Image processing method, program, and image processing device

Abstract

This image processing method includes: a step in which a vascular mask creating unit creates a vascular mask image by subjecting an image, obtained from a magnetic resonance signal obtained by enhancing regions corresponding to blood vessels in an MRI image, to removal of dot-shaped low signals, removal of linear low signals, extraction of linear low signals, removal of dot-shaped high signals, removal of linear high signals, and extraction of linear high signals; and a step in which a mask processing unit generates an image obtained by employing the vascular mask image to remove a vascular structure from a phase difference enhanced image created from the MRI image.

Description

Image processing method, program, and image processing device

[0001] The present invention relates to an image processing method, a program, and an image processing device. This application claims priority to Japanese Patent Application No. 2021-029161, filed on February 25, 2021, the contents of which are incorporated herein by reference.

[0002] Amyloid beta is the causative agent of Alzheimer's disease, the leading cause of dementia. Amyloid beta deposits in the brains of Alzheimer's disease patients, for example, 20 to 30 years before the onset of dementia. Therefore, detecting amyloid beta during the asymptomatic stage is highly desirable from the perspectives of early treatment, prevention of dementia, and monitoring in drug development. A conventional technique for diagnosing intracerebral amyloid beta deposition is positron emission tomography (PET) testing (amyloid PET), which uses radioactive agents that bind to amyloid beta, such as PiB (Pittsburgh compound-B). Amyloid PET is a useful imaging test that can visualize amyloid beta deposition in the brain, but has drawbacks such as patient exposure, high testing costs, and a limited number of facilities where it can be performed. Therefore, as a diagnostic method without radiation exposure, research has been vigorously conducted on predicting cerebral amyloid β deposition using MRI (Magnetic Resonance Imaging). For example, research has been conducted to predict the cerebral deposition of amyloid β and age-related iron based on a quantitative value called QSM (Quantitative Susceptibility Mapping) obtained from the phase component information of MRI images. In addition, a method has been proposed in which a phase corresponding to the tissue contrast to be enhanced is selected and enhanced using PADRE (Phase Difference Enhancement Imaging; phase difference enhanced imaging method) (see, for example, Patent Document 1). The PADRE method can improve the tissue resolution of target tissues such as tissues and blood vessels, such as amyloid plaques and blood vessels.

[0003] Furthermore, visual assessment using AP-PADRE (Amyloid Plaque-PADRE) has been proposed. Significant differences were observed in the distribution of low signals in the cerebral cortex in visual assessment using AP-PADRE between patients clinically diagnosed with Alzheimer's disease (patients with Alzheimer's dementia) and a control group, and it has been reported that the distribution of low signals in the superior temporal gyrus in patients with Alzheimer's disease is positively correlated with the Mini-Mental State Examination (MMSE) score (see, for example, Non-Patent Document 1).

[0004] International Publication No. 2010 / 073923

[0005] Tateishi M. et al., “Differentiating between Alzheimer Disease Patients and Controls with Phase-difference-enhanced Imaging at 3T: A Feasibility Study”, Magn Reson Med Sci. 2018

[0006] However, although the evaluation in Non-Patent Document 1 involved visual evaluation, it did not extract and comparatively evaluate amyloid beta, the cause of Alzheimer's disease, but rather relied solely on clinical symptoms to assess discrimination ability. Furthermore, with the method using AP-PADRE images, it is difficult to distinguish between PiB-positive and PiB-negative cases based solely on the simple distribution of low signal intensity in the cortex. Note that PiB-positive refers to a positive diagnosis in PiB-PET testing, while PiB-negative refers to a negative diagnosis in PiB-PET testing. Furthermore, although PiB-positive cases tend to exhibit unclear corticomedullary boundaries and regional differences in signal intensity in the created color map, these were not reliable. Thus, visual evaluation of amyloid beta deposition was difficult with conventional techniques.

[0007] The present invention has been made in consideration of the above-mentioned problems, and aims to provide an image processing method, program, and image processing device that can visually determine and quantitatively evaluate the presence or absence of amyloid beta deposition, and diagnose the presence or absence of Alzheimer's disease.

[0008] (1) In order to achieve the above object, an image processing method according to one aspect of the present invention includes a step in which a vascular mask creation unit creates a vascular mask image by removing point-like low signals, removing linear low signals, and extracting linear low signals, and removing point-like high signals, removing linear high signals, and extracting linear high signals from an image obtained from magnetic resonance signals that have enhanced areas corresponding to blood vessels in an MRI image; and a step in which a mask processing unit uses the vascular mask image to generate an image in which vascular structures have been removed from a phase-contrast-weighted image created from the MRI image.

[0009] (2) In addition, the image processing method according to one aspect of the present invention may further include a step in which a separation unit separates local low signal intensity and diffuse components from the image from which the vascular structure has been removed, and a step in which a brain surface projection image output unit projects the diffuse components onto the brain surface to generate a brain surface projection image.

[0010] (3) In addition, in an image processing method according to one aspect of the present invention, the vascular mask creation unit may create the vascular mask image by morphological calculation processing, and the separation unit may separate the local low signal intensity and diffuse component image by the morphological calculation processing.

[0011] (4) In addition, in an image processing method according to one aspect of the present invention, the vascular mask creation unit may include the steps of: removing the point-like hypointensities by morphological calculation processing using a 3×3×3 linear kernel in 13 directions; removing the linear hypointensities by morphological calculation processing using a 3×3×3 planar kernel in 13 directions on the image from which the point-like hypointensities have been removed; extracting the linear hypointensities using the image from which the point-like hypointensities have been removed and the image from which the linear hypointensities have been removed; removing the point-like hyperintensities by morphological calculation processing using a 3×3×3 linear kernel in 13 directions; removing the linear hyperintensities by morphological calculation processing using a 3×3×3 planar kernel in 13 directions on the image from which the point-like hyperintensities have been removed; and extracting the linear hyperintensities using the image from which the point-like hyperintensities have been removed and the image from which the linear hyperintensities have been removed.

[0012] (5) In addition, in an image processing method according to one aspect of the present invention, the vascular mask creation unit may perform, for different pixel widths, a process of removing the point-like low signals, a process of removing the linear low signals, a process of extracting the linear low signals, a process of removing the point-like high signals, a process of removing the linear high signals, and a process of extracting the linear high signals.

[0013] (6) In addition, in an image processing method according to one aspect of the present invention, the separation unit may perform a closing process by morphological operation using a spherical kernel of a radius p (p is an integer of 1 to n) pixels on the image from which the vascular structure has been removed, to separate the local low signal of size n, and separate a diffuse image using p results obtained by performing the closing process on the image from which the vascular structure has been removed using a spherical kernel of a radius n pixels.

[0014] (7) In the image processing method according to one aspect of the present invention, the phase-contrast weighted image may be an AP-PADRE image created by applying amyloid beta-bound iron-phase contrast weighted imaging to the MRI image.

[0015] (8) In the image processing method according to one aspect of the present invention, the vascular mask creating unit may extract information about a region of interest using the vascular mask image.

[0016] (9) In addition, in an image processing method according to one aspect of the present invention, the vascular mask creation unit may extract information about the region of interest by performing mask processing based on the vascular mask image, a template of the region of interest, information obtained by aligning the acquired enhanced image with the phase-contrast enhanced image to align both images, and information obtained by converting standardized information into the individual brain coordinates based on information obtained by determining a transformation vector field between the individual brain coordinates and the standard brain coordinates by transforming the enhanced image by nonlinear transformation and aligning it with a template image of the enhanced image on standard brain coordinates.

[0017] (10) In order to achieve the above object, a program according to one aspect of the present invention causes a computer to create a vascular mask image by removing point-like low signal intensity, removing linear low signal intensity, and extracting linear low signal intensity, and removing point-like high signal intensity, removing linear high signal intensity, and extracting linear high signal intensity from an image obtained from magnetic resonance signals that have enhanced areas corresponding to blood vessels in an MRI image, and then uses the vascular mask image to generate an image from a phase-contrast-weighted image created from the MRI image in which vascular structures have been removed.

