Image analysis device, image analysis method, and computer program product

Abstract

The present invention relates to an image analysis device, an image analysis method, and an image analysis program. These provide an image analysis device and method capable of automatically measuring tissue with high accuracy, even with images from different models and imaging conditions. The device and method perform image normalization and regional segmentation, including contrast, and then, based on the characteristics of the tissue of interest in the segmented region, determine the measurement section that best matches the measurement section determined to be appropriate for measuring a given indicator. The measurement value serving as the indicator is calculated within this measurement section.

Description

Technical Field

[0001] The present invention relates to an image analysis device that presents diagnostic indicators such as diseases based on images acquired by a medical imaging device, and particularly to a technology that presents important brain abnormalities as indicators in the determination of dementia. Background Art

[0002] Medical imaging devices, such as X-ray cameras, CT scanners, ultrasound cameras, and MRI machines, are widely used as effective tools for diagnosing various diseases. Doctors perform diagnoses by examining images of the subject obtained from these medical imaging devices. However, they also sometimes measure changes in the shape of the tissues reflected in the images and use these values ​​as diagnostic indicators. Therefore, technologies have been developed that perform measurements and output diagnostic indicators within medical imaging devices or image processing devices that receive image data from them. When performing measurements within a device, for example, an image of the subject tissue is displayed on a display device. A doctor, technician, or other such user (hereinafter referred to as a user) then inputs information such as the location to be measured into the displayed image via an input device. The device receives this information, calculates the measured values, and displays the results.

[0003] Furthermore, technologies have been proposed that automatically determine the region of interest (ROI) to be measured from multiple cross-sections and calculate measurement values, eliminating the need for user input and time (Patent Document 1, etc.). In the technology described in Patent Document 1, when automatically determining the ROI, the target tissue (here, the brain) is anatomically standardized to eliminate measurement variations caused by individual differences in the imaged subject. Data on anatomical regions assigned to the standard brain are then presented as ROI candidates.

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Publication No. 2019-74343

[0007] However, for diagnostic images, the resolution and contrast of the images vary significantly, for example, depending on the imaging device modality or, in the case of MRI devices, the imaging conditions (imaging parameters). Therefore, it is difficult to ensure the accuracy of determining the measurement section and measurement position for various images with different image qualities simply by relying on anatomical standardization.

[0008] In recent years, research on brain findings related to dementia and brain diseases with similar symptoms has advanced, and the process for diagnosing dementia based on various brain findings is becoming clearer. For example, normal-pressure hydrocephalus, including secondary normal-pressure hydrocephalus (sNPH) and incipient normal-pressure hydrocephalus (iNPH), is difficult to distinguish from dementia and has led to the development of diagnostic guidelines for these conditions using imaging. Guidelines for iNPH include the Evans index, the corpus callosum angle, and asymptomatic ventriculomegaly with DESH findings. The Evans index uses the ratio of the maximum width between the anterior horns of the lateral ventricles to the width of the skull at that location to be greater than 0.3 as a diagnostic indicator for iNPH. Furthermore, regarding the corpus callosum angle, a steep angle (less than 90 degrees) on a coronal MRI section that passes through the posterior commissure and is perpendicular to the anterior-posterior commissure is a diagnostic indicator for Alzheimer's dementia. DESH observations show minimization of the high fornix accompanied by uneven expansion of the subarachnoid space, making it a diagnostic indicator for Alzheimer's dementia due to its high sensitivity and specificity.

[0009] When doctors measure and calculate these diagnostic indices from images, they use images with appropriate modes and contrasts corresponding to the indices. For example, in the case of MR images, the Evans index uses T2-weighted images or FLAIR images, which depict the cerebrospinal fluid as the highest or lowest signal areas, while the corpus callosum angle is preferably measured using T1-weighted images. Therefore, automatically measuring the index from input diagnostic images requires preparing diagnostic images corresponding to the indices, which is a significant burden.

[0010] Furthermore, since these indices are calculated based on measurements of the brain's finer shape, their values ​​can vary significantly depending on slight differences in the measurement section (slice location), making the selection of an appropriate slice location crucial. However, selecting an appropriate measurement section (slice location) is difficult, and even when performed by a physician, it can be difficult to determine an appropriate section for measuring the corpus callosum angle using only a single coronal section (COR section). Similarly, determining an appropriate section for each of the other indices is difficult for individuals with varying ages, symptoms, and other factors. Summary of the Invention

[0011] An object of the present invention is to provide an image analysis device and method that can accurately measure tissue even when images are obtained using different modes, models, and imaging conditions.

[0012] In order to solve the above-mentioned problems, the present invention performs standardization and regional segmentation of an image including contrast, and determines a measurement section that is most consistent with the measurement section that is determined to be appropriate for measuring a given indicator based on the characteristics of the tissue of interest in the segmented area, and calculates the measurement value that becomes the indicator in this measurement section.

