阅读说明：本技术 对阿尔茨海默病的扩散磁共振成像处理方法 (Diffusion magnetic resonance imaging processing method for Alzheimer disease ) 是由 马国林 杜雷 高文文 刘秀秀 王依格 于 2020-05-25 设计创作，主要内容包括：对阿尔茨海默病的扩散磁共振成像处理方法,首先在3D T1图像上勾画出感兴趣区ROI,将勾画好的T1图像配准到dMRI图像中,随后计算出双侧海马的α,H,μ和ADC的平均值,然后使用FSL工具对获得的图像进行头部运动和涡流畸变校正,使用b值为0和954s/mm～(2)时获得的图像计算ADC值,采用FM模型对图像进行分析以获取异常扩散参数,根据基于FM模型的dMRI理论获得由扩散引起的信号衰减,再进行统计学分析获得结果,因此能够较好地反映生物活细胞扩散过程,达到真正信号衰减曲线和拟合曲线之间的最佳一致性。(A diffusion magnetic resonance imaging processing method for Alzheimer disease comprises the steps of firstly delineating a region of interest ROI on a 3D T1 image, registering the delineated T1 image into a dMRI image, then calculating the average values of alpha, H, mu and ADC of bilateral hippocampus, then using an FSL tool to correct head movement and eddy current distortion of the obtained image, and using b values of 0 and 954s/mm 2 The ADC value of the obtained image is calculated, the FM model is adopted to analyze the image so as to obtain abnormal diffusion parameters, signal attenuation caused by diffusion is obtained according to a dMRI theory based on the FM model, and then statistical analysis is carried out to obtain a result, so that the diffusion process of the living biological cells can be better reflected, and the optimal consistency between a real signal attenuation curve and a fitting curve is achieved.)

1. A diffusion magnetic resonance imaging processing method for Alzheimer's disease is characterized in that: which comprises the following steps:

(1) image acquisition: on a magnetic resonance scanner, using an 8-channel head coil, obtaining dMRI images of all subjects using a Stejskal-Tanner single spin echo planar imaging sequence;

(2) selecting a left hippocampus and a right hippocampus as interested regions ROI, firstly delineating the ROI on a 3D T1 image, then registering the delineated T1 image into a dMRI image, then calculating the average values of alpha, H, mu and ADC of bilateral hippocampus, wherein alpha refers to Noah index, the fluctuation of random process is quantified, and the increment is Gaussian distribution when alpha is 2 and is Lnevy distribution when 0< alpha < 2; h is a hester index, describing the self-similarity of molecular orbitals; μ is a memory parameter and μ — h-1/α, increments are positively correlated and long range dependent when μ >0 and negatively correlated and short range dependent when μ < 0;

(3) the obtained image was corrected for head movement and eddy current distortion using an FSL tool using b values of 0 and 954s/mm²Calculating an ADC value of the obtained image, analyzing the image by adopting an FM model to obtain an abnormal diffusion parameter, and calculating the signal attenuation caused by diffusion according to a dMRI theory based on the FM model by using a formula (1):

D_α,Ηdiffusion coefficient of anomalous diffusion, gamma gyromagnetic ratio, G₀The diffusion gradient amplitude is referred, delta is the gradient separation time, eta is a dimensionless number and is determined by alpha, H, delta and delta;

(4) the differences in age, education and MMSE scores of the mild AD patients, the moderate AD patients and the healthy controls were compared by one-way analysis of variance, the data were expressed as mean. + -. standard deviation, the gender was compared by chi-square test, and the difference was statistically significant when the P-value was < 0.05.

2. The diffuse magnetic resonance imaging treatment method for alzheimer's disease according to claim 1, characterized in that: said step (c) is(1) In the conventional dMRI sequence scanning process, the diffusion gradient separation time delta is 27.060ms, 39.560ms and 52.060ms in sequence, and the diffusion gradient amplitude G of each delta value₀15.67, 19.68, 24.73, 31.06, 39.01 and 49.00mT/m in sequence, and a gradient time constant delta is fixed to 20.676 ms; thus, 18 non-zero b-values are obtained in each gradient direction: 151. 239, 377, 595, 939, 1481, 245, 387, 611, 964, 1521, 2399, 339, 535, 845, 1333, 2103 and 3317s/mm²(ii) a In order to reduce the influence of diffusion anisotropy, diffusion gradients were applied in three orthogonal directions, x-axis, y-axis and z-axis, and a total of 12 images were obtained with b being 0.