[0018] (11) In order to achieve the above object, an image processing device according to one aspect of the present invention includes a vascular mask creation unit that creates a vascular mask image by removing point-like low signals, removing linear low signals, and extracting linear low signals, and removing point-like high signals, removing linear high signals, and extracting linear high signals from an image obtained from a magnetic resonance signal that has enhanced areas corresponding to blood vessels in an MRI image, and a mask processing unit that uses the vascular mask image to generate an image from which vascular structures have been removed from a phase-contrast-enhanced image created from the MRI image.

[0019] (12) In the image processing device according to one aspect of the present invention, the vascular mask creating unit may extract information about a region of interest using the vascular mask image.

[0020] According to the present invention, it is possible to visually determine and quantitatively evaluate the presence or absence of amyloid β deposition, and to diagnose the presence or absence of Alzheimer's disease.

[0021] FIG. 1 is a diagram showing an example of the configuration of an image processing device according to an embodiment. FIG. 2 is a diagram showing a two-component model for explaining an overview of AP-PADRE. FIG. 3 is a diagram for explaining morphological operations. FIG. 4 is a diagram showing an example of a process for extracting vascular structures according to an embodiment. FIG. 5 is a diagram showing an example of extraction for each vascular diameter. FIG. 6 is a diagram showing an example of a process for creating a vascular mask image according to an embodiment. FIG. 7 is a diagram showing examples of an image before application of a vascular mask image according to an embodiment and an image obtained by removing vascular structures from an AP-PADRE image using a vascular mask and further removing extraparenchymal components using a brain parenchyma mask created from MPRAGE. FIG. 8 is a flowchart of a vascular mask image creation procedure according to an embodiment. FIG. 9 is a flowchart of a process for removing point-like hypointensities with a one-pixel width according to an embodiment. FIG. 10 is a flowchart of a process for removing linear hypointensities with a one-pixel width according to an embodiment. FIG. 11 is a diagram showing a linear kernel of 3×3×3 pixels in 13 directions. FIG. 12 is a diagram showing a planar kernel of 3×3×3 pixels in 13 directions. FIG. 13 is a diagram showing an example of an image used for comparison between a PADRE image from which vascular structures and extraparenchymal components have been removed for a PiB-negative group in a comparative example and PiB-PET. 1 shows a PADRE image in which the vascular structure and extra-brain parenchymal components of the PiB-positive group in the comparative example have been removed, and an example image used for comparison with PiB-PET. It shows an example AAL atlas image set on standard brain coordinates, and an example image in which an ROI based on the AAL atlas is superimposed on the PADRE image. It shows the average values ​​of the PADRE signal normalized by the cerebellar average value for each of the PiB-positive and negative groups in the cortex analyzed by ROI using the AAL atlas. It shows the average values ​​of the PADRE signal normalized by the cerebellar average value for each of the PiB-positive and negative groups in the cortex analyzed by ROI using the AAL atlas. It shows the average values ​​of the PADRE signal normalized by the cerebellar average value for each of the PiB-positive and negative groups in the cortex analyzed by ROI using the AAL atlas. It shows three-dimensional visualization of regions where there was a statistically significant difference between the average values ​​of the PiB-positive and negative groups of the normalized PADRE signal in the cortex analyzed by ROI using the AAL atlas, mapped onto the brain surface. FIG. 10 is a diagram showing color-coded histograms of PiB-positive and -negative groups of PADRE signals for voxel values ​​contained in local brain cortex analyzed by ROI analysis using the AAL atlas.1 is a flowchart of a process for separating a local low signal intensity component and a diffuse component according to an embodiment. FIG. 2 is a diagram showing the average values ​​of normalized PADRE signal values ​​for each of a PiB-positive group and a PiB-negative group within an ROI of a diffuse component image. FIG. 3 is a diagram showing the average values ​​of normalized PADRE signal values ​​for each of a PiB-positive group and a PiB-negative group within an ROI of a diffuse component image. FIG. 4 is a diagram showing the average values ​​of normalized PADRE signal values ​​for each of a PiB-positive group and a PiB-negative group within an ROI of a diffuse component image. FIG. 5 is a diagram showing an example of an image in which average values ​​of diffuse components extracted from a PADRE image in the PiB-negative group and a PiB-positive group are projected onto the brain surface. FIG. 6 is an image in which blood vessels and low signal intensity are removed from an MRI image of a PiB-negative patient with spinocerebellar degeneration using a method according to an embodiment, and the extracted diffuse component is mapped onto the brain surface. 27 to 30 are diagrams showing the results of interpretation of 11 subjects.

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

[0058]

[0059]

[0060]

[0061]

[0062]

[0063]

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074]

[0075]

[0076]

[0077]

[0078]

[0079]

[0080]

[0081]

[0082]

[0083]

[0084]

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095]

[0096]

[0097]

[0098]

[0099]

[0100]

[100]

[1101]

[1112]

[1123]

[113]

[114]

[114]

[115]

[115]

[116]

[117]

[118]

[119]

[120]

[121]

[122]

[123] 1 is a diagram showing a color map projection onto the brain surface of the cerebellum-normalized signal of the AP-PADRE diffuse component image in each ROI, in which the average signal difference between the PiB-positive group and the PiB-negative group was calculated, and the result was used to calculate the average signal difference between the PiB-positive group and the PiB-negative group. This diagram shows the results of the Non-AD group versus the MCI group + Alzheimer's dementia group in the fifth verification. This diagram shows the results of the Non-AD group versus the MCI group + Alzheimer's dementia group in the fifth verification. This diagram shows the results of the Non-AD group versus the MCI group + Alzheimer's dementia group in the fifth verification. This diagram shows the results of the Non-AD group versus the MCI group + Alzheimer's dementia group in the fifth verification.FIG. 10 is a diagram showing the results of a one-tailed t-test between the non-AD group and the Alzheimer's dementia group in the fifth verification. FIG. 11 is a diagram showing the results of a one-tailed t-test between the non-AD group and the Alzheimer's dementia group in the fifth verification. FIG. 12 is a diagram showing the results of a one-tailed t-test between the non-AD group and the Alzheimer's dementia group in the fifth verification. FIG. 13 is a diagram showing the results of a one-tailed t-test of similar ROI analysis results in two groups, a PiB-negative group and a PiB-positive group, for 11 cases having PiB data in the fifth verification. FIG. 14 is a diagram showing the results of a one-tailed t-test of similar ROI analysis results in two groups, a PiB-negative group and a PiB-positive group, for 11 cases having PiB data in the fifth verification. FIG. 15 is a diagram showing the results of a one-tailed t-test of similar ROI analysis results in two groups, a PiB-negative group and a PiB-positive group, for 11 cases having PiB data in the fifth verification.

[0022] Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the drawings used in the following description, the scale of each component is appropriately changed so that each component can be recognized.

[0023] 1 is a diagram showing an example of the configuration of an image processing device 1 according to this embodiment. As shown in FIG. 1, the image processing device 1 includes an image acquisition unit 11, a vascular-weighted image generation unit 12, a Bias Field correction unit 13, a vascular mask creation unit 14, an AP-PADRE image generation unit 15, a Bias Field correction unit 16, a mask processing unit 17, a separation unit 18, a brain surface projection image output unit 19, a 3D-T1-weighted image generation unit 20, a registration unit 21, an anatomical standardization unit 22, a coordinate conversion unit 23, a statistical image analysis result output unit 24, a template storage unit 25, a coordinate conversion unit 26, a mask processing unit 27, and a region-of-interest analysis result output unit 28.

[0024] The image processing device 1 performs image processing on the acquired MRI image, and outputs a brain surface projection image, a statistical image analysis result, and a region of interest analysis result.

[0025] The image acquisition unit 11 acquires two types of images: an original PADRE image based on an MRI image of a patient obtained from magnetic resonance signals, and an original image for generating a 3D-T1 weighted image. The image acquisition unit 11 outputs the acquired original PADRE image to the vascular-weighted image generation unit 12 and the AP-PADRE image generation unit 15, and outputs the acquired original image for generating a 3D-T1 weighted image to the 3D-T1 weighted image generation unit 20. The image acquisition unit 11 may acquire the vascular-weighted image, the AP-PADRE image, and the 3D-T1 weighted image from an external device.

