JP2017136134A - Medical support device, image processing device, medical support method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a medical support device, etc. that sets a region of interest using a new criterion in an ROI method.SOLUTION: A medical support device 1 sets a region of interest in subject image data acquired by imaging a body tissue of a subject. A subject gene information storage part 15 stores information on a gene of the subject. A region-of-interest setting part 21 refers to the subject gene information storage part 15, and sets a region-of-interest according to a type of the gene. For example, when a disease to be diagnosed is Alzheimer type dementia, attention is paid to APOEε4 gene as one of the corresponding genes. For the gene type with APOEε4 gene, the region of interest is hippocampus. For the gene type with no APOEε4 gene, the region of interest is parietal lobe. An analysis part analyzes atrophia in the region of interest in the subject image data. Or, by reading a change pattern of the image from an onset example, relationships between the change pattern and the gene type may be automatically analyzed.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a medical support apparatus, an image processing apparatus, a medical support method, and a program, and more particularly to a medical support apparatus that sets a region of interest in subject image data obtained by imaging a subject's body tissue.

  In Patent Document 1, when the ROI method is used, a region of interest (ROI) for each disease in a brain image is determined in advance by a statistical method, and a test is performed using the region of interest corresponding to the disease. And presenting a diagnosis result by performing a statistical comparison between the brain image of a healthy person and the brain image of a healthy person.

JP 2005-237441 A

  In the ROI method, the diagnosis result greatly varies depending on the setting of the region of interest. Therefore, in the ROI method, the setting of the region of interest has an important meaning.

  However, conventionally, the region of interest is set corresponding to the disease, and the same region of interest has been set if the disease is the same. For this reason, the study on the setting of the region of interest has been insufficient.

  Therefore, an object of the present invention is to propose a medical support device or the like that sets a region of interest using a new standard.

  A first aspect of the present invention is a medical support device that sets a region of interest in subject image data obtained by imaging a subject's body tissue, and stores information on the subject's genes Subject gene information storage means, region-of-interest setting means for setting different regions of interest corresponding to a part of the gene type of the subject, and analysis of the region of interest of the subject image data Analyzing means is provided.

  A second aspect of the present invention is the medical support apparatus according to the first aspect, wherein the region of interest setting means sets a first region of interest when there is a specific gene in the gene of the subject. In the absence of the specific gene, a second region of interest that is at least partially different from the first region of interest is set.

  A third aspect of the present invention is the medical support apparatus according to the second aspect, wherein the subject data is obtained by photographing the brain of the subject, and the specific gene is: When there is an APOEε4 gene, the region of interest setting means sets the hippocampus as the first region of interest, and when there is no APOEε4 gene, sets the parietal lobe as the second region of interest.

  A fourth aspect of the present invention is the medical support apparatus according to any one of the first to third aspects, wherein a plurality of reference image data obtained by imaging a body tissue at the same location as the subject image data. And standardization means for obtaining standard image data and standard subject image data by setting a plurality of parts common to the respective standard image data and the subject image data. The analyzing unit analyzes the difference in the region of interest set by the region of interest setting unit by comparing the standard image data and the standard subject image data.

  According to a fifth aspect of the present invention, there is provided an image analysis apparatus for obtaining a region of interest candidate set by the medical support apparatus according to any one of the first to fourth aspects. Standardization means for obtaining standard onset image data by setting a plurality of sites for each of the onset image data obtained by imaging the body tissue of the person, and the genetic information of each onset and the onset Corresponding analysis target data storage means and one or a plurality of the standard onset image data not similar to the other standard onset image data in the candidate region of interest that is a part of the set part Extraction means for extracting, and gene types that are included in the gene information corresponding to the extracted standard symptom image data and are not included in the gene information corresponding to the standard symptom image data not extracted are candidate genes And a candidate display unit for displaying a combination of the candidate region of interest and the candidate gene type as a candidate for a combination of the region of interest and the gene type used by the region of interest setting unit. is there.

  A sixth aspect of the present invention is a medical support method in a medical support apparatus that analyzes image data obtained by imaging a body tissue, and the medical support apparatus is obtained by imaging the same part of a body tissue. Reference image data storage means for storing a plurality of reference image data, and subject image data storage means for storing subject image data obtained by imaging the same location as the reference image data of the subject's body tissue And a subject gene information storage means for storing information on the gene of the subject, and a standardization means provided in the medical support device sets a plurality of common portions for each reference image data A standardization step of obtaining standard image data, setting the same plurality of parts as the standard image data for the subject image data to obtain standard subject image data, and a region of interest setting provided in the medical support device Means A region-of-interest setting step for setting different regions of interest corresponding to a part of the gene type of the subject, and an analysis means provided in the medical support device includes the standard image data and the standard subject image The method includes an analysis step of comparing data and analyzing a difference in the region of interest set by the region-of-interest setting means.