3. The diffuse magnetic resonance imaging treatment method for alzheimer's disease according to claim 2, characterized in that: in the step (1), dMRI sequence parameters: the repetition time TR is 3800 ms; echo time TE is 110 ms; the flip angle FA is 90 °; the layer thickness is 5.0 mm; the FOV is 240mm multiplied by 240 mm; the matrix size is 128 × 128; the acceleration factor is 2; the excitation times are 2; the total scanning time is 8 minutes and 33 seconds; t1 imaging parameters: TR is 6.7 ms; TE Min Full; FA-12 °; the layer thickness is 1.0 mm; the FOV is 256mm multiplied by 256 mm; the matrix size is 256 × 256, and the scan time is 4 minutes and 10 seconds.

4. The diffuse magnetic resonance imaging treatment method for alzheimer's disease according to claim 3, characterized in that: in step (2), the ROI was manually delineated on the 3D T1 image by an experienced radiologist using MRICRON software.

5. The diffuse magnetic resonance imaging treatment method for alzheimer's disease according to claim 4, characterized in that: in the step (3), all fitting processes are carried out by adopting a trust-region-reflection nonlinear fitting algorithm in MATLAB.

6. The diffuse magnetic resonance imaging treatment method for alzheimer's disease according to claim 5, characterized in that: in said step (4), the difference between α, Η, μ and ADC values between AD group and healthy control group and between mild AD group and moderate AD group was compared using a two-sample t-test.

7. The diffuse magnetic resonance imaging treatment method for alzheimer's disease according to claim 6, characterized in that: in the step (4), a receiver operating characteristic curve ROC is adopted for evaluation, and an area AUC under the curve is calculated.

8. The diffuse magnetic resonance imaging treatment method for alzheimer's disease according to claim 7, characterized in that: in step (5), correlation between α, Η, μ and ADC values and MMSE and MoCA scale scores for AD patients was obtained using pearson partial correlation analysis.

Technical Field

The invention relates to the technical field of medical image processing, in particular to a diffusion magnetic resonance imaging processing method for Alzheimer's disease.

Background

Alzheimer's Disease (AD) is a complex and common neurodegenerative disease characterized by memory loss and cognitive decline. According to World Health Organization (WHO) data, dementia affects nearly 4750 thousands of people worldwide and there are still approximately 770 new cases increasing each year. Alzheimer's disease is the most common type of dementia, and may account for 60-70% of these cases, and has a major impact on the patient's daily life. The pathogenesis of AD is very complex, mainly involving the formation of beta amyloid (a β) and hyperphosphorylated tau protein leading to the formation of a β plaques and neurofibrillary tangles, respectively, leading to neuronal death. At present, the diagnosis of AD is complex and has low accuracy, so that the development of an effective AD diagnosis method is very important in clinical research.

Diffusion magnetic resonance imaging (dMRI) is a powerful tool that can describe the random movement of water molecules in biological tissues, can provide unique information about the microscopic properties of the tissues, and is highly sensitive to detect changes in grey and white matter of the brain. The diffusion process of water molecules in the nervous system is directional. This directional dependence, i.e., anisotropy, is primarily due to the intrinsic axonal membrane impeding the diffusion of water molecules and the dense packing of axons. Measuring the anisotropy of water molecule diffusion at the micrometer level can reflect changes in the microstructure.

Compared with other magnetic resonance imaging methods, dMRI researches on the horizontal diffusion process of cells are far superior to traditional millimeter-level resolution images. One of the most widely used dMRI techniques in clinical practice at present is the Apparent Diffusion Coefficient (ADC), and the b value obtained by dMRI is 0-1000 s/mm²In the meantime. The clinical application of ADC is wide, and the ADC can be used for identifying brain tumors and grading the tumors, and can also be used for identifying patients with Mild Cognitive Impairment (MCI) and AD (Alzheimer's disease). Diffusion Tensor Imaging (DTI) is one of the widely used dMRI techniques in research. Fractional Anisotropy (FA) and Mean Diffusivity (MD) obtained from DTI are potential biomarkers of brain abnormalities in MCI and AD patients. However, both ADC and DTI haveCertain limitations are imposed. First, conventional dMRI employs a single exponential model, which assumes that there is a regular diffusion process in biological tissue. However, many studies have found that the observed dMRI signal decay curve deviates from the single exponential form. Whereas both ADC and DTI are based on conventional single exponential diffusion models, the observed diffusion time reflects the non-gaussian nature of diffusion. The second is that the DTI tensor model is too simple, which means that its indices are affected by many features of the microstructure. To solve this problem, many models based on different theories and anomalous diffusion processes were developed to find the best agreement between the true signal attenuation curve and the fitted curve. These models allow more detailed detection of disease type and grade, but their signal attenuation curves are still poorly matched.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention provides a diffusion magnetic resonance imaging processing method for Alzheimer's disease, which can better reflect the diffusion process of living biological cells and achieve the optimal consistency between a real signal attenuation curve and a fitting curve.