[0013] Specifically, the image analysis device of the present invention includes: an image receiving unit that inputs a diagnostic image and imaging conditions for the diagnostic image; an image normalizing unit that normalizes the diagnostic image; a region segmenting unit that extracts a region of interest from the diagnostic image normalized by the image normalizing unit; a slice position determining unit that determines a slice position in the normalized diagnostic image based on the characteristics of the tissue of interest obtained through region segmentation; a measuring unit that measures the tissue of interest at the slice position determined by the slice position determining unit; and an index calculating unit that calculates an index using the measurement value of the tissue of interest measured by the measuring unit. The region segmenting unit segments the region based on the imaging conditions of the diagnostic image so that the pixel values ​​of the segmented region conform to a predetermined standard pixel value (label).

[0014] In addition, the image analysis method of the present invention inputs a diagnostic image to calculate an index of a specific disease, and the image analysis method includes: a step of inputting a diagnostic image and the imaging conditions of the diagnostic image; an image standardization step of standardizing the diagnostic image; a regional segmentation step of extracting an area of ​​a tissue of interest from the standardized diagnostic image; a step of determining a slice position in the standardized diagnostic image based on the characteristics of the tissue of interest obtained by regional segmentation; a step of measuring a given tissue of interest at the determined slice position; and a step of calculating an index using the measured value of the measured tissue of interest, wherein in the regional segmentation step, regional segmentation is performed according to the imaging conditions so that the pixel value of the segmented area becomes a predetermined standard pixel value (label).

[0015] Furthermore, the image analysis program of the present invention is a program that causes a computer to execute the above-mentioned steps.

[0016] In this specification, the term "cross section" includes cross sections such as an axial cross section, a coronal cross section, and a sagittal cross section, as well as a cross section at a given slice position.

[0017] Effects of the Invention

[0018] According to the present invention, by using imaging conditions to standardize images, including image quality, it is possible to perform highly accurate measurements regardless of differences in imaging mode or imaging conditions. Furthermore, by determining the slice position based on the standardized characteristics of the tissue of interest, it is possible to determine the most appropriate slice position for calculating an index, thereby preventing erroneous measurement of the index due to slight differences in slice position. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a diagram showing the outline of the configuration of a medical imaging device and an image analysis device.

[0020] Figure 2 This is a functional block diagram showing the image processing unit of the image analysis device.

[0021] Figure 3 This is a diagram showing the processing flow of the image analysis device.

[0022] Figure 4 The figures illustrate indices measured from brain images. (A) shows an image of an axial section for measuring the Evans index, and (B) shows an image of a coronal section for measuring the corpus callosum angle.

[0023] Figure 5 This is a diagram showing the process of image normalization processing.

[0024] Figure 6 This is a diagram showing the process of determining the slice position.

[0025] Figure 7 This is a diagram illustrating in detail one process of determining a slice position.

[0026] Figure 8 This is a diagram illustrating the determination of the slice position.

[0027] Figure 9 These are diagrams explaining the calculation of indices according to a modified example, wherein (A) shows an axial cross section and (B) shows a coronal cross section.

[0028] Description of reference numerals:

[0029] 100: Image analysis device, 10: Image acquisition unit, 20: Image processing unit, 30: Statistical analysis unit, 40: Storage device, 50: UI unit, 210: Image information extraction unit, 220: Image normalization unit, 230: Region segmentation unit, 240: Slice position determination unit, 250: Tissue of interest extraction unit, 260: Line segment calculation unit, 270: Diagnostic index calculation unit, 300: Medical imaging device, 500: Medical image database DETAILED DESCRIPTION

[0030] Hereinafter, embodiments of the image analysis device of the present invention will be described with reference to the drawings.

[0031] <Implementation Method 1>

[0032] Image analysis device 100 Figure 1As shown, the medical device comprises: an image capture unit (image receiving unit) 10 for capturing images captured by a medical imaging device 300; an image processing unit 20 for performing various processes using the captured images; and a statistical analysis unit 30 for performing statistical processing using a large number of images. The device also comprises: a storage device 40 for storing data required for the processing, processing results, etc.; and a UI unit 50 including a display unit and an input unit for displaying images, measurement values, and a GUI as processing results.

[0033] The image analysis device 100 can be connected to a medical imaging device 300 and a database 500 of a large number of images. The medical imaging device 300 is not limited to a single device but can also be multiple devices installed in different locations, or multiple devices with different modes. Examples of devices with different modes include MRI devices, X-ray imaging devices, ultrasound imaging devices, CT devices, and PET devices.

[0034] The image input unit 10 is equivalent to an image receiving unit that inputs a diagnostic image and the imaging conditions of the diagnostic image. It inputs the image data to be processed and its DICOM information (including various information including the subject information and imaging conditions attached to the image) via wired, wireless, removable media or network, and hands it over to the image processing unit 20 or the statistical analysis unit 30.

