can be represented as:

[0087]

[0088] Here, w(x) is a weighting function, which takes a high value in the blood vessel wall and edema area (to strengthen biomechanical constraints), and a low value in other areas (allowing for flexible adjustment).

[0089] Time smoothness term L temp It can be represented as:

[0090]

[0091] The time smoothness term ensures the continuity of the deformation field between adjacent time frames, avoiding physically infeasible abrupt changes.

[0092] Step 303, iterate in the following steps until the energy function converges or reaches the preset accuracy threshold: estimate the global rigid motion through physiological gating signals and load the biomechanical prior: use the predicted spatiotemporal deformation field as the initial value of the local nonlinear deformation, use the gradient descent algorithm to minimize the energy function, and update the global rigid motion and local nonlinear deformation;

[0093] Among them, physiological gating signals such as respiratory triggers and gradient descent algorithms such as the Adam optimizer.

[0094] Step 304: Generate a spatially aligned brain multimodal image sequence based on the deformation field after the loop iteration ends.

[0095] In step 103, spatially aligned three-dimensional contrast-free vascular imaging and plaque hemorrhage image sequences and spatially aligned time-of-flight imaging sequences are fused to obtain cerebral vascular imaging data.

[0096] Figure 4 This is a flowchart illustrating the process of obtaining cerebral vascular imaging data in an embodiment of the present invention. In one embodiment, spatially aligned three-dimensional contrast-free vascular imaging sequences, plaque hemorrhage imaging sequences, and spatially aligned time-of-flight imaging sequences are fused to obtain cerebral vascular imaging data, including:

[0097] Step 401: Unify the grayscale values ​​of the spatially aligned 3D contrast-free vascular imaging and plaque hemorrhage image sequences and the time-of-flight imaging sequence.

[0098] The grayscale values ​​of spatially aligned 3D contrast-free angiography and plaque hemorrhage image sequences and time-of-flight imaging sequences were unified to the range of [0,1].

[0099] For 3D SNAP-MRA sequences, because they can present details of bleeding in the vessel wall and plaque, the grayscale distribution of different tissues is relatively wide.

[0100] For 3D TOF-MRA sequences, which mainly highlight the vascular lumen and whose grayscale distribution is relatively concentrated in the high signal area of ​​blood flow, a similar formula is used for standardization to ensure the grayscale comparability of the two sequences in subsequent processing.

[0101] Step 402: Enhance the three-dimensional contrast-free angiography and plaque hemorrhage image sequence after grayscale value unification; enhance the vascular orientation features of the time-of-flight imaging image sequence after grayscale value unification.

[0102] Specifically, for 3D SNAP-MRA sequences, the Marr-Hildreth operator (a second-order Gaussian derivative operator) is used to enhance the vessel wall edge. This operator can effectively highlight the boundary between the vessel wall and the surrounding tissue, especially the edge enhancement effect of plaque sites, providing clearer structural features for subsequent fusion with 3D TOF-MRA sequences.

[0103] For 3D TOF-MRA sequences, a gradient-based directional filter is used to enhance the vascular orientation features. This is because 3D TOF-MRA sequences display the overall morphology of the vascular lumen well, but the details of the vascular edge are relatively blurred. This filter can highlight the directionality of the vascular vessels, making it easier to match the vascular wall features of 3D SNAP-MRA sequences.

[0104] Step 403: In the enhanced three-dimensional contrast-free vascular imaging and the intraplaque hemorrhage image sequence, the vessel wall region and the plaque region are segmented; in the enhanced time-of-flight imaging image sequence, the vessel lumen region is segmented.

[0105] Specifically, for 3D SNAP-MRA sequences, threshold segmentation combined with morphological operations is used to segment the vessel wall and plaque regions. Since intraplaque hemorrhage presents a high signal in this sequence, an empirical initial threshold (e.g., a grayscale value of 0.6, which can be adjusted according to actual data) is first set, and areas above the threshold are initially identified as plaque and hemorrhage areas. Then, morphological closing operations are used to fill small cavities and smooth boundaries to obtain an accurate vessel wall and plaque region mask, Mwall SNAP.

[0106] For 3D TOF-MRA sequences, a region growing algorithm is used, with the high signal point in the center of the blood vessel lumen as the seed point, and the region growing is performed based on the continuity and similarity of the blood flow signal to obtain the blood vessel lumen region mask MlumenTOF.

[0107] Step 404: Weight allocation is performed on the segmented vessel wall region, plaque region, and vessel lumen region;

[0108] In the vessel wall and plaque regions (i.e., the regions corresponding to MwallSNAP), 3D SNAP-MRA is given a higher weight, such as 0.8, because this sequence can provide richer information on plaque composition and vessel wall integrity in this region; 3D TOF-MRA is given a weight of 0.2 in this region because it is not good at displaying vessel wall details.

[0109] In the vascular lumen region (the region corresponding to MlumenTOF), the weight of the 3D TOF-MRA sequence is set to 0.7 because it has advantages in displaying the morphology of the vascular lumen and the enhancement effect of blood flow inflow; the weight of the 3D SNAP-MRA sequence in this region is set to 0.3, mainly to supplement some possible fine structural information of the junction between the vascular wall and the lumen.

[0110] In other background areas, both weights are set to 0 to eliminate background noise interference, and the final weight assignment W is generated, which has the same dimension as the image, with each pixel corresponding to a weight value.

[0111] Step 405: Perform multi-scale decomposition on the enhanced three-dimensional contrast-free vascular imaging sequence and the plaque hemorrhage imaging sequence and the time-of-flight imaging sequence to form image pyramids of different scales.

