Image processing method and system for asl cerebral perfusion kinetics and associated markers

Abstract

The image processing method and system of ASL cerebral perfusion kinetics, the associated marker, can sensitively reflect the cerebral microvascular function and hemodynamic abnormalities in the early stage of AD disease, and provide reliable technical means for early diagnosis, risk assessment and mechanism research of AD. The method comprises: (1) grouping the subjects; (2) collecting magnetic resonance imaging data; (3) preprocessing the magnetic resonance imaging data; (4) segmenting the brain regions of the T1 structural image by using an automatic segmentation algorithm to obtain the gray matter, white matter, hippocampus and each brain lobe of the region of interest; (5) calculating the cerebral perfusion related kinetic parameters based on the multi-delay ASL data, generating the parameter distribution map at the voxel level, and registering it to the structural space; (6) extracting the average parameter value in each region of interest, and performing volume normalization processing; (7) using statistical analysis method to compare and evaluate the differences between different groups.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and more particularly to an image processing method for ASL brain perfusion dynamics, an image processing system for ASL brain perfusion dynamics, and biomarkers associated with image indicators. Background Technology

[0002] Alzheimer's disease (AD) is the most common type of dementia, and its biological definition is based on the abnormal deposition of amyloid-β (Aβ, A) and tau protein (T) and their downstream neurodegeneration (N). In addition to the classic ATN framework, increasing evidence suggests that cerebrovascular dysfunction, neuroinflammation, and impaired brain clearance pathways also play important roles in the occurrence and progression of the disease.

[0003] Currently, brain perfusion imaging technology is mainly based on arterial spin labeling (ASL) to assess cerebral blood flow (CBF) status by quantitatively calculating cerebral blood flow. Multi-delay ASL technology, by introducing multiple post-labeling delays (PLDs), improves the sensitivity of traditional single-delay ASL to arterial transit time (ATT) to some extent, thus enhancing the accuracy of CBF measurement.

[0004] However, the above methods still have significant limitations. First, CBF, as an indicator of blood flow "volume," is easily affected by the cerebral vascular autoregulation mechanism. It may remain relatively stable in the early stages of disease, thus masking subtle abnormalities at the microvascular level, making it difficult to detect significant differences in the subject cognitive decline (SCD) stage. Second, existing technologies mainly focus on changes in perfusion volume, lacking the ability to characterize changes in blood transport efficiency and microvascular network structure, and cannot comprehensively reflect abnormalities in intracerebral material exchange and transport processes. In addition, CBF measurements are affected by various physiological factors, such as hematocrit, sex differences, and individual hemodynamic status, which limits its stability and specificity in cross-population comparisons. Regarding sex difference studies, although existing research has found that women generally have higher cerebral perfusion levels than men, existing indicators are difficult to distinguish between physiological differences and pathological changes, and also difficult to describe the dynamic changes of sex effects at different disease stages. Furthermore, the correlation between traditional perfusion indicators and plasma biomarkers (such as neurofilament light chain protein NfL and glial fibrillary acidic protein GFAP) is weak, making it difficult to support the explanation of neuroinflammation and neurodegeneration mechanisms. Therefore, existing technologies are insufficient in terms of early sensitivity, mechanistic explanatory power, and multimodal integrated analysis, and new technical solutions are urgently needed to improve them.

[0005] A long-standing and clinically significant phenomenon is the substantial sex difference in Alzheimer's disease (AD): women account for approximately two-thirds of AD patients and have shown a faster rate of cognitive decline and a higher risk of disease progression in multiple studies. Although studies have reported sex differences in amyloid-beta and tau load, the mechanisms underlying the higher susceptibility in women remain not fully elucidated. One possible mechanism is the difference in vulnerability of cerebral microvessels between the sexes, as effective clearance of Aβ depends on intact blood-brain barrier (BBB) ​​function and perivascular / lymphoid transport processes. Early microvascular damage may promote amyloid-beta accumulation and accelerate downstream neurodegeneration, but these early changes are difficult to detect with conventional perfusion parameters due to the partially stabilizing effect of autoregulatory mechanisms on cerebral blood flow.

[0006] In Alzheimer's disease (AD), sex differences in cerebrovascular parameters have been reported across various imaging modalities. For example, white matter hyperintensities (WMH), a key MRI finding in cerebral small vessel disease, are typically higher in women than in men, especially in the elderly, and their burden is closely associated with faster cognitive decline in women. Cerebral blood flow (CBF) obtained from arterial spin labeling (ASL) MRI is generally higher in women in normal aging, but this apparent "advantage" may diminish with increased vascular risk or disease progression, increasing the complexity of interpretation, as CBF is significantly influenced by sex-related baseline physiology and self-regulation. Cerebrovascular reactivity (CVR), reflecting the dilatational reserve capacity of resistance vessels, has been considered a sensitive indicator of microvascular health and endothelial function; multiple studies suggest that even with small differences in resting CBF, CVR exhibits specific variations across sexes and AD spectrum populations. Overall, these findings support a model that women may experience a different trajectory of microvascular damage and compensation than men. However, there is still a lack of non-invasive imaging methods that can sensitively detect microvascular and hemodynamic changes and assess microvascular integrity in the very early stages of disease. Summary of the Invention

[0007] To overcome the shortcomings of existing technologies, the technical problem to be solved by this invention is to provide an imaging processing method for ASL brain perfusion dynamics, which can sensitively reflect abnormalities in brain microvascular function and hemodynamics in the early stage of AD, quantitatively assess continuous changes in different disease stages (including normal, SCD, mild cognitive impairment and AD), construct an analytical framework that can reveal gender differences and their role in disease progression, and realize comprehensive correlation analysis between imaging indicators, plasma biomarkers and cognitive function, thereby providing a reliable technical means for the early diagnosis, risk assessment and mechanism research of AD.

[0008] The technical solution of this invention is: an image processing method for ASL brain perfusion dynamics, which includes the following steps: (1) The subjects were divided into groups according to the clinical diagnostic criteria: normal cognitive NC group, subjective cognitive decline SCD group, mild cognitive impairment MCI group, and Alzheimer's disease AD group. Their demographic information and neuropsychological scores were collected. (2) Acquire magnetic resonance imaging data, including T1-weighted structural images, T2-FLAIR sequences, and multiple delayed ASL sequences, to obtain information related to brain structure and perfusion; (3) Preprocess the magnetic resonance imaging data, including motion correction, spatial registration, and standard spatial transformation, to ensure spatial consistency between different modal data; (4) Use an automatic segmentation algorithm to segment the brain regions of the T1 structural image to obtain the regions of interest of gray matter, white matter, hippocampus and each lobe. (5) Calculate brain perfusion-related dynamic parameters based on multi-delay ASL data, generate a voxel-level parameter distribution map, and register it to the structural space; (6) Extract the average parameter values ​​in each region of interest and perform volume normalization to reduce the impact of individual differences; (7) Statistical analysis was used to compare the differences between different groups, assess the interaction between gender and disease stage, analyze the correlation between imaging indicators and plasma biomarkers and cognitive function, and explore the relationship between brain structural changes, perfusion abnormalities and cognitive decline through mediation analysis.

