A method and system for perivascular cerebrospinal fluid quantitative analysis based on diffusion imaging

Abstract

The application provides a perivascular cerebrospinal fluid quantitative analysis method and system based on diffusion imaging, relates to the technical field of medical image processing and magnetic resonance imaging, and comprises the following steps: acquiring a reference diffusion image and diffusion images under multiple b values of a target object; extracting a target signal region based on the reference diffusion image and acquiring a blood vessel spatial distribution atlas and a blood vessel scale atlas; generating a perivascular cerebrospinal fluid region mask by spatial intersection of the target signal region and the blood vessel spatial distribution atlas, and obtaining multiple sub-regions of different scales based on the blood vessel scale atlas; and performing diffusion model fitting based on the diffusion images under multiple b values in the sub-regions to acquire diffusion quantitative parameters. The application effectively reduces partial volume effects and realizes multi-scale fine quantitative evaluation.

Description

Technical Field

[0001] This application relates to the fields of medical image processing and magnetic resonance imaging technology, specifically to a method and system for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging. Background Technology

[0002] In the functional assessment of the brain's lymphatic system, the flow characteristics of the perivascular cerebrospinal fluid (pCSF) region are crucial observational indicators. Currently, multi-b-value diffusion imaging techniques, such as intravoxel incoherent motion (IVIM), have been used to detect microcirculation perfusion and flow-related signals within tissues. However, when applying diffusion imaging to quantitatively analyze the pCSF region, existing techniques are limited by image resolution. Signals from intravascular blood flow, brain parenchyma, and non-perivascular cerebrospinal fluid are often mixed into the voxel, leading to severe partial volume effects and distorting the extracted flow-related parameters such as the fast diffusion coefficient. Furthermore, traditional analysis methods typically only perform statistical analysis within the whole brain or a coarse region of interest, lacking refined assessment methods for perivascular spaces at different scales, making it difficult to comprehensively reflect the physiological state of the brain's lymphatic system at different vascular levels. Therefore, there is an urgent need for a perivascular cerebrospinal fluid region analysis method that can accurately eliminate partial volume effects and achieve multi-scale quantitative analysis. Summary of the Invention

[0003] To address the problems of severe partial volume effect and lack of multi-scale fine assessment methods in the diffusion imaging analysis of perivascular cerebrospinal fluid regions in existing technologies, this application proposes a diffusion imaging-based quantitative analysis method and system for perivascular cerebrospinal fluid, which enables accurate and hierarchical quantitative assessment of the flow characteristics of pCSF regions.

[0004] To achieve the above objectives, this application adopts the following technical solution: A method for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging includes: acquiring a baseline diffusion image and diffusion images at multiple b-values ​​of a target object; extracting a target signal region based on the baseline diffusion image and acquiring a vascular spatial distribution map and a vascular scale map; spatially intersecting the target signal region with the vascular spatial distribution map to generate a perivascular cerebrospinal fluid region mask, and performing scale layering on the perivascular cerebrospinal fluid region mask based on the vascular scale map to obtain multiple sub-regions of different scales; within the multiple sub-regions of different scales, performing diffusion model fitting based on the diffusion images at the multiple b-values ​​to obtain corresponding quantitative diffusion parameters.

[0005] Preferably, the step of extracting the target signal region based on the reference diffusion image includes: the target signal region being a cerebrospinal fluid-like signal region; obtaining a brain mask corresponding to the reference diffusion image; and within the brain mask, extracting a high-signal region from the reference diffusion image as the target signal region based on a preset brightness threshold or user interaction command.

[0006] Preferably, the acquisition of the spatial distribution map and the scale map of blood vessels includes: acquiring an initial spatial distribution map and an initial scale map of blood vessels in a standard space; registering the initial spatial distribution map and the initial scale map of blood vessels to the space where the structural image of the target object is located to obtain an intermediate map; and registering the intermediate map to the space where the reference diffuse image is located to obtain the spatial distribution map of blood vessels and the scale map of blood vessels.

[0007] Preferably, the step of performing scale-based stratification of the perivascular cerebrospinal fluid region mask based on the vascular scale map to obtain multiple sub-regions of different scales includes: dividing the perivascular cerebrospinal fluid region mask into large vessel scale sub-regions, medium vessel scale sub-regions, and small vessel scale sub-regions based on the vessel radius information in the vascular scale map.

[0008] Preferably, the step of fitting diffusion models based on diffusion images at multiple b-values ​​within the multiple sub-regions of different scales to obtain corresponding quantitative diffusion parameters includes: fitting the signals of diffusion images at multiple b-values ​​using a bi-exponential model within the large vessel scale sub-region, the medium vessel scale sub-region, and the small vessel scale sub-region, respectively; the bi-exponential model represents the diffusion-weighted signal intensity under different diffusion gradient factors as a weighted sum of slow diffusion components and fast diffusion components; wherein, the slow diffusion component is determined by the slow diffusion coefficient Dslow and the pseudo-diffusion fraction f, characterizing the diffusion characteristics of molecules within the tissue; the fast diffusion component is determined by the fast diffusion coefficient Dfast and the pseudo-diffusion fraction f, characterizing the flow-related pseudo-diffusion characteristics in the perivascular cerebrospinal fluid region; the pseudo-diffusion fraction f reflects the proportion of perfusion or flow-related components; and the fitted fast diffusion coefficient, the slow diffusion coefficient, and the pseudo-diffusion fraction are used as the quantitative diffusion parameters.

