JP7834281B2 - Image processing program, image processing method, and image processing apparatus - Google Patents

Description

Article 30, Paragraph 2 of the Patent Act applies Journal of the Neurological Sciences Vol. 423 (2021) doi:10.1016/j. jns. 2021. Article 117363

This invention relates to an image processing program, an image processing method, and an image processing apparatus.

Dopamine transporters (DATs) are highly expressed in the terminals of substantia nigra-striatal dopaminergic neurons located in the striatum of the brain. In Parkinson's disease (PD), Parkinsonian syndromes (PS) such as ^progressive supranuclear palsy (PSP), and Lewy body dementia (DLB), the expression level of DATs is known to be decreased. Therefore, in the diagnosis of these diseases, the brain distribution of dopamine transporters is evaluated using SPECT (Single-photon emission computed tomography) or PET (Positron emission tomography) scans, which use radiopharmaceuticals that image dopamine transporters, such as I-FP-CIT.

As a quantitative indicator for dopamine transporter imaging, the ratio of radioactivity due to specific binding in the striatum to radioactivity due to nonspecific binding is used. One such method is the Specific Binding Ratio (SBR), proposed by Bolt et al. This method involves setting a Volume of Interest (VOI) in the striatum, obtaining the VOI count value, and then calculating the SBR by taking the ratio of this VOI count value to the count value of nonspecific accumulation sites where radiopharmaceutical accumulation is less affected by disease, such as the cerebellum.

Striatal SBR and the asymmetry index (AI) of striatal SBR have been used as diagnostic biomarkers for diseases accompanied by a decrease in DAT, such as Parkinson's disease, Parkinsonian syndromes, and Lewy body dementia.

Japanese Patent Publication No. 2008-26144 Japanese Patent Publication No. 2013-250085 Japanese Patent Publication No. 2014-508340 Japanese Patent Publication No. 2020-043927

However, in differentiating between diseases involving a decrease in DAT, such as Parkinson's disease and progressive supranuclear palsy, accurate differentiation was difficult because striatal SBR and AI showed abnormalities in both diseases.

Therefore, in view of the above background, the present invention aims to provide an image processing technology that is useful for the appropriate differentiation of diseases accompanied by a decrease in DAT.

The present invention provides an image processing program for dopamine transporter images, which causes a computer to perform the following steps: an input step of inputting a dopamine transporter image of a subject; an SBR image generation step of generating a brain SBR image by determining the dopamine transporter accumulation ratio to the nonspecific accumulation, treating the accumulation of dopamine transporters in a predetermined brain region other than the striatum as nonspecific accumulation; an inversion step of generating an inverted SBR image by inverting the SBR image with respect to the symmetry axis of the brain in the SBR image; an AI image generation step of generating an AI image showing the asymmetry of the subject's SBR image based on the difference between the SBR image and the inverted SBR image; and an output step of outputting the AI image.

By generating AI images that demonstrate the asymmetry of SBR images across the entire brain, information can be provided to differentiate between diseases accompanied by DAT reduction.

The present invention provides an image processing method for a dopamine transporter image using an image processing device, comprising: an input step of inputting a dopamine transporter image of a subject; a standardization step of performing anatomical standardization of the dopamine transporter image; an SBR image generation step of generating a brain SBR image by determining the dopamine transporter accumulation ratio to the nonspecific accumulation, treating the accumulation of dopamine transporters in a predetermined brain region other than the striatum as nonspecific accumulation; an inversion step of generating an inverted SBR image by inverting the SBR image with respect to the symmetry axis of the brain in the SBR image; an AI image generation step of generating an AI image showing the asymmetry of the subject's SBR image based on the difference between the SBR image and the inverted SBR image; and an output step of outputting the AI image.

The present invention relates to an image processing apparatus for dopamine transporter images, comprising: an input unit for inputting a subject's dopamine transporter image; an SBR image generation unit that generates a brain SBR image by determining the dopamine transporter accumulation ratio to the nonspecific accumulation, treating the accumulation of dopamine transporters in a predetermined brain region other than the striatum as nonspecific accumulation; an AI image generation unit that generates an inverted SBR image by inverting the SBR image with respect to the brain's symmetry axis of the SBR image, and generates an AI image showing the asymmetry of the subject's SBR image based on the difference between the SBR image and the inverted SBR image; and an output unit that outputs the AI image.

