IMAGE PROCESSING APPARATUS, IMAGE PROCESSING METHOD, AND PROGRAM

Abstract

An image processing device includes: an image data acquisition unit for acquiring SPECT image data of a brain; a brain-region ROI definition unit for defining a brain-region ROI in the SPECT image; a striatum ROI definition unit for defining a striatum ROI in the SPECT image; and a threshold determination unit for, based on counts in the SPECT image's background which is the brain-region ROI except the striatum ROI, determining a threshold for distinguishing ventricles and sulci in the SPECT image; a region distinction unit for distinguishing between a region whose number of counts is smaller than or equal to the threshold and a region whose number of counts is larger than the threshold.

Claims

1. An image processing device comprising: an image data acquisition unit for acquiring nuclear medicine image data of a brain; a brain-region ROI definition unit for defining a brain-region ROI in the nuclear medicine image; a striatum ROI definition unit for defining a striatum ROI containing the striatum in the nuclear medicine image; and a threshold determination unit for determining a threshold for distinguishing ventricles and sulci in the nuclear medicine image based on counts in the nuclear medicine image's background which is the brain-region ROI except the striatum ROI, wherein a region whose number of counts is smaller than or equal to the threshold and a region whose number of counts is larger than the threshold are distinguished in the nuclear medicine image.

2. The image processing device according to claim 1, comprising a histogram generation unit for determining the number of counts for each unit region in the background and generating a histogram whose horizontal axis represents the class of counts and whose vertical axis represents the frequency of unit region for each class, wherein the threshold determination unit fits a normal distribution to the histogram and determines the threshold based on at least one of the median and standard deviation of the normal distribution.

3. The image processing device according to claim 2, wherein the threshold determination unit determines the threshold by (the median of the normal distribution)−(the standard deviation of the normal distribution)×(a predetermined coefficient).

4. The image processing device according to claim 3, wherein the predetermined coefficient is 1.

5. The image processing device according to claim 1, comprising a histogram generation unit for determining the number of counts for each unit region in the background and generating a histogram whose horizontal axis represents the class of counts and whose vertical axis represents the frequency of unit region for each class, wherein the threshold determination unit determines as the threshold the number of counts for a class which corresponds to a frequency that is smaller than a maximum frequency multiplied by a predetermined coefficient, the class being smaller than the class that gives the maximum frequency.

6. The image processing device according to claim 1, comprising a calculation unit for calculating an index of specific binding of a radiopharmaceutical using data of a region whose number of counts is larger than the threshold.

7. An image processing method for nuclear medicine image data of a brain, comprising the steps of: an image processing device acquiring nuclear medicine image data of a brain; the image processing device defining a brain-region ROI in the nuclear medicine image; the image processing device defining a striatum ROI containing the striatum in the nuclear medicine image; the image processing device determining a threshold for distinguishing ventricles and sulci in the nuclear medicine image based on counts in the nuclear medicine image's background which is the brain-region ROI except the striatum ROI; and the image processing device distinguishing in the nuclear medicine image a region whose number of counts is smaller than or equal to the threshold and a region whose number of counts is larger than the threshold.

8. The image processing method according to claim 7, comprising the step of the image processing device determining the number of counts for each unit region in the background and generating a histogram whose horizontal axis represents the class of counts and whose vertical axis represents the frequency of unit region for each class, wherein the step of determining the threshold involves fitting a normal distribution to the histogram and determining the threshold based on at least one of the median and standard deviation of the normal distribution.

9. The image processing method according to claim 8, wherein the step of determining the threshold involves determining the threshold by (the median of the normal distribution)−(the standard deviation of the normal distribution)×(a predetermined coefficient).

10. The image processing method according to claim 9, wherein the predetermined coefficient is 1.

11. The image processing method according to claim 7, comprising the step of the image processing device determining the number of counts for each unit region in the background and generating a histogram whose horizontal axis represents the class of counts and whose vertical axis represents the frequency of unit region for each class, wherein the step of determining the threshold involves determining as the threshold the number of counts for a class which corresponds to a frequency that is smaller than a maximum frequency multiplied by a predetermined coefficient, the class being smaller than the class that gives the maximum frequency.

12. The image processing method according to claim 7, comprising the step of the image processing device calculating an index of specific binding of a radiopharmaceutical using data of a region whose number of counts is larger than the threshold.