[0026] The vascular image generator 12 generates a vascular image by enhancing the vascular portion of the original PADRE image using a susceptibility change, for example, using a reconstruction method such as SWI (Susceptibility Weighted Imaging). When the image acquirer 11 acquires a vascular image, the vascular image generator 12 outputs the vascular image output by the image acquirer 11 to the bias field corrector 13.

[0027] The bias field correction unit 13 performs bias field correction on the vessel-enhanced image using information output by the registration unit 21. Note that bias field correction is a correction for homogenizing the influence of a magnetic field.

[0028] The vascular mask creating unit 14 creates a vascular mask image for removing vascular structures from the bias field corrected vascular enhancement image. The method for creating the vascular mask image will be described later.

[0029] The AP-PADRE image generator 15 generates an AP-PADRE image by a reconstruction method for the original PADRE image. When the image acquirer 11 acquires an AP-PADRE image, the AP-PADRE image generator 15 outputs the AP-PADRE image output by the image acquirer 11 to the bias field corrector 16.

[0030] The bias field correction unit 16 obtains a corrected image by performing bias field correction on the AP-PADRE image using information output by the registration unit 21. At the same time, a gray matter component image is obtained by tissue segmentation.

[0031] The mask processing unit 17 applies the gray matter mask image obtained by binarizing the gray matter image output by the Bias Field correction unit 16 and the vascular mask image created by the vascular mask creation unit 14 to the Bias Field corrected image output by the Bias Field correction unit 16, and removes from the image portions corresponding to blood vessels and portions other than the gray matter.

[0032] The separation unit 18 separates the image from which the blood vessels have been removed into local low signal components and diffuse components. The separation process will be described later.

[0033] The brain surface projection image output unit 19 outputs the brain surface projection image to an external device (for example, an image display device, a printing device, etc.) based on the results of separation by the separation unit 18.

[0034] The 3D-T1-weighted image generating unit 20 generates a T1-weighted image including the whole brain region by, for example, imaging using the gradient echo method. The 3D-T1-weighted image is used for the mutual registration with PADRE images (vascular-weighted images and AP-PADRE images) and the anatomical standardization process of the images. When the image acquiring unit 11 acquires a 3D-T1-weighted image, the 3D-T1-weighted image generating unit 20 outputs the 3D-T1-weighted image output by the image acquiring unit 11 to the registration unit 21 and the anatomical standardization unit 22.

[0035] The registration unit 21 registers the 3D-T1 weighted image with the AP-PADRE image, and aligns the two images.

[0036] The anatomical standardization unit 22 transforms the 3D-T1 weighted image by nonlinear transformation and aligns it with a 3D-T1 template image (T1 weighted image template) on the standard brain coordinates, thereby determining a transformation vector field between the individual brain coordinates and the standard brain coordinates.

[0037] The coordinate conversion unit 23 converts the information output by the separation unit 18 into standard brain coordinates based on the information output by the anatomical standardization unit 22 .

[0038] The statistical image analysis result output unit 24 performs a voxel-based statistical image analysis using a known method based on the information output by the coordinate transformation unit 23, and outputs the analysis results to an external device.

[0039] The template storage unit 25 stores templates of regions of interest. The templates of regions of interest are used, for example, for ROI analysis, which will be described later.

[0040] The coordinate conversion unit 26 converts the standardized information into individual brain coordinates based on the template of the region of interest stored in the template storage unit 25, the information output by the alignment unit 21, and the information output by the anatomical standardization unit 22.

[0041] The mask processing unit 27 extracts information about the region of interest by performing mask processing based on the vascular mask image created by the vascular mask creating unit 14 and the information output by the coordinate transformation unit 26 .

[0042] The region of interest analysis result output unit 28 analyzes the region of interest using a known method based on the information output by the mask processing unit 27, and outputs the analysis result to an external device.

[0043] 1 is an example and is not limiting. The image processing device 1 does not necessarily have to include, for example, the 3D-T1 weighted image generating unit 20, the registration unit 21, the anatomical standardization unit 22, the coordinate conversion unit 23, the statistical image analysis result output unit 24, the template storage unit 25, the coordinate conversion unit 26, the mask processing unit 27, and the region of interest analysis result output unit 28.

[0044] (AP-PADRE) Here, an overview of AP-PADRE will be explained. FIG. 2 is a diagram showing a two-component model for explaining an overview of AP-PADRE. In FIG. 2, the horizontal axis represents phase (rad), and the vertical axis represents the number of voxels (frequency of distribution). The dashed line g31 represents the distribution of phase image data collected from a ROI (Region of Interest), the dashed line g32 represents the phase distribution of iron deposits (age-related iron) that physiologically accumulate with aging, which is a background component relative to the phase distribution of iron (amyloid iron) in the AP to be detected, the dashed line g33 represents the phase distribution of iron accumulated in the AP, and the threshold g34 is the phase threshold. Each open square (□) represents actual measurement data. The threshold g34 represents the phase value at the intersection of the distribution of the dashed line g31 and the distribution of the dashed line g33, and serves as a threshold at which the iron distribution in the AP becomes dominant at phase values ​​lower than the intersection. As shown in Figure 2, the entire phase distribution is mostly occupied by a negative region with a phase value of 0 or less. Furthermore, as shown in Figure 2, the phase distribution of the AP is biased toward the negative side (to the left) of the overall distribution. The two-component model assumes that the phase distribution within the ROI is the sum of the phase distribution of iron corresponding to the background component and the phase distribution of iron accumulated in the AP.

[0045] First, an overview of PADRE (phase contrast enhanced imaging) will be described. PADRE is a post-processing technique based on conventional SWI (susceptibility weighted imaging), and enhances the imaging capability of target tissues by selecting and enhancing a phase according to the tissue to be enhanced. By using PADRE, target tissues, such as tissues, blood vessels, and amyloid beta, can be enhanced by improving the contrast in MRI head images (see Patent Document 1).

[0046] Next, AP-PADRE used in this embodiment will be described. AP-PADRE is an image that emphasizes amyloid plaques. AP-PADRE is an image reconstruction method that enables visualization of amyloid β accumulation by selecting a phase difference specific to iron that polymerizes with amyloid β protein, one of the causes of Alzheimer's disease, and enhancing that contrast. Iron present in the cerebral cortex is classified into physiological iron, including aging iron, and non-physiological iron, including amyloid-associated iron. AP-PADRE emphasizes and visualizes amyloid β-bound iron based on the difference in phase difference.

[0047] Visual assessment using AP-PADRE revealed significant differences in the distribution of low signals in the cerebral cortex between patients with Alzheimer's disease and control groups. Furthermore, visual assessment using AP-PADRE has been reported to positively correlate the distribution of low signals in the superior temporal gyrus with MMSE in patients with Alzheimer's disease (see Non-Patent Document 1). Note that this assessment was a clinical evaluation and did not evaluate actual amyloid beta. PiB-PET is an imaging technique that can visualize amyloid beta accumulation in the brain. Visual assessments of AP-PADRE and PiB-PET in patients with Alzheimer's disease, mild cognitive impairment, healthy individuals, and non-Alzheimer's dementia patients showed that it was difficult to distinguish between PiB-positive and PiB-negative cases based solely on the simple distribution of low signals in the cortex in AP-PADRE images. Note that patients with amyloid beta accumulation in the brain are referred to as "Alzheimer's disease," and patients exhibiting symptoms of dementia due to this are referred to as "Alzheimer's dementia."

[0048] The poor contrast within the brain parenchyma of AP-PADRE images, which is thought to be a factor in the challenges of clinical diagnosis using AP-PADRE images, is thought to be due in part to the fact that the numerous fine vascular structures running near the cerebral cortex, the region of interest, are depicted as significantly low signals on AP-PADRE images. Furthermore, patchy low-signal structures are observed within the cortex in AP-PADRE images, but it was unclear whether these patchy structures or the signal value of the cortex itself, which serves as the background, reflect amyloid-β deposition. For this reason, in this embodiment, the removal of vascular structures that hinder quantitative evaluation and the separation of the cortical patchy structures from the diffuse components are performed during the image processing process.