[0009] This invention utilizes 4D time-resolved multi-delay ASL data to estimate voxel-level perfusion velocity. By fitting the spatiotemporal evolution of the ASL tracer in tissues, a three-dimensional perfusion velocity field is reconstructed, thus supplementing the traditional perfusion amplitude index CBF. By leveraging the complete spatiotemporal information of multi-delay ASL and avoiding dependence on the arterial input function AIF, it provides an index centered on the transport process, with less dependence on upstream injection conditions and potentially greater sensitivity to microvascular damage. Therefore, it can sensitively reflect abnormalities in cerebral microvascular function and hemodynamics in the early stages of Alzheimer's disease (AD). It can quantitatively assess the continuous changes in different disease stages (including normal, SCD, mild cognitive impairment, and AD), construct an analytical framework that can reveal gender differences and their role in disease progression, and achieve comprehensive correlation analysis between imaging indicators, plasma biomarkers, and cognitive function. This provides a reliable technical means for the early diagnosis, risk assessment, and mechanistic research of AD.

[0010] An image processing system for ASL brain perfusion dynamics is also provided, comprising: The grouping module is configured to group subjects according to clinical diagnostic criteria into the normal cognitive NC group, the subjective cognitive decline SCD group, the mild cognitive impairment MCI group, and the Alzheimer's disease AD group, and collect their demographic information and neuropsychological scores. The data acquisition module is configured to acquire magnetic resonance imaging data, including T1-weighted structural images, T2-FLAIR sequences, and multi-delay ASL sequences, to obtain information related to brain structure and perfusion. The preprocessing module is configured to preprocess magnetic resonance imaging data, including motion correction, spatial registration, and standard spatial transformation, to ensure spatial consistency between different modal data. The brain region segmentation module is configured to use an automatic segmentation algorithm to segment brain regions on T1 structural images, obtaining regions of interest for gray matter, white matter, hippocampus, and each lobe. The generation module is configured to calculate brain perfusion-related dynamic parameters based on multi-delay ASL data, generate a voxel-level parameter distribution map, and register it to the structural space. The extraction module is configured to extract average parameter values ​​in each region of interest and perform volume normalization to reduce the impact of individual differences. The statistical analysis module is configured to use statistical analysis methods to compare differences between different groups, assess the interaction between gender and disease stage, analyze the correlation between imaging indicators and plasma biomarkers and cognitive function, and explore the relationship between brain structural changes, perfusion abnormalities and cognitive decline through mediation analysis.

[0011] The study also provided biomarkers associated with the imaging processing system of ASL brain perfusion dynamics. Spearman correlation analysis was used to assess the correlation between plasma biomarkers and regional CBF and QTM velocity in the overall cohort and sex stratification. FDR correction was performed independently for each biomarker and imaging modality to avoid cross-interference. As a sensitivity analysis, a covariate-corrected linear regression model was used to verify the robustness of significant associations. QTM velocity was robustly negatively correlated with the biomarker NfL in all brain regions. The correlation between QTM velocity and GFAP showed brain region dependence. In the overall sample, hippocampal QTM velocity was negatively correlated with the biomarker GFAP, and this association was driven by females. Attached Figure Description

[0012] Figure 1 A flowchart of the image processing method for ASL brain perfusion dynamics according to the present invention is shown.

[0013] Figure 2The following diagrams show the group-mean QTM velocity and CBF plots for each diagnostic group in the MNI template space. (A)–(C) show the group-mean QTM velocity plots for the NC, SCD, MCI, and AD groups in the coronal hippocampal view (Cor-Hippo), axial / transverse hippocampal view (Tran-Hippo), and cortical view (Cortex), respectively. The QTM velocity generally decreases from NC to AD, a trend observed in both the hippocampal and cortical views. Red arrows (coronal) and red outlines (axial) mark the regions of interest (ROIs) in the hippocampus, highlighting the differences in hippocampal QTM velocity among the groups. (D)–(F) show the group-mean CBF plots for the NC, SCD, MCI, and AD groups in the same views (Cor-Hippo, Tran-Hippo, and Cortex), respectively. CBF also shows a decreasing trend with group progression, with a more pronounced decrease in the hippocampus from SCD to MCI / AD. The color bars indicate the range of QTM velocity and CBF values. NC, normal cognition; SCD, subjective cognitive decline; MCI, mild cognitive impairment; AD, Alzheimer's disease; CBF, cerebral blood flow; QTM, quantitative transport imaging. Detailed Implementation

[0014] This invention proposes a quantitative transport mapping (QTM) model based on biophysical principles, which can estimate voxel-level perfusion velocity using 4D time-resolved multi-delay ASL data. This method reconstructs a three-dimensional perfusion velocity field by fitting the spatiotemporal evolution of the ASL tracer in tissues, thus supplementing traditional perfusion amplitude indicators (such as CBF). This method has been validated in numerical simulations at the microvascular scale and in porcine liver fluid model experiments. By utilizing the complete spatiotemporal information of multi-delay ASL and avoiding dependence on the arterial input function (AIF), this method provides an indicator centered on the transport process, with less dependence on upstream injection conditions and potentially greater sensitivity to microvascular damage. In previous studies, this method has shown superior sensitivity to CBF in various diseases, including benign and malignant breast cancer differentiation, nasopharyngeal carcinoma gene expression characterization, pulmonary shunt fraction estimation, liver disease staging, and early-stage AD abnormality detection (such as distinguishing between cognitively normal and mild cognitive impairment).

[0015] like Figure 1 As shown, this image processing method for ASL brain perfusion dynamics includes the following steps: (1) The subjects were divided into groups according to the clinical diagnostic criteria: normal cognitive NC group, subjective cognitive decline SCD group, mild cognitive impairment MCI group, and Alzheimer's disease AD group. Their demographic information and neuropsychological scores were collected. (2) Acquire magnetic resonance imaging data, including T1-weighted structural images, T2-FLAIR sequences, and multiple delayed ASL sequences, to obtain information related to brain structure and perfusion; (3) Preprocess the magnetic resonance imaging data, including motion correction, spatial registration, and standard spatial transformation, to ensure spatial consistency between different modal data; (4) Use an automatic segmentation algorithm to segment the brain regions of the T1 structural image to obtain the regions of interest of gray matter, white matter, hippocampus and each lobe. (5) Calculate brain perfusion-related dynamic parameters based on multi-delay ASL data, generate a voxel-level parameter distribution map, and register it to the structural space; (6) Extract the average parameter values ​​in each region of interest and perform volume normalization to reduce the impact of individual differences; (7) Statistical analysis was used to compare the differences between different groups, assess the interaction between gender and disease stage, analyze the correlation between imaging indicators and plasma biomarkers and cognitive function, and explore the relationship between brain structural changes, perfusion abnormalities and cognitive decline through mediation analysis.