  A seventh aspect of the present invention is a program for causing a computer to function as the medical support apparatus according to any one of the first to fourth aspects.

  Note that the present invention may be regarded as a program for causing a computer to function as the image analysis apparatus according to the fifth aspect. Further, the present invention may be regarded as a computer-readable recording medium that regularly records a program.

  The inventors have revealed that atrophy appears in different regions depending on the presence or absence of APOEε4, which is one of the genes related to Alzheimer's disease (AD). That is, when APOEε4 gene is present, atrophy appears in the hippocampus, and when APOEε4 gene is absent, atrophy appears in the parietal lobe. Since atrophy is a spatial change, it can be detected by image processing.

  According to the present invention, by setting a region of interest using a subject's gene type and performing image analysis, a patient at a genetic risk is carefully observed, and the onset of the disease is detected very early. It becomes possible. In this way, by integrating and analyzing gene information and image information, next generation CAD (Computer-Aided Diagnosis) technology that supports personalized medicine can be realized.

  In particular, according to the third aspect, a lesion caused by AD that currently accounts for about half of all dementia can be detected very early. In particular, the present invention only uses the genetic information and image information of the subject. It is not necessary for a doctor or the like to check symptoms in advance for setting the region of interest. Therefore, the diagnosis according to the present invention can be performed even when a brain image is taken for diagnosing a human dock or other diseases. And according to this invention, onset can be started at an early stage in order to detect the onset of AD very early and to suppress the progress of the symptom of AD.

  Furthermore, according to the fourth aspect of the present invention, it is possible to perform comparison with high accuracy by standardizing and comparing the image data of the subject and, for example, the image data of the normal case by standardization processing. .

  Furthermore, according to the fifth aspect of the present invention, by automatically classifying a pattern of spatial change such as atrophy and extracting a gene related to this pattern, a gene related to the disease and its spatial change Pattern candidates can be automatically extracted.

  The image data of the present invention includes 2D images and 3D images, and also includes still images and moving images. The present invention can be applied to technical fields such as HIS (Hospital Information System) and PACS (Picture Archiving and Communication System).

It is a figure which shows the grade of the atrophy | shrink of the area | region of the hippocampus and parietal lobe by the presence or absence of APOE (epsilon) 4 gene in experiment of this invention. It is a block diagram which shows an example of a structure of the medical assistance apparatus which concerns on embodiment of this invention. It is a flowchart which shows an example of a process of the medical assistance apparatus 1 of FIG. It is a figure of Tarailach Atlas. It is a block diagram which shows an example of a structure of the image processing apparatus which concerns on embodiment of this invention. It is a flowchart which shows an example of a process of the image processing apparatus 51 of FIG. It is a figure which shows the classification result by the similarity by the image processing apparatus 51 of FIG. It is explanatory drawing of the scatter diagram analysis for visualizing the regularity and abnormality of an atrophy pattern. It is a figure which shows the example of detection of the specific atrophy pattern by a scatter diagram analysis. It is a figure which shows the classification result by the hierarchical clustering of all 40 AD cases.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. The present invention is not limited to this embodiment.

  First, experiments conducted by the inventors using the medical support apparatus according to the embodiment of the present invention will be described. FIG. 1A shows an overall image of processing in the experiment. In this experiment, the Alzheimer's Disease Neuroimaging Initiative (ADNI) database was used. The ADNI database contains 1879 data. Of these, 30 cases (71 to 79 years) were selected for use in the experiment so that the age range was limited from 52 cases prepared as a normal set, and the number of samples that could approximate the normal distribution was secured. Moreover, 27 cases (56-91 years old) prepared as AD sets were used as subjects of AD cases.

  The images used were taken with a 3.0T MR device using the Magnetization-Prepared Rapid Acquisition with Gradient Echo (MP-RAGE) imaging method. The matrix size was 256 × 256, and the voxel size was 1.0 × 1.0 × 1.2 mm.

  In addition, genetic information of these cases is provided in EXCEL files.