[0009] Preferably, the method further includes: acquiring a map of major arterial branches, including a spatial distribution mask of the middle cerebral artery, posterior cerebral artery, and frontal lobe artery; combining the map of major arterial branches with segment labels or radius intervals, and spatially intersecting it with the target signal region to generate a sub-region of major vascular branches; within the sub-region of major vascular branches, performing diffusion model fitting based on the diffusion images under the multiple b values ​​to obtain the corresponding branch diffusion quantitative parameters.

[0010] Preferably, the method further includes: extracting a diffusion image brain mask from the reference diffusion image based on the high-intensity central connected component method; processing the structural image of the target object based on the de-skull algorithm to obtain a de-skull structural image; and performing spatial registration between the reference diffusion image and the structural image based on the de-skull structural image and the diffusion image brain mask.

[0011] Preferably, the step of fitting the diffusion image signals at the multiple b values ​​using a double exponential model within the large vessel scale sub-region, the medium vessel scale sub-region, and the small vessel scale sub-region includes: calculating the average signal intensity of all voxels within each sub-region under different diffusion gradient factors; and using the double exponential model to fit the signals based on the average signal intensity to generate a signal attenuation curve for the corresponding sub-region.

[0012] In addition, this application also provides a diffusion imaging-based quantitative analysis device for perivascular cerebrospinal fluid, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the diffusion imaging-based quantitative analysis method for perivascular cerebrospinal fluid as described above.

[0013] In addition, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the quantitative analysis method for perivascular cerebrospinal fluid based on diffusion imaging as described above.

[0014] Beneficial effects: The diffusion imaging-based quantitative analysis method for perivascular cerebrospinal fluid (CSF) provided in this application constructs a CSF region mask by spatially intersecting the target signal region extracted from the baseline diffusion image with the spatial distribution map of blood vessels. Utilizing dual logical verification based on actual signal characteristics and anatomical location constraints, it eliminates non-CSF tissues that might be introduced by relying solely on map localization, and non-CSF regions that might be mistakenly included by relying solely on signal thresholds. This minimizes the partial volume effect caused by the mixing of intravascular blood flow, brain parenchyma, and background noise, significantly improving the accuracy and specificity of diffusion quantification parameters. Furthermore, this method further performs scale stratification of the generated mask based on vascular scale maps and independently fits diffusion models within each sub-region. This overcomes the limitations of traditional methods that can only perform general statistics on the whole brain or a single region, enabling differentiated and refined quantitative assessment of the flow characteristics of perivascular spaces at different scales (large, medium, and small). This provides reliable multi-dimensional data support for a comprehensive analysis of the physiological and pathological mechanisms of the brain's lymphatic system.

[0015] Furthermore, this application employs a continuous registration mapping strategy from standard space to individual structural images and then to diffusion image space, ensuring precise alignment between the prior atlas and the actual individual images in physical space, thus avoiding positioning deviations caused by individual anatomical differences. By introducing a double-exponential model and limiting the fitting to a pure sub-region after double constraints and scale stratification, the pseudo-diffusion components related to tissue molecular diffusion and flow are effectively separated, enabling the obtained fast diffusion coefficient, slow diffusion coefficient, and pseudo-diffusion fraction to truly reflect the hydrodynamic characteristics of the perivascular space at a specific scale. By averaging the voxel signals within the sub-region before fitting, the interference of individual voxel noise on parameter estimation is further suppressed, improving the robustness and reproducibility of the quantitative analysis results. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging, according to an embodiment of this application. Figure 2 This is a sagittal MNI view of the standard spatial three-scale arterial probability map mask in an embodiment of this application; Figure 3 This is a coronal MNI view of the standard spatial three-scale arterial probability map mask according to an embodiment of this application; Figure 4 This is an axial view of the MNI space of the standard spatial three-scale arterial probability map mask according to an embodiment of this application; Figure 5 This is a schematic diagram of the left middle cerebral artery portion of the standard spatial main major blood vessel branch mask in an embodiment of this application; Figure 6 This is a schematic diagram of the right middle cerebral artery portion of the standard spatial main major blood vessel branch mask in an embodiment of this application; Figure 7 This is a schematic diagram of the left posterior cerebral artery portion of the standard spatial main major blood vessel branch mask in an embodiment of this application; Figure 8 This is a schematic diagram of the right posterior cerebral artery portion of the standard spatial main major blood vessel branch mask in an embodiment of this application; Figure 9 This is a schematic diagram of the left frontal lobe artery portion of the standard spatial main major blood vessel branch mask according to an embodiment of this application; Figure 10 This is a schematic diagram of the right frontal lobe artery portion of the standard spatial main major blood vessel branch mask according to an embodiment of this application; Figure 11 This is a fitting curve of IVIM signal attenuation in pCSF regions at different vascular scales according to an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0019] Example 1: This example provides a quantitative analysis method for perivascular cerebrospinal fluid based on diffusion imaging. This method combines imaging signal features with prior anatomical knowledge to achieve precise localization and multi-scale quantitative assessment of the perivascular cerebrospinal fluid (pCSF) region. Specifically, as... Figure 1 As shown, the method includes the following steps: Step S1: Acquire a baseline diffusion image and diffusion images at multiple b-values ​​for the target object. The baseline diffusion image refers to a diffusion-weighted image used to provide a reference for the basic signal intensity. In practice, an image with a b-value of 0 (or the lowest b-value image in the acquisition sequence) is typically selected as the baseline because it has the highest signal-to-noise ratio and is not affected by diffusion gradient attenuation, clearly reflecting the T2-weighted signal characteristics of the tissue, making it suitable as an anatomical base map for subsequent extraction of high-signal regions. Diffusion images at multiple b-values ​​refer to a series of images acquired under different diffusion-sensitive gradient factors. These images collectively constitute a dataset of signal attenuation as the b-value changes, serving as the physical basis for subsequent diffusion model fitting and separation of different motion components. It should be understood that although this embodiment uses an IVIM (intravoxel incoherent motion imaging) sequence as an example, other diffusion imaging sequences capable of acquiring multiple b-value data are also applicable.