The technical scheme of the invention is as follows: the diffusion magnetic resonance imaging processing method for the Alzheimer disease comprises the following steps:

(1) image acquisition: on a magnetic resonance scanner, using an 8-channel head coil, obtaining dMRI images of all subjects using a Stejskal-Tanner single spin echo planar imaging sequence;

(2) selecting a left hippocampus and a right hippocampus as interested regions ROI, firstly delineating the ROI on a 3D T1 image, then registering the delineated T1 image into a dMRI image, then calculating the average values of alpha, H, mu and ADC of bilateral hippocampus, wherein alpha refers to Noah index, the fluctuation of random process is quantified, and the increment is Gaussian distribution when alpha is 2 and is Lnevy distribution when 0< alpha < 2; h is a hester index, describing the self-similarity of molecular orbitals; μ is a memory parameter and μ — h-1/α, increments are positively correlated and long range dependent when μ >0 and negatively correlated and short range dependent when μ < 0;

(3) image acquisition using FSL tool pairCorrection of head movement and eddy current distortion using b values of 0 and 954s/mm²Calculating an ADC value of the obtained image, analyzing the image by adopting an FM model to obtain an abnormal diffusion parameter, and calculating the signal attenuation caused by diffusion according to a dMRI theory based on the FM model by using a formula (1):

D_α,Ηdiffusion coefficient of anomalous diffusion, gamma gyromagnetic ratio, G₀The diffusion gradient amplitude is referred, delta is the gradient separation time, eta is a dimensionless number and is determined by alpha, H, delta and delta;

(4) the differences in age, education and MMSE scores of the mild AD patients, the moderate AD patients and the healthy controls were compared by one-way analysis of variance, the data were expressed as mean. + -. standard deviation, the gender was compared by chi-square test, and the difference was statistically significant when the P-value was < 0.05.

The invention firstly delineates the ROI of interest on a 3D T1 image, registers the delineated T1 image into a dMRI image, then calculates the average values of alpha, H, mu and ADC of bilateral hippocampus, and then uses an FSL tool to correct head movement and eddy current distortion of the obtained image, and uses b values of 0 and 954s/mm²The ADC value of the obtained image is calculated, the FM model is adopted to analyze the image so as to obtain abnormal diffusion parameters, signal attenuation caused by diffusion is obtained according to a dMRI theory based on the FM model, and then statistical analysis is carried out to obtain a result, so that the diffusion process of the living biological cells can be better reflected, and the optimal consistency between a real signal attenuation curve and a fitting curve is achieved.

Drawings

Fig. 1 shows that the left and right hippocampus are selected as the regions of interest.

Fig. 2 shows axial T1-weighted images, alpha, H, ADC plots, of an AD patient (first row, 58 year old male) and a healthy control group (second row, 60 year old male), with no significant difference in bilateral hippocampus seen with the naked eye between AD patient and healthy control.

Figure 3 shows a comparison between AD patients and healthy controls. Scatter plots (a) and table (B) show that alpha and ADC values can identify AD patients and healthy controls, while H and μ values are not statistically significantly different when identifying the two groups. The AD group had 24 persons, and the healthy control group had 11 persons. With the two-sample t test, P <0.05 is statistically significant.

Figure 4 shows a comparison between patients with mild AD and moderate AD. Scatter plots (a) and table (B) show that alpha and ADC values can identify two groups, whereas H and μ values do not show significant statistical differences when identifying two groups. There were 12 patients with mild AD and 12 patients with moderate AD. With the two-sample t test, P <0.05 is statistically significant.

Figure 5 shows ROC plots consisting of α, ADC, α + ADC to identify AD patients (n-24) and healthy controls (n-11). The area under the curve for α + ADC (AUC 0.848, left ROI; AUC 0.856, right ROI) is shown to be higher than α and ADC alone.