[0049] (Extraction and Removal of Blood Vessel Structure) Next, the process of removing blood vessel structures that interfere with quantitative evaluation will be described with reference to FIGS. 3 and 4. FIG. 3 is a diagram for explaining morphological operations. FIG. 4 is a diagram showing an example of a process of extracting blood vessel structures according to this embodiment. In this embodiment, for example, linear blood vessel structures are extracted using morphological operations. As shown in FIG. 3, morphological transformation performs image processing such as "dilation" and "erosion" on a grayscale image. In morphological transformation, two inputs are provided: an input image and a structural element (kernel). Basic morphological processing includes erosion (g52, g61) and dilation (g51, g62), as well as opening processing g50 and closing processing g60, which combine these two processes. The morphological transformation involves a black hat transformation process g71 that calculates the difference between the input image g61 and the closing-processed image g53, and a top hat transformation process g72 that calculates the difference between the input image g61 and the opening-processed image g63. As will be described later, a linear kernel is used for removing point components, and a planar kernel is used for removing linear components. The pixel values ​​shown in FIG. 3 are merely examples, and are not limiting. The separation unit 18 uses the erosion, dilation, closing, and black hat transformations described with reference to FIG. 3.

[0050] The image processing device 1 removes noise in the background of the input image by the processing of Figure 3 and fills in small black dots contained in the object, and then extracts a linear structure g84 by extracting point components g82 from the input image g81 using a linear kernel and extracting line components g83 using a planar kernel, as shown in Figure 4.

[0051] Next, the process of creating a vascular mask image using morphological calculations will be described with reference to FIGS. 5 to 7. FIG. 5 shows an example of extraction for each vascular diameter. Image g101 is an image obtained by calculation with a 3-pixel width, image g102 is an image obtained by calculation with a 4-pixel width, and image g103 is an image obtained by calculation with a 5-pixel width. In this way, by changing the pixel width used for calculation, vascular images to be removed from the image according to the thickness of the blood vessel can be extracted and removed. Note that a low signal intensity is a low signal intensity (black in the image) in an enhanced image. Also, a high signal intensity is a high signal intensity (white in the image) in an enhanced image. Note that the pixel width shown in FIG. 5 is an example and is not limited to this, and a width of 6 or more pixels may also be used.

[0052] 6 is a diagram showing an example of a process for creating a vascular mask image according to this embodiment. As shown in FIG. 6, the image processing device 1 first performs binarization processing g122 on an input image g121. The threshold value used in the binarization processing is set based on the volume ratio relative to the intracranial volume. Note that the volume ratio of the arterial and venous parts within the cranium is approximately 2% of the intracranial volume, and the threshold value is set based on this. In this embodiment, the mask processing unit 17 creates a vascular mask image including a margin and removes the vascular image.

[0053] Next, the image processing device 1 performs expansion processing g124 (g125) on the binarized image g123, and removes the vascular structure from the MRI image using a vascular mask image such as that shown in Figure 7. Figure 7 shows an image g121 before the application of the vascular mask image according to this embodiment, and an image g126 obtained by removing the vascular structure from the AP-PADRE image using the vascular mask and then removing extraparenchymal components using a brain parenchyma mask created from MPRAGE. Note that for visual evaluation, image g126, which also includes gray matter and white matter, is used. For quantitative analysis, only the gray matter portion is used as the ROI. By removing the vascular structure from the MRI image as shown in Figure 7, it becomes possible to visually evaluate and analyze low-signal structures within the cortex.

[0054] (Vessel Mask Creation Procedure) Next, an example of a vascular mask creation procedure will be described. Fig. 8 is a flowchart of a vascular mask image creation procedure according to this embodiment.

[0055] (Step S11) The vascular mask creating unit 14 removes one-pixel-wide point-like low signals from the bias field corrected vascular image. The processing method for removing low signals will be described with reference to FIG. 9.

[0056] (Step S12) The vascular mask creation unit 14 removes linear hypointensities of one pixel width from the image from which point hypointensities of one pixel width have been removed, and extracts linear hypointensities of one pixel width. The processing method for removing and extracting linear hypointensities will be described with reference to FIG. 10 .

[0057] (Step S13) The vascular mask creating unit 14 removes a 2-pixel wide point hypointensity from the image from which the 1-pixel wide point hypointensity and the 1-pixel wide linear hypointensity have been removed.

[0058] (Step S14) The vascular mask creation unit 14 removes the 2-pixel-wide linear hypointensity signal from the image in which the 1-pixel-wide point-like hypointensity signal, the linear hypointensity signal, and the 2-pixel-wide point-like hypointensity signal have been removed, and extracts the 2-pixel-wide linear hypointensity signal.

[0059] (Step S15) The vascular mask creation unit 14 removes a 3-pixel wide point-like low signal intensity from the image in which the 1-pixel wide point-like low signal intensity, the 1-pixel wide linear low signal intensity, the 2-pixel wide point-like low signal intensity, and the 2-pixel wide linear low signal intensity have been removed.

[0060] (Step S16) The vascular mask creation unit 14 removes a 3-pixel wide linear hypointensity from the image in which the 1-pixel wide point-like hypointensity, the 1-pixel wide linear hypointensity, the 2-pixel wide point-like hypointensity, the 2-pixel wide linear hypointensity, and the 3-pixel wide point-like hypointensity have been removed, and extracts a 3-pixel wide linear hypointensity.

[0061] (Step S17) The vascular mask creation unit 14 creates a linear low signal intensity image using the extracted 1-pixel wide linear low signal intensity extraction image, the 2-pixel wide linear low signal intensity extraction image, and the 3-pixel wide linear low signal intensity extraction image. Through the processing of steps S11 to S17, the vascular mask creation unit 14 extracts low signal intensity components, mainly venous components.

[0062] (Step S18) The vascular mask creation unit 14 performs binarization processing on the linear low-signal image. The threshold used in the binarization processing is set based on the volume ratio to the intracranial volume. After processing, the mask processing unit 17 uses the vascular mask image to remove vascular components from the AP-PADRE image.

[0063] (Steps S21 to S26) The vascular mask creation unit 14 removes point-like hyperintensities with a width of one pixel from the bias field corrected vascular enhancement image. After this, the vascular mask creation unit 14 processes "hyperintensities" instead of "hypointensities" in steps S11 to S16.

[0064] (Step S27) The vascular mask creation unit 14 creates a linear high-intensity image using the extracted 1-pixel-wide linear high-intensity extracted image, the 2-pixel-wide linear high-intensity extracted image, and the 3-pixel-wide linear high-intensity extracted image. The reason for processing the high-intensity components is that when creating a vascular mask image, arteries are also extracted, and high signals appear in arteries with flow. When making judgments based on the image, these components also affect the interpretation. Therefore, by processing steps S21 to S27, the vascular mask creation unit 14 removes the high-intensity components of the arterial components.

[0065] (Step S28) The vascular mask creating unit 14 performs binarization processing on the linear high signal image. After the processing, the mask processing unit 27 performs mask processing to extract information on the region of interest.

[0066] (Step S31) The vascular mask creating unit 14 performs dilation processing on the binarized linear low signal intensity image and linear high signal intensity image.

[0067] (Step S32) The vascular mask creating unit 14 creates and outputs a vascular mask image based on the dilated image.

[0068] 8, the vascular mask creation unit 14 may perform the process of step S10 (S11 to S16) and the process of step S20 (S21 to S26) simultaneously, in parallel, or in a time-division manner. Also, the vascular mask creation unit 14 may perform the process of steps S17 and S18 and the process of steps S27 and S28 simultaneously, in parallel, or in a time-division manner.

[0069] 8 shows an example in which the pixel widths are 1 to 3 and the removal process has three stages, but this is not limiting. The pixel width and the number of stages of the removal process depend on the resolution of the original image, etc., and the removal process may be performed with any pixel width. In this case, the process of removing point-like low signals with an n-pixel width and removing / extracting linear low signals with an n-pixel width may be performed n times, and the process of removing point-like high signals with an n-pixel width and removing / extracting linear high signals with an n-pixel width may be performed n times.

[0070] (Processing Method for Removing Point-Like Signals) Next, a processing method for removing point-like signals will be described. As a representative example, a processing procedure for removing point-like low signals at a width of one pixel will be described. Fig. 9 is a flowchart of the processing procedure for removing point-like low signals at a width of one pixel according to this embodiment.

[0071] (Step S101) The vascular mask creating unit 14 acquires a bias field corrected vascular enhancement image (A).