This invention provides information for differentiating between diseases involving DAT reduction by generating AI images that show asymmetry in DAT accumulation.

This is a diagram showing the configuration of an image processing apparatus according to an embodiment. This is a flowchart showing the image processing method performed by the image processing device. This is a diagram showing the procedure for anatomical standardization. This diagram shows the procedure for AI image generation. This figure shows an example of group comparison in AI images of patients and healthy control groups displayed within Glassbrain. This figure shows the ROC curve for the data in group A.

In the present invention, "dopamine transporter image" refers to a brain image obtained by administering a radioactive dopamine transporter imaging agent to a subject and performing a nuclear medicine examination. Examples of radioactive dopamine transporter imaging agents include ¹²³ I-FP-CIT, [ ¹⁸ F]FP-CIT ([ ¹⁸ F]fluoropropylcarbomethoxyiodophenylnortropane), [ ^99m Tc]TRODAT-1 ([ ^99m Tc][2-[[2-[[[3-(4-chlorophenyl)-8-methyl-8-azabiscis[3.2.1]octo-2-yl]methyl](2-mercaptoethyl)amino]ethyl]-amino]ethanethiolate-(3-)-N2,N2',S2,S2']oxo-[1R-(exo-exo)]), [ ¹¹ Examples include [C]CFT ((1S,5R)-3β-(4-fluorophenyl)-8-[11C]methyl-8-azabicyclo[3.2.1]octane-2β-methylcarboxylate) and [ ^123I ]β-CIT ((2β-carbomethoxy-3β-(4-[ ^123I ]iodophenyl)tropane).

The following describes an image processing program, an image processing method, and an image processing apparatus 1 according to embodiments of the present invention. The image processing apparatus 1 will be described first in the following description.
Figure 1 shows the configuration of an image processing device 1 according to an embodiment. The image processing device 1 includes an input unit 10, a calculation processing unit 11, an output unit 12, and an MRI standard brain template 13.

The input unit 10 accepts input ^of I-FP-CIT images (hereinafter referred to as "FP-CIT images") and CT images acquired by a SPECT/CT device. Since the FP-CIT images and CT images are acquired simultaneously by the same device, the positions of each part of the brain image are aligned. The input unit 10 also accepts input of MRI images of the same subject.

The arithmetic processing unit 11 has the function of generating an AI image showing the asymmetry of the SBR image based on the input FP-CIT image, CT image, and MRI image, and also calculating an AI index value. The arithmetic processing unit 11 includes the functions of a standardization unit 20, an SBR image generation unit 21, an AI image generation unit 22, a region of interest setting unit 23, and an AI index value calculation unit 24. Details of each function will be described later with reference to Figure 2.

The output unit 12 has the function of outputting the AI image and AI index values generated by the calculation processing unit 11. The hardware of the output unit 12 includes, for example, a display for showing the calculation results and an interface for outputting the calculation results to an external device.

The MRI standard brain template 13 stores a standard brain template that represents the tissues at specific locations in the brain using coordinates.

Figure 2 is a flowchart showing the image processing method performed by the image processing device 1. Referring to Figure 2, the processing performed by the image processing device 1 will be explained, along with a description of each component of the image processing device 1.

(Step S10)
The input unit 10 of the image processing device 1 receives input of FP-CIT images and CT images acquired simultaneously by a SPECT/CT device, as well as MRI images of the same subject acquired separately.

(Steps S11-S13)
The standardization unit 20 of the image processing device 1 performs anatomical standardization of FP-CIT images. Because FP-CIT images are images in which DAT is specifically accumulated, the morphology of the brain is not clear. The standardization unit 20 performs anatomical standardization of FP-CIT images using CT images acquired simultaneously with the FP-CIT images and MRI images acquired separately.