Figure 6 shows ROC plots consisting of α, ADC, α + ADC to identify patients with mild AD (n-12) and moderate AD (n-12). The area under the curve for the α + ADC (AUC 0.861, left ROI; AUC 0.868, right ROI) is shown to be higher than for α and ADC alone.

Figure 7 shows the correlation of alpha values and MMSE scores for the right hippocampus of AD patients. Pearson partial correlation analysis was performed. n is 24. P < 0.05.

Figure 8 shows the correlation of alpha and MoCA scores for bilateral hippocampus of AD patients. Pearson partial correlation analysis was performed. n is 24. P < 0.05.

Figure 9 shows a flow chart of a method of diffuse magnetic resonance imaging treatment of alzheimer's disease according to the invention.

Detailed Description

As shown in figure 1, the method for treating the diffusion magnetic resonance imaging of the Alzheimer disease comprises the following steps:

(1) image acquisition: dMRI images of all subjects were obtained on a magnetic resonance scanner (3.0T magnetic resonance scanner, GE Healthcare, Discovery MR750, USA) using a Stejskal-Tanner single spin echo planar imaging sequence using an 8-channel head coil;

(2) the left and right hippocampus were selected as ROI-of-interest, the ROI was first delineated on the 3D T1 image, and the delineated T1 image was then registered to the dmiri image in order to more accurately determine the hippocampal boundaries, excluding ambiguous voxels, and then the mean values of α, H, μ and ADC of the bilateral hippocampus were calculated, α referring to the norian index, quantifying the fluctuation of the stochastic process, the increment was gaussian when α is 2 and the increment was when 0< α <2

L virtual allocation; h is a hester index, describing the self-similarity of molecular orbitals; μ is a memory parameter and μ — h-1/α, increments are positively correlated and long range dependent when μ >0 and negatively correlated and short range dependent when μ < 0;

(3) the obtained image was corrected for head movement and eddy current distortion using an FSL tool using b values of 0 and 954s/mm²Calculating an ADC value of the obtained image, analyzing the image by adopting an FM model to obtain an abnormal diffusion parameter, and calculating the signal attenuation caused by diffusion according to a dMRI theory based on the FM model by using a formula (1):

D_α,Ηdiffusion coefficient of anomalous diffusion, gamma gyromagnetic ratio, G₀The diffusion gradient amplitude is referred, delta is the gradient separation time, eta is a dimensionless number and is determined by alpha, H, delta and delta;

(4) the differences in age, education and MMSE scores of the mild AD patients, the moderate AD patients and the healthy controls were compared by one-way analysis of variance, the data were expressed as mean. + -. standard deviation, the gender was compared by chi-square test, and the difference was statistically significant when the P-value was < 0.05.

The method comprises the steps of firstly delineating a region of interest ROI on a 3D T1 image, registering the delineated T1 image into a dMRI image, then calculating the average values of alpha, H, mu and ADC of bilateral hippocampus, and then using an FSL tool to correct head movement and eddy current distortion of the obtained image, wherein the value b is used as0 and 954s/mm²The ADC value of the obtained image is calculated, the FM model is adopted to analyze the image so as to obtain abnormal diffusion parameters, signal attenuation caused by diffusion is obtained according to a dMRI theory based on the FM model, and then statistical analysis is carried out to obtain a result, so that the diffusion process of the living biological cells can be better reflected, and the optimal consistency between a real signal attenuation curve and a fitting curve is achieved.

Preferably, in step (1), during a conventional dMRI sequence scan, in order to fit the FM model without fixing the diffusion gradient separation time, the diffusion gradient separation time Δ is 27.060ms, 39.560ms, 52.060ms in turn, and the diffusion gradient amplitude G of each Δ value₀15.67, 19.68, 24.73, 31.06, 39.01 and 49.00mT/m in sequence, with the aim of approximately uniform spacing on the logarithmic axis, and the gradient time constant δ fixed at 20.676 ms; thus, 18 non-zero b-values are obtained in each gradient direction: 151. 239, 377, 595, 939, 1481, 245, 387, 611, 964, 1521, 2399, 339, 535, 845, 1333, 2103 and 3317s/mm²(ii) a In order to reduce the influence of diffusion anisotropy, diffusion gradients were applied in three orthogonal directions, x-axis, y-axis and z-axis, and a total of 12 images were obtained with b being 0.