[0112] Gaussian pyramid decomposition was performed on 3D SNAP-MRA and 3D TOF-MRA sequences to construct image pyramids at different scales. First, a Gaussian kernel was defined, and convolved with the original image to obtain blurred images at different scales. For example, the decomposition was performed at three scales: the original image was the finest scale, an intermediate scale image was obtained after one Gaussian blur, and the coarsest scale image was obtained after another Gaussian blur.

[0113] Images at each scale contain different levels of structural information. The coarse scale contains the overall direction and general shape of the blood vessels, while the fine scale contains the detailed features of the vessel walls and lumens.

[0114] Step 406: At different scales, the image pyramids are fused according to the weight allocation.

[0115] At the coarsest scale, based on the weight allocation W, corresponding pixels of the 3D SNAP-MRA sequence and the 3D TOF-MRA sequence are weighted and fused, i.e.:

[0116]

[0117] Among them I SNAP，coarse and I TOF，coarse These are images of 3D SNAP-MRA and 3D TOF-MRA sequences at the coarsest scale, respectively. It is a coarse-scale image after fusion.

[0118] For intermediate and fine scales, the same formula is used for fusion. However, before fusion, the image fused at the previous scale (coarser scale) is upsampled (e.g., by bilinear interpolation) to make its size consistent with the current scale image, and then weighted fusion is performed. This allows the global information at the coarse scale to be gradually fused with the local detail information at the fine scale, resulting in a more comprehensive fused image at different scales.

[0119] Finally, the fused image at the finest scale was used as the final fused backbone image data. This image not only contains the advantages of 3D TOF-MRA sequence in terms of the overall morphology and blood flow of the vascular lumen, but also combines the detailed features of 3DSNAP-MRA sequence in terms of the vascular wall and plaque hemorrhage, providing a richer and more accurate data foundation for subsequent operations such as physiological feature tree construction.

[0120] In step 104, cerebral vascular imaging data are used as the root and trunk nodes of the tree, hemodynamic parameter data in spatially aligned phase-contrast vascular imaging sequences are used as the branch nodes representing blood supply status, neural structure parameter data in spatially aligned diffusion magnetic resonance imaging sequences are used as the branch nodes representing neural structures, and physicochemical parameter data in spatially aligned fluid attenuation inversion recovery sequences are used as the leaf nodes representing functional metabolism. The hemodynamic parameter data in spatially aligned magnetic resonance brain perfusion imaging sequences are refined to the pixel level and used as the leaf nodes representing metabolic activity, thus constructing a brain biometric tree of the target object.

[0121] Among them, cerebral vascular imaging data, as the backbone information of the tree structure, integrates the vascular wall structure (SNAP) and lumen morphology (TOF), covering the complete vascular tree from the heart to the terminal cerebral vessels. Its biological significance is that the vascular system is the material basis of brain function, and the backbone layer represents the "main trunk of nutrient supply," providing spatial positioning benchmarks for subsequent parameters.

[0122] Hemodynamic tree nodes: Hemodynamic parameters such as blood flow velocity and direction from the PCA sequence are introduced as "secondary branches" of the trunk node, i.e. tree nodes, to reflect the intravascular blood supply status.

[0123] Neural structural tree nodes: Neural structural parameters such as fiber bundle orientation and anisotropy (FA) included in the DTI sequence are used as another set of tree nodes to characterize the integrity of neural conduction pathways.

[0124] Functional metabolic leaf nodes: using physicochemical data such as iron deposition and calcification from QSM sequences as leaf node quality parameters to correlate with neurodegenerative changes;

[0125] Metabolic activity data: The metabolic activity of “leaf nodes” is simulated by refining hemodynamic parameters (such as CBF) of the pCASL perfusion sequence to the pixel level.

[0126] Figure 5 This is a flowchart illustrating the construction of a brain biometric tree for a target object in an embodiment of the present invention. In one embodiment, cerebral vascular imaging data are used as the root and trunk nodes, hemodynamic parameter data from spatially aligned phase-contrast vascular imaging sequences are used as branch nodes representing blood supply status, neural structure parameter data from spatially aligned diffusion magnetic resonance imaging sequences are used as branch nodes representing neural structure, and physicochemical parameter data from spatially aligned fluid attenuation inversion recovery sequences are used as leaf nodes representing functional metabolism. Hemodynamic parameter data from spatially aligned magnetic resonance brain perfusion imaging sequences, refined to the pixel level, are used as leaf nodes representing metabolic activity. The construction of the brain biometric tree for the target object includes:

[0127] Step 501: Extract the large cervical blood vessel structure from the cerebral vascular imaging data, use it as the root node of the tree, perform spatial localization, and extract the global geometric parameters of the root node.

[0128] The major vascular structures in the neck include the aortic arch, common carotid artery, vertebral artery, and other initial segments of the vessels. A region growing algorithm combined with manual correction can be used to ensure segmentation accuracy (Dice coefficient > 0.95), mark the coordinates of the vessel origin (such as the bifurcation of the common carotid artery), and map them to the standard space (MNI coordinates).

[0129] Calculate global geometric parameters:

[0130] Vessel diameter (D_root): Average diameter of the proximal end of the common carotid artery;

[0131] Blood vessel length (L_root): The length of the main blood vessel from the aortic arch to the base of the skull;

[0132] Plaque burden (PL_root): The percentage of neck vascular plaque volume in a SNAP sequence.

[0133] Create the root node attributes: RootNode = {ID:Root,Type:Carotid Artery,D_root,L_root,PL_root}.

[0134] Biological significance: As the "source" of cerebral blood supply, the parameters of the root nodes reflect the basic state of vascular health throughout the body.