Figure 3 shows the procedure for anatomical standardization. The operation shown in Figure 2 will be explained with reference to Figure 3. In Figure 3, the CT image and FP-CIT image are images simultaneously acquired by a SPECT/CT device. The MRI image is an MRI image of the same subject acquired separately. The standardization unit 20 aligns (co-registration) the CT image with the MRI image (S11).

As shown in Figure 3, the standardization unit 20 anatomically standardizes the MRI images and obtains the conversion parameters used for anatomical standardization (S12). The standardization unit 20 then uses the conversion parameters obtained from the anatomical standardization of the MRI images to perform anatomical standardization of the FP-CIT images (S13).

Since FP-CIT images and CT images are acquired simultaneously and the brain images are aligned, FP-CIT images can be anatomically standardized by using conversion parameters from MRI images aligned with the CT images.

This example demonstrates anatomical standardization using MRI images, but the use of MRI images is not mandatory. Alternatively, anatomical standardization of CT images may be performed to obtain conversion parameters, and these conversion parameters may then be used to perform anatomical standardization of FP-CIT images. In this embodiment, FP-CIT images and CT images acquired with a SPECT/CT system are used; however, if only FP-CIT images are acquired, anatomical standardization may be performed using a different method.

(Step S14)
The SBR image generation unit 21 of the image processing device 1 generates an SBR image based on a standardized FP-CIT image. The SBR image is an image that shows the SBR of DAT in the whole brain on a pixel-by-pixel basis. DAT accumulates not only specifically in the striatum but also non-specifically in other areas, but the SBR image is an image generated by calculating the accumulation ratio of DAT to non-specific accumulation.

The SBR image generation unit 21 determines the accumulation of DAT in a predetermined brain region other than the striatum as nonspecific accumulation. In this embodiment, the cerebellar region is used as the predetermined region. The SBR image generation unit 21 generates an SBR image by performing calculations according to the following formula.
SBR image = (original image - mean cerebellar count) / (mean cerebellar count)

(Steps S15-S16)
The AI image generation unit 22 of the image processing device 1 generates an AI image that shows the asymmetry of the SBR image based on the SBR image.

Figure 4 shows the procedure for AI image generation. The operation shown in Figure 2 will be explained with reference to Figure 4. The AI image generation unit 22 inverts the SBR image using the symmetry axis of the brain in the SBR image as a reference, and generates an inverted SBR image (S15). In Figure 4, the SBR image is the SBR image generated by the SBR image generation unit 21. To the right of the SBR image is the inverted SBR image, which is the SBR image inverted.

The AI image generation unit 22 generates an AI image showing the asymmetry of the subject's SBR image based on the difference between the SBR image and the inverted SBR image, according to the following formula (S16).
AI image = |SBR image - inverted SBR image| / {(SBR image + inverted SBR image) / 2}

(Step S17)
In this embodiment, the region of interest setting unit 23 of the image processing device 1 sets the region of interest to the subthalamic nucleus region in the AI image. Since the AI image is a standard brain that has undergone anatomical standardization, as an example, the coordinate data of the subthalamic nucleus region can be read from the MRI standard brain template 13, and the region of interest can be set to the subthalamic nucleus region using this coordinate data. Alternatively, the user may be allowed to set the region of interest. In this case, the region of interest setting unit 23 displays a screen for specifying the region of interest and accepts the setting from the user. At this time, the subthalamic nucleus region read from the MRI standard brain template 13 may be displayed by default. In this embodiment, the region of interest is set to the subthalamic nucleus region, but the region of interest may be set to a region other than the subthalamic nucleus region.

(Step S18)
The AI index value calculation unit 24 of the image processing device 1 calculates the AI index value of the subthalamic nucleus region, which is set as the region of interest. The AI index value is an index that evaluates the asymmetry of DAT accumulation in the subthalamic nucleus region. The AI index value calculation unit 24 obtains the AI index value by taking the average of the AI of the region of interest within the subthalamic nucleus region.

(Step S19)
The output unit 12 of the image processing device 1 outputs AI image and AI index value data to the display. This allows the user to receive information that is useful for differentiating between diseases that involve a decrease in DAT based on the AI image and AI index value data.