Preferably, in step (1), the dMRI sequence parameters: the repetition time TR is 3800 ms; echo time TE is 110 ms; the flip angle FA is 90 °; the layer thickness is 5.0 mm; the FOV is 240mm multiplied by 240 mm; the matrix size is 128 × 128; the acceleration factor is 2; the excitation times are 2; the total scanning time is 8 minutes and 33 seconds; t1 imaging parameters: TR is 6.7 ms; TE Min Full; FA-12 °; the layer thickness is 1.0 mm; the FOV is 256mm multiplied by 256 mm; the matrix size is 256 × 256, and the scan time is 4 minutes and 10 seconds.

Preferably, in step (2), the ROI is manually delineated on the 3D T1 image by an experienced radiologist using MRICRON software.

Preferably, in the step (3), all fitting processes are performed by using a trust-region-reflection nonlinear fitting algorithm in MATLAB.

Preferably, in step (4), the difference between α, Η, μ and ADC values between AD group and healthy control group and between mild AD group and moderate AD group is compared using a two-sample t-test.

Preferably, in said step (4), in order to quantify the sensitivity and specificity of α, Η, μ and ADC values in the identification of AD groups and healthy control groups and in the identification of mild AD groups and moderate AD groups, the subject working characteristic curve roc (receiver operating characteristics) is used for evaluation and the area under the curve auc (area under the curve) is calculated.

Preferably, in step (5), correlation between α, Η, μ and ADC values and MMSE and MoCA scale scores for AD patients is obtained using pearson partial correlation analysis.

Examples of the present invention are described in more detail below.

Demographic and clinical information for all subjects is summarized in table 1. The present invention included 24 AD patients (9 males, 15 females, mean age 68.96 years, range 50-79 years) and 11 healthy controls (2 males, 9 females, mean age 65.27 years, range 54-78 years). The AD groups were divided into mild AD group and moderate AD group. The three groups of patients had no statistical difference in age, education (P > 0.05). The MMSE scores of three groups of patients were statistically different (P < 0.05).

TABLE 1

Fig. 1 shows bilateral hippocampus in an axial T1 image. Fig. 2 shows α, h, ADC images and T1 images for AD and healthy controls, with no significant difference seen to the naked eye in bilateral hippocampus between AD patients and healthy controls.

TABLE 2

The values of FM-related parameters and ADC in bilateral hippocampus of all subjects are summarized in Table 2 and expressed as mean. + -. standard deviation. Fig. 3 and 4 show scatter plots identifying AD patients versus healthy controls and mild AD and moderate AD patients, respectively. As shown in fig. 3, the a value (left ROI, P ═ 0.011; right ROI, P ═ 0.001) and the ADC value (left ROI, P ═ 0.001; right ROI, P ═ 0.001) can identify AD patients and healthy controls. In addition, fig. 4 also shows that alpha values (right ROI, P ═ 0.015) and ADC values (left ROI, P ═ 0.011; right ROI, P ═ 0.022) can identify mild AD and moderate AD patients. Whereas h and μ values failed to distinguish between AD groups and healthy controls, mild AD and moderate AD patients (P > 0.05).

ROC curve analysis also shows the ability to identify AD patients from healthy controls. Fig. 5 is a ROC curve identifying AD patients from healthy controls, as seen, α + ADC (AUC ═ 0.848 left ROI, AUC ═ 0.856 right ROI) shows larger area under the curve compared to α (AUC ═ 0.811 left ROI) and ADC (AUC ═ 0.847 left ROI, AUC ═ 0.833 right ROI), showing the highest approximately dengue index (sensitivity + specificity-1) when α is thresholded at 1.5939 (left ROI) and 1.58415 (right ROI), with sensitivity and specificity at 0.917 and 0.636 (left ROI), 0.917 and 0.727 (right ROI), respectively. Similarly, for patients identified with mild AD and moderate AD (fig. 6), a + ADC (AUC 0.861 left ROI, AUC 0.868 right ROI) gave larger area under the curve compared to α (AUC 0.813 right ROI) and ADC (AUC 0.792 left ROI, AUC 0.806 right ROI); the highest john index was obtained when α is 1.55115 (right ROI), where sensitivity and specificity were 0.917 and 0.75, respectively.

The value of the FM associated parameter alpha in AD patients is positively correlated with MMSE scores (P < 0.05; FIG. 7) and MoCA scores (P < 0.05; FIG. 8). Other FM-related parameter values and ADC values have no significant correlation with MMSE scores and MoCA scores.

The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention in any way, and all simple modifications, equivalent variations and modifications made to the above embodiment according to the technical spirit of the present invention still belong to the protection scope of the technical solution of the present invention.