[0020] Step S2 involves extracting the target signal region based on the baseline diffuse image and obtaining a vascular spatial distribution map and a vascular scale map. Specifically, the target signal region refers to a set of voxels exhibiting high signal intensity similar to cerebrospinal fluid on the baseline diffuse image, representing a potential fluid-filled space. The vascular spatial distribution map is prior knowledge data describing the probability or location of cerebral blood vessels in space, such as a whole-cerebral artery probability map, which is not dependent on the actual scan signal of the current subject but is constructed based on population anatomical statistics. The vascular scale map is a dataset containing information on the geometric dimensions of blood vessels, used to characterize the thickness of blood vessels at different locations.

[0021] Step S3 involves spatially intersecting the target signal region with the vascular spatial distribution map to generate a perivascular cerebrospinal fluid (PSF) region mask. Based on the vascular scale map, this mask is then scale-layered to obtain multiple sub-regions at different scales. This is the core processing logic of this application. The spatial intersection essentially performs a logical AND operation; only voxels that belong to both a high-signal region on the reference image and are located within the perivascular region defined by the vascular spatial distribution map are retained in the final pCSF region mask. This dual constraint mechanism effectively eliminates two types of interference: first, it eliminates brain parenchyma or partially mixed voxels that appear as low signal on the reference image despite being close to blood vessels; second, it eliminates ventricles or subarachnoid cerebrospinal fluid that appear as high signal on the reference image but are far from blood vessels, thereby minimizing partial volume effects and ensuring the purity of the analyzed region. Building upon this, the scale stratification operation is applied directly to the pCSF region mask generated after intersection. Utilizing radius or size information from the vascular scale atlas, the mask is further subdivided into large vessel scale sub-regions, medium vessel scale sub-regions, and small vessel scale sub-regions, etc. It is important to emphasize that this stratification is performed within the pure pCSF mask, rather than directly on the whole-brain atlas. This ensures that the voxels within each scale sub-region definitively belong to the perivascular space, avoiding errors caused by calculating scale parameters in non-target tissues.

[0022] Step S4 involves fitting diffusion models to diffusion images at multiple b-values ​​within sub-regions of different scales to obtain corresponding quantitative diffusion parameters. Specifically, quantitative diffusion parameters are indicators characterizing tissue microstructure and hydrodynamic properties, such as the fast diffusion coefficient (Dfast), slow diffusion coefficient (Dslow), and pseudo-diffusion fraction (f). By fitting models to each specific scale sub-region obtained in Step S3, specific quantitative parameters for the pCSF region at that scale can be obtained. This hierarchical fitting strategy ensures that the analysis results are not merely averages of the entire brain or a single region of interest, but rather reflect the differentiated physiological information regarding the flow characteristics of perivascular spaces of different diameters, providing refined data support for a comprehensive assessment of brain-like lymphatic system function.

[0023] Example 2: Based on Example 1, this example further elaborates on the extraction method of the target signal region and the spatial mapping process of the vascular atlas.

[0024] In some embodiments, the specific implementation of extracting the target signal region based on a reference diffusion image includes: the target signal region being a cerebrospinal fluid-like signal region; obtaining a brain mask corresponding to the reference diffusion image; and within the brain mask, extracting a high-signal region from the reference diffusion image as the target signal region based on a preset brightness threshold or user interaction command. Specifically, a cerebrospinal fluid-like signal region refers to a set of voxels exhibiting a high-intensity signal similar to the diffusion characteristics of free water molecules on the reference diffusion image, whose physical nature reflects a potential fluid-filled space. However, segmentation based solely on signal intensity is easily interfered with by high signals from non-brain tissues, such as scalp fat, vitreous humor, or background noise, which may also present high-brightness features on the diffusion-weighted image. Therefore, this embodiment introduces a brain mask as a pre-constraint condition, forcibly limiting the signal extraction range to the intracranial anatomical structure. The brain mask can be obtained using various existing technologies, such as automatic extraction based on high-intensity central connected domain methods, or segmentation using deep learning models; this application does not limit this approach. Within the effective range of the brain mask, the preset brightness threshold can be set as an adaptive percentile threshold, such as selecting the 90th or 95th percentile of the signal intensity within the brain mask as the cutoff value. This method can automatically adjust according to the dynamic range of different subjects' images, avoiding the problem of poor generalization caused by a fixed absolute threshold. As another optional implementation, user interaction commands can be represented as slider adjustment controls on a graphical user interface, allowing operators to manually fine-tune the threshold based on real-time preview results to address situations where cerebrospinal fluid signals are abnormal or image artifacts are severe in individual pathological conditions. Through the dual mechanism of brain mask constraint and signal threshold / interactive confirmation, this embodiment ensures that the extracted target signal region belongs to intracranial tissue in terms of anatomical location and conforms to the appearance of cerebrospinal fluid in terms of imaging characteristics, thereby effectively excluding the intrusion of high-signal and non-fluid tissues outside the brain, laying a reliable data foundation for the subsequent construction of a pure pCSF mask.