[0035] The image processing unit 20 mainly processes the image to be measured (measurement image) by standardizing the image, segmenting the tissue, determining the slice position, performing given measurements, and calculating indices. Figure 2 As shown, the apparatus comprises: an image information extraction unit 210 that extracts image information from data taken in by the image input unit 10; an image normalization unit 220 that normalizes the diagnostic image; a region segmentation unit 230 that extracts a region of tissue of interest from the normalized diagnostic image; a slice position determination unit 240 that determines a slice position in the normalized diagnostic image based on characteristics of the tissue of interest obtained by region segmentation; a measurement unit (here, a tissue of interest extraction unit 250 and a line segment calculation unit 260) that measures the tissue of interest at the slice position determined by the slice position determination unit; and a diagnostic index calculation unit (hereinafter referred to as the index calculation unit) 270 that calculates an index using the measurement value of the tissue of interest measured by the measurement unit.

[0036] The statistical analysis unit 30 performs statistical processing of indices and creates standard images for a large number of images including images processed by the image processing unit 20. In this embodiment, the image processing unit 20 selects or creates standard images for reference when selecting slice positions for measurement images.

[0037] The main components of the image analysis device 100 described above can be implemented in a computer (workstation) equipped with memory and a CPU or GPU. The functions of these components are implemented by programs loaded into the computer. However, some functions of the image analysis device 100 can also be implemented using hardware such as a programmable logic device (PLD).

[0038] Next, the outline of the operation of the image analysis device 100 having the above configuration will be described. Figure 3 Shows the flow of actions.

[0039] First, the image acquisition unit 10 acquires DICOM information along with the image data. The image information extraction unit 210 extracts information such as the mode (MR / CT), 2D / 3D, imaging section, contrast, field of view (FOV), and resolution from the DICOM information (S1). Next, the image normalization unit 220 performs processing to bring image parameters and positions, such as FOV and resolution, which vary from image to image, into alignment with predetermined values ​​and positions (standard values ​​or standard positions) (S2).

[0040] Next, the region segmentation unit 230 extracts the region of interest from the normalized image through segmentation (S3). Segmentation involves segmenting the region based on differences in the pixel values ​​of the tissues, using the mode (e.g., MR image or CT image) and contrast information extracted from the DICOM information. During this region segmentation (region extraction), the region segmentation unit 230 assigns a predetermined pixel value to each tissue in the segmented image, regardless of the mode or image type (contrast differences) of the original image. This results in a segmented image with a contrast that is independent of the contrast of the original image.

[0041] Next, the slice position determination unit 240 uses the segmented image to determine the slice cross section to be measured (S4). The slice position determination unit 240 creates a binary image of the given tissue of interest from the segmented image, aligns it with a pre-established standard image for which slice positions are empirically determined to be suitable for index calculation, and determines the slice position. The pre-established standard image (hereinafter simply referred to as the standard image) is an image obtained by the statistical analysis unit 30 by averaging images acquired from a large number of subjects and determining the slice positions used for index calculation. The image normalization unit 220 then normalizes the image to standard values ​​for the FOV and resolution used for standardization, as well as the standard position of the subject.

[0042] Once the slice position is determined, the tissue of interest extraction unit 250 extracts the tissue of interest to be measured from the segmented image at that slice position (S5). The tissue of interest to be extracted may or may not be the same as the tissue of interest (region) used in the slice determination process. Since the tissue of interest to be measured is determined based on the desired index to be calculated, and each region segmented by the region segmentation unit 230 is assigned a predetermined pixel value, the tissue of interest can be automatically extracted once the index is determined. Specifically, the region of the tissue of interest is determined based on the position, shape, and size of the tissue of interest.

[0043] The line segment calculation unit 260 calculates a predetermined line segment for the extracted tissue of interest (S6). The line segment varies depending on the tissue of interest and the index calculated by the index calculation unit 270. A specific technique will be described later. In response to the doctor's work of measuring the width and angle of a predetermined location on the tissue of interest from an image, the line segment at the location of maximum width and the line segment at the location of tangent to the tissue of interest are calculated based on the characteristics of the tissue of interest's shape.

[0044] Finally, the index calculation unit 270 calculates an index using the calculated line segment and outputs the result to an output device such as a display device ( S7 ).

[0045] According to this embodiment, after image standardization, regional segmentation is performed to unify the pixel values ​​of each region. This minimizes the pattern dependency, imaging condition dependency, and imaging position dependency when calculating indices. Furthermore, according to this embodiment, slices are determined based on the shape characteristics of the tissue of interest, and the tissue of interest is determined based on its position, shape, and size. This minimizes the subject dependency of the slice position and tissue of interest. Consequently, indices can be automatically calculated with high accuracy.

[0046] <Implementation Method 2>

[0047] Next, an embodiment will be described in which the image is a brain image and the calculated indices are the Evans index and the corpus callosum angle.