[0072] (Step S102) The vascular mask creation unit 14 creates a removed image (B) from the vascular-enhanced image (A) by removing point-like low signals of one pixel width using a 13-directional 3×3×3 pixel linear kernel (Fi1) for removing point-like low signals by morphological calculation processing according to the following equation (1). The 13-directional 3×3×3 pixel linear kernel will be described later.

[0073]

[0074] In addition, in formula (1), the operator "●" is a closing process.

[0075] (Step S103) The vascular mask creating unit 14 outputs a removed image (B) in which the created one-pixel wide point low signal intensity has been removed.

[0076] The vascular mask creation unit 14 performs calculations on an image from which point-like low signal intensity components of 2 pixels in width have been removed, using Fi2 instead of Fi1 in equation (1) as a linear structural unit. The vascular mask creation unit 14 performs calculations on an image from which point-like low signal intensity components of 3 pixels in width have been removed, using Fi3 instead of Fi1 in equation (1) as a linear structural unit.

[0077] In addition, the vascular mask creation unit 14 calculates an image from which point high signals of one pixel width have been removed, using a 3 x 3 x 3 pixel linear kernel (Fi1) in 13 directions for removing high signals, and creates a removed image (B) from which linear low signals of one pixel width have been removed by morphological calculation processing using the following equation (2).

[0078]

[0079] In the formula (2), the operator "o" indicates an opening process.

[0080] The vascular mask creation unit 14 performs calculations on an image from which point-like high signals of two pixels wide have been removed, using Fi2 instead of Fi1 in equation (2) as a linear structural unit.The vascular mask creation unit 14 performs calculations on an image from which point-like high signals of three pixels wide have been removed, using Fi3 instead of Fi1 in equation (2) as a linear structural unit.

[0081] (Processing Method for Removing and Extracting Linear Signals) Next, a processing method for removing and extracting linear signals will be described. As a representative example, a processing procedure for removing and extracting linear low signals with a width of one pixel will be described. Fig. 10 is a flowchart of the processing procedure for removing linear low signals with a width of one pixel according to this embodiment.

[0082] (Step S201) The vascular mask creating unit 14 acquires a removed image (B) in which point low signals of one pixel width have been removed.

[0083] (Step S202) The vascular mask creation unit 14 uses a 13-directional 3×3×3 pixel planar kernel (Gi1) for removing linear low signals from the removed image (B) to create a removed image (C) by performing morphological calculation processing according to the following equation (3) to remove linear low signals of 1 pixel width. The 13-directional 3×3×3 pixel planar kernel will be described later.

[0084]

[0085] (Step S203) The vascular mask creating unit 14 outputs a removed image (C) from which the created linear low signal intensity of one pixel width has been removed.

[0086] (Step S204) The vascular mask creating unit 14 subtracts the removed image (C) from the removed image (B) to extract a linear low signal intensity of one pixel width, and outputs the extracted image (B-C).

[0087] The vascular mask creation unit 14 performs calculations on the image from which linear low signal intensity components of 2 pixels in width have been removed, using Gi2 instead of Gi1 in equation (3) as a planar structural unit. The vascular mask creation unit 14 performs calculations on the image from which linear low signal intensity components of 3 pixels in width have been removed, using Gi2 instead of Gi1 in equation (3) as a planar structural unit.

[0088] The vascular mask creation unit 14 also performs morphological calculations on the image from which linear high signals of one pixel width have been removed using a 13-directional 3×3×3 pixel planar kernel (Gi1) for high signal removal to create a removed image (C) from which planar low signals of one pixel width have been removed by the following formula (4): Note that the vascular mask creation unit 14 subtracts the removed image (B) from the removed image (C) to extract linear low signals of one pixel width, and outputs the extracted image (C-B).

[0089]

[0090] The vascular mask creation unit 14 also performs calculations on an image from which linear high signals of two pixels in width have been removed, substituting Gi2 for Gi1 in equation (4) as a planar structural unit.The vascular mask creation unit 14 also performs calculations on an image from which linear high signals of three pixels in width have been removed, substituting Gi3 for Gi1 in equation (4) as a planar structural unit.

[0091] (3x3x3 Kernel in 13 Directions) Next, a 3x3x3 pixel kernel in 13 directions will be described. FIG. 11 is a diagram showing a 3x3x3 pixel linear kernel in 13 directions. As shown in FIG. 11, each of the 13-directional linear kernels g201 to g213 is three adjacent linear pixels in a 3x3x3 pixel space. In this embodiment, point components are removed using such a linear kernel.

[0092] Fig. 12 is a diagram showing a 13-direction, 3x3x3 pixel planar kernel. As shown in Fig. 12, each of the 13-direction planar kernels g301 to g313 is nine adjacent planar pixels in a 3x3x3 pixel space. In this embodiment, point components and line components can be classified by performing processing using linear structural units and planar structural units in the order shown in Figs. 8 to 10.

[0093] In the above example, local low signal components are removed from an AP-PADRE image by using a vascular mask image to remove vascular structures, and then diffuse components are extracted. However, the present invention is not limited to this. For example, the image processing device 1 may remove local low signal components from an AP-PADRE image and then extract diffuse components.

[0094] <Verification> Below, examples of verification results of visual evaluation using images created by the above-described embodiment and visual evaluation using images created by conventional technology will be described.

[0095] (First Verification) First, a doctor visually evaluated images obtained by PADRE processing of MRI images and removing vascular structures using a vascular mask image, and the results of comparing the results between a positive group and a negative group in PiB-PET (Positron Emission Tomography) will be described with reference to FIGS. 13 and 14.

[0096] FIG. 13 shows a PADRE image (MRI-PADRE image) from which the vascular structure and extra-brain parenchymal components of a PiB-negative group (Normal Control) in a comparative example have been removed, and an example image used for comparison with PiB-PET. FIG. 14 shows a PADRE image (MRI-PADRE image) from which the vascular structure and extra-brain parenchymal components of a PiB-positive group (Alzheimer's disease (AD)) in a comparative example have been removed, and an example image used for comparison with PiB-PET. Images g401 and g411 are MRI-PADRE images from which the vascular image and extra-brain parenchymal components have been removed using the method described above. Images g402 and g412 are PiB-PET images.

[0097] In a comparative example, 11 MRI-PADRE images of PiB-PET-negative and -positive cases were visually evaluated by a physician. As a result, when vascular areas and extracerebral components were removed from the PADRE images, it was easier to evaluate than before the vascular areas were removed. However, there were cases where visual evaluation of whether the PADRE images were negative or positive was difficult using only the images from which the vascular areas had been removed.

[0098] (Second Verification) Next, examples of the results of ROI (Region of Interest) analysis of cortical signals using an AAL (Automated Anatomical Labeling (see Reference 1)) atlas will be described with reference to Figures 15 to 18. In this verification, PADRE signals in each brain region were compared between the PiB-PET-positive and -negative groups. Figure 15 shows an example of an AAL atlas image set on standard brain coordinates, and an example of an image in which an ROI based on the AAL atlas is superimposed on the PADRE image. Image g451 is an example of an image based on the AAL atlas. Image g452 is an example of an image in which an AAL atlas image set on a standard brain was converted to individual brain coordinates and then processed with a vascular mask and a gray matter mask. Figure 16 shows that signals are distributed in the cortical region.

[0099] Reference 1: N. Tzourio-Mazoyer, B. Landeau, al, ``Automated Anatomical Labeling of Activations in SPM Using a Macroscopic Anatomical Parcellation of the MNI MRI Single-Subject Brain'', NeuroImage, Volume 15, Issue 1, January 2002, Pages 273-289

[0100] Figures 16 to 18 show the average values ​​of PADRE signals normalized to the cerebellar average value for the PiB-positive and -negative groups in the cortex, where ROI analysis was performed using the AAL atlas. Note that in each of graphs g511 to g522 in Figures 16 to 18, item g501 on the left represents the results for the negative group, and item g502 on the right represents the results for the positive group. In each of graphs g511 to g522 in Figures 16 to 18, the vertical axis represents the relative signal value when the average value for the cerebellar gray matter is set to 1.