[0135] Step 502: Extract the branch structure of the circle of arteries at the base of the brain from the cerebral vascular imaging data as the trunk node, and form the center line of the vascular tree. Extract the local geometric features of each trunk node, and establish the connection between the root node and the trunk node through vascular continuity.

[0136] The branching structures of the circle of arteries at the base of the brain include the internal carotid siphon segment, the M1 segment of the middle cerebral artery, and the basilar artery. The vascular skeleton extraction algorithm can be used to generate the vascular tree centerline; each branch is named according to anatomical standards (such as Fischer segmentation) (e.g., MCA M1, ACA A2).

[0137] Calculate the following local geometric features:

[0138] Stenosis ratio (DS_trunk): The ratio of the minimum lumen diameter to the normal proximal diameter measured by TOF-MRA;

[0139] Vessel wall thickness (WT_trunk): The thickness of the vessel wall at the plaque site in the SNAP sequence (automatically segmented PISR low signal region);

[0140] Intraplaque hemorrhage (PB_trunk): The volume of the hemorrhage region in a SNAP sequence (high signal is segmented using a threshold method).

[0141] Finally, for each backbone segment (e.g., MCA M1), construct the attributes:

[0142] TrunkNode={ID:MCA-M1,DS_trunk,WT_trunk,PB_trunk}.

[0143] Parent-child relationship: The trunk node is a child node of the root node (e.g., internal carotid artery → MCA M1), and the connection is established through the continuity of blood vessels.

[0144] Step 503: For each trunk node, extract the vascular segment structure of the trunk node from the phase-contrast angiography sequence as a branch node, and extract the hemodynamic parameter data of the branch node.

[0145] The hemodynamic parameters included: peak velocity (Vpeak), mean velocity (Vmean), direction of flow consistency index (DOI, reflecting laminar / turbulent flow), and wall shear stress (WSS) and oscillating shear index (OSI) derived from the fluid dynamics model.

[0146] Constructing bloodline tree nodes:

[0147] HemodynamicsNode={ID:MCA-M1-blood flow, Vpeak_branch, DOI_branch, WSS_branch}.

[0148] Hierarchical relationship: As a child node of the trunk node, the naming rule is "trunk ID-bloodline".

[0149] Step 504: The white matter region around the trunk node is taken as the branch node, and the neural structural parameter data of the branch node is extracted from the diffusion magnetic resonance imaging sequence.

[0150] White matter regions around tree trunk nodes, such as the posterior limb of the internal capsule.

[0151] In one embodiment, extracting neural structural parameter data of tree nodes from a diffusion magnetic resonance imaging sequence includes:

[0152] The white matter fiber bundles are reconstructed using a probabilistic fiber bundle tracing algorithm, and neural structural parameter data of the tree nodes in the diffusion magnetic resonance imaging sequence are extracted. The neural structural parameter data includes geometric parameter data and microscopic parameter data.

[0153] Among them, the geometric parameters in the neural structural parameter data include fiber bundle length, branch density, and crossing angle;

[0154] Microscopic parameter data include FA value (reflecting fiber bundle integrity) and ADC value (diffusion coefficient, indicating myelin sheath status).

[0155] For example, the corticospinal tract and corpus callosum branches in neural structural parameter data can be used as tree branch nodes to establish spatial proximity associations with vascular nodes in the trunk node. For example, if the FA value of a white matter fiber tract near a stenotic segment of a blood vessel decreases, it can be marked as a "structural-neural interaction node".

[0156] Construct neural structure tree nodes: NeuralStructureNode = {ID:MCA-M1-fiber bundle,FA_branch,ADC_branch,FL_branch}.

[0157] Alongside the blood supply branch node, it is a child node of the trunk node, and its naming rule is "trunk ID-fiber bundle".

[0158] Step 505: Extract the pathological regions of the tree nodes from the liquid decay inversion recovery sequence as leaf nodes, and extract the physicochemical parameter data of the pathological regions of the leaf nodes.

[0159] Pathological areas include edema, demyelination, or infarction areas; physicochemical parameter data, including iron deposition density and calcification volume, can be extracted using the Otsu thresholding method combined with morphological opening and closing operations.

[0160] Construct a functional metabolic leaf node: MetabolismNode1 = {ID:Left Basal Ganglion-Edema,LesionVol_leaf,Loc_leaf}.

[0161] Hierarchical association: Sub-nodes that are nodes in the tree branches of neural structures (e.g., posterior limb fiber bundles of the internal capsule → edema area of ​​the basal ganglia).

[0162] Step 506: Extract the blood flow regions of the tree branch nodes from the magnetic resonance brain perfusion imaging sequence as leaf nodes, and extract pixel-level hemodynamic parameter data of the blood flow regions of the leaf nodes.

[0163] Voxel-level cerebral blood flow (CBF) calculations were performed on magnetic resonance brain perfusion imaging sequences. CBF heatmaps were generated using statistical dynamic parameter mapping (SPM), and pixel-level hemodynamic parameter data were extracted, including:

[0164] Local CBF value (CBF_leaf): Average blood flow in the target area (mL / 100g / min);

[0165] Perfusion coefficient of variation (CV_CBF): reflects the homogeneity of blood flow within the region (CV = standard deviation / mean).

[0166] Construct a leaf node for metabolic activity: MetabolismNode2 = {ID: left prefrontal cortex-BA9, ​​CBF_leaf, CV_CBF}.

[0167] Alongside the leaf nodes of functional metabolism, they serve as child nodes of the blood supply tree nodes (such as branches of the anterior cerebral artery → prefrontal cortex).