[0048] Device structure and processing overview ( Figure 3 ) is the same as embodiment 1, so the following mainly describes the functions and details of the operations of each part of the image processing unit 20.

[0049] Figure 4 (A) and (B) are diagrams showing the axial and coronal planes of brain images. The Evans index is as follows: Figure 4As shown in (A), a ratio (W401 / W402) of the width of the anterior angle of the lateral ventricle W401 to the width of the cranial cavity W402 exceeding 0.3 is an indicator for the diagnosis of iNPH. Furthermore, regarding the corpus callosum angle, a steep angle in an MRI coronal section passing through the posterior commissure (a section perpendicular to the line connecting the anterior and posterior commissures), specifically a callosal angle of 90 degrees or less, is an indicator for the diagnosis of iNPH. Therefore, for the Evans index, the slice position is determined in the axial section, and the ventricles and brain parenchyma are treated as the tissue of interest. For the callosal angle, the slice position is determined in the coronal section, and the ventricles are extracted.

[0050] In this embodiment, the image acquisition unit 10 acquires DICOM information together with the image data, and the image information extraction unit 210 extracts information such as mode (MR / CT), 2D / 3D, imaging section, contrast, FOV, resolution, etc. from the DICOM information (S1) in the same manner as in the first embodiment. Figure 3 Details of the processing after S2.

[0051] <Image Normalization Processing: S2>

[0052] The image normalization unit 220 is as follows Figure 5 As shown, the FOV is adjusted (S21), the resolution is adjusted (S22), and the position of the object is adjusted (S23). As for the parameter values ​​that become the standard, the values ​​that are most suitable for measuring the indicators to be measured are set in advance. If the measurement object is, for example, the Evans index, FOV: 300mm, resolution: 1mm, etc. are set as standard values, and the FOV and resolution of the measurement image are adjusted to make them the standard values. Regarding the adjustment of FOV, the redundant area is cut off or the surrounding area is filled with zeros in accordance with the standard value (S21). In addition, regarding the resolution, for example, when the resolution is larger than the standard value (for example, 2mm), the pixels are filled by linear interpolation, etc. to make it consistent with the standard value (S22). The above processing corresponds to whether the image data is a 2D multiple slice image or a 3D image, and becomes either a two-dimensional processing or a three-dimensional processing.

[0053] The image normalization unit 210 further adjusts the position of the subject in the image (S23). For example, the position of the brain center is adjusted by performing parallel translation and angular correction so that the center of the brain becomes the center of the image. Regarding the position of the brain center, any of the following methods can be used: creating a head shape mask (an image obtained by binarizing the inner and outer sides of the head shape) from the subject's image (diagnostic image) and determining the center of the brain as the center of gravity of the mask; or performing an elliptical approximation on the head shape mask in the axial plane and determining the center of the ellipse as the center of the brain. For the coronal plane, the brain center can also be determined using landmarks such as the top of the head and the eyeballs, as well as the symmetry of the head.

[0054] Furthermore, for 3D image data, the center position can be determined for multiple cross-sections (e.g., axial and coronal planes), and the intersection of lines passing through these center positions and perpendicular to the cross-sections can be used as the center of the 3D image data. This process normalizes the measured image to a uniform FOV and resolution, with the center of the brain at the center of the image.

[0055] <Region Division Processing: S3>

[0056] When performing region segmentation through segmentation, the region segmentation unit 230 first obtains the mode (e.g., MR image or CT image) and contrast information from the DICOM information and then segments the regions based on differences in pixel values ​​between tissues. Contrast information in MR images varies depending on the image type, which is determined by the imaging conditions. For T2-enhanced images, cerebrospinal fluid is depicted as the highest signal area, and bone as the lowest signal area. Therefore, region segmentation can be performed by treating the highest signal area as cerebrospinal fluid, the lowest signal area as bone, and the rest of the area as brain parenchyma. Similarly, for T1-enhanced images, cerebrospinal fluid is depicted as the central signal area, and bone as the lowest signal area. For FLAIR images, since cerebrospinal fluid and bone appear dark, they can be segmented as the lowest signal areas. For CT images, contrast is determined by the CT value of each tissue, so segmentation can be performed based on that value.

[0057] Segmentation can be performed using well-known techniques such as k-means, region expansion, or semantic segmentation, a deep learning method that associates classes with all pixels within an image. This yields an image (segmented image) in which each brain tissue, including cerebrospinal fluid, bone, and brain parenchyma (white matter and gray matter), is represented by a single pixel value (brightness value). In this case, the region segmentation unit 230 assigns a predetermined pixel value (label) to the pixel values ​​of each segmented tissue within the segmented image, regardless of the original image's mode or image type (contrast differences). Specifically, while the pixel value resolution varies depending on the image, the image is segmented into a predetermined number of stages (e.g., the number of tissues of interest). Regardless of the original image type, the pixel values ​​are consistently assigned a value of 1 for white matter, 2 for gray matter, 3 for cerebrospinal fluid, and so on. This yields a segmented image with uniform contrast, enabling the following processing regardless of image type.