[0016] This invention utilizes 4D time-resolved multi-delay ASL data to estimate voxel-level perfusion velocity. By fitting the spatiotemporal evolution of the ASL tracer in tissues, a three-dimensional perfusion velocity field is reconstructed, thus supplementing the traditional perfusion amplitude index CBF. By leveraging the complete spatiotemporal information of multi-delay ASL and avoiding dependence on the arterial input function AIF, it provides an index centered on the transport process, with less dependence on upstream injection conditions and potentially greater sensitivity to microvascular damage. Therefore, it can sensitively reflect abnormalities in cerebral microvascular function and hemodynamics in the early stages of Alzheimer's disease (AD). It can quantitatively assess the continuous changes in different disease stages (including normal, SCD, mild cognitive impairment, and AD), construct an analytical framework that can reveal gender differences and their role in disease progression, and achieve comprehensive correlation analysis between imaging indicators, plasma biomarkers, and cognitive function. This provides a reliable technical means for the early diagnosis, risk assessment, and mechanistic research of AD.

[0017] Preferably, in step (1), the NC group is strictly defined as an individual with no abnormalities in clinical interviews, a Mini-Mental State Examination (MMSE) score ≥27, and a Clinical Dementia Rating Scale (CDR) global score of 0; under feasible conditions, the diagnostic classification of the SCD, MCI, and AD groups is verified by amyloid PET imaging or plasma biomarker analysis to confirm their amyloid positive status.

[0018] Preferably, step (2) is performed using a 3.0T magnetic resonance scanner equipped with an 8-channel head coil. The core sequence includes: (1) 3D T1-weighted imaging for region of interest (ROI) segmentation, with repetition time / echo time = 6.7 / 2.9 ms, flip angle = 12°, slice thickness = 1.0 mm, no gap, matrix = 256×256, and field of view = 256×256 mm²; (2) multi-label post-delay ASL based on enhanced ASL sequence for quantification of cerebral blood flow (CBF) and arterial transit time (QTM) velocities, encoding 7 different label post-delays [800, 1098, 1427, 1797, 2228, 2761, 3492] ms into a single acquisition, with an effective label duration of [298, 328, 370, 432, 533, 731, 1308]. ms, repetition time / echo time = 6273 / 11.4ms, flip angle = 111°, voxel size = 1.72×1.72×4.5 mm³, matrix = 128×128, field of view = 220×220 mm², number of excitations = 2; (3) T2-FLAIR sequence used to quantify white matter high signal WMH representing small vessel injury, repetition time / echo time = 8668 / 166 ms, flip angle = 111°, voxel size = 0.5×0.5×5 mm³, slice interval = 1.0 mm, matrix = 256×256, field of view = 240×240 mm².

[0019] Preferably, in step (4), FreeSurfer v7.1 is used to automatically segment the T1-weighted images into cortical and subcortical regions to generate individual-specific ROIs; these ROIs are bilaterally merged, including the whole-brain cortical gray matter (GM), white matter (WM), hippocampus, and frontal, parietal, temporal, and occipital lobes; to extract ROI values, the CBF map and QTM velocity map derived from mPLD-ASL are registered to the T1w image in the FreeSurfer space for each participant; to mitigate the partial volume effect at tissue boundaries, all ROIs undergo 1 mm morphological erosion before extracting the regional average; a validated deep learning-based neural network is used to segment the WMH from the T2-FLAIR image and is used to evaluate CBF, QTM velocity, and their association with cognition; all ROI volume values ​​are standardized by intracranial volume to correct for head circumference differences.

[0020] Preferably, in step (5), the CBF map is reconstructed on the scanner from data acquired by mPLD-ASL using a dynamic model with arterial transit time (ATT) correction; ATT correction is used to reduce the CBF estimation bias caused by the extended label delay time; and FLIRT (FMRIB linear image registration tool) is used to register the CBF map to the FreeSurfer T1w space of each subject; to improve the registration accuracy, image registration is performed within the cortical mask area, and CBF values ​​are extracted in the FreeSurfer T1w space using the corresponding FreeSurfer partition.

[0021] Preferably, in step (6), QTM models the kinetics of the labeled tracer using a mass conservation equation to estimate the perfusion rate: , Where C(x,t) is the tracer concentration at position x and time t, v(x) is the time-invariant tracer velocity field, D is the apparent diffusion coefficient, and R is the signal attenuation rate; for ASL-labeled endogenous water, R is set to 1 / T1b, where T1b is the blood time T1, and the diffusion term is ignored in the perfusion estimation. Because the diffusion rate is much slower than the infusion-related transport process; The velocity field v is reconstructed by minimizing the objective function: , Where λ is the regularization parameter, λ=0.1, used to promote the smoothness of the velocity field space. The reconstruction of motion-corrected mPLD-ASL data is completed using MATLAB internal code. v represents the tracer velocity of the capillaries on the voxel surface, with typical units in the image space being mm / s, and its amplitude... Defined as the average perfusion rate at the voxel level in the QTM model; The QTM velocity map is registered to the FreeSurfer T1w space of each subject using the same transformation matrix as the CBF map, and the ROI velocity value is extracted using the same method as CBF.

[0022] Preferably, in step (7), continuous variables are represented by the median interquartile range, and categorical variables are represented by the count %; all tests are two-tailed tests, the significance threshold is set to p<0.05, and when comparing multiple regions, the Benjamini-Hochberg false discovery rate (FDR) is used to correct the p value.

[0023] Preferably, in step (7), the demographic, clinical, and biomarker characteristics among the four diagnostic groups are compared using the Kruskal-Wallis test or the χ² test. For those with significant Kruskal-Wallis results, the FDR-corrected Dunn post-hoc test is performed. The Kruskal-Wallis test and the FDR-corrected Dunn test are used to assess the inter-group differences in the mean CBF and QTM velocity of the ROI. The Wilcoxon rank-sum test is used to analyze the gender differences in the whole sample and in specific diagnostic stages. In order to explore whether the gender differences change with disease progression, an age-corrected linear interaction model is fitted in the symptomatic cohorts SCD, MCI, and AD, incorporating age, gender, ordered diagnostic stage, and gender × diagnosis interaction terms to assess the attenuation or aggravation trend of gender differences in clinical stages.