[0101] Graph g511 shows the results of the inferior frontal lobe orbital region (right) (Frontal Inf Orb R). Graph g512 shows the results of the Rolandic area operculum (left) (Rolandic Oper L). Graph g513 shows the results of the Rolandic area operculum (right) (Rolandic Oper L). Graph g514 shows the results of the cuneus (left) (Cunrus L).

[0102] Graph g515 shows the results for the right superior occipital gyrus (Occipital Sup R). Graph g516 shows the results for the left middle occipital gyrus (Occipital Mid L). Graph g517 shows the results for the right middle occipital gyrus (Occipital Mid R). Graph g518 shows the results for the superior parietal lobule (left) (Parietal Sup L).

[0103] Graph g519 shows the results for the superior parietal lobule (right) (Parietal Sup R). Graph g520 shows the results for the supramarginal gyrus (right) (SupraMarginal R). Graph g521 shows the results for the right angular gyrus (Angular R). Graph g522 shows the results for the paracentral lobule (left) (Paracentral Lob L).

[0104] In each of graphs g511 to g522 in Figures 16 to 18, "*", "**", and "***" indicate evaluations based on P values, and the greater the number of asterisks, the more statistically significant the difference between the positive and negative groups is.

[0105] Figure 19 shows three-dimensional visualization of regions on the brain surface where there was a statistically significant difference between the mean values ​​of the PiB-positive and -negative groups in the cortical PADRE signal obtained by ROI analysis using the AAL atlas. Note that in Figure 19, the darker the region, the clearer the difference between the groups. Image g561 is an external image of the left hemisphere, and image g562 is an internal cross-sectional image of the left hemisphere. Image g571 is an external image of the right hemisphere, and image g562 is an internal cross-sectional image of the right hemisphere. In images g561, g562, g571, and g572, the darkest region has a p<0.001 value, for example, for region g551, and a p<0.01 value, for example, for region g552.

[0106] 16 to 18, the PADRE signal tends to be lower in the positive group than in the negative group to some extent. Also, as shown in Fig. 19, the PiB positive group has a low signal near the parietal-occipital lobe.

[0107] Figure 20 shows color-coded histograms of PiB-positive and -negative PADRE signal voxel values ​​in the local brain cortex analyzed using the AAL atlas. Graph g591 shows the distribution of the negative and positive groups in the right superior occipital gyrus, graph g592 shows the distribution of the negative and positive groups in the right medial occipital gyrus, and graph g593 shows the distribution of the negative and positive groups in the superior parietal lobule (left). In graphs g591 to g593, distribution g581 shows the distribution of the positive group, and distribution g582 shows the distribution of the negative group. In graphs g591 to g593, the horizontal axis represents the normalized PADRE signal value, and the vertical axis represents the number of pixels. In graphs g591 to g593, each distribution line represents data for one subject. As shown in Figure 20, the positive group has a lower signal value distribution than the negative group.

[0108] 20 suggests that either the punctate component or the diffuse component (background component) in the cortex may reflect amyloid β. For this reason, in this embodiment, the punctate component and the diffuse component in the cortex are separated.

[0109] (Separation of Intracortical Punctate Components and Diffuse Components) Next, a procedure for separating local low signal intensity components and diffuse components will be described. Fig. 21 is a flowchart of the process for separating local low signal intensity components and diffuse components according to this embodiment.

[0110] (Step S301) The separating unit 18 acquires the AP-PADRE image (D) that has been subjected to the masking process by the masking processing unit 17.

[0111] (Step S302-p) (p is an integer from 1 to n) The separation unit 18 performs closing processing using the following equation (5) through morphological calculation processing using a spherical kernel (Hp) with a radius of p pixels. Note that the separation unit 18 may perform the processing of steps S302-1 to S302-n in parallel or in a time-division manner.

[0112]

[0113] (Step S303-1) The separating unit 18 separates the local low signal of size 1 by performing morphological calculation processing to calculate the local low signal of size 1 using the following equation (6).

[0114]

[0115] (Step S303-n) The separating unit 18 separates the local low signal of size n by performing morphological calculation processing to calculate the local low signal of size n using the following equation (7).

[0116]

[0117] (Step S304) The separating unit 18 outputs the n images S1, . . . , Sn calculated in steps S302-1 to S302-n as a diffuse component image.

[0118] (Third Verification) Next, the results of the verification in which the PADRE signals of the diffuse components in each ROI region were compared between the PiB-PET positive group and the PiB-PET negative group will be described with reference to FIGS. 22 to 27.

[0119] Figures 22 to 25 are diagrams showing the average values ​​of normalized PADRE signal values ​​within the ROI of the diffuse component image for the PiB-positive group and the PiB-negative group. In each of graphs g611 to g624 in Figures 22 to 25, the left item g601 represents the results for the negative group, and the right item g602 represents the results for the positive group. In each of graphs g611 to g624 in Figures 22 to 25, the vertical axis represents the normalized signal value of the diffuse component PADRE image. Each of graphs g611 to g624 in Figures 22 to 25 represents the ROI analysis results of an image containing only the diffuse component, with the patchy component (low signal component) of each area removed.

[0120] Graph g611 shows the results of the superior frontal lobe orbital region (right) (Frontal Sup Orb R). Graph g612 shows the results of the inferior frontal lobe orbital region (right) (Frontal Inf Orb R). Graph g613 shows the results of the Rolandic area operculum (left) (Rolandic Oper L). Graph g614 shows the results of the Rolandic area operculum (right) (Rolandic Oper L).

[0121] Graph g615 shows the results for the cuneus (left) (Cunrus L). Graph g616 shows the results for the right superior occipital gyrus (Occipital Sup R). Graph g617 shows the results for the left middle occipital gyrus (Occipital Mid L). Graph g618 shows the results for the right middle occipital gyrus (Occipital Mid R).

[0122] Graph g619 shows the results for the superior parietal lobule (left) (Parietal Sup L). Graph g620 shows the results for the superior parietal lobule (right) (Parietal Sup L). Graph g621 shows the results for the supramarginal gyrus (right) (SupraMarginal R). Graph g622 shows the results for the left angular gyrus (Angular L).

[0123] Graph g623 shows the results for the paracentral lobule (left) (Paracentral Lob L). Graph g624 shows the results for the inferior temporal gyrus (right) (Temporal Inf L).

[0124] 22 to 25 show a clearer difference between the negative and positive groups compared to Figures 16 to 18. This indicates that the diffuse component tends to reflect amyloid β.

[0125] (Verification 4) Based on the results of Verification 3, the vascular image portion was masked from the PADRE image and patchy low-signal components were removed. Then, only the diffuse PADRE signal was extracted and projected onto the brain surface for display, and an interpretation experiment was conducted. Figure 26 shows example images in which the average values ​​of the diffuse components extracted from the PADRE images in the PiB-negative and -positive groups were projected onto the brain surface. In Figure 26, image g701 is an image in which the average value of the negative group is projected, and image g702 is an image in which the average value of the positive group is projected. Furthermore, region g711 is a region with a high signal (i.e., low amyloid beta). Region g712 is a region with a relatively low PADRE signal. From Figure 26, it can be visually seen that the parietal-occipital lobe signal is high in the negative group, and the parietal-occipital lobe signal is lower in the positive group than in the negative group.

[0126] (Evaluation of Each Case) Next, examples of applying the method of this embodiment to each case will be described with reference to FIGS. 27 to 30. FIG. 27 shows an MRI image of a PiB-negative patient with spinocerebellar degeneration (SCD), in which blood vessels and low signal intensity were removed using the method of this embodiment, and the extracted diffuse components were mapped onto the brain surface. Three physicians (a diagnostic radiologist, a neurosurgeon, and an internist) interpreted this image, and all three judged it to be negative for amyloid-β deposition.

[0127] 28 shows an MRI image of a healthy (NC) PiB-negative individual, in which blood vessels and low signals were removed using the method of this embodiment, and the extracted diffuse components were mapped onto the brain surface. Three doctors interpreted this image and determined that amyloid-β deposition was negative in all three cases.

[0128] 29 shows an MRI image of a PiB-positive individual with mild cognitive impairment (MCI due to AD) in which blood vessels and low signals were removed using the method of this embodiment, and the extracted diffuse components were mapped onto the brain surface. Three doctors interpreted this image, and three of them judged it to be positive for amyloid-β deposition.