[0168] Figure 6 This is a flowchart illustrating the verification of a brain biometric tree in an embodiment of the present invention; in one embodiment, the method further includes:

[0169] Step 601: Establish cross-modal associations for each node in the brain biometric tree through coordinate mapping;

[0170] For example, the stenosis rate (DS_trunk) of the MCA M1 segment of the trunk node is associated with an increase in WSS_branch (>4Pa) of the corresponding vascular branch node; the decrease in FA_branch (<0.3) of the neural structural branch node and the decrease in CBF_leaf (<30mL / 100g / min) of the leaf node spatially overlap.

[0171] Step 602: Verify the brain biometric tree according to the preset parameter chain rules;

[0172] Parameter chain rules include:

[0173] Positive rules: For example, root node PL_root > 20% → trunk node DS_trunk > 50% → bloodline branch node DOI_branch < 0.4 (turbulence);

[0174] Reverse rule: For example, leaf node LesionVol_leaf > 5mL → neural structure branch node ADC_branch > 0.0015mm 2 / s (indicating edema).

[0175] Step 603: After the brain biometric tree passes the verification, the brain biometric tree is visualized.

[0176] Tree root / trunk: represented by pipe diameter D_root / D_trunk, color-coded PL_root / DS_trunk;

[0177] Tree branches: Blood supply nodes are represented by arrow velocity vectors, and neural nodes are rendered using fiber bundle trajectories;

[0178] Leaves: LesionVol_leaf is represented by bubble size, and CBF_leaf is represented by color gradient.

[0179] Figure 7 This is a visualization of the brain biometric tree in an embodiment of the present invention, which clearly shows the structure of branches, trunk, roots and leaves (forming leaves).

[0180] In one embodiment, the method further includes:

[0181] The target object's brain biometric tree is input into the abnormal data analysis module to obtain abnormal data of the target brain. The abnormal data analysis module is trained based on the brain biometric trees of multiple target objects.

[0182] In one embodiment, the method further includes:

[0183] An attention heatmap generator is embedded in the brain biometrics tree to visualize the brain biometrics tree.

[0184] After embedding the attention heatmap generator, the brain biometric tree can be enhanced with the following display effects:

[0185] First color zone: TOF-MRA shows luminal stenosis >70%;

[0186] First color region: SNAP-MRA detected intraplaque hemorrhage signals;

[0187] The third color area: DTI shows the interruption of nerve fiber bundles in the corresponding brain region.

[0188] The colors mentioned above can be determined according to the actual situation; no restrictions are imposed here.

[0189] The above embodiments address the black box problem of traditional AI and increase doctors' trust in algorithmic decision-making.

[0190] In this embodiment of the invention, a brain biometrics tree and gene association network can be established to promote the "precision medicine model" of imaging genomics. For example:

[0191] Individuals carrying the APOEε4 allele showed significantly higher levels of iron deposition in the hippocampus in their QSM sequences compared to wild-type individuals (p < 0.05).

[0192] Matrix metalloproteinase-9 (MMP-9) gene polymorphism was associated with the thickness of the SNAP-MRA plaque fibrous cap (r = 0.53).

[0193] In this embodiment of the invention, a 3D physiological feature tree VR interactive platform can also be developed. Doctors can navigate the multi-dimensional data space of blood vessels, nerves, and metabolism through gestures, realizing a closed loop of the entire process from image acquisition to analysis, visualization, and treatment planning, thereby improving the efficiency and accuracy of multidisciplinary team (MDT) consultations. For example:

[0194] Penetration-sensor visualization: Simultaneous observation of vascular wall plaques and deep brain white matter fiber bundles;

[0195] Dynamic simulation: Simulate the impact of different treatment options (such as stent placement vs. drug therapy) on hemodynamic parameters.

[0196] This invention also proposes a nutrient-phenotype analysis theory for analyzing brain function. The core idea is that an organism's phenotype (such as organ functional state) is jointly determined by the amount of nutrients it receives and the structural efficiency of the supply pathway.

[0197] In brain function analysis, nutrients are cerebral blood flow perfusion parameters (such as CBF of pCASL), representing the energy and oxygen supply to nerve cells; the supply pathway structure is the geometry of cerebral blood vessels (such as stenosis rate of SNAP / TOF and plaque load), determining nutrient delivery efficiency; and phenotypes are neural structural and metabolic parameters (such as FA value of neural DTI sequences and iron deposition in QSM), reflecting the ultimate impact of insufficient nutrient supply. The specific analytical steps are given below.

[0198] Figure 8 This is a flowchart of brain function analysis in an embodiment of the present invention. In one embodiment, the method further includes:

[0199] Step 801: Using the root node of the brain biometric tree as the structural layer, the branch node as the nutrient layer, and the leaf node as the phenotypic layer, calculate the structural efficiency from the trunk node to the branch node and the effective nutrient from the branch node to the leaf node.

[0200] For each trunk node (e.g., MCA M1 segment), calculate the structural efficiency (SLF) expressed as the structural loss factor: SLF = 1 - exp(-k × DS) (k is a constant, calibrated according to the vessel diameter)

[0201] Significance: The closer SLF is to 1, the greater the blood flow loss caused by stenosis (e.g., when DS = 50%, SLF ≈ 0.39, which means blood flow is reduced by 39%).

[0202] The effective nutrients (CBF_eff_leaf) of a leaf node = the CBF of a branch node × (1 - the sum of the SLF of all upstream trunk nodes);

[0203] Example: If the SLF of the MCA M1 segment is 0.3 and the SLF of the lenticulostriate artery is 0.2, then the CBF_eff_leaf of the basal ganglia region is equal to the CBF of the common carotid artery × (1-0.3-0.2) = the CBF of the common carotid artery × 0.5.