[0024] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium. When executed, the program includes the steps of the methods of the above embodiments. The storage medium can be ROM / RAM, magnetic disk, optical disk, memory card, etc. Therefore, corresponding to the method of the present invention, the present invention also includes an ASL brain perfusion dynamics image processing system, which is typically represented in the form of functional modules corresponding to the steps of the method. The system includes: The grouping module is configured to group subjects according to clinical diagnostic criteria into the normal cognitive NC group, the subjective cognitive decline SCD group, the mild cognitive impairment MCI group, and the Alzheimer's disease AD group, and collect their demographic information and neuropsychological scores. The data acquisition module is configured to acquire magnetic resonance imaging data, including T1-weighted structural images, T2-FLAIR sequences, and multi-delay ASL sequences, to obtain information related to brain structure and perfusion. The preprocessing module is configured to preprocess magnetic resonance imaging data, including motion correction, spatial registration, and standard spatial transformation, to ensure spatial consistency between different modal data. The brain region segmentation module is configured to use an automatic segmentation algorithm to segment brain regions on T1 structural images, obtaining regions of interest for gray matter, white matter, hippocampus, and each lobe. The generation module is configured to calculate brain perfusion-related dynamic parameters based on multi-delay ASL data, generate a voxel-level parameter distribution map, and register it to the structural space. The extraction module is configured to extract average parameter values ​​in each region of interest and perform volume normalization to reduce the impact of individual differences. The statistical analysis module is configured to use statistical analysis methods to compare differences between different groups, assess the interaction between gender and disease stage, analyze the correlation between imaging indicators and plasma biomarkers and cognitive function, and explore the relationship between brain structural changes, perfusion abnormalities and cognitive decline through mediation analysis.

[0025] The study also provided biomarkers associated with the imaging processing system of ASL brain perfusion dynamics. Spearman correlation analysis was used to assess the correlation between plasma biomarkers and regional CBF and QTM velocity in the overall cohort and sex stratification. FDR correction was performed independently for each biomarker and imaging modality to avoid cross-interference. As a sensitivity analysis, a covariate-corrected linear regression model was used to verify the robustness of significant associations. QTM velocity was robustly negatively correlated with the biomarker NfL in all brain regions. The correlation between QTM velocity and GFAP showed brain region dependence. In the overall sample, hippocampal QTM velocity was negatively correlated with the biomarker GFAP, and this association was driven by females.

[0026] The research scheme and results of this invention are described in detail below.

[0027] Research Design and Participants

[0028] This study was approved by the Ethics Committee of the China-Japan Friendship Hospital and strictly followed the principles of the Declaration of Helsinki. All participants or their legal authorized representatives signed written informed consent forms. Study participants were recruited from the Department of Neurology at the China-Japan Friendship Hospital. The final analysis sample included 182 eligible participants. After excluding individuals who did not complete neuropsychological testing or had severe MRI artifacts / head movement problems, they were divided into four clinical groups: normal cognition (NC, n=53), subjective cognitive decline (SCD, n=48), mild cognitive impairment (MCI, n=31), and Alzheimer's disease (AD, n=50). The NC group was strictly defined as individuals with normal clinical interviews, a Mini-Mental State Examination (MMSE) score ≥27, and a Clinical Dementia Rating Scale (CDR) global score of 0. Where feasible, the diagnosis of SCD, MCI, and AD groups was verified by amyloid (Aβ) PET imaging (F-florbetaben) or plasma biomarker analysis to confirm their amyloid positivity.

[0029] Neuropsychological assessment

[0030] A standardized neuropsychological test suite was used to assess overall cognitive function and disease severity. The MMSE, a comprehensive screening tool encompassing 30 components including orientation, memory, attention, and language abilities, was used as the primary cognitive outcome in association and mediation analyses. Clinical severity was graded using the CDR global score. Furthermore, the Ray Auditory-Verbal Learning Test (RAVLT) immediate recall score was used to assess episodic memory and short-term learning abilities.

[0031] Quantitative analysis of plasma biomarkers

[0032] Fasting blood samples were collected from 69 participants (8 in the NC group, 24 in the SCD group, 13 in the MCI group, and 24 in the AD group) using the ultrasensitive Simoa platform. Biomarkers detected included Aβ40, Aβ42, phosphorylated tau217 (p-tau217), glial fibrillary acidic protein (GFAP), and neurofilament light chains (NfL). One case of Aβ42 was excluded due to exceeding the signal intensity limit, resulting in a final sample size of 68 for Aβ42 and its related ratio analysis.

[0033] MRI data acquisition

[0034] All MRI examinations were performed at the China-Japan Friendship Hospital using a standardized protocol and a 3.0T MRI scanner (Discovery MR 750, GE Healthcare, USA) equipped with an 8-channel head coil. The core sequences included: (1) 3D T1-weighted imaging (T1w) for region of interest (ROI) segmentation (repetition time / echo time = 6.7 / 2.9 ms, flip angle = 12°, slice thickness = 1.0 mm, no gap, matrix = 256×256, field of view = 256×256 mm²); and (2) multilabeled post-delayed ASL (mPLD-ASL) based on enhanced ASL sequences for cerebral blood flow (CBF) and arterial transit time (QTM) velocity quantification. The scheme encodes 7 different post-mark delays ([800, 1098, 1427, 1797, 2228, 2761, 3492] ms) into a single acquisition, with an effective marker duration of [298, 328, 370, 432, 533, 731, 1308] ms, repetition time / echo time = 6273 / 11.4 ms, flip angle = 111°, voxel size = 1.72×1.72×4.5 mm³, matrix = 128×128, field of view = 220×220 mm², number of excitations = 2; (3) mainly used to quantify the white matter high signal (FLIRT) representing small vessel damage using T2-FLAIR sequences (repetition time / echo time = 8668 / 166 ms, flip angle = 111°, voxel size = 0.5×0.5×5 mm³, layer spacing = 1.0 mm, matrix = 256×256, field of view = 240×240 mm².

[0035] MRI Data Processing

[0036] The workflow of this study is as follows: Figure 1 As shown.

[0037] Brain Region ROI Segmentation

[0038] FreeSurfer v7.1 was used to automatically segment T1-weighted images into cortical and subcortical regions, generating individual-specific areas of interest (ROIs). These ROIs were bilaterally merged, including whole-cortical gray matter (GM), white matter (WM), hippocampus, and the four major cortical lobes (frontal, parietal, temporal, and occipital). To extract ROI values, mPLD-ASL-derived CBF and QTM velocity maps were registered to each participant's T1w image in FreeSurfer space. To mitigate partial volume effects at tissue boundaries, all ROIs underwent 1 mm morphological erosion before averaging. A validated deep learning-based neural network was used to segment WMH (as a component of vascular risk) from T2-FLAIR images, and this segmentation was used to assess CBF, QTM velocity, and their association with cognition. Given that periventricular WMH (pWMH) dominates the overall WMH burden (approximately 90%), this study only analyzed pWMH. All ROI volume values ​​were normalized to intracranial volume (ICV) to correct for head circumference differences.