[0058] <Slice Position Determination: S4>

[0059] The slice position determination unit 240 uses the segmented image to determine the slice section to be measured. In the case of 2D image data, since it is composed of multiple slice images, the slice image most suitable for measurement is selected from these slice images. In the case of 3D image data, multiple slices are cut out at a given slice interval for the section corresponding to the measurement object, and the slice position most suitable for measurement is determined. Whether the image data is 2D or 3D can be determined from the DICOM information extracted by the image information extraction unit 210, and the above-mentioned processing is performed accordingly. In addition, for example, in the case of the Evans index, the slice position of the axial section is determined, and in the case of the corpus callosum angle, the slice position of the coronal section is determined.

[0060] Therefore, the slice position determination unit 240 is as follows Figure 6 As shown, the standard image is read ( S41 ), a binary image is generated ( S42 ), compared with the standard binary image ( S43 ), and a slice position at which the error is minimized is determined ( S44 ).

[0061] As described in Embodiment 1, a standard image is an image created from a large number of previously captured images, in which the slice positions used to calculate a given indicator are known. Similar to the regional segmentation process used for diagnostic images, the standard image is segmented so that the pixel values ​​of each region correspond to the reference pixel values. In other words, the segmented image represents the average value of the positions and pixel values ​​of each region across the large number of images. The standard image or its segmented image is pre-created by the statistical analysis unit 30 and stored in the storage device 40 or database 500. Standard images can also be created for each age group, such as children, adults, and the elderly, and the standard images for the corresponding age groups are read in based on the DICOM subject information (S41).

[0062] Next, the slice position determination unit 240 creates a reference image from the read standard image and a binary image of the tissue of interest (an image in which the pixel values ​​of the tissue of interest are set to 1 and the pixel values ​​of all other tissues are set to 0) from the segmented image obtained through region segmentation (S3) (S42). The tissue of interest is a tissue whose shape changes clearly at each slice position and for which slice position determination is easy, such as the cerebrospinal fluid region or the eyeball. By representing the tissue of interest and all other tissues in binary form in the segmented image, a binary image is obtained. Alternatively, the reference image can be pre-created for each given tissue of interest in the statistical analysis unit 30 based on the standard image and read in S41.

[0063] The binary image is an image that extracts the characteristics of the shape of the tissue of interest. The slice position determination unit 240 uses these characteristics to determine the slice position. Therefore, first, a reference image created from the standard image is compared with the binary image of the diagnostic image to be measured, and the error is calculated (S43). Here, a specific example of the technique will be described using the case where the tissue of interest is the cerebrospinal fluid region.

[0064] Figure 7 The upper side is a reference image created by averaging a large number of images of the cerebrospinal fluid region captured in the past. Figure 7 The lower side is a binary image of the cerebrospinal fluid region of the image to be measured. Images A1, A2, A3 at slice positions at two or more (here, three) from the reference image are compared with binary images I1, I2, I3 of two or more measurement images at the same slice interval, and their difference (In-An) is calculated (n is any one of 1 to 3). The root mean square error (RMSE) shown in formula (1) or the normalized RMSE shown in formula (2) is calculated based on the difference between the corresponding images (binary images of the reference image and the diagnostic image). In addition, normalization is not necessary in the calculation of RMSE, but by performing normalization, the influence of the error change due to the size of the tissue of interest can be eliminated, that is, the influence on the calculation result caused by the different sets to be compared can be eliminated.

[0065]

[0066] exist Figure 7 In the figure, three images are compared. One of the reference images is an image at slice position S0, used for index measurement, and the other two are images at slice positions spaced a predetermined distance from S0. The binary images of the diagnostic images are composed of three images at slice positions spaced the same distance from the reference image. The image spacing can be the same as or different from the central image, and can be adjacent or spaced to some extent. However, the spacing between the reference image and the binary images of the diagnostic image is assumed to be the same.

[0067] While maintaining the same slice interval as the previous three images, the slice positions of the diagnostic image are shifted sequentially, changing the set of three images that differ from the three reference images, and calculating the root mean square error for each set. If the root mean square error is calculated for multiple sets, it is as follows Figure 8 As shown, a set of three images with the minimum value is obtained. In this set, the central slice position corresponds to the slice position S0 of the reference image, so this slice position is set as the slice position of the diagnostic image (S44).

[0068] In addition, Figure 7 In the example shown, the reference image set includes a set of slice positions with slice position S0 at the center. However, in this process, since the positional relationship between the reference image that defines slice position S0 and the binary image of the diagnostic image is sufficient, it is not necessary to set the set so that slice position S0 is at the center. Furthermore, while the positional relationship can be accurately determined using images at three slice positions, it can also be determined using two, and the number of images is not limited to three.