[0129] 30 shows an MRI image of a PiB-positive individual with mild cognitive impairment (MCI due to AD), in which blood vessels and low signals were removed using the method of this embodiment, and the extracted diffuse components were mapped onto the brain surface. Three doctors interpreted this image, and three of them judged it to be positive.

[0130] FIG. 31 shows the interpretation results for 11 subjects, including those shown in FIGS. 27 to 30. In FIG. 31, table g801 in the upper row shows the interpretation judgment results. In table g801, "-" indicates a result judged as negative, and "+" indicates a result judged as positive. Note that "positive" refers to a judgment that "amyloid β is deposited." In FIG. 31, table g802 in the lower row shows the results of evaluating the judgment results of each doctor in table g801. As shown in FIG. 31, as a result of doctors interpreting images using the method of this embodiment, amyloid β deposition can be predicted with high accuracy.

[0131] Here, the AP-PADRE used in this embodiment will be further explained. A two-component model can be adopted, which assumes that iron present in the cerebral cortex is composed of two components: age-related iron that accumulates with aging, and amyloid iron that accumulates with amyloid accumulation. In this embodiment, AP-PADRE uses this two-component model to emphasize and visualize amyloid β-bound iron based on the difference in phase difference.

[0132] Regarding age-related iron, research has been conducted to evaluate and predict the total amount of iron associated with amyloid beta (amyloid iron) and iron accumulated with age (age-related iron) based on a quantitative value called QSM obtained from phase component information of MRI images, without distinguishing between them. Thus, the prior art did not distinguish between amyloid iron and age-related iron. In contrast, the present embodiment distinguishes between the two and detects information based on amyloid iron.

[0133] In the above formulas (1) and (3), examples using max have been described, but sup may be used instead of max. Also, in the above formulas (2) and (4), examples using mix have been described, but inf may be used instead of mix.

[0134] The verification results will be further explained. As an example of extracting components other than diffuse components from an AP-PADRE image, a point component image will be used. In this case, the point component image is extracted from an AP-PADRE image in which the vascular structure has been removed using a vascular mask image.

[0135] FIG. 32 shows a punctate component image and a diffuse component image created from an AP-PADRE image of a patient with mild cognitive impairment (MCI) due to Alzheimer's disease. It can be seen that the punctate component image g811 has strong contrast and can be easily recognized from the original AP-PADRE image. On the other hand, the diffuse component image g812 has a large signal difference, but its contrast is low, making it difficult to recognize in the original AP-PADRE image.

[0136] Figure 33 shows a color map projected onto the brain surface, where the average signal difference between the PiB-positive group and the PiB-negative group was calculated for the cerebellum-normalized signal of the AP-PADRE punctate component image in each ROI. Figure 34 shows a color map projected onto the brain surface, where the average signal difference between the PiB-positive group and the PiB-negative group was calculated for the cerebellum-normalized signal of the AP-PADRE diffuse component image in each ROI. Note that Figures 33 and 34 show grayscaled color images. Note that in Figures 33 and 34, images g831, g832, g841, and g842 are side views. Images g833, g834, g843, and g844 are cross-sectional views. Furthermore, the dark-colored area g845 indicates that the PiB(-) value of the negative group is higher than the PiB(+) value of the positive group, indicating a large signal difference.

[0137] As shown in Figure 33, there is little difference between the groups in the punctate component image. In contrast, as shown in Figure 34, in the diffuse component image, it can be seen that the PiB-positive group tends to have lower signal values ​​overall compared to the PiB-negative group. Note that in each embodiment, verification, etc., the terms focal hypointensity, punctate, and patchy are used interchangeably.

[0138] (Fifth Validation) In the fifth validation, quantitative analysis was performed using AAL atlas on AP-PADRE images of 191 patients who visited the Department of Aging and Geriatrics at Tohoku University Hospital with the chief complaint of forgetfulness. In this validation, patients were divided into three groups based on comprehensive diagnoses obtained from the symptoms and course of the cases, medical history and comorbidities, psychological tests, blood tests, brain MRI tests, and cerebral blood flow SPECT (single photon emission computed tomography) tests. The first group, the non-AD group, included cognitively normal individuals and patients with dementia due to causes other than Alzheimer's disease. The second group consisted of patients with MCI due to Alzheimer's disease. The third group consisted of patients with Alzheimer's disease (AD).

[0139] In this verification, AP-PADRE images were subjected to morphological analysis to create diffuse component images, and the signals were normalized by dividing them by the signal from the cerebellar cortex or hippocampus. Furthermore, in this verification, the average signal value of the image in 90 gray matter regions using ROIs from AAL atlas was calculated, and a one-tailed t-test was performed between the two groups to verify whether there was a difference in the mean values ​​of each group. Figures 35 to 37 show the results of the non-AD group versus the MCI group + Alzheimer's dementia group in the fifth verification. Figures 38 to 40 show the results of a one-tailed t-test of the non-AD group versus the Alzheimer's dementia group in the fifth verification. Figures 41 to 43 show the results of a one-tailed t-test of the results of similar ROI analysis in two groups, the PiB-negative group and the PiB-positive group, for 11 cases with PiB data in the fifth verification. 35 to 43, the notation is p-value, and p<0.05 is the statistical significance level. The AD group is a group with Alzheimer's dementia, the Non-AD group is a group with non-Alzheimer's dementia, and the MCI group is a group with mild cognitive impairment.

[0140] In addition, in Figures 35 to 43, the values ​​on the left are values ​​normalized by the cerebellum, and the values ​​on the right are values ​​normalized by the hippocampus. Table g851 in Figure 35, Table g861 in Figure 38, and Table g871 in Figure 41 show the results for the region Precentral L to Rolandic Oper L. Table g852 in Figure 35, Table g862 in Figure 38, and Table g872 in Figure 41 show the results for the region Rolandic Oper R to Hipcampus L. Table g853 in Figure 36, Table g863 in Figure 39, and Table g873 in Figure 42 show the results for the region Hipcampus R to Postcentral L. Table g854 in Figure 36, table g864 in Figure 39, and table g874 in Figure 42 are results for the region from Postcentral R to Thalamus L. Table g855 in Figure 37, table g865 in Figure 40, and table g875 in Figure 43 are results for the region from Thalamus R to Temporal R.

[0141] As shown in Figures 35 to 37, when comparing the non-AD group and the MCI + AD group, when normalized to the cerebellum, signal reduction was observed in 21 of 90 cerebral regions in the MCI + AD group. When normalized to the hippocampus, signal reduction was observed in 69 of 90 regions in the AD group. Next, when comparing the non-AD group and the AD group, as shown in Figures 38 to 40, signal reduction was observed in the AD group at 23 of 90 regions in the cerebellum-normalized images and at 79 of 90 regions in the hippocampus-normalized images.

[0142] As shown in Figures 41 to 43, in the normalized images of the cerebellum, signal reduction was observed in 14 of 90 areas in the PiB-positive group, while in the normalized images of the hippocampus, signal reduction was observed in 3 of 90 areas in the PiB-positive group.

[0143] In clinical case studies, the hippocampal signal in AP-PADRE diffuse component images is highly correlated with age and MMSE, and it is expected that signal increases primarily reflecting structural atrophy of the hippocampus are observed. In Alzheimer's disease, pathological atrophy occurs selectively in the area surrounding the hippocampus. Studies based on clinical diagnosis included many patients with advanced AD who clearly had pathological atrophy of the hippocampus. Therefore, when normalization was performed using the hippocampal signal, it is thought that the ability to distinguish Alzheimer's disease improved due to the synergistic effect of the amyloid-related signal decrease by AP-PADRE and the signal increase due to hippocampal atrophy specific to Alzheimer's disease. On the other hand, in examining the relationship with PiB in a small number of cases, including patients with MCI and early Alzheimer's dementia with little hippocampal atrophy, patients with non-Alzheimer's dementia with hippocampal atrophy, and healthy individuals, it was thought that the amyloid-specific signal decrease in AP-PADRE was offset by the inclusion of information about hippocampal atrophy through normalization of the hippocampal signal, resulting in a decrease in discrimination ability.