[0039] CBF map construction based on mPLD-ASL

[0040] CBF maps were reconstructed from data acquired using mPLD-ASL on a scanner (GE Discovery MR750) using a kinetic model corrected for arterial transit time (ATT). ATT correction was employed to reduce CBF estimation bias caused by prolonged label delay time; previous studies have confirmed that this method is more accurate than CBF derived from single-delay ASL. Subsequently, CBF maps were registered to each subject's FreeSurfer T1w space using FLIRT (FSL). To improve registration accuracy, image registration was performed within the cortical mask area. Within the FreeSurfer T1w space, CBF values ​​were extracted using the corresponding FreeSurfer partitions.

[0041] QTM speed quantization based on mPLD-ASL

[0042] QTM models labeled tracer kinetics to estimate infusion rates using mass conservation (transport) equations: Where C(x,t) is the tracer concentration at location x and time t, v(x) is the time-invariant tracer velocity field, D is the apparent diffusion coefficient, and R is the signal attenuation rate. For ASL-labeled endogenous water, R is set to 1 / T1b (T1b is the blood time T1). The diffusion term is ignored in the perfusion estimation. (This is because the diffusion rate is much slower than the infusion-related transport process.)

[0043] The velocity field v is reconstructed by minimizing the objective function: λ is a regularization parameter (λ=0.1, selected through L-curve analysis) used to promote spatial smoothness of the velocity field. Motion-corrected mPLD-ASL data was reconstructed using MATLAB internal code. Referring to previous QTM studies, v represents the tracer velocity of capillaries on the voxel surface, typically in mm / s in the image space. Its amplitude... The L2 norm is defined as the average perfusion rate at the voxel level in the QTM model.

[0044] The QTM velocity map is registered to the FreeSurfer T1w space of each subject using the same transformation matrix as the CBF map, and the ROI velocity value is extracted using the same method as CBF.

[0045] Statistical analysis

[0046] Statistical analyses were performed in R v4.5.2 (RStudio 2025.09.2 environment, Posit Software). Continuous variables were expressed as median [interquartile range], and categorical variables as count (%). All tests were two-tailed, with a significance threshold of p < 0.05. For comparisons across multiple regions, the Benjamini-Hochberg false discovery rate (FDR) was used to correct for p-values.

[0047] Comparison between diagnostic groups

[0048] Demographic, clinical, and biomarker characteristics among the four diagnostic groups were compared using the Kruskal-Wallis test (for continuous / ordinal variables) or the χ² test (for categorical variables, with Fisher's exact test used when necessary). For those with significant Kruskal-Wallis results, a post-hoc Dunn test with FDR correction was performed.

[0049] The impact of diagnosis and sex on CBF and QTM speed

[0050] The Kruskal-Wallis test and the FDR-corrected Dunn test were used to assess the intergroup differences in mean CBF and QTM speed within the ROI. The Wilcoxon rank-sum test was used to analyze gender differences across the entire sample and within specific diagnostic stages. To investigate whether gender differences change with disease progression, an age-corrected linear interaction model was fitted in symptomatic cohorts (SCD, MCI, AD), incorporating age, gender, ordered diagnostic stage, and a gender × diagnosis interaction term to assess the attenuation or exacerbation trends of gender differences at clinical stages.

[0051] Association of CBF and QTM rates with plasma biomarkers

[0052] In the overall cohort and sex stratified, Spearman correlation analysis was used to assess the correlation between plasma biomarkers and regional CBF and QTM velocities. FDR correction was performed independently for each biomarker and imaging modality to avoid cross-contamination. As a sensitivity analysis, covariate-adjusted linear regression models (adjusted for age and sex) were used to verify the robustness of significant associations.

[0053] Correlation between CBF and QTM speeds and overall perception

[0054] Spearman partial correlation analysis, controlling for age and sex, was used to preliminarily assess the relationship between perfusion-related indicators and overall cognitive function (MMSE). This analysis was performed separately in the general cohort and the symptomatic cohort, and adjusted for FDR. Subsequently, a multivariate linear regression model was constructed to include age, sex, diagnosis, and a sex × (CBF or QTM velocity) interaction term to identify sex-dependent effects in cognitive associations. Sensitivity analysis for the symptomatic cohort further adjusted for ICV-normalized hippocampal volume to separate the independent contribution of CBF / QTM velocity to cognition (independent of the effect of structural atrophy).

[0055] Mediation effect analysis

[0056] To investigate whether regional CBF or QTM velocity mediates the known atrophy-cognitive association, a mediation effect analysis was conducted. All variables were Z-score standardized before modeling. Two covariate-adjusted linear models (age and sex) were constructed: (1) a mediation variable model (QTM ~ age + sex + hippocampal volume), with parietal QTM velocity as the dependent variable and hippocampal volume as the predictor variable; (2) an outcome variable model (MMSE ~ age + sex + hippocampal volume + QTM), with MMSE as the dependent variable and hippocampal volume and parietal QTM velocity as predictors. The average causal mediation effect (ACME), average direct effect (ADE), total effect, and mediation proportion were estimated using a nonparametric bootstrap method based on 5,000 simulations.

[0057] result

[0058] Demographic characteristics and key indicators

[0059] This invention summarizes the basic characteristics of 182 participants (53 in the NC group, 48 in the SCD group, 31 in the MCI group, and 50 in the AD group). There was no significant difference in gender distribution among the four groups (p=0.351), but significant differences existed in age and education level (p<0.001 and p=0.001, respectively), with the AD group showing an overall older age and shorter years of education. With disease progression, cognitive performance (MMSE and RAVLT scores) progressively declined, while clinical severity (CDR) gradually increased (all p<0.001). Regarding plasma biomarkers, significant differences were observed in p-tau217 (p=0.040), NfL (p<0.001), and GFAP (p=0.047) levels among the groups, while there were no significant differences between Aβ42 and Aβ40 (p=0.298 and 0.457, respectively).

[0060] Regional CBF and QTM speed in different diagnostic groups

[0061] CBF and QTM velocity maps were generated, and the median maps of each group were mapped to the MNI space for visualization. Figure 2 ).

[0062] The Kruskal-Wallis test was used to compare the four diagnostic categories, and the results showed that regional CBF and QTM velocity had significant diagnostic effects in all assessed regions (including gray matter, white matter, and hippocampus) (all p < 0.01). Post-hoc pairwise comparisons corrected for Benjamini-Hochberg FCR revealed differential patterns of change in the AD spectrum.