[0069] Furthermore, since the position with the minimum difference only needs to be found, it is not necessary to repeat the above calculations for all diagnostic images. Instead, the slice positions included in the entire set can be discretely selected on either side of the image center. Furthermore, while the example here uses the root mean square error (RMSE) for the difference, other difference metrics, such as the mean square error (MSE), can also be used.

[0070] Furthermore, in the above description, the slice position is determined using the error of the tissue of interest from the average shape (standard image), but an index representing features other than shape, such as an index representing features such as area and contour length, may also be used.

[0071] Regardless of whether the measurement section is an axial section or a coronal section, the same method can be used to set a given slice position. In addition, in 3D image data, the slice position can also be determined by comparing the binary images of the axial section and the coronal section.

[0072] <Group of Interest Extraction: S5>

[0073] Once the slice position is determined by the slice position determination unit 240 as described above, the tissue of interest for measurement is extracted from the image at the slice position determined by the tissue of interest extraction unit 250. For example, when measuring the Evans index, the width of the ventricles and the width of the cranial cavity are measured, thereby extracting the ventricles and the brain parenchyma (cerebral cavity). Furthermore, tissue of interest extraction is performed based on the signal values ​​of each region segmented by segmentation. For the corpus callosum angle, only the ventricles are extracted.

[0074] The tissue of interest extraction can be performed only on the image of a determined slice position, but in this embodiment, in order to improve the accuracy of the index calculation, it is preferred to also extract the tissue of interest on the images of the slice positions nearby, such as the images of the slice positions before and after it.

[0075] <Line Segment Calculation: S6>

[0076] Next, predetermined measurements are performed on the tissue of interest (the predetermined slice position and its adjacent slice positions) extracted by the line segment calculation unit 260 .

[0077] In the case of the Evans index, as Figure 4 As shown in (A), the maximum width between the anterior horns of the lateral ventricles to be measured is known to be located in the posterior half of the brain center. Therefore, for the extracted ventricles, the line segment with the maximum ventricle width is calculated for the posterior half (upper half) of the axial section, and this is used as the maximum width between the anterior horns of the lateral ventricles. Furthermore, since the width of the skull cavity is known to be located in the anterior half, the width of the brain parenchyma is measured in the lower half of the axial section, and the line segment with the maximum width is calculated, and this is used as the skull cavity width.

[0078] If tissue of interest is also extracted for nearby slices, the maximum width between the anterior horns of the lateral ventricles is first calculated for these slices as well. The slice with the largest maximum width among the given slice and its nearby slices is then determined as the measurement target, and the width of the skull cavity at that slice's location is calculated. Typically, there is a 6-7 mm gap between slices in MR images. Therefore, the maximum width between the anterior horns of the lateral ventricles may deviate somewhat from the slice position determined in the slice position determination step S4. By also calculating the maximum width between the anterior horns of the lateral ventricles for nearby slices, the slice position to be measured can be determined from multiple slices, including the nearby slices, and the maximum width between the anterior horns of the lateral ventricles and the skull cavity width can be measured at that slice position.

[0079] In the case of the corpus callosum angle, the ventricle extracted in the coronal section was determined along the centerline of the sulcus ( Figure 4 (B) The point line L) and the closest point of the left and right ventricles ( Figure 4 For point P (B), left and right tangents to the ventricle shape from this point to the top of the head are obtained, and two line segments are calculated. Regarding the tangents, for example, by initially setting point P and then varying the angle of the point line L, the line segment tangent to the ventricle contour can be determined as the tangent. While only coronal cross-sections are used to calculate the two line segments here, if the image data read by the image processing unit 20 is a 3D image and also includes sagittal cross-section image data, the sagittal image can also be used to determine point P, for example, to identify the centerline. Furthermore, while the above description uses tangents, other techniques such as the following can be used to determine line segments by connecting one point on each side of the ventricle shape at a given distance from point P to point P.

[0080] <Indicator calculation: S7>

[0081] Finally, the index calculation unit 270 uses the calculation results of the line segment calculation unit 260 to calculate an index. Specifically, the maximum width between the anterior horns of the lateral ventricles on both sides is divided by the width of the cranial cavity to obtain the Evans index. Furthermore, the angle formed by the two line segments calculated from point P is calculated to obtain the corpus callosum angle.

[0082] The index calculated by the index calculation unit 270 can be output as a numerical value via the UI unit 50 included with the image analysis device 100. Alternatively, the index can be sent to a separate image processing device within the image analysis device 100 and output as a medical observation result, integrated with other iNPH observation results, etc. Furthermore, the statistical analysis unit 30 can display a map of the Evans index and corpus callosum angle for each age group, along with results previously calculated from a large number of diagnostic images, and the position of the subject on the map can be clearly displayed.

[0083] The above processing completes the calculation of the index for the diagnostic image being measured. However, the data for the diagnostic image (segmented image or binary image, and information on the determined slice position) may be sent to the statistical analysis unit 30. The statistical analysis unit 30 can use the sent data to update the already created standard image (averaged segmented image, binary image of a given area).