[0025] Furthermore, to accurately map prior anatomical knowledge to the individual's diffuse image space, a continuous registration mapping strategy is employed in the process of acquiring the vascular spatial distribution atlas and vascular scale atlas. This includes: acquiring the initial vascular spatial distribution atlas and initial vascular scale atlas in the standard space; registering the initial vascular spatial distribution atlas and initial vascular scale atlas to the space of the target object's structural image to obtain the intermediate atlas; and registering the intermediate atlas to the space of the reference diffuse image to obtain the vascular spatial distribution atlas and vascular scale atlas. This process constructs a three-level mapping chain from the standard space to the individual's structural image space and then to the reference diffuse image space. The reason for not directly registering the standard spatial atlas to the diffuse image space is that diffuse imaging typically has low resolution, large geometric distortion, and significant differences in tissue contrast compared to the standard template. Direct cross-modal registration is highly susceptible to getting trapped in local optima or producing large nonlinear deformation errors. High-resolution three-dimensional structural images, with clear gray-white matter boundaries and anatomical landmarks, are an ideal bridge connecting standardized prior knowledge with individual functional images. In practice, a nonlinear registration algorithm is first used to accurately transform the initial atlas in the standard space to the individual T1 space, generating an intermediate atlas. Then, a rigid body or affine registration algorithm is used to align the individual T1 space with the reference diffusion image space, and the intermediate atlas is synchronously resampled into the voxel grid of the diffusion image. This resampling operation ensures that every probability value or radius value in the atlas corresponds strictly to a specific voxel in the diffusion image in physical coordinates, which is the geometric prerequisite for subsequent spatial intersection logic operations. Through this hierarchical registration strategy, this embodiment effectively overcomes the significant differences between cross-modal images, ensuring the positioning accuracy of vascular prior information in the individual diffusion space. This allows the pCSF mask generated based on the atlas to accurately reflect the pervasive space distribution characteristics of the subject's own blood vessels.

[0026] Example 3: Based on Example 1, this example further details the specific implementation of scale stratification of the perivascular cerebrospinal fluid region mask based on vascular scale atlas, as well as the preferred strategy for diffusion model fitting and signal processing in the stratified sub-regions.

[0027] In some embodiments, the perivascular cerebrospinal fluid region mask is scale-stratified based on the vascular scale map to obtain multiple sub-regions of different scales. Specifically, based on the vascular radius information in the vascular scale map, the perivascular cerebrospinal fluid region mask is divided into large vessel scale sub-regions, medium vessel scale sub-regions, and small vessel scale sub-regions. Figures 2-4This provides continuous or discrete distribution information of vascular geometry across the entire brain, where red represents large vessel scale sub-regions, yellow represents medium vessel scale sub-regions, and blue represents small vessel scale sub-regions. Specifically, the segmentation process is not performed directly on the whole-brain atlas, but is strictly limited to the pure perivascular cerebrospinal fluid region mask generated by spatial intersection in Example 1. This approach ensures that each segmented sub-region voxel definitively belongs to the perivascular space, avoiding errors caused by scale classification in non-target tissues. The radius threshold is typically based on population anatomical statistical distributions or prior clinical knowledge. For example, regions with a radius greater than 1.0 mm can be defined as large vessel scale sub-regions, mainly covering the spaces around large vessels such as the main trunk of the Circle of Willis (arterial anastomosis) and the proximal end of the middle cerebral artery; regions with a radius between 0.5 mm and 1.0 mm are defined as medium vessel scale sub-regions; and regions with a radius less than 0.5 mm are defined as small vessel scale sub-regions, mainly corresponding to the spaces around perforating arteries in the cortical surface and deep white matter. It should be understood that the aforementioned radius threshold is merely illustrative. In practical applications, it can be adaptively adjusted or reset according to the specific imaging resolution, research objectives, or anatomical characteristics of a particular pathological group, as long as the goal of functionally differentiating the perivascular cerebrospinal fluid region based on vessel size can be achieved. Through this scale stratification, this application can refine the general whole-brain pCSF analysis into a specific assessment of the perivascular spaces of different diameters, thereby capturing the differential flow characteristics that may exist in the brain's lymphatic system at different vascular levels.

[0028] Furthermore, within multiple sub-regions of different scales, diffusion models are fitted based on diffusion images at multiple b-values ​​to obtain corresponding quantitative diffusion parameters. Specifically, this includes fitting the signals of diffusion images at multiple b-values ​​using a bi-exponential model within the large vessel scale sub-region, medium vessel scale sub-region, and small vessel scale sub-region, respectively. In this embodiment, the diffusion model is preferably an intravoxel incoherent motion (IVIM) bi-exponential model, which can decompose the diffusion-weighted signal attenuation into two independent physical components. The specific mathematical expression is as follows: ; in, This indicates that when the diffusion gradient factor is The diffusion-weighted signal strength at that time, express The baseline signal strength at that time. It is the slow diffusion coefficient, which characterizes the pure diffusion motion of water molecules within tissues. In the perivascular cerebrospinal fluid region, it reflects the diffusion capacity of interstitial fluid or restricted water molecules. The rapid diffusion coefficient, typically used in traditional IVIM applications to characterize spurious diffusion caused by blood perfusion within microvessels, is different in the perivascular cerebrospinal fluid region analysis scenario of this application. It specifically characterizes pseudo-diffusion features associated with lymphoid flow in the perivascular space and is a key and sensitive indicator for assessing the clearance function of the brain's lymphoid system. The pseudo-diffusion fraction reflects the proportion of fast-flowing components in the total signal, i.e., the volume fraction of perfusion or flow-related components. By limiting the fitting process to a pure sub-region after dual constraints and scale stratification, this embodiment effectively separates tissue molecular diffusion from flow-related pseudo-diffusion components, resulting in the acquisition of... , and It can accurately reflect the hydrodynamic characteristics of the perivascular space at a specific scale, avoiding parameter estimation bias caused by partial volume effects.