13. A program for image processing of nuclear medicine image data of a brain, the program causing a computer to execute the steps of: acquiring nuclear medicine image data of a brain; defining a brain-region ROI in the nuclear medicine image; defining a striatum ROI containing the striatum in the nuclear medicine image; determining a threshold for distinguishing ventricles and sulci in the nuclear medicine image based on counts in the nuclear medicine image's background which is the brain-region ROI except the striatum ROI; and distinguishing in the nuclear medicine image a region whose number of counts is smaller than or equal to the threshold and a region whose number of counts is larger than the threshold.

14. The program according to claim 13, causing a computer to execute the step of determining the number of counts for each unit region in the background and generating a histogram whose horizontal axis represents the class of counts and whose vertical axis represents the frequency of unit region for each class, wherein the step of determining the threshold involves fitting a normal distribution to the histogram and determining the threshold based on at least one of the median and standard deviation of the normal distribution.

15. The program according to claim 14, wherein the step of determining the threshold involves determining the threshold by (the median of the normal distribution)−(the standard deviation of the normal distribution)×(a predetermined coefficient).

16. The program according to claim 15, wherein the predetermined coefficient is 1.

17. The program according to claim 13, causing a computer to execute the step of determining the number of counts for each unit region in the background and generating a histogram whose horizontal axis represents the class of counts and whose vertical axis represents the frequency of unit region for each class, wherein the step of determining the threshold involves determining as the threshold the number of counts for a class which corresponds to a frequency that is smaller than a maximum frequency multiplied by a predetermined coefficient, the class being smaller than the class that gives the maximum frequency.

18. The program according to claim 13, causing a computer to execute the step of calculating an index of specific binding of a radiopharmaceutical using data of a region whose number of counts is larger than the threshold.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0016] FIG. 1 shows the configuration of an image processing device of a first embodiment;

[0017] FIG. 2 shows an example of a striatum ROI containing the striatum;

[0018] FIG. 3A shows an example of a histogram of counts in a background;

[0019] FIG. 3B shows an example in which a normal distribution is fitted to the histogram;

[0020] FIG. 4 shows how to determine a threshold in the first embodiment;

[0021] FIG. 5 shows an example of an image processing program;

[0022] FIG. 6 shows an operation of the image processing device of the first embodiment;

[0023] FIG. 7A shows an example of a histogram of counts in a background;

[0024] FIG. 7B shows how to determine a threshold in a second embodiment;

[0025] FIG. 8 shows a histogram created from a background and its normal distribution;

[0026] FIG. 9A shows a result of a comparison between an SBR determined without eliminating the influence of ventricles and sulci and the true value of the SBR; and

[0027] FIG. 9B shows a result of a comparison between an SBR determined by the method of the invention and the true value of the SBR.

MODES OF EMBODYING THE INVENTION

[0028] Now, the image processing device of embodiments of the invention will be described with reference to the drawings. The examples described below are only exemplary of preferred modes and are not intended to limit the subject matter of the invention.

First Embodiment

[0029] FIG. 1 shows the configuration of an image processing device 1 of a first embodiment of the invention. The image processing device 1 has: an image data acquisition unit 10 for acquiring brain SPECT image data; a control unit 11 for processing the SPECT image data to distinguish the region of ventricles and sulci and calculating an SBR which is a quantitative index of specific binding in the striatum; and an output unit 12 for outputting the calculation result.

[0030] SPECT image data used in this example is obtained by SPECT examination using .sup.123I-ioflupane. Radiation emitted from .sup.123I-ioflupane is detected by using a dedicated gamma camera, and the image data is made from the distribution. Each pixel has the number of radiation events, i.e. counts.

[0031] In a preferred mode, the control unit 11 has a brain-region ROI definition unit 13, a striatum ROI definition unit 14, a histogram generation unit 15, a threshold determination unit 16, a region distinction unit 17, and an SBR calculation unit 18.

[0032] The brain-region ROI definition unit 13 locates and defines in a SPECT image a region where a brain region exists. Methods of defining the brain-region ROI include, for example, a method in which a user uses a mouse, a stylus pen, or the like to manually enclose the brain parenchyma to define the region, and a method in which the region is defined based on an image contour determined by a threshold method. When the region is defined based on an image contour, the brain-region ROI may be the image contour itself, or may be a region defined by pixels a few pixels (actually about 20 mm) inside the contour. There is also a method in which the brain-region ROI is defined by a circle, an ellipse, or other standard shape figures containing the brain parenchyma. In this case, it is desirable for the brain-region ROI not to extend off the brain parenchyma.

[0033] As shown in FIG. 2, the striatum ROI definition unit 14 defines as the striatum ROI a predetermined region containing the striatum in a brain SPECT image. The region of the brain-region ROI except the striatum region in SPECT image data is herein called the “background (B.G.).” The SBR calculation unit 18 is of course illustrative, and may be any that has the function to calculate an index of specific binding of a radiopharmaceutical. The term “SBR calculation unit” is only intended to show a representative calculation to facilitate understanding of the summary of the invention.