Furthermore, when outputting the AI index value, the cutoff value for ROC analysis, which is an evaluation criterion for differentiating between progressive supranuclear palsy and Parkinson's disease, may also be output. In this case, the cutoff value obtained by performing ROC analysis on the AI index values of patients with progressive supranuclear palsy and Parkinson's disease is stored in the memory unit beforehand. The output unit 12 reads the cutoff value from the memory unit and outputs the cutoff value along with the AI index value. By outputting the cutoff value, which serves as a guideline for differentiation, along with the subject's AI index value in this way, it becomes useful in differentiating between progressive supranuclear palsy and Parkinson's disease.

The image processing apparatus 1 and image processing method of this embodiment have been described above. An example of the hardware of the image processing apparatus 1 described above is a computer equipped with a CPU, RAM, ROM, hard disk, display, keyboard, mouse, communication interface, etc. The image processing apparatus 1 is realized by storing a program containing modules that realize each function of the arithmetic processing unit 11 in RAM or ROM, and executing this program by the CPU. The program of this embodiment is a program that causes the computer to execute each step of the image processing method described above.

Next, an example of image processing using the image processing apparatus of this embodiment will be described.
(Example 1)
[subject]
The study included patients who met the following criteria (1) and (2).
(1) Between April 2018 and December 2019, DAT scintigraphy using ¹²³ I-FP-CIT was performed at Hyogo Prefectural Rehabilitation Hospital Nishiharima for patients with PD who met the diagnostic criteria for probable PD of the International Association for Movement Disorders, and patients with PSP who met the diagnostic criteria for probable PSP with Richardson syndrome.
(2) Patients for whom the following information is available from their medical records: year of birth, sex, results of clinical diagnosis, results of 3D brain MRI scans, AI images created from DAT scintigraphy, results of the Mini-Mental State Examination (MMSE), and results of the Hoehn-Yahr severity classification.

The subjects selected based on the above criteria included 20 healthy individuals, 64 patients with Lewy body dementia, 111 patients with Parkinson's disease, and 18 patients with progressive supranuclear palsy due to Richardson's syndrome.

[SPECT/CT]
Subjects were intravenously injected with 167 MBq of ¹²³ I-FP-CIT. 180 minutes after the injection of ¹²³ I-FP-CIT, FP-CIT and CT images were simultaneously acquired. A dual-head Optima NM/CT 640 gamma camera (GE, Boston, MA) equipped with an extended low-energy general-purpose collimator was used for this imaging. Subsequently, scattering correction and attenuation correction using CT images were performed on the acquired image data. The pixel size was 3.05 mm × 3.05 mm, and the matrix was 128 × 128.

[Image Processing]
Anatomical alignment, anatomical standardization, and statistical processing of FP-CIT images were performed using Statistical Parametric Mapping for Windows, version 12 (SPM 12; Wellcome Department of Cognitive Neurology, London, United Kingdom).

Calculations and image matrix manipulation were performed using MATLAB® R2015a (MathWorks Inc., Natick, MA, USA). ROI-based SBR and AI for each patient were determined using the Tossici-Bolt method (SBRBolt).

[Generation of SBR and AI images]
The image processing device aligned the CT image with the corresponding MRI image. Next, the image processing device transformed the MRI image into an anatomically standardized brain using transformation parameters created through the DARTEL (diffeomorphic anatomical registration through exponentiated Lie algebra) transformation process. The image processing device then performed anatomical standardization of the FP-CIT image using the transformation parameters obtained from the anatomical standardization of the MRI image.

Simultaneously, the image processing device created an SBR image in the standard stereotactic space using the following formula. The average count of the cerebellum was used as the nonspecific accumulation site of DAT.
SBR image = (original image - mean cerebellar count) / (mean cerebellar count)

Next, the image processing device inverted the normalized SBR image in the X-axis direction to create an inverted SBR image, and then created an AI image according to the following formula.
AI image = |SBR image - inverted SBR image| / {(SBR image + inverted SBR image) / 2}

For pixel-level comparison, AI images were smoothed using an isotropic 8mm Gaussian kernel to improve the signal-to-noise ratio and correct for interpersonal anatomical differences. SBR images were not smoothed because the original images achieved an acceptable signal-to-noise ratio for intergroup comparison. One-way ANOVA was performed with age, sex, and disease duration as disturbance variables for intergroup comparison. Cortical regions were excluded from the analysis by masking because DAT was sparsely present and could not be accurately measured in FP-CIT images.