[0029] To further improve the robustness and accuracy of parameter fitting, a double-exponential model is used to fit the signals of diffusion images at multiple b-values ​​within the large vessel scale sub-region, medium vessel scale sub-region, and small vessel scale sub-region. This process also includes: calculating the average signal intensity of all voxels within each sub-region under different diffusion gradient factors; and using a double-exponential model to fit the average signal intensity, generating the signal attenuation curve for the corresponding sub-region. Because diffusion imaging, especially IVIM sequences containing low b-value sampling, typically has a low signal-to-noise ratio (SNR) in its original images, directly fitting a single voxel with a double-exponential model is highly susceptible to noise interference, leading to parameter estimation divergence or non-physiologically significant outliers. This embodiment adopts a strategy of averaging first, then fitting. Specifically, within each scale sub-region, the signals of all voxels are first arithmetically averaged according to their b-values. Spatial redundancy information is used to suppress random noise, obtaining a regionally representative signal attenuation curve with a high SNR. Then, this curve is fitted with a model.

[0030] Example 4: This example, based on Example 1, further provides a refined analysis scheme based on the anatomical branch dimension. In some embodiments, the method further includes: acquiring a major arterial branch atlas; combining the major arterial branch atlas with segment labels or radius intervals, and spatially intersecting it with the target signal region to generate major vascular branch sub-regions; within the major vascular branch sub-regions, performing diffusion model fitting based on diffusion images at multiple b-values ​​to obtain corresponding branch diffusion quantitative parameters. Specifically, the major arterial branch atlas is prior data containing specific cerebral vascular anatomical nomenclature information, such as a spatial distribution mask of key blood supply vessels like the middle cerebral artery (MCA), posterior cerebral artery (PCA), and frontal lobe artery. Figures 5-10 As shown, this atlas can break down the entire cerebral vascular system into independent units with clear physiological significance, allowing analysis to move beyond the limitations of the whole brain or general scale classifications and delve into specific vascular basins. Segment labels refer to the segmentation markings of the same vascular branch along the blood flow direction or anatomical course, such as proximal and distal, main trunk and branch; radius intervals refer to further subdivisions within the branch based on vessel diameter. By combining one or more of the three—the major arterial branch atlas, segment labels, or radius intervals—and spatially intersecting them with the target signal region extracted in Example 1, finer-grained sub-regions of major vascular branches can be generated. This intersection operation also follows a dual-constraint logic, ensuring that the voxels within the generated sub-regions belong to specific anatomical branches and possess the signal characteristics of the perivascular cerebrospinal fluid, thereby effectively eliminating interference from blood flow within the branch's vascular lumen and the surrounding brain parenchyma.

[0031] It should be understood that the anatomical branch-based analysis dimension described in this embodiment and the vascular scale-based analysis dimension described in Embodiment 3 are independent and complementary. In practical applications, the two can be selectively executed or executed in parallel according to specific research objectives or clinical needs. For example, when assessing the overall clearance capacity of the whole-brain lymphoid system, the scale stratification strategy of Embodiment 3 can be preferentially adopted; while when studying specific stroke risk areas or early changes in neurodegenerative diseases, the branch analysis strategy of this embodiment can be combined to focus on the pCSF flow characteristics of key blood supply areas such as the MCA or PCA. After generating the main vascular branch sub-regions, the aforementioned diffusion model is also used to fit the multi-b-value diffusion signal in the region to obtain the corresponding branch diffusion quantitative parameters. These parameters can specifically reflect the lymphoid clearance function status in a specific vascular domain, providing important quantitative basis for accurately locating the anatomical location of brain metabolic waste clearance disorders in clinical practice. For example, by comparing the Dfast difference between the left and right MCA-pCSF regions, the asymmetric impact of unilateral cerebrovascular lesions on lymphoid system function can be assessed, which cannot be achieved by traditional whole-brain averaging analysis methods.

[0032] Example 5: Based on Example 1, this example further provides a preferred image preprocessing and cross-modal registration implementation scheme, aiming to provide a high-precision geometric basis for the spatial mapping in the aforementioned examples. In some embodiments, the method further includes: extracting the brain mask of the diffusion image from the reference diffusion image based on the high-intensity central connected component method; processing the structural image of the target object based on the de-braining algorithm to obtain the de-braining structural image; and performing spatial registration between the reference diffusion image and the structural image based on the de-braining structural image and the diffusion image brain mask.