[0034] The histogram generation unit 15 generates a histogram of counts in a background. FIG. 3A shows an example of a histogram generated by the histogram generation unit 15. The horizontal axis of the histogram represents the class of counts, and the vertical axis of the histogram represents the number of pixels having counts for the relevant class. Though the frequency is determined in terms of pixels in the example, the frequency may be determined for each class in terms of certain regions including, for example, 2×2 pixels instead.

[0035] The threshold determination unit 16 has a function to determine a threshold based on a histogram. In a preferred mode, the threshold determination unit 16 first fits a normal distribution to a histogram. FIG. 3B shows an example in which a normal distribution is fitted to the histogram shown in FIG. 3A. The threshold determination unit 16 then subtracts the standard deviation of the normal distribution from the median of the normal distribution to determine the threshold as shown in FIG. 4. This threshold is intended to regard regions whose number of counts is smaller than or equal to the threshold as ventricles or sulci. The region distinction unit 17 identifies regions whose number of counts is smaller than or equal to the threshold as regions of ventricles or sulci, and regions whose number of counts is larger than the threshold as regions other than those.

[0036] The SBR calculation unit 18 eliminates data of the regions whose number of counts is smaller than or equal to the above-described threshold (i.e. regions of ventricles and sulci) distinguished by the region distinction unit 17, and calculates the SBR. The calculation of the SBR is the same as the conventional method except for the elimination of data of parts determined to be ventricles and sulci by using the above-described threshold. Specifically, the SBR is determined by the following equation (1).

SBR=[(ROI average counts)−(B.G. average counts)]/(B.G. average counts)  (1)

[0037] The image processing device 1 is composed of a computer comprising: a CPU; RAM; ROM; a display; a keyboard; a mouse; a communications interface; and the like. The computer calculates the SBR based on SPECT image data in a manner that a program for calculating the SBR is stored in ROM in advance and the CPU reads the program from ROM and runs it.

[0038] FIG. 5 shows an example of an image processing program 20 stored in ROM. The image processing program 20 has a brain-region ROI definition module 21, a striatum ROI definition module 22, a histogram generation module 23, a threshold determination module 24, a region distinction module 25, and an SBR calculation module 26. Each of the modules 21 through 26 causes the computer to execute the program 20 to realize each function of the control unit 11 described above. While the image processing program 20 stored in ROM has been described here, the image processing described in the present embodiment may be realized by recording the image processing program on a computer-readable recording medium and causing a computer to read the recording medium and run the program. The image processing program may also be downloaded to a computer via a communications line.

[0039] FIG. 6 is a flowchart showing an operation of the image processing device 1. The image processing device 1 first acquires SPECT image data (S10). The image processing device 1 may acquire the SPECT image data directly from a SPECT device, or from a database in which the SPECT image data is stored.

[0040] The image processing device 1 then defines a brain-region ROI (S11) and a striatum ROI in a region containing the striatum (S12) in the SPECT image, and generates a histogram of counts in a region of the brain-region ROI except the striatum ROI (background) (S13). The image processing device 1 subsequently fits a normal distribution to the histogram, and determines a threshold by subtracting the standard deviation from the median (S14).

[0041] The image processing device 1 then distinguishes regions in the SPECT image whose number of counts is smaller than or equal to the threshold, and generates a mask for excluding the regions (S15). That is, the regions whose number of counts is smaller than or equal to the threshold are not included in the subsequent calculation. The image processing device 1 uses the masked data to determine the average counts for the striatum ROI and the average counts for the background, and calculates the SBR using the equation (1) described above (S16). The image processing device 1 outputs the calculated SBR (S17).

[0042] There has been described the image processing device 1 and image processing method of the first embodiment of the invention. The image processing device 1 of the first embodiment uses counts in the background to determine the threshold, and can therefore appropriately determine the threshold for distinguishing ventricles and sulci using only SPECT image data. This threshold can be used to eliminate the influence of ventricles and sulci to determine the SBR.

[0043] While in the present embodiment the threshold is determined by (the median)−(the standard deviation), the threshold may be determined by subtracting the standard deviation multiplied by a predetermined coefficient from the median. It is preferable that this predetermined coefficient is determined so that the SBR comes closest to the true value determined by a predetermined method.

Second Embodiment

[0044] An image processing device of a second embodiment of the invention will now be described. The configuration of the image processing device of the second embodiment is basically the same as the image processing device 1 of the first embodiment, but is different therefrom in that it determines the threshold for distinguishing ventricles and sulci in a different manner.