[0204] Step 802: For each leaf node, fit a nutrient-phenotype regression equation based on the available nutrients;

[0205] The nutrient-phenotypic regression equation for each leaf node can be expressed as:

[0206] FA_leaf=α0+α1×CBF_eff_leaf+α2×Fe_content+∈

[0207] Variable description:

[0208] FA_leaf: Neural fiber integrity (phenotype) of leaf nodes;

[0209] CBF_eff_leaf: Effective nutrient supply;

[0210] Fe_content: Iron deposition amount (reflects neurodegenerative changes and interferes with nutrient utilization efficiency).

[0211] Step 803: Calculate the nutrient attenuation rate and phenotypic abnormality index for each nutrient transport path, and determine the significant nutrient-phenotype association paths, where each path from the root node to the leaf node represents a nutrient transport path.

[0212] For example, the nutrient transport pathway can be: common carotid artery (root) → MCA M1 (trunk) → insular leaf branch (twig) → insular leaf cortex (leaf).

[0213] Nutrient decay rate = (CBF_root - CBF_eff_leaf) / CBF_root; where CBF_root is the effective nutrient supply to the root node;

[0214] Phenotypic abnormality index = |FA_leaf-FA_normal| / FA_normal, where FA_normal is a constant.

[0215] If the nutrient attenuation rate of a certain nutrient transport path is greater than the first nutrient attenuation rate threshold (e.g., 40%) and the phenotypic aberration index is greater than the first phenotypic aberration index threshold (e.g., 20%), then it is determined to be a nutrient-phenotype significant path.

[0216] Step 804: Perform causal attribution analysis on significant pathways of nutrient-phenotype association to obtain the brain function analysis results of the target subjects;

[0217] The following are typical examples of causal attribution analysis:

[0218] Structure-dominant anomalies:

[0219] Characteristics: Nutrient decay rate > first nutrient decay rate threshold (e.g., 30%), and trunk node DS > 50% or PL > 30%;

[0220] Attribution: Vascular stenosis / plaque leads to insufficient nutrient delivery, such as hypothalamic regional phenotypic abnormalities accompanied by basilar artery stenosis.

[0221] Nutrient utilization disorder:

[0222] Characteristics: Nutrient decay rate <20%, but leaf node Fe_content > normal mean +1 SD;

[0223] Attribution: Abnormal neurometabolism (such as iron deposition) leads to decreased nutrient utilization efficiency, which is common in Alzheimer's disease.

[0224] Mixed-factor anomalies:

[0225] Characteristics: Nutrient decay rate > 20% and Fe_content > normal mean + 1 SD;

[0226] Attribution: The combined effects of vascular structural damage and abnormal neurometabolism, such as white matter lesions in the elderly.

[0227] The percentages above are for illustrative purposes only and may be modified according to actual circumstances.

[0228] After completing the above analysis, visualizations can be created, including:

[0229] Tree-structured weighted mapping:

[0230] Nutrient-phenotypic association strength of nodes encoded by color:

[0231] First color: Structural damage dominant (DS / PL parameter weight > 60%);

[0232] The second color: dominated by nutrient deficiency (CBF_eff parameter weight > 60%);

[0233] The third color: Mixed factors (the weights of each parameter are close).

[0234] For each abnormal leaf node, the output includes an analysis report containing the following: ranking of key influencing factors (e.g., MCAM1 stenosis (35%) → reduced blood flow in the lenticulostriate arteries (28%) → iron deposition in the basal ganglia (22%)); a parameter comparison heatmap with the normal population; and recommended intervention directions (e.g., vascular reconstruction, neuroprotective therapy).

[0235] In this embodiment of the invention, the correlation of parameter changes can be calculated for two scans of the same target at an interval of 6-12 months: Δphenotypic index = β × Δ nutrient effective supply + γ × Δ structural loss factor. If β or γ is significant (p < 0.05), the effectiveness of the correlation model is verified.

[0236] Based on a tree-like parameter set from multiple patients, a causal inference model is constructed using the random forest algorithm. Inputting structural / nutrient parameters, it predicts the risk of phenotypic abnormalities.

[0237] Risk probability = f(DS,PL,CBF_eff,Fe_content)

[0238] The model can be further used for prospective prediction (such as assessment of neurological function recovery after stroke). These brain function analysis results, such as the risk of phenotypic abnormalities and prospective predictions, can be used as a reference for doctors to assist them in making decisions.

[0239] This invention also proposes a brain multimodal image data processing device, the principle of which is similar to the brain multimodal image data processing method, and will not be described in detail here.

[0240] Figure 9 This is a schematic diagram of the brain multimodal imaging data processing device in an embodiment of the present invention, comprising:

[0241] The brain multimodal image sequence acquisition module 901 is used to acquire the brain multimodal image sequence of the target object. The brain multimodal images include three-dimensional contrast-free angiography and plaque hemorrhage image sequence, time-of-flight imaging image sequence, magnetic resonance brain perfusion imaging sequence, diffusion nuclear magnetic resonance data sequence, fluid attenuation inversion recovery sequence, and phase contrast angiography sequence.

[0242] Spatial alignment module 902 is used for spatial alignment of brain multimodal image sequences;

[0243] The fusion module 903 is used to fuse spatially aligned three-dimensional contrast-free vascular imaging sequences with plaque hemorrhage image sequences and spatially aligned time-of-flight imaging sequences to obtain cerebral vascular imaging data.

[0244] The brain biometric tree construction module 904 is used to construct a brain biometric tree for a target object by using cerebral vascular imaging data as the root and trunk nodes, hemodynamic parameter data from spatially aligned phase-contrast vascular imaging sequences as branch nodes representing blood supply status, neural structure parameter data from spatially aligned diffusion magnetic resonance imaging sequences as branch nodes representing neural structures, physicochemical parameter data from spatially aligned fluid attenuation inversion recovery sequences as leaf nodes representing functional metabolism, and refining hemodynamic parameter data from spatially aligned magnetic resonance brain perfusion imaging sequences to the pixel level as leaf nodes representing metabolic activity. The brain biometric tree of the target object is used to perform brain function analysis on the target object.