[0033] Specifically, this embodiment employs a high-intensity central connected component method for brain mask extraction from a baseline diffusion-weighted image. The physical basis of this method lies in the fact that in the baseline image of diffusion-weighted imaging, the brain parenchyma and cerebrospinal fluid regions typically exhibit relatively uniform high signal intensity and constitute a maximally connected component in three-dimensional space. In contrast, background noise, vitreous humor, or other non-brain tissue high-signal artifacts are often spatially discrete or have unevenly distributed signal intensity. In practice, the baseline diffusion image is first subjected to preliminary thresholding to obtain candidate high-signal voxels. Then, a three-dimensional connected component labeling algorithm is used to identify the largest connected component containing the image's central region as the main brain tissue. Finally, morphological closing operations are used to fill internal holes and smooth boundaries, thereby generating a diffusion image brain mask that accurately encapsulates the brain tissue. Compared to a simple global thresholding method, this method effectively eliminates non-brain high-signal interference such as orbital fat and subcutaneous fluid, ensuring that subsequent analysis is strictly limited to the intracranial anatomical region. It should be understood that although this embodiment takes the high-intensity central connected region method as an example, in other embodiments, a semantic segmentation model based on deep learning or a joint segmentation strategy combining T2-weighted images can also be used to obtain the brain mask of the diffusion image, as long as the purpose of accurately separating brain tissue regions from the baseline diffusion image can be achieved.

[0034] For the structural image of the target object, this embodiment employs a de-braining algorithm to obtain a clean de-braining structural image. A de-braining algorithm refers to an image processing technique capable of automatically removing non-brain tissue structures such as the scalp, skull, dura mater, and venous sinuses, such as the ROBEX (Robust Brain Extraction) algorithm, the BET (Brain Extraction Tool) algorithm, or graph cut-based optimization algorithms. Since structural images typically have millimeter-level or even sub-millimeter-level high resolution, non-brain tissues occupy a significant proportion of the image and have complex texture features. If directly used for registration with low-resolution, low-contrast diffuse images, the registration algorithm is easily dominated by the strong signal features of these non-brain tissues, leading to alignment deviations in the brain parenchyma regions. Through de-braining processing, the forced registration process focuses only on core intrabrain structures such as gray matter, white matter, and the ventricular system, significantly improving the robustness and accuracy of cross-modal registration. It should be noted that the choice of de-braining algorithm is not limiting; any technique that can effectively remove non-brain tissues from structural images while preserving the complete brain parenchyma is applicable to this application.

[0035] When performing spatial registration between the baseline diffuse image and the structural image, this embodiment explicitly uses the processed descaling structural image and the brain mask of the diffuse image as registration inputs. Specifically, the descaling structural image can be used as the fixed image, and the baseline diffuse image cropped by the brain mask can be used as the moving image. Mutual information or normalized cross-correlation can be used as similarity metrics for rigid body or affine transformation solutions. This registration strategy based on pure brain tissue images is a key prerequisite for the reliable execution of the continuous registration mapping chain of standard space, individual structural image, and baseline diffuse image described in Embodiment 2. It not only ensures that the vascular spatial distribution map can be accurately transferred from the standard space to the individual diffuse image space, but the preprocessing step itself has independent technical value. Even in scenarios where subsequent pCSF quantitative analysis is not performed, it can provide a higher quality data foundation for conventional diffusion tensor imaging (DTI) or functional magnetic resonance imaging (fMRI) analysis.

[0036] Example 6: This example provides a device for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging. The device employs a physical hardware architecture, specifically including a memory and a processor. The memory stores a computer program containing executable instructions for implementing any one of the diffusion imaging-based quantitative analysis methods for perivascular cerebrospinal fluid described in Examples 1 to 5. The processor is communicatively connected to the memory and, when executing the computer program stored in the memory, implements the diffusion imaging-based quantitative analysis method for perivascular cerebrospinal fluid as described in any of the preceding examples.

[0037] Specifically, the memory in this embodiment can be volatile or non-volatile memory, such as random access memory (RAM), read-only memory (ROM), flash memory, disk, or optical disk. As a computer-readable storage medium, it carries code entities that transform abstract analysis methods into machine-readable instructions. The processor can be a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device. It reads the instruction sequence from the memory through a bus or other internal interconnection mechanism and drives hardware resources to complete specific operations such as data acquisition, spatial intersection operations, scale layering, and model fitting.

[0038] In practical applications, this device can take on various physical forms. For example, it can be integrated into the scanning console or built-in workstation of a magnetic resonance imaging (MRI) device to perform real-time quantitative analysis of the perivascular cerebrospinal fluid region while acquiring diffuse images; it can also serve as a standalone medical image post-processing workstation, receiving raw DICOM data transmitted externally for offline analysis; or it can be deployed as a cloud server cluster to provide high-concurrency pCSF quantitative calculation services to multiple terminals via a remote interface. It should be understood that regardless of the specific hardware model, deployment location, or network topology of the device, as long as it possesses a memory and processor and implements the technical solution of any of the aforementioned method embodiments of this application by executing the stored program, it falls within the protection scope of this application.

[0039] Example 7: This example provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the quantitative analysis method for perivascular cerebrospinal fluid based on diffusion imaging as described in any of Examples 1 to 6 above. Specifically, the computer-readable storage medium referred to in this application refers to a physical entity capable of persistently or temporarily storing data instructions and readable by a computer system to perform specific operations. For example, the storage medium may include read-only memory (ROM), random access memory (RAM), disk storage media (such as hard disk drives, floppy disks), optical disk storage media (such as CD-ROMs, DVDs), flash memory devices (such as USB flash drives, solid-state drives (SSDs), memory cards), and any other form of non-volatile or volatile data storage medium. It should be understood that the subject matter protected by this application is strictly limited to the storage medium itself with a physical structure, and explicitly excludes computer program products, carrier signals, transient electrical signals, or other intangible media that are not subject to patent protection.