[0063] Decreased CBF mainly occurred in the later stages of the disease. Compared with the NC group, the CBF in the gray matter, hippocampus and white matter regions of the MCI and AD groups were significantly decreased, while no significant differences were found between the NC and SCD groups, or between the MCI and AD groups (p>0.05 after all adjustments). Specifically, compared to the NC group (grey matter: 42.95 ml / 100 g / min; hippocampus: 36.27 ml / 100 g / min; white matter: 25.11 ml / 100 g / min), the CBF of the MCI group (grey matter: 36.23 ml / 100 g / min, a decrease of 15.65%; hippocampus: 31.28 ml / 100 g / min, a decrease of 13.76%; white matter: 22.07 ml / 100 g / min, a decrease of 12.11%) and the AD group (grey matter: 36.40 ml / 100 g / min, a decrease of 15.25%; hippocampus: 33.28 ml / 100 g / min, a decrease of 8.24%; white matter: 22.85 ml / 100 g / min, a decrease of 9.00%) was significantly reduced (corrected p<0.01). In contrast, the changes in the SCD group were minimal and not statistically significant (gray matter: 41.26 ml / 100 g / min, a decrease of 3.93%, p=0.358; hippocampus: 36.09 ml / 100 g / min, a decrease of 0.50%, p=0.773; white matter: 24.84 ml / 100 g / min, a decrease of 1.08%, p=0.455). Notably, compared with the NC group, the white matter CBF of both the MCI and AD groups was significantly reduced, while there was no significant difference between the MCI and AD groups.

[0064] QTM showed a sensitive decrease in perfusion velocity in the early stages of the AD spectrum. Overall, QTM velocity progressively decreased across ordered diagnosis groups, plateauing at the MCI to AD stage. Pairwise comparisons revealed significant differences between all paired groups except for the MCI and AD groups. Compared to the NC group, the SCD, MCI, and AD groups showed significantly reduced QTM velocity in the hippocampus, with marginally significant reductions in gray and white matter regions. Baseline values ​​for the NC group (gray matter 5.87 mm / s, hippocampus 5.86 mm / s, white matter 4.47 mm / s) showed that the QTM velocities of the SCD group decreased in gray matter (5.48 mm / s, a decrease of 6.63%, p=0.057), hippocampus (5.26 mm / s, a decrease of 10.20%, p=0.034), and white matter (4.16 mm / s, a decrease of 7.01%, p=0.056). The MCI group showed a wider decrease across the entire region (gray matter 4.74 mm / s, a decrease of 19.29%; hippocampus 4.98 mm / s, a decrease of 14.90%; white matter 3.79 mm / s, a decrease of 15.29%; all p<0.001). The AD group maintained a decreasing level (gray matter 4.67 mm / s, a decrease of 20.41%; hippocampus 4.66 mm / s, a decrease of 20.47%; white matter 3.64 mm / s). mm / s, a decrease of 18.71%; all p<0.001). In the gray matter, white matter, and hippocampus, QTM showed a more significant stage-dependent decrease than CBF, especially in the NC to SCD stage, which may be related to microvascular changes in the AD process.

[0065] Gender differences in CBF and QTM speeds

[0066] The study examined gender differences in CBF and QTM speeds, finding that females had significantly higher CBF and QTM speeds in gray matter, hippocampus, and white matter than males (Wilcoxon rank-sum test: all p < 0.001). Compared to males, females showed a median increase in CBF of 7.24 ml / 100g / min (20.88%) in gray matter, 3.73 ml / 100g / min (17.76%) in white matter, and 4.59 ml / 100g / min (14.61%) in the hippocampus. QTM speed also showed a female-associated increase, with a median increase of 0.71 mm / s (14.97%) in gray matter, 0.47 mm / s (12.76%) in white matter, and 0.56 mm / s (12.13%) in the hippocampus. Overall, significant gender differences were observed in both CBF and QTM speeds.

[0067] Interaction effect of gender and diagnosis on QTM speed and CBF

[0068] Stratified analysis of CBF and QTM velocities by diagnostic stage revealed a stage-dependent pattern for both indicators. For CBF, significant gender differences primarily occurred in later clinical stages (gray matter: SCD-AD stage; white matter: SCD and MCI stage; hippocampus: MCI stage). In contrast, significant gender differences in QTM velocities were more concentrated in earlier stages, particularly NC and SCD stages (e.g., gray matter: NC and SCD stage; white matter: SCD stage; hippocampus: SCD and MCI stage).

[0069] To quantify the changes in gender effects throughout the disease progression, the gender-specific relative changes in CBF and QTM velocity between adjacent stages were calculated. Gray matter CBF decreased from NC to SCD (female: -1.40%; male: -15.15%) and from SCD to MCI (female: -6.86%; male: -17.90%). Hippocampal CBF showed a similar pattern: from NC to SCD (female: +0.28%; male: -7.82%) and from SCD to MCI (female: -7.78%; male: -14.71%). Similarly, gray matter QTM velocity decreased from NC to SCD (female: -1.40%; male: -15.15%) and from SCD to MCI (female: -6.86%; male: -7.42%). Hippocampal QTM velocity followed a trend consistent with gray matter: NC to SCD stage (female: -3.16%; male: -15.44%), SCD to MCI stage (female: -10.55%; male: -15.44%). In white matter, sex differences in CBF and QTM velocity were similar across stages. Data showed that the decline in QTM velocity was faster in females from SCD to MCI than from NC to SCD, while the decline was comparable in males across the two transition periods, suggesting that sex may modulate the pattern of CBF and QTM velocity changes during AD progression.

[0070] In the gray matter region, a significant sex-diagnosis interaction was observed in QTM velocity, while no such effect was observed in CBF. Age-adjusted linear regression models (QTM velocity or CBF ~ age + sex + diagnosis + sex × diagnosis) showed significant main sex effects in all regions of interest for QTM velocity and CBF (all p < 0.05), indicating higher overall values ​​in females. Age was negatively correlated with QTM velocity in gray matter, white matter, and hippocampus (all p < 0.05), while no significant age effect was found in CBF (all p > 0.05). Importantly, a significant sex × diagnosis interaction was detected only in QTM velocity in gray matter (β = -0.465, p = 0.022) and white matter (β = -0.340, p = 0.022). The negative interaction coefficients indicate that the decrease in female-specific QTM velocity was more significant as the diagnostic stage progressed. In contrast, no significant sex × diagnosis interaction was found in any CBF index (all p > 0.05).

[0071] Association analysis of QTM speed and CBF with plasma biomarkers

[0072] To explore the potential mechanisms of QTM velocity and CBF in the AD lineage, we analyzed its association with common plasma biomarkers (including Aβ40, Aβ42, Aβ42 / 40, p-tau217, NfL, and GFAP).

[0073] The study found that QTM velocity was robustly negatively correlated with plasma NfL in all brain regions (gray matter: r = -0.536, p < 0.001; white matter: r = -0.522, p < 0.001; hippocampus: r = -0.510, p < 0.001). After stratification by sex, the negative correlation remained significant and the correlation coefficient was higher in the female population (gray matter: r = -0.600; white matter: r = -0.622; hippocampus: r = -0.605; all p < 0.001), while no significant association was found in the male population (all p > 0.220), suggesting the existence of sex-specific neurovascular interactions.