[0084] According to this embodiment, the slice position and tissue of interest for measuring the Evans index and corpus callosum angle, which are diagnostic indicators of iNPH, can be determined independently of the type of input diagnostic image or individual differences in the subject, thereby automatically and accurately calculating the diagnostic indicators and providing prompts.

[0085] Furthermore, after determining a slice, the accuracy of determining the maximum width can be improved by measuring nearby slices at the same time.

[0086] Furthermore, depending on the facility or the examination procedure, it is not necessary to obtain sagittal cross-sectional image data. However, in this embodiment, since the index can be calculated based on one cross-section, it is also possible to cope with a situation where not all data sets are available.

[0087] Furthermore, although the present embodiment describes the case where two diagnostic indices of iNPH are calculated, the present invention naturally includes the case where only one is calculated.

[0088] <Variation of Implementation Example 2>

[0089] In the second embodiment, the case of obtaining the Evans index and the corpus callosum angle as numerical indices has been described, but information related to the DESH observation results can also be obtained by applying the same method. The following describes how to obtain information related to the DESH observation results.

[0090] The pattern of brain atrophy differs between Alzheimer's disease and iNPH. The former is characterized by global brain parenchymal atrophy, while the latter is associated with uneven expansion of the subarachnoid space. This observational result is referred to as a DESH observation. Although no numerical indicator is provided in the guideline for the DESH observation, images show that the brain parenchyma at the top of the head is rarefied (more voids) in Alzheimer's disease, whereas this observational result is not seen in iNPH. In this embodiment, to identify the characteristics of this DESH observation, the ventricles and brain parenchyma (intracranial cavity) are extracted as tissues of interest, and the area of ​​the brain parenchyma is calculated. The section for extracting the tissue of interest is preferably a coronal section, but an axial section may also be used, or information from both sections may be used.

[0091] In this modification, the basic processing flow (image information extraction (S1), image normalization (S2), region segmentation (S3), slice position determination (S4), tissue of interest extraction (S5)) is the same as Figure 3 The process flow is the same as that of the embodiment 2. Hereinafter, this modification will be described focusing on the differences from the embodiment 2.

[0092] After determining the slice position using the binary image created from the segmented image ( S1 - S4 ), the tissue of interest extraction unit 250 extracts the ventricles and brain parenchyma using the slice image ( S5 ). Similar processing can be performed on nearby slice images.

[0093] Next, the line segment calculation unit 260 calculates line segments for further dividing the brain parenchyma of the extracted cross section. Figure 9 As shown in (A), for example, four line segments are determined to divide the brain parenchyma area surrounded by the outer shape of the ventricle and the reduced shape located at a given distance from the brain center into four small areas. The four line segments can be set to be equally spaced from the brain center, or can be determined based on other shape criteria. For the coronal section, as shown in Figure 9 As shown in (B), the area of ​​the brain parenchyma surrounded by the outer shape of the brain parenchyma and a similar shape at a predetermined distance is divided into three small areas using four line segments (line segments that separate the areas and line segments that define the ends of the areas).

[0094] Furthermore, in this variation, the line segment calculation unit 260 (measurement unit) calculates the area (volume) of the segmented small regions. The area can be calculated as the sum of the pixel values. When calculating for a given slice and its surroundings, the average value can be used as the area of ​​the small region.

[0095] Index calculation unit 270 provides information related to the DESH observation results based on the area (volume) of each small region. Information related to the DESH observation results can be calculated as an index by calculating the area ratio or volume ratio of the small regions, or by comparing the ratio with the average value of normal subjects or patients diagnosed with Alzheimer's disease.

[0096] This modification performs specialized processing on DESH observation results, but by performing the processing in conjunction with the second embodiment and presenting the results together with the Evans index and the corpus callosum angle index, it can contribute to the diagnosis of iNPH.

[0097] In addition, in embodiment 2 and its variations, an example of calculating the iNPH index from a brain image is described. However, the present invention can also be applied to brain diseases other than iNPH and parts other than the brain, as long as the technology is used to determine the cross-section and automatically measure the part of interest. As a result, measurement can be performed while avoiding pattern dependence, image type dependence, subject dependence, etc.

Claims

1. An image analysis device, characterized in that: have: An image receiving unit for inputting a diagnostic image and imaging conditions of the diagnostic image; an image standardization unit, configured to standardize the diagnostic image; a region segmentation unit for extracting a region of tissue of interest from the diagnostic image normalized by the image normalization unit; a slice position determination unit that determines a slice position in the standardized diagnostic image based on features of the tissue of interest obtained by regional segmentation; a measuring unit that determines a tissue of interest to be measured based on at least one of the position, shape, length, and size of the tissue at the slice position determined by the slice position determining unit, and measures the tissue; and an index calculation unit that calculates an index using the measurement value of the tissue of interest measured by the measurement unit, The region segmentation unit extracts regions according to the imaging conditions, so that the pixel values ​​of the segmented regions are assigned predetermined labels. The slice position determination unit determines, in a prior standard image in which a reference slice position for measuring the indicator is predetermined, a slice position of the diagnostic image corresponding to the reference slice position of the prior standard image as a slice position based on an evaluation value obtained from the prior standard image at multiple slice positions and the diagnostic images at multiple slice positions.