[0040] The computer program stored on this storage medium contains a series of executable instructions, which constitute the coded expression of the technical solution of this application. When the processor reads and executes these instructions from the storage medium, it drives hardware resources to complete a series of data processing operations, including acquiring a baseline diffusion image and a multi-b-value diffusion image, extracting the target signal region, acquiring the spatial distribution and scale map of blood vessels, performing spatial intersection to generate a mask of the perivascular cerebrospinal fluid region, performing scale layering, and fitting quantitative diffusion parameters within the layered sub-regions. In this way, this embodiment solidifies the abstract algorithm logic in the aforementioned method embodiments into a circulating and reproducible software product carrier.

[0041] Example 8: This example provides a specific application scenario of a diffusion imaging-based quantitative analysis method for perivascular cerebrospinal fluid in the functional assessment of the brain's lymphoid system. The aim is to verify the effectiveness of the methods described in Examples 1 to 5 using actual subject data, particularly to verify their technical advantages in excluding partial volume effects and achieving multi-scale refined analysis. In this application scenario, the input data consists of a subject's 3D T1 structural image and IVIM multi-b-value diffusion image. The system automatically processes and analyzes the data according to the complete workflow described in the preceding examples.

[0042] Specifically, after steps such as target signal region extraction from the baseline diffusion image, registration of the vascular spatial distribution map, mask generation through spatial intersection, and scale layering, the system performed quantitative diffusion parameters within a pure perivascular cerebrospinal fluid region mask constrained by b0-CSF. The statistical results showed significant differences in the number of voxels and quantitative parameters in perivascular cerebrospinal fluid regions of different scales. For example, in the analysis of this subject, the number of voxels in the large vessel scale sub-region was 375, with a mean Dfast coefficient (Dfast) of 0.018840 mm² / s and a median of 0.020000 mm² / s; the number of voxels in the medium vessel scale sub-region was 10425, with a mean Dfast coefficient (Dfast) of 0.012118 mm² / s and a median of 0.011684 mm² / s; and the number of voxels in the small vessel scale sub-region was 20453, with a mean Dfast coefficient (Dfast) of 0.005270 mm² / s and a median of 0.003382 mm² / s. Meanwhile, the slow diffusion coefficient Dslow and the pseudo-diffusion fraction f also obtained stable estimates in each scale sub-region. For example, the mean value of Dslow in the large blood vessel scale sub-region was 0.000507 mm² / s, and the mean value of f was 0.743576.

[0043] The above data strongly demonstrates the technical effectiveness of the proposed solution. First, from the perspective of the physiological rationality of the parameter values, the Dfast value of the large vessel scale sub-region is significantly higher than that of the medium and small vessel scale sub-regions. This aligns with the physiological expectation that the fluid flow velocity in the perivascular space of the brain-like lymphatic system is faster and the pseudo-diffusion effect is stronger. If the average value is calculated only within the whole brain or a coarse region of interest without scale stratification, this crucial hierarchical difference will be masked, making it impossible to accurately assess the clearance function of different vascular levels. Second, all parameter statistics are strictly limited to a mask generated by the intersection of b0-CSF signal and vascular atlas, effectively eliminating brain parenchyma voxels that may be introduced by relying solely on atlas localization, and non-perivascular cerebrospinal fluid voxels that may be mistakenly included by relying solely on signal thresholds. This strict control of partial volumetric effects ensures that the obtained flow-related parameters such as Dfast truly reflect the hydrodynamic characteristics within the perivascular space, rather than being a mixed result of intravascular blood flow or tissue diffusion signals.

[0044] Furthermore, to visually demonstrate the signal attenuation characteristics and discriminative power of perivascular cerebrospinal fluid regions at different scales, this embodiment also generates IVIM bi-exponential fitting curves based on the average signal intensity within each scale sub-region. For example... Figure 4 As shown, the horizontal axis represents the b-value (s / mm²), and the vertical axis represents the normalized average signal S(b) / S0. The figure displays three signal attenuation curves representing large, medium, and small vessel scale sub-regions, as well as the signal attenuation curves of the left and right MCA-pCSF. The red curve representing the large vessel scale sub-region shows the steepest drop in the low b-value range (0-200 s / mm²), followed by a flattening, reflecting the presence of a significant fast-flow component in this region. The blue curve representing the small vessel scale sub-region shows the gentlest overall decline, indicating that its signal attenuation is mainly dominated by slow diffusion components. The orange curve representing the medium vessel scale sub-region falls between the two. This significant separation of curve shapes not only verifies the robustness of the bi-exponential model fitting across different scale sub-regions but also corroborates the necessity of scale stratification analysis from the perspective of the original signal. It should be understood that although... Figure 4 Only the fitting curves at the whole-brain scale are shown. However, in practical applications, corresponding fitting curves can also be generated for sub-regions of specific anatomical branches to assess the lymphoid functional status of specific blood supply areas.

[0045] Furthermore, the output of this application scenario also includes quantitative analysis data of major arterial branches. For example, the mean Dfast value of the pCSF region of the left middle cerebral artery is 0.020000 mm² / s, the mean Dfast value of the pCSF region of the right middle cerebral artery is 0.019352 mm² / s, and the mean Dfast value of the pCSF region of the left posterior cerebral artery is 0.018000 mm² / s. These refined data at the branch level complement the whole-brain scale analysis data, jointly constructing a multi-dimensional brain lymphatic system functional assessment system. In summary, this embodiment, through the processing and demonstration of real data, fully verifies the effectiveness and practicality of the diffusion imaging-based perivascular cerebrospinal fluid quantitative analysis method proposed in this application in overcoming the partial volume effect and achieving multi-scale and multi-branch refined quantification, providing a reliable technical means for clinical research.