Brain regions were identified using the Neuromorphometrics atlas implemented in SPM12. The subthalamic nucleus region was identified using the approach described in the Montreal Neurological Institute's review of subthalamic nucleus coordinates (G. de Hollander, M.C. Keuken, B.U. Forstmann, "The subcortical cocktail problem; mixed signals from the subthalamic nucleus and substantia nigra," PLoS One 10 (3) (2015), e0120572, https://doi.org/10.1371/journal.pone.0120572).

[result]
Figure 5 shows an example of intergroup comparison in AI images of patients and healthy control groups displayed in Glassbrain. Figure 5(a) shows a comparison between Parkinson's disease patients and healthy control groups. In Figure 5(a), pixels with higher values in Parkinson's disease patients are displayed compared to healthy control groups. In Parkinson's disease, an increase in rAI was observed from both anterior insulaes to the putamen and basal forebrain.

Figure 5(b) shows a comparison between patients with progressive supranuclear palsy (SPP) and healthy controls. Figure 5(b) displays pixels where the SPP patients had higher values compared to healthy controls. In SPP patients, an increase in rAI was observed in the subthalamic nucleus and the putamen extending into the midbrain.

Figure 5(c) shows a comparison between patients with Parkinson's disease and patients with supranuclear palsy. Figure 5(b) compares patients with progressive supranuclear palsy and patients with Parkinson's disease, showing pixels where the values were higher in patients with progressive supranuclear palsy. Compared to patients with Parkinson's disease, patients with progressive supranuclear palsy showed an increase in rAI in the subthalamic nucleus.

(Example 2)
[subject]
This is the same as Example 1.
[SPECT/CT]
This is the same as Example 1.
[Generation of SBR and AI images]
This is the same as Example 1.

[ROC analysis]
AI index values in the subthalamic nucleus region were determined from patients with progressive supranuclear palsy (PD) and Parkinson's disease (PSP), and ROC analysis was performed. ROC curves for differentiating between the two diseases were created using the AI index values from the PD and PSP groups, and the cutoff value, sensitivity, and specificity of the AI index value for differentiation were calculated. EZR (Easy R) (Bone Marrow Transplantation 2013: 48, 452-458) was used for the ROC analysis. Alternatively, Statistical Package for the Social Sciences (SPSS, version 23, IBM, Armonk, NY, USA) or JMP may be used for the ROC analysis.

From 111 PD cases and 18 PSP cases, two groups were randomly selected: Group A (56 PD cases and 9 PSP cases) and Group B (55 PD cases and 9 PSP cases). The cutoff value was determined based on the ROC analysis results for Group A.

Figure 6 shows the ROC curve for the data of Group A. At a cutoff value of 0.283, the sensitivity was 100%, the specificity was 78.6%, and the AUC was 0.887. The sensitivity and specificity for differentiating between PSP and PD in Group B were calculated using the cutoff value (0.283) set for Group A.

[result]
Table 1 shows the differential diagnostic accuracy using the cutoff value (0.283) set based on the ROC analysis results for Group A.
As shown in these results, a good differential diagnostic accuracy was achieved, with a sensitivity of 88.9% and a specificity of 72.7% for PSP positivity.

This invention is useful for analyzing dopamine transporter images.