[0074] The correlation between QTM velocity and GFAP showed a brain region-dependent relationship. In the overall sample, hippocampal QTM velocity was negatively correlated with GFAP (r = -0.328, p = 0.032). Similar to NfL, this association was primarily driven by females (r = -0.404, p = 0.005), with no significant correlation in males (p = 0.170). Without multiple comparison correction, plasma GFAP was negatively correlated with gray matter, white matter, and hippocampal QTM velocity. A comprehensive correlation assessment was performed between QTM velocity in all cortical and subcortical regions of interest as defined by FreeSurfer and plasma biomarkers (including Aβ40, Aβ42, Aβ42 / 40, p-tau217, NfL, and GFAP).

[0075] A linear regression model adjusted for age and sex (QTM ~ biomarker + age + sex) confirmed that the association between QTM velocity and NfL was independent of age and sex. NfL remained a significant predictor of QTM velocity in gray matter (β = -0.049, adjusted R² = 0.285, p < 0.001), white matter (β = -0.041, adjusted R² = 0.277, p < 0.001), and hippocampus (β = -0.041, adjusted R² = 0.290, p = 0.001). However, the association between GFAP and hippocampal QTM was not significant after adjusting for age and sex (β = -0.001, p = 0.098).

[0076] No significant association was found between CBF and plasma biomarkers. Using the same FDR-corrected correlation analysis framework, no statistically significant paired associations were found between CBF and any plasma biomarker in any of the assessment regions.

[0077] Correlation between radiographic indicators and MMSE

[0078] Given that perfusion and vascular integrity defects can lead to cognitive impairment, this study analyzed the association between global cognitive function (MMSE) and CBF or QTM velocity to reveal their potential differential contributions to cognition. Spearman partial correlation analysis, controlling for age and sex, revealed a positive correlation between hippocampal QTM velocity and MMSE in the whole sample (r=0.220, p=0.020). In the symptomatic cohort, this association was stronger and extended to more brain regions, including gray matter (r=0.275, p=0.007), hippocampus (r=0.296, p=0.006), and periventricular white matter hyperintensity (pWMH) (r=0.197, p=0.043). A moderate positive correlation between MMSE and CBF was also observed in the symptomatic cohort (gray matter CBF: r=0.199, p=0.043; hippocampus: r=0.198, p=0.062; pWMH area CBF: r=0.192, p=0.043).

[0079] In the multiple regression model incorporating age, sex, diagnosis, and the sex × (QTM velocity or CBF) interaction term, no significant interaction was found across the entire sample. However, in the symptomatic cohort, hippocampal QTM velocity had a significant main effect on MMSE (β=2.160, p=0.007), and the sex × hippocampal QTM velocity interaction term was significant (β=-1.760, p=0.049), indicating a sex-dependent association.

[0080] As a sensitivity analysis, further sex-stratified linear model analysis was performed in the symptomatic cohort, adjusting for age and standardized hippocampal volume. In the female group, QTM velocity in the gray matter and pWMH regions remained significantly associated with MMSE (gray matter: β=1.546, p=0.002, R²=0.335; pWMH: β=1.654, p=0.011, R²=0.331), while no similar effect was found in the male group. In all models, hippocampal volume showed a persistent negative correlation with MMSE, reflecting the close association between hippocampal atrophy and cognitive decline. In summary, QTM velocity (especially in the hippocampus and gray matter regions) is associated with cognitive performance in symptomatic participants and is sex-dependent, while the association with CBF is weaker.

[0081] Mediation effect analysis

[0082] Hippocampal atrophy is a classic structural marker closely associated with cognitive decline in the course of Alzheimer's disease (AD). Based on the finding of an interaction between QTM velocity and disease stage and cognitive performance, we hypothesized that regional QTM velocity might mediate the atrophy-cognitive association. Our initial model, using whole-cortical QTM velocity as a mediating variable, only reached marginal significance (p=0.074). Subsequent lobe-level analysis identified parietal QTM velocity as a specific functional mediator.

[0083] In the covariate-adjusted model (age and sex), hippocampal volume was positively correlated with parietal QTM velocity (path a: β=0.283, p<0.001), and after controlling for hippocampal volume, parietal QTM velocity was positively correlated with MMSE (path b: β=0.197, p=0.007). Even after including mediating variables, the direct effect of hippocampal volume on MMSE remained significant (path c′: β=0.461, p<0.001), suggesting partial mediation. Bootstrap mediation analysis (5000 simulations) showed significant indirect effects (ACME=0.056, p=0.034), direct effects (ADE=0.462, p<0.001), and total effects (total effect=0.518, p<0.001). Approximately 10.2% of the total association between hippocampal volume and MMSE was mediated by parietal QTM velocity (p=0.034).

[0084] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An image processing method of ASL cerebral perfusion kinetics, characterized in that: It includes the following steps: (1) The subjects were divided into groups according to the clinical diagnostic criteria: normal cognitive NC group, subjective cognitive decline SCD group, mild cognitive impairment MCI group, and Alzheimer's disease AD group. Their demographic information and neuropsychological scores were collected. (2) Acquire magnetic resonance imaging data, including T1-weighted structural images, T2-FLAIR sequences, and multiple delayed ASL sequences, to obtain information related to brain structure and perfusion; (3) Preprocess the magnetic resonance imaging data, including motion correction, spatial registration, and standard spatial transformation, to ensure spatial consistency between different modal data; (4) Use an automatic segmentation algorithm to segment the brain regions of the T1 structural image to obtain the regions of interest of gray matter, white matter, hippocampus and each lobe. (5) Calculate brain perfusion-related dynamic parameters based on multi-delay ASL data, generate a voxel-level parameter distribution map, and register it to the structural space; (6) Extract the average parameter values ​​in each region of interest and perform volume normalization to reduce the impact of individual differences; (7) Statistical analysis was used to compare the differences between different groups, assess the interaction between gender and disease stage, analyze the correlation between imaging indicators and plasma biomarkers and cognitive function, and explore the relationship between brain structural changes, perfusion abnormalities and cognitive decline through mediation analysis.

2. The method of image processing of ASL cerebral perfusion kinetics according to claim 1, characterized in that: In step (1), the NC group is strictly defined as an individual with no abnormalities in clinical interviews, a Mini-Mental State Examination (MMSE) score ≥27 and a Clinical Dementia Rating Scale (CDR) global score of 0; where feasible, the diagnostic classification of the SCD, MCI and AD groups is verified by amyloid PET imaging or plasma biomarker analysis to confirm their amyloid positive status.