[0245] In one embodiment, the space alignment module is used for:

[0246] Unify multimodal brain imaging into a pre-defined standardized space;

[0247] Biomechanically relevant anatomical features were extracted from three-dimensional contrast-free angiography and plaque hemorrhage imaging sequences, diffusion magnetic resonance imaging sequences, and fluid attenuation inversion recovery sequences.

[0248] Dynamic parameters were extracted from magnetic resonance brain perfusion imaging sequences and phase-contrast angiography sequences according to time frames to form spatiotemporal features;

[0249] The brain multimodal images in the pre-defined standardized space are segmented into multiple regions. Based on the anatomical and spatiotemporal characteristics of each region, a finite element model based on tetrahedral mesh is constructed as a biomechanical deformation model.

[0250] Define biomechanical constraints;

[0251] Under biomechanical constraints, a biomechanical deformation model is used to predict the spatiotemporal deformation field;

[0252] Based on the predicted spatiotemporal deformation field, brain multimodal images in a pre-defined standardized space are registered to obtain a spatially aligned brain multimodal image sequence.

[0253] In one embodiment, the spatial alignment module is used to: decompose the predicted spatiotemporal deformation field into global rigid motion and local nonlinear deformation;

[0254] Construct the energy function;

[0255] The following steps are iterated until the energy function converges or reaches a preset accuracy threshold: estimate the global rigid motion through physiological gating signals and load biomechanical priors; use the predicted spatiotemporal deformation field as the initial value of the local nonlinear deformation, minimize the energy function using the gradient descent algorithm, and update the global rigid motion and local nonlinear deformation.

[0256] Based on the deformation field after the loop iteration ends, a spatially aligned brain multimodal image sequence is generated.

[0257] In one embodiment, the fusion module is configured to:

[0258] Gray values ​​were unified for spatially aligned 3D contrast-free angiography and plaque hemorrhage imaging sequences and time-of-flight imaging sequences.

[0259] Enhancement processing was performed on the three-dimensional contrast-free angiography and plaque hemorrhage image sequences after grayscale value unification; enhancement processing was performed on the vascular orientation features of the time-of-flight imaging image sequences after grayscale value unification.

[0260] In contrast-free 3D angiography and intraplaque hemorrhage image sequences after enhancement, the vessel wall region and plaque region were segmented; in time-of-flight imaging sequences after enhancement, the vessel lumen region was segmented.

[0261] Weights are assigned to the segmented vessel wall region, plaque region, and vessel lumen region.

[0262] Multi-scale decomposition was performed on the enhanced three-dimensional contrast-free angiography image sequence and the time-of-flight imaging image sequence to form image pyramids of different scales.

[0263] At different scales, the image pyramids are fused according to the weight allocation.

[0264] In one embodiment, the brain biometric tree is used for:

[0265] Physicochemical parameter data are used as tree nodes representing blood supply status; neural structural parameter data from diffusion magnetic resonance imaging sequences are used as tree nodes representing neural structure; physicochemical parameter data from fluid attenuation inversion recovery sequences are used as leaf nodes representing functional metabolism; and hemodynamic parameter data from magnetic resonance brain perfusion imaging sequences, refined to the pixel level, are used as leaf nodes representing metabolic activity. This constructs a brain biometric tree for the target object, including:

[0266] The large cervical blood vessel structure was extracted from cerebral vascular imaging data and used as the root node of the tree. The tree was then spatially located, and the global geometric parameters of the root node were extracted.

[0267] The branching structures of the circle of arteries at the base of the brain are extracted from cerebral vascular imaging data as trunk nodes, and the central line of the vascular tree is formed. The local geometric features of each trunk node are extracted, and the connection between the root node and the trunk node is established through vascular continuity.

[0268] For each trunk node, the vascular segment structure of that trunk node is extracted from the phase-contrast angiography sequence and used as a branch node. The hemodynamic parameter data of the branch node are then extracted.

[0269] The white matter region surrounding the trunk node is used as the branch node, and neural structural parameter data of the branch node are extracted from the diffusion magnetic resonance imaging sequence.

[0270] The pathological regions of the tree nodes were extracted from the liquid decay inversion recovery sequence and used as leaf nodes. The physicochemical parameter data of the pathological regions of the leaf nodes were then extracted.

[0271] Blood flow regions of tree nodes are extracted from magnetic resonance brain perfusion imaging sequences and used as leaf nodes. Pixel-level hemodynamic parameter data of the blood flow regions of the leaf nodes are then extracted.

[0272] Based on the root node, trunk node, branch node, and leaf node, a brain biometric tree of the target object is constructed.

[0273] In one embodiment, the brain biometric tree is used for:

[0274] The white matter fiber bundles are reconstructed using a probabilistic fiber bundle tracing algorithm, and neural structural parameter data of the tree nodes in the diffusion magnetic resonance imaging sequence are extracted. The neural structural parameter data includes geometric parameter data and microscopic parameter data.

[0275] In one embodiment, the brain biometric tree is used for:

[0276] Cross-modal associations of nodes in the brain biometrics tree are established through coordinate mapping;

[0277] The brain biometric tree is validated according to the preset parameter chain rules.

[0278] After the brain biometrics tree is verified and approved, it is then visualized.

[0279] In one embodiment, the device further includes a brain function analysis module for:

[0280] Using the root nodes of the brain biometric tree as the structural layer, the branch nodes as the nutrient layer, and the leaf nodes as the phenotypic layer, we calculate the structural efficiency from the trunk node to the branch node and the effective nutrient from the branch node to the leaf node.