[0046] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Those skilled in the art should understand that, without departing from the essence of the technical solution of this application, various combinations, splits, omissions, or equivalent substitutions can be made to the technical features in the above embodiments. For example, the spatial intersection logic between the vascular spatial distribution map and the target signal region can be extended to weighted fusion or probability threshold screening; in addition to the double exponential model, the diffusion model fitting can also adopt a stretched exponential model or a multi-compartment diffusion model; the basis for scale stratification can be based not only on the vascular radius but also on vascular length, tortuosity, or hemodynamic parameters for multi-dimensional division; the image registration strategy can also be adjusted according to the differences in imaging modalities to end-to-end registration based on deep learning or other nonlinear transformation methods. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging, characterized in that, include: Obtain the baseline diffusion image and diffusion images at multiple b values ​​for the target object; Based on the reference diffuse image, the target signal region is extracted, and the spatial distribution map and scale map of blood vessels are obtained; The target signal region is spatially intersected with the vascular spatial distribution map to generate a perivascular cerebrospinal fluid region mask. Based on the vascular scale map, the perivascular cerebrospinal fluid region mask is scale-layered to obtain multiple sub-regions of different scales. Within the multiple sub-regions of different scales, diffusion models are fitted based on diffusion images at the multiple b values ​​to obtain the corresponding quantitative diffusion parameters.

2. The method for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging according to claim 1, characterized in that, The extraction of the target signal region based on the reference diffuse image includes: Obtain the brain mask corresponding to the baseline diffusion image; Within the brain mask, based on a preset brightness threshold or user interaction commands, a high-signal region is extracted from the reference diffuse image as the target signal region.

3. The method for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging according to claim 2, characterized in that, The acquisition of the spatial distribution map and scale map of blood vessels includes: Obtain the initial spatial distribution map and initial scale map of blood vessels in standard space; The initial spatial distribution map of blood vessels and the initial scale map of blood vessels are registered to the space where the structural image of the target object is located to obtain an intermediate map. The intermediate atlas is registered to the space where the reference diffuse image is located to obtain the spatial distribution atlas of blood vessels and the scale atlas of blood vessels.

4. The method for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging according to claim 1, characterized in that, The process of performing scale-based stratification of the perivascular cerebrospinal fluid region mask based on the vascular scale atlas yields multiple sub-regions of different scales, including: Based on the vascular radius information in the vascular scale atlas, the cerebrospinal fluid region mask around the blood vessels is divided into large vessel scale sub-regions, medium vessel scale sub-regions, and small vessel scale sub-regions.

5. The method for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging according to claim 4, characterized in that, Within the multiple sub-regions at different scales, a diffusion model is fitted based on the diffusion images at the multiple b-values ​​to obtain the corresponding quantitative diffusion parameters, including: Within the large vessel scale sub-region, the medium vessel scale sub-region, and the small vessel scale sub-region, a double exponential model is used to fit the signals of the diffusion images at the multiple b values, respectively. The bi-exponential model represents the diffusion-weighted signal intensity under different diffusion gradient factors as a weighted sum of slow diffusion components and fast diffusion components; wherein, the slow diffusion component is determined by the slow diffusion coefficient Dslow and the pseudo-diffusion fraction f, characterizing the diffusion characteristics of molecules within the tissue; the fast diffusion component is determined by the fast diffusion coefficient Dfast and the pseudo-diffusion fraction f, characterizing the flow-related pseudo-diffusion characteristics in the perivascular cerebrospinal fluid region; the pseudo-diffusion fraction f reflects the proportion of perfusion or flow-related components; The fast diffusion coefficient, the slow diffusion coefficient, and the pseudo-diffusion fraction obtained from the fitting are used as the diffusion quantification parameters.

6. The method for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging according to claim 4, characterized in that, The method further includes: Obtain a map of the major arterial branches, including a spatial distribution mask of the middle cerebral artery, posterior cerebral artery, and frontal lobe artery; The main arterial branch atlas is combined with segment labels or radius intervals and spatially intersected with the target signal region to generate main vascular branch sub-regions; Within the main vascular branch sub-region, diffusion model fitting is performed based on the diffusion images under the multiple b values ​​to obtain the corresponding branch diffusion quantitative parameters.

7. The method for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging according to claim 1, characterized in that, The method further includes: Based on the high-intensity central connected region method, the brain mask of the diffusion image is extracted from the benchmark diffusion image; The structural image of the target object is processed based on the de-braining algorithm to obtain a de-braining structural image; Based on the descaling structural image and the diffused image brain mask, spatial registration is performed between the reference diffused image and the structural image.

8. The method for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging according to claim 5, characterized in that, The process involves fitting the signals of the diffusion images at the multiple b-values ​​using a bi-exponential model within the large vessel scale sub-region, the medium vessel scale sub-region, and the small vessel scale sub-region, respectively. For each sub-region of the large vessel scale sub-region, the medium vessel scale sub-region, and the small vessel scale sub-region, calculate the average signal intensity of all voxels within that sub-region under different diffusion gradient factors; Based on the average signal strength, the double exponential model is used for fitting to generate the signal attenuation curve for the corresponding sub-region.

9. A quantitative analysis system for perivascular cerebrospinal fluid based on diffusion imaging, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement a method for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a method for quantitative analysis of perivascular cerebrospinal fluid based on diffusion imaging as described in any one of claims 1 to 8.

Images