3. The method of image processing of ASL cerebral perfusion kinetics according to claim 2, characterized in that: In step (2), a 3.0T MRI scanner equipped with an 8-channel head coil is used. The core sequence includes: (1) 3D T1-weighted imaging for region of interest (ROI) segmentation, with repetition time / echo time = 6.7 / 2.9 ms, flip angle = 12°, slice thickness = 1.0 mm, no gap, matrix = 256×256, and field of view = 256×256 mm²; (2) multi-label post-delayed ASL based on enhanced ASL sequence for cerebral blood flow (CBF) and quantitative transport mapping (QTM) perfusion velocity quantification. Seven different label post-delays [800, 1098, 1427, 1797, 2228, 2761, 3492] ms are encoded into a single acquisition, and the effective label duration is [298, 328, 370, 432, 533, 731, 1308]. ms, repetition time / echo time = 6273 / 11.4 ms, flip angle = 111°, voxel size = 1.72×1.72×4.5 mm³, matrix = 128×128, field of view = 220×220 mm², number of excitations = 2; (3) T2-FLAIR sequence used to quantify white matter high signal WMH representing small vessel injury, repetition time / echo time = 8668 / 166 ms, flip angle = 111°, voxel size = 0.5×0.5×5 mm³, slice interval = 1.0 mm, matrix = 256×256, field of view = 240×240 mm².

4. The method of image processing of ASL cerebral perfusion kinetics according to claim 3, characterized in that: In step (4), FreeSurfer v7.1 was used to automatically segment the T1-weighted images into cortical and subcortical regions to generate individual-specific ROIs. These ROIs were bilaterally merged, including the whole cortical gray matter (GM), white matter (WM), hippocampus, and frontal, parietal, temporal, and occipital lobes. To extract ROI values, the CBF map and QTM velocity map derived from mPLD-ASL were registered to the T1w image in the FreeSurfer space for each participant. To mitigate the partial volume effect at tissue boundaries, all ROIs underwent 1 mm morphological erosion before extracting the regional average. A validated deep learning-based neural network was used to segment the WMH from the T2-FLAIR images and was used to evaluate CBF, QTM velocity, and their association with cognition. All ROI volume values ​​were normalized by intracranial volume to correct for head circumference differences.

5. The image processing method of ASL cerebral perfusion kinetics according to claim 4, characterized in that: In step (5), the CBF map is reconstructed on the scanner from data acquired by mPLD-ASL using a dynamic model with arterial transit time ATT correction. ATT correction is used to reduce the CBF estimation bias caused by the extended labeling delay time. The CBF map is registered to the FreeSurfer T1w space of each subject using the FMRIB linear image registration tool FLIRT. To improve registration accuracy, image registration is performed within the cortical mask area, and CBF values ​​are extracted in the FreeSurfer T1w space using the corresponding FreeSurfer partition.

6. The image processing method of ASL cerebral perfusion kinetics according to claim 5, characterized in that: In step (6), the QTM models tracer kinetics through a mass conservation equation to estimate perfusion velocity: , Where C(x,t) is the tracer concentration at position x and time t, v(x) is the time-invariant tracer velocity field, D is the apparent diffusion coefficient, and R is the signal attenuation rate; for ASL-labeled endogenous water, R is set to 1 / T1b, where T1b is the blood time T1, and the diffusion term is ignored in the perfusion estimation. Because the diffusion rate is much slower than the infusion-related transport process; The velocity field v is reconstructed by minimizing an objective function: , where λ is a regularization parameter, λ = 0.1, used to promote spatial smoothness of the velocity field, the reconstruction of motion-corrected mPLD-ASL data was done using in-house MATLAB code, v denotes the tracer velocity of the capillary surface at a voxel, with typical units of mm / s, whose magnitude is defined as the average perfusion velocity at the voxel level in the QTM model; The QTM velocity map is registered to the FreeSurfer T1w space of each subject using the same transformation matrix as the CBF map, and the ROI velocity value is extracted using the same method as CBF.

7. The image processing method of ASL cerebral perfusion kinetics according to claim 6, characterized in that: In step (7), continuous variables are represented by the median interquartile range, and categorical variables are represented by the count %; All tests were two-tailed, with a significance threshold of p<0.

05. For comparisons across multiple regions, the Benjamini-Hochberg false discovery rate (FDR) was used to correct the p-value.

8. The method of image processing of ASL cerebral perfusion kinetics according to claim 7, characterized in that: In step (7), demographic, clinical, and biomarker characteristics among the four diagnostic groups were compared using the Kruskal-Wallis test or the χ² test. For those with significant Kruskal-Wallis results, the FDR-corrected Dunn post-hoc test was performed. The Kruskal-Wallis test and the FDR-corrected Dunn test were used to assess the inter-group differences in mean CBF and QTM velocity of the ROI. The Wilcoxon rank-sum test was used to analyze gender differences in the whole sample and in specific diagnostic stages. To explore whether gender differences change with disease progression, an age-corrected linear interaction model was fitted in the symptomatic cohorts SCD, MCI, and AD, incorporating age, gender, ordered diagnostic stage, and gender × diagnosis interaction terms to assess the attenuation or exacerbation trend of gender differences in clinical stages.

9. An image processing system for ASL cerebral perfusion kinetics, characterized in that: It includes: The grouping module is configured to group subjects according to clinical diagnostic criteria into the normal cognitive NC group, the subjective cognitive decline SCD group, the mild cognitive impairment MCI group, and the Alzheimer's disease AD group, and collect their demographic information and neuropsychological scores. The data acquisition module is configured to acquire magnetic resonance imaging data, including T1-weighted structural images, T2-FLAIR sequences, and multi-delay ASL sequences, to obtain information related to brain structure and perfusion. The preprocessing module is configured to preprocess magnetic resonance imaging data, including motion correction, spatial registration, and standard spatial transformation, to ensure spatial consistency between different modal data. The brain region segmentation module is configured to use an automatic segmentation algorithm to segment brain regions on T1 structural images, obtaining regions of interest for gray matter, white matter, hippocampus, and each lobe. The generation module is configured to calculate brain perfusion-related dynamic parameters based on multi-delay ASL data, generate a voxel-level parameter distribution map, and register it to the structural space. The extraction module is configured to extract average parameter values ​​in each region of interest and perform volume normalization to reduce the impact of individual differences. The statistical analysis module is configured to use statistical analysis methods to compare differences between different groups, assess the interaction between gender and disease stage, analyze the correlation between imaging indicators and plasma biomarkers and cognitive function, and explore the relationship between brain structural changes, perfusion abnormalities and cognitive decline through mediation analysis.

10. The image processing system for ASL cerebral perfusion kinetics associated with the marker of claim 9, wherein: In the overall cohort and sex stratification, Spearman correlation analysis was used to assess the correlation between plasma biomarkers and regional CBF and QTM velocity. For each biomarker and imaging modality, FDR correction was performed independently to avoid cross-interference. As a sensitivity analysis, a covariate-adjusted linear regression model was used to verify the robustness of significant associations. QTM velocity showed a robust negative correlation with the biomarker NfL in all brain regions. The correlation between QTM velocity and GFAP showed brain region dependence; in the overall sample, hippocampal QTM velocity was negatively correlated with the biomarker GFAP, and this association was driven by females.

Images