[0281] For each leaf node, a nutrient-phenotype regression equation is fitted based on the available nutrients.

[0282] Calculate the nutrient depletion rate and phenotypic anomalous index for each nutrient transport path, and identify significant nutrient-phenotype associated paths, where each path from the root node to the leaf node represents a nutrient transport path.

[0283] Causal attribution analysis was performed on significant pathways associated with nutrients and phenotypes to obtain brain function analysis results for the target subjects.

[0284] In summary, the method and apparatus proposed in this invention obtain a multimodal brain image sequence of the target object. The multimodal brain images include a three-dimensional contrast-free angiography and plaque hemorrhage image sequence, a time-of-flight imaging image sequence, a magnetic resonance brain perfusion imaging sequence, a diffusion magnetic resonance imaging (MRI) data sequence, a fluid attenuation inversion recovery (FIR) sequence, and a phase-contrast angiography sequence. The multimodal brain image sequences are spatially aligned. The spatially aligned three-dimensional contrast-free angiography and plaque hemorrhage image sequence and the spatially aligned time-of-flight imaging image sequence are fused to obtain cerebral vascular image data. The cerebral vascular image data is used as the root node and... The tree structure uses hemodynamic parameters from spatially aligned phase-contrast angiography sequences as branch nodes representing blood supply status, neural structural parameters from spatially aligned diffusion magnetic resonance imaging sequences as branch nodes representing neural structures, and physicochemical parameters from spatially aligned fluid attenuation inversion recovery sequences as leaf nodes representing functional metabolism. Hemodynamic parameters from spatially aligned magnetic resonance brain perfusion imaging sequences, refined to pixel level, are then used as leaf nodes representing metabolic activity. This process constructs a brain biometric tree for the target subject, which is then used for brain function analysis. Through this process, the acquired multimodal brain image sequences are given biological meaning, forming a quantified brain biometric tree for use in brain function evaluation from a naturalistic perspective.

[0285] This invention also provides a computer device. Figure 10 This is a schematic diagram of a computer device in an embodiment of the present invention. The computer device 1000 includes a memory 1010, a processor 1020, and a computer program 1030 stored in the memory 1010 and executable on the processor 1020. When the processor 1020 executes the computer program 1030, it implements the above-mentioned brain multimodal image data processing method.

[0286] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described brain multimodal image data processing method.

[0287] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described brain multimodal image data processing method.

[0288] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0289] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0290] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0291] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0292] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for processing multimodal brain imaging data, characterized in that, include: Obtain a multimodal brain imaging sequence of the target object, wherein the multimodal brain imaging includes a three-dimensional contrast-free angiography and plaque hemorrhage imaging sequence, a time-of-flight imaging sequence, a magnetic resonance brain perfusion imaging sequence, a diffusion-weighted magnetic resonance imaging data sequence, a fluid attenuation inversion recovery sequence, and a phase-contrast angiography sequence. Spatial alignment of brain multimodal image sequences; Cerebrovascular imaging data were obtained by fusing spatially aligned 3D contrast-free vascular imaging sequences with intraplaque hemorrhage imaging sequences and spatially aligned time-of-flight imaging sequences. Using cerebral vascular imaging data as root and trunk nodes, hemodynamic parameters from spatially aligned phase-contrast vascular imaging sequences as branch nodes representing blood supply status, neural structural parameters from spatially aligned diffusion magnetic resonance imaging sequences as branch nodes representing neural structures, and physicochemical parameters from spatially aligned fluid attenuation inversion recovery sequences as leaf nodes representing functional metabolism, and refining hemodynamic parameters from spatially aligned magnetic resonance brain perfusion imaging sequences to pixel level as leaf nodes representing metabolic activity, a brain biometric tree of the target object is constructed. This brain biometric tree of the target object is used for brain function analysis of the target object.

2. The method as described in claim 1, characterized in that, Spatial alignment of brain multimodal image sequences, including: Unify multimodal brain imaging into a pre-defined standardized space; Biomechanically relevant anatomical features were extracted from three-dimensional contrast-free angiography and plaque hemorrhage imaging sequences, diffusion magnetic resonance imaging sequences, and fluid attenuation inversion recovery sequences. Dynamic parameters were extracted from magnetic resonance brain perfusion imaging sequences and phase-contrast angiography sequences according to time frames to form spatiotemporal features; The brain multimodal images in the pre-defined standardized space are segmented into multiple regions. Based on the anatomical and spatiotemporal characteristics of each region, a finite element model based on tetrahedral mesh is constructed as a biomechanical deformation model. Define biomechanical constraints; Under biomechanical constraints, a biomechanical deformation model is used to predict the spatiotemporal deformation field; Based on the predicted spatiotemporal deformation field, brain multimodal images in a pre-defined standardized space are registered to obtain a spatially aligned brain multimodal image sequence.

3. The method as described in claim 2, characterized in that, Based on the predicted spatiotemporal deformation field, brain multimodal images in a pre-defined standardized space are registered to obtain a spatially aligned brain multimodal image sequence, including: The predicted spatiotemporal deformation field is decomposed into global rigid motion and local nonlinear deformation; Construct the energy function; The following steps are iterated until the energy function converges or reaches a preset accuracy threshold: estimate the global rigid motion through physiological gating signals and load biomechanical priors; use the predicted spatiotemporal deformation field as the initial value of the local nonlinear deformation, minimize the energy function using the gradient descent algorithm, and update the global rigid motion and local nonlinear deformation. Based on the deformation field after the loop iteration ends, a spatially aligned brain multimodal image sequence is generated.