[0144] In the above-described embodiment, an AP-PADRE image created by the amyloid beta-bound iron-phase contrast enhanced imaging method, which emphasizes amyloid beta-bound iron, was described as an example of a phase contrast enhanced image created from an MRI image, but phase contrast enhanced images are not limited to this.

[0145] As described above, in this embodiment, morphological calculation processing is performed on the AP-PADRE image to mask the vascular structure and separate local low signal components (punctate components) from diffuse components, thereby creating an image that allows visual evaluation of the presence or absence of amyloid beta.

[0146] Currently, evaluation of intracerebral amyloid-β deposition involves cerebrospinal fluid testing and amyloid PET, which are invasive, expensive, and involve radiation exposure. In contrast, according to this embodiment, image processing is performed when capturing MRI images, eliminating the radiation exposure associated with PET and providing an inexpensive test. For example, this embodiment can be applied to brain checkups and other tests, potentially enabling early and efficient detection of pre-symptomatic Alzheimer's disease, which could be a major breakthrough in dementia treatment.

[0147] Furthermore, in the above example, image interpretation was performed on Alzheimer's dementia, but this is not limited to this. Images processed by the method of this embodiment could also be used to distinguish non-Alzheimer's dementia (CBDS: corticobasal degeneration) and the like. In clinical dementia settings, effective treatments vary depending on whether the condition is Alzheimer's disease or another pathology, and incorrect treatments may even cause side effects. If the cause of dementia in patients can be estimated using images from this embodiment, this would be extremely useful in determining dementia treatment strategies.

[0148] Although the above explanation uses the term "amyloid β," in each embodiment, each example, each verification example, etc., "amyloid β" is also used to mean "amyloid plaque."

[0149] The image processing device 1 of this embodiment may be provided in an MRI imaging device, etc. The image processing device 1 may also be, for example, a personal computer, etc.

[0150] A program for implementing all or part of the functions of the image processing device 1 of the present invention may be recorded on a computer-readable recording medium, and the program may be loaded into a computer system and executed to perform all or part of the processing of the image processing device 1. The term "computer system" as used herein includes hardware such as an OS and peripheral devices. The term "computer system" also includes a WWW system equipped with a homepage provision environment (or display environment). The term "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into computer systems. The term "computer-readable recording medium" also includes devices that retain a program for a certain period of time, such as volatile memory (RAM) within a computer system that acts as a server or client when the program is transmitted via a network such as the Internet or a communication line such as a telephone line.

[0151] The program may also be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by transmission waves in the transmission medium. Here, the "transmission medium" that transmits the program refers to a medium that has the function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may also be a program that realizes part of the above-mentioned functions. Furthermore, the program may be a so-called differential file (differential program) that can realize the above-mentioned functions in combination with a program already recorded in the computer system.

[0152] The above describes the form for carrying out the present invention using an embodiment, but the present invention is not limited to such an embodiment, and various modifications and substitutions can be made within the scope that does not deviate from the gist of the present invention.

[0153] 1...image processing device, 11...image acquisition unit, 12...vascular enhancement image generation unit, 13...Bias Field correction unit, 14...vascular mask creation unit, 15...AP-PADRE image generation unit, 16...Bias Field correction unit, 17...mask processing unit, 18...separation unit, 19...brain surface projection image output unit, 20...3D-T1 weighted image generation unit, 21...alignment unit, 22...anatomical standardization unit, 23...coordinate conversion unit, 24...statistical image analysis result output unit, 25...template storage unit, 26...coordinate conversion unit, 27...mask processing unit, 28...region of interest analysis result output unit

Claims

1. a step in which a vascular mask creating unit creates a vascular mask image by removing point-like low signals, removing linear low signals, extracting linear low signals, and removing point-like high signals, removing linear high signals, and extracting linear high signals from an image obtained from a magnetic resonance signal in which regions corresponding to blood vessels are highlighted in an MRI image or a vascular highlighted image acquired by an image acquiring unit; a step of a mask processing unit generating an image by removing a vascular structure from an image obtained from the magnetic resonance signal or a vascular enhancement image acquired by the image acquisition unit, using the vascular mask image; An image processing method comprising:

2. A step of separating a local low signal intensity and a diffuse component from the image from which the vascular structure has been removed by a separation unit; A step in which a brain surface projection image output unit projects the diffuse component onto a brain surface to generate a brain surface projection image; The image processing method of claim 1 further comprising:

3. the vascular mask creating unit creates the vascular mask image by morphological calculation processing; The separation unit separates the local low signal intensity image and the diffuse component image by the morphological operation processing. The image processing method according to claim 2 .

4. The vascular mask preparation unit: removing the point-like low signals by morphological calculation processing using a 3×3×3 linear kernel in 13 directions; removing the linear low signal intensity from the image from which the point low signal intensity has been removed by performing a morphological calculation process using a 3×3×3 planar kernel in 13 directions; extracting the linear hypointensity using the image from which the point hypointensity has been removed and the image from which the linear hypointensity has been removed; removing the point-like high signals by morphological calculation processing using a 3×3×3 linear kernel in 13 directions; removing the linear high intensity signals by performing morphological calculation processing on the image from which the point high intensity signals have been removed using a 3×3×3 planar kernel in 13 directions; extracting the linear hyperintensity signal using the image from which the point hyperintensity signal has been removed and the image from which the linear hyperintensity signal has been removed; The image processing method according to claim 1 , further comprising:

5. The vascular mask preparation unit: performing a process of removing the point-like low signals, a process of removing the linear low signals, a process of extracting the linear low signals, a process of removing the point-like high signals, a process of removing the linear high signals, and a process of extracting the linear high signals for different pixel widths; The image processing method according to claim 3.

6. The separation unit is A closing process is performed on the image from which the vascular structure has been removed by a morphological operation process using a spherical kernel of radius p (p is an integer from 1 to n) pixels to separate the local low signal of size n; A step of performing a closing process using a spherical kernel having a radius of n pixels on the image from which the vascular structure has been removed, and separating the diffuse image using p results; The image processing method according to claim 2 , comprising:

7. An image processing method described in any one of claims 1 to 6, comprising a step in which the mask processing unit generates an image in which vascular structures have been removed from a phase contrast weighted image created from the MRI image using the vascular mask image.

8. 8. The image processing method according to claim 7, wherein the phase-contrast weighted image is an AP-PADRE image created by amyloid-β-bound iron-phase-contrast weighted imaging of the MRI image.

9. The image processing method according to claim 1 , wherein the vascular mask creating unit extracts information about a region of interest by using the vascular mask image.

10. The vascular mask preparation unit:

9. The image processing method according to claim 1 , wherein information on the region of interest is extracted by performing mask processing based on the vascular mask image, a template of the region of interest, information obtained by aligning the acquired enhanced image with the image obtained from the magnetic resonance signal or the vascular enhanced image acquired by the image acquisition unit to align both images, and information obtained by converting standardized information into the individual brain coordinates based on information obtained by determining a transformation vector field between individual brain coordinates and standard brain coordinates by transforming the enhanced image by nonlinear transformation and aligning it with a template image of the enhanced image on standard brain coordinates.

11. On the computer, creating a vascular mask image by removing point-like low signals, removing linear low signals, extracting linear low signals, and removing point-like high signals, removing linear high signals, and extracting linear high signals from an image obtained from a magnetic resonance signal in which regions corresponding to blood vessels are emphasized in an MRI image or a vascular-emphasized image acquired by an image acquisition unit; generating an image in which a vascular structure is removed from an image obtained from the magnetic resonance signal or a vascular emphasis image obtained by the image acquisition unit, using the vascular mask image; program.

12. a vascular mask creating unit that creates a vascular mask image by removing point-like low signals, removing linear low signals, extracting linear low signals, and removing point-like high signals, removing linear high signals, and extracting linear high signals from an image obtained from a magnetic resonance signal that has been obtained by highlighting regions corresponding to blood vessels in an MRI image or a vascular-enhanced image acquired by an image acquiring unit; a mask processing unit that generates an image by removing a vascular structure from an image obtained from the magnetic resonance signal or a vascular emphasis image acquired by the image acquisition unit, using the vascular mask image; An image processing device comprising:

13. The image processing device according to claim 12 , wherein the vascular mask creating unit extracts information of a region of interest by using the vascular mask image.