[0045] FIGS. 7A and 7B illustrate how to determine the threshold in the second embodiment. As shown in FIG. 7A, the maximum frequency is determined in the histogram in the second embodiment. This maximum frequency is then multiplied by a predetermined coefficient, which is 0.05 in the present embodiment. The maximum frequency in the example shown in FIG. 7A is 2500, which is multiplied by 0.05, resulting in 125. The threshold determination unit 16 determines as the threshold the number of counts for a class whose frequency is smaller than or equal to 125. In this regard, a class whose frequency is smaller than or equal to 125 also exists among classes with a large number of counts, but the threshold is chosen from classes with a small number of counts. That is, a class is chosen that is smaller than the class that gives the maximum frequency. More specifically, classes are checked in descending order from the class with the maximum frequency, and the number of counts for the class that first falls below 125 is determined to be the threshold.

[0046] The image processing device of the second embodiment also has the advantage of being able to distinguish regions of ventricles and sulci using only SPECT image data and determining the SBR exclusive of the influence of ventricles and sulci. In addition, since the image processing device of the second embodiment does not have to fit a normal distribution, its process is simple.

EXAMPLES

[0047] Shown below is the fact that the method of the invention in which the SBR is calculated by using only SPECT image data can provide a calculation close to the true SBR.

[0048] (True Value of SBR)

[0049] A calculation of SBR determined by creating a brain parenchyma mask from an MR image shall be the true value to be compared with the method of the invention. The true value is determined according to the following procedure:

[0050] 1. A SPECT image is aligned to an MR image;

[0051] 2. SPM (statistical parametric mapping) is used to extract gray matter and white matter from the MR image;

[0052] 3. Gray matter and white matter are added up and converted to a binary representation with a threshold of 50%;

[0053] 4. The binary-converted data is set to be a mask; and

[0054] 5. Only masked SPECT image data is calculated to determine the SBR.

[0055] (SBR Without Mask)

[0056] In order to examine how ventricles and sulci affect the SBR, striatum ROI (and background) data defined in a SPECT image was used to directly determine the SBR without using the masked image described above. The determined SBR is compared with the true value. Each ROI was defined by using “DaTView” in “Nou Toukei Kaiseki Package” (Brain statistical analysis package) contained in AZE VirtualPlace HAYABUSA (made by AZE, Ltd., Medical device approval number: 22000BZX00379000).

[0057] (SBR Determined by the Invention)

[0058] The SBR is determined by the method described in the above first embodiment.

[0059] FIG. 8 shows a histogram of counts in a background determined by the method described in the first embodiment for one of the used cases. When this histogram is fitted with a normal distribution, the median is 51.65 and the standard deviation is 13.52. The standard deviation of the normal distribution is subtracted from the median of the normal distribution to determine the threshold of 38.03. The SBR is calculated excluding regions whose number of counts is smaller than or equal to this threshold. In this example, the above-described process was executed for each used case, and their own thresholds were determined.

[0060] (Comparison Result)

[0061] FIG. 9A is a graph showing a comparison between an SBR determined by using striatum ROI (and background) data defined in a SPECT image without using a mask image and the true value of the SBR. The vertical axis of the graph represents the SBR determined by using striatum ROI (and background) data defined in a SPECT image without using a mask image, and the horizontal axis represents the true value of the SBR. As shown in FIG. 9A, a strong correlation was seen between the SBR determined without using a mask and the true value, but the SBR is overvalued by approximately 20%.

[0062] FIG. 9B is a graph showing a comparison between an SBR determined by the method of the invention and the true value of the SBR. As shown in FIG. 9B, the slope of the obtained line is almost 1, which indicates that a result extremely close to the true value can be obtained by using the method of the invention.

INDUSTRIAL APPLICABILITY

[0063] The invention allows for distinguishing regions of ventricles and sulci using only nuclear medicine image data, and is useful for SPECT image processing.

DESCRIPTION OF THE SYMBOLS

[0064] 1: Image processing device [0065] 10: Image data acquisition unit [0066] 11: Control unit [0067] 12: Output unit [0068] 13: Brain-region ROI definition unit [0069] 14: Striatum ROI definition unit [0070] 15: Histogram generation unit [0071] 16: Threshold determination unit [0072] 17: Region distinction unit [0073] 18: SBR calculation unit [0074] 20: Image processing program [0075] 21: Brain-region ROI definition module [0076] 22: Striatum ROI definition module [0077] 23: Histogram generation module [0078] 24: Threshold determination module [0079] 25: Region distinction module [0080] 26: SBR calculation module