4. The method as described in claim 1, characterized in that, Cerebrovascular imaging data were obtained by fusing spatially aligned 3D contrast-free angiography sequences with intraplaque hemorrhage images and spatially aligned time-of-flight imaging sequences, including: Gray values ​​were unified for spatially aligned 3D contrast-free angiography and plaque hemorrhage imaging sequences and time-of-flight imaging sequences. Enhancement processing was performed on the three-dimensional contrast-free angiography and plaque hemorrhage image sequences after grayscale value unification; enhancement processing was performed on the vascular orientation features of the time-of-flight imaging image sequences after grayscale value unification. In contrast-free 3D angiography and intraplaque hemorrhage image sequences after enhancement, the vessel wall region and plaque region were segmented; in time-of-flight imaging sequences after enhancement, the vessel lumen region was segmented. Weights are assigned to the segmented vessel wall region, plaque region, and vessel lumen region. Multi-scale decomposition was performed on the enhanced three-dimensional contrast-free angiography image sequence and the time-of-flight imaging image sequence to form image pyramids of different scales. At different scales, the image pyramids are fused according to the weight allocation.

5. The method as described in claim 1, characterized in that, Physicochemical parameter data are used as tree nodes representing blood supply status; neural structural parameter data from diffusion magnetic resonance imaging sequences are used as tree nodes representing neural structure; physicochemical parameter data from fluid attenuation inversion recovery sequences are used as leaf nodes representing functional metabolism; and hemodynamic parameter data from magnetic resonance brain perfusion imaging sequences, refined to the pixel level, are used as leaf nodes representing metabolic activity. This constructs a brain biometric tree for the target object, including: The large cervical blood vessel structure was extracted from cerebral vascular imaging data and used as the root node of the tree. The tree was then spatially located, and the global geometric parameters of the root node were extracted. The branching structures of the circle of arteries at the base of the brain are extracted from cerebral vascular imaging data as trunk nodes, and the central line of the vascular tree is formed. The local geometric features of each trunk node are extracted, and the connection between the root node and the trunk node is established through vascular continuity. For each trunk node, the vascular segment structure of that trunk node is extracted from the phase-contrast angiography sequence and used as a branch node. The hemodynamic parameter data of the branch node are then extracted. The white matter region surrounding the trunk node is used as the branch node, and neural structural parameter data of the branch node are extracted from the diffusion magnetic resonance imaging sequence. The pathological regions of the tree nodes were extracted from the liquid decay inversion recovery sequence and used as leaf nodes. The physicochemical parameter data of the pathological regions of the leaf nodes were then extracted. Blood flow regions of tree nodes are extracted from magnetic resonance brain perfusion imaging sequences and used as leaf nodes. Pixel-level hemodynamic parameter data of the blood flow regions of the leaf nodes are then extracted. Based on the root node, trunk node, branch node, and leaf node, a brain biometric tree of the target object is constructed.

6. The method as described in claim 5, characterized in that, Neural structural parameter data were extracted from diffusion magnetic resonance imaging sequences, including: The white matter fiber bundles are reconstructed using a probabilistic fiber bundle tracing algorithm, and neural structural parameter data of the tree nodes in the diffusion magnetic resonance imaging sequence are extracted. The neural structural parameter data includes geometric parameter data and microscopic parameter data.

7. The method as described in claim 1, characterized in that, Also includes: Cross-modal associations of nodes in the brain biometrics tree are established through coordinate mapping; The brain biometric tree is validated according to the preset parameter chain rules. After the brain biometrics tree is verified and approved, it is then visualized.

8. The method as described in claim 1, characterized in that, Also includes: Using the root nodes of the brain biometric tree as the structural layer, the branch nodes as the nutrient layer, and the leaf nodes as the phenotypic layer, we calculate the structural efficiency from the trunk node to the branch node and the effective nutrient from the branch node to the leaf node. For each leaf node, a nutrient-phenotype regression equation is fitted based on the available nutrients. Calculate the nutrient depletion rate and phenotypic anomalous index for each nutrient transport path, and identify significant nutrient-phenotype associated paths, where each path from the root node to the leaf node represents a nutrient transport path. Causal attribution analysis was performed on significant pathways associated with nutrients and phenotypes to obtain brain function analysis results for the target subjects.

9. A brain multimodal imaging data processing device, characterized in that, include: The brain multimodal image sequence acquisition module is used to acquire brain multimodal image sequences of the target object. The brain multimodal images include three-dimensional contrast-free angiography and plaque hemorrhage image sequences, time-of-flight imaging image sequences, magnetic resonance brain perfusion imaging sequences, diffusion nuclear magnetic resonance data sequences, fluid attenuation inversion recovery sequences, and phase-contrast angiography sequences. The spatial alignment module is used to spatially align brain multimodal image sequences; The fusion module is used to fuse spatially aligned 3D contrast-free vascular imaging sequences with plaque hemorrhage imaging sequences and spatially aligned time-of-flight imaging sequences to obtain cerebral vascular imaging data. A brain biometric tree construction module is used to construct a brain biometric tree for a target object. This tree uses cerebral vascular imaging data as root and trunk nodes, hemodynamic parameters from spatially aligned phase-contrast vascular imaging sequences as branch nodes representing blood supply status, neural structural parameters from spatially aligned diffusion magnetic resonance imaging sequences as branch nodes representing neural structures, and physicochemical parameters from spatially aligned fluid attenuation inversion recovery sequences as leaf nodes representing functional metabolism. Furthermore, hemodynamic parameters from spatially aligned magnetic resonance brain perfusion imaging sequences are refined to the pixel level and used as leaf nodes representing metabolic activity. This brain biometric tree is then used to perform brain function analysis on the target object.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

12. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

Images