1 Image processing device 10 Input unit 11 Calculation processing unit 12 Output unit 13 MRI standard brain template 20 Standardization unit 21 SBR image generation unit 22 AI image generation unit 23 Region of interest setting unit 24 AI index value calculation unit

Claims (11)

A program for image processing of dopamine transporter images, which is used on a computer.
An input step involves inputting the subject's dopamine transporter image,
The SBR image generation step involves determining the ratio of dopamine transporter accumulation to nonspecific accumulation in a predetermined brain region other than the striatum, and generating an SBR image of the brain.
The inversion step involves inverting the SBR image with respect to the axis of symmetry of the brain in the aforementioned SBR image to generate an inverted SBR image,
An AI image generation step that generates an AI image showing the asymmetry of the subject's SBR image based on the difference between the SBR image and the inverted SBR image,
The output step of outputting the aforementioned AI image,
An image processing program that executes [this]. In the AI image generation step,
|SBR image - Inverted SBR image| / {(SBR image + Inverted SBR image) / 2}
The image processing program according to claim 1, which generates the AI image by performing the calculation. The steps include setting a region of interest in the subthalamic nucleus region in the aforementioned AI image,
The steps include calculating the AI index value for the subthalamic nucleus region,
An image processing program according to claim 1 or 2 that causes to perform the following. The process involves reading out cutoff values from a memory unit, which are obtained by performing ROC analysis on AI index values for patients with progressive supranuclear palsy and Parkinson's disease, and
The steps include outputting the subject's AI index value and the cutoff value read from the memory unit,
The image processing program according to claim 3, which causes the program to execute. In the input step described above, in addition to the dopamine transporter image, the system accepts input of a CT image acquired together with the dopamine transporter image by SPECT/CT, and a separately acquired MRI image.
The steps include aligning the CT image and the MRI image,
The process involves anatomically standardizing MRI images and storing the conversion parameters used during anatomical standardization.
The steps include: performing anatomical standardization of dopamine transporter images using the aforementioned conversion parameters;
An image processing program according to any one of claims 1 to 4, which causes the following to be executed. A method for performing image processing on a dopamine transporter image using an image processing device,
An input step involves inputting the subject's dopamine transporter image,
A standardization step to perform anatomical standardization of the dopamine transporter images,
The SBR image generation step involves determining the ratio of dopamine transporter accumulation to nonspecific accumulation in a predetermined brain region other than the striatum, and generating an SBR image of the brain.
The inversion step involves inverting the SBR image with respect to the axis of symmetry of the brain in the aforementioned SBR image to generate an inverted SBR image,
An AI image generation step that generates an AI image showing the asymmetry of the subject's SBR image based on the difference between the SBR image and the inverted SBR image,
The output step of outputting the aforementioned AI image,
An image processing method comprising: In the AI image generation step,
|SBR image - Inverted SBR image| / {(SBR image + Inverted SBR image) / 2}
The image processing method according to claim 6, which generates the AI image by performing the calculation. The steps include setting a region of interest in the subthalamic nucleus region in the aforementioned AI image,
The steps include calculating the AI index value for the subthalamic nucleus region,
The image processing method according to claim 6 or 7, comprising: The process involves reading out cutoff values from a memory unit, which are obtained by performing ROC analysis on AI index values for patients with progressive supranuclear palsy and Parkinson's disease, and
The steps include outputting the subject's AI index value and the cutoff value read from the memory unit,
The image processing method according to claim 8, comprising: In the input step described above, in addition to the dopamine transporter image, the system accepts input of a CT image acquired together with the dopamine transporter image by SPECT/CT, and a separately acquired MRI image.
The standardization step described above is:
The steps include aligning the CT image and the MRI image,
The process involves anatomically standardizing MRI images and storing the conversion parameters used during anatomical standardization.
The steps include: performing anatomical standardization of dopamine transporter images using the aforementioned conversion parameters;
An image processing method according to any one of claims 6 to 9, comprising: A device for image processing of dopamine transporter images,
An input unit for inputting the subject's dopamine transporter image,
An SBR image generation unit generates an SBR image of the brain by determining the ratio of dopamine transporter accumulation to nonspecific accumulation, and by defining the accumulation of dopamine transporters in a predetermined brain region other than the striatum as nonspecific accumulation.
An AI image generation unit generates an AI image that shows the asymmetry of the subject's SBR image based on the difference between the SBR image and the inverted SBR image, by inverting the SBR image with respect to the symmetry axis of the brain in the SBR image,
The output unit outputs the aforementioned AI image,
An image processing device equipped with the following features.