2. The image analysis device according to claim 1, wherein The imaging conditions include the type of imaging device used to obtain the diagnostic image, the image dimensions, the image contrast, the FOV, and the resolution.

3. The image analysis device according to claim 1, wherein The image normalization unit performs normalization so that the FOV, resolution, and subject position included in the imaging conditions conform to predetermined standard values ​​and standard positions.

4. The image analysis device according to claim 1, wherein The multiple slice positions are different slice positions of the same cross section.

5. The image analysis device according to claim 1, wherein The multiple slice positions are slice positions of different cross sections.

6. The image analysis device according to claim 1, wherein The image analysis device further comprises: The statistical analysis unit creates the aforementioned pre-standard image using a large number of diagnostic images.

7. The image analysis device according to claim 1, wherein The measuring unit sets a line segment for calculating the index for the tissue of interest.

8. The image analysis device according to claim 7, wherein: The tissue of interest includes at least a portion of a cerebral ventricle and brain parenchyma, and the measuring unit calculates a line segment having a maximum width in the anterior horn of the ventricle and a line segment having a maximum width in the brain parenchyma.

9. The image analysis device according to claim 7, wherein: The tissue of interest includes at least a portion of a cerebral ventricle, and the measurement unit calculates tangent lines of the cerebral ventricle as two line segments for determining a corpus callosum angle.

10. The image analysis device according to claim 1, wherein The measuring unit divides the tissue of interest into a plurality of small regions and calculates the volumes of the small regions.

11. The image analysis device according to claim 10, wherein The index calculation unit calculates a qualitative or quantitative index associated with a DESH observation result of sudden normal pressure hydrocephalus based on the volume of the small region.

12. The image analysis device according to claim 1, wherein The diagnostic image is a brain image, The index calculated by the index calculation unit includes at least one of an Evans index and a corpus callosum angle, which are diagnostic indicators of sudden normal pressure hydrocephalus.

13. An image analysis method for calculating an index of a specific disease by inputting a diagnostic image, the image analysis method comprising: a step of inputting the diagnostic image and imaging conditions of the diagnostic image; an image standardization step of standardizing the diagnostic image; a region segmentation step of extracting a region of tissue of interest from the normalized diagnostic image; The step of determining a slice position in the standardized diagnostic image based on the characteristics of the tissue of interest obtained by regional segmentation; The step of performing measurement of a given tissue of interest at the determined slice position; and Steps to calculate indicators using the measured values ​​of the organization of interest, In the region segmentation step, region extraction is performed according to the imaging conditions so that the pixel values ​​of the segmented regions are assigned predetermined labels. In a prior standard image in which a reference slice position for measuring the indicator is predetermined, the slice position of the diagnostic image corresponding to the reference slice position of the prior standard image is determined as the slice position based on an evaluation value obtained based on the prior standard image of multiple slice positions and the diagnostic images of multiple slice positions.

14. The image analysis method according to claim 13, wherein: The diagnostic image is an image obtained by an MRI device. The imaging conditions include the type of imaging device used to obtain the diagnostic image, the image dimensions, the image contrast, FOV, and resolution. The image normalization step performs normalization so that the FOV, resolution, and object position conform to predetermined standard values ​​and positions.

15. The image analysis method according to claim 13, wherein: The diagnostic image is a brain image, The index calculated in the step of calculating the index includes at least one of the Evans index and the corpus callosum angle, which are diagnostic indicators of sudden normal pressure hydrocephalus, and a qualitative or quantitative index associated with a DESH observation result.

16. A computer program product comprising an image analysis program, characterized in that: The image analysis program causes the computer to perform: a step of receiving a diagnostic image and imaging conditions for the diagnostic image; an image standardization step of standardizing the diagnostic image; A region segmentation step is performed on the standardized diagnostic image, wherein the contrast is normalized according to the imaging conditions so that the pixel values ​​of the segmented regions are assigned predetermined labels; The step of determining a slice position in the standardized image based on a feature of at least one tissue of interest obtained by regional segmentation; The step of performing measurement of a given tissue of interest at the determined slice position; and Steps to calculate indicators using the measured values ​​of the organization of interest, In a prior standard image in which a reference slice position for measuring the indicator is predetermined, the slice position of the diagnostic image corresponding to the reference slice position of the prior standard image is determined as the slice position based on an evaluation value obtained based on the prior standard image of multiple slice positions and the diagnostic images of multiple slice positions.

Images