  These images cannot be used as they are because they have different shooting conditions. To deal with this, a standard set is provided in the ADNI database. First, the size and anatomical position were made the same by performing brain morphology standardization using SPM8. First, realignment processing was performed in order to compensate for body movement and pulsation during MRI. Realignment processing assumes that the head position in the image time series does not change, so that the head position matches between the reference image (the first image in the session) and the image at a certain point in time, Rigid body transformation is performed by translation and rotation in the Y-axis and Z-axis directions. Furthermore, resampling was performed on the image data on which the motion was corrected, and a slice was reconstructed. Next, normalization processing was performed. The motion-corrected data is converted to match the Montreal Neurological Institute (MNI) standard brain (T1 template image) prepared as a standard brain in SPM8. Brain morphology standardization processing was performed by adding the above processing to all MR images targeted in this experiment. The standardized image data size was 68 × 95 × 79. By performing this brain shape standardization process, the size and anatomical position of the brain become the same, so that morphological changes can be analyzed by comparing pixel values.

  Next, a normal standard brain to be compared was created from 30 normal cases. The average values and standard deviations of 30 normal patient MR images with standardized brain morphology were calculated for each voxel, and standard brain mean images and standard deviation images were created.

  In addition, 27 AD patient MR images were classified into two AD groups according to the presence or absence of the APOEε4 gene, which is one of AD-related genes. There were 17 cases in the APOEε4 (+) group and 10 cases in the APOEε4 (−) group. An average image was created for each group.

  Then, in order to visually analyze the characteristics of brain morphology changes in AD patients depending on the presence or absence of the APOEε4 gene, differences from the normal standard brain were displayed in a Z score map that will be described later. By this processing, it is possible to visually analyze in what part the AD group classified according to the presence or absence of a gene is changing.

  FIGS. 1B to 1E show an average image created for each group as a Z score map. (B) and (d) are of the APOEε4 (+) group, and (c) and (e) are of the APOEε4 (−) group. The anatomical position of all the image data becomes the same by the brain morphology standardization process. Therefore, if the region of interest is set using the normal standard brain, the degree of atrophy of the region of interest can be quantified.

  In this experiment, regions of interest were set in the hippocampus and parietal lobe to quantitatively evaluate the degree of atrophy. Pixels in which the Z score map has a negative value (hereinafter referred to as “negative pixels”) indicate a region with atrophy as compared to the normal standard brain. In the visual evaluation, the negative pixel portion was displayed in blue. In the quantitative evaluation, the degree of atrophy in the region of interest was quantified by calculating the average value of the negative pixels in the Z score map.

  The numerical values in FIGS. 1B and 1C are values obtained by setting the region of interest in the hippocampal region and quantifying the components in the negative region of the Z score map. In visual evaluation, APOEε4 (−) has the impression that hippocampal negative pixel components are small and atrophy is mild. In quantitative evaluation, APOEε4 (+) is −0.330, and APOEε4 (−) is −0.297. Therefore, APOEε4 (+) has a greater degree of contraction.

  The numerical values in FIGS. 1D and 1E are values obtained by setting a region of interest in the parietal lobe region and quantifying components in the region where the Z score map is negative. In the visual evaluation, APOEε4 (−) has the impression that atrophy is spreading in the parietal and temporal lobes. Even quantitative evaluation can be judged to be strong.

  Therefore, according to this experiment, patients with APOEε4 gene have strong hippocampal atrophy but mild atrophy of the whole brain, and patients without APOEε4 gene have lesions spreading to the parietal and temporal lobes. The trend was confirmed. As described above, the inventors have clarified through experiments that the lesion can be confirmed with high accuracy by changing the region of interest depending on the presence or absence of the gene.

  FIG. 2 is a block diagram showing an example of the configuration of the medical support apparatus of the present invention. FIG. 3 is a flowchart showing an example of processing of the medical support apparatus 1 of FIG.

  With reference to FIG. 2, the structure of the medical assistance apparatus 1 is demonstrated. The medical support apparatus 1 includes an imaging unit 3, a reference image data storage unit 5, a standardization unit 7, a standard image data storage unit 9, a subject image data storage unit 11, and a standard subject image data storage unit. 13, a subject gene information storage unit 15, a disease input unit 17, a combination storage unit 19, a region of interest setting unit 21, an analysis unit 23, and a diagnosis result display unit 25.

  The imaging unit 3 captures the body tissues of the healthy subjects 27 and the subject 29 and obtains image data. Hereinafter, a case where brain image data of a subject is obtained by imaging with MRI (Magnetic Resonance Imaging) will be described. The brain image data of the healthy persons 27 is stored in the reference image data storage unit 5. The brain image data of the subject 29 is stored in the subject image data storage unit 11.

  The standardization unit 7 performs a brain form standardization process for matching the brain size and anatomical position in order to compare the brain image data. The brain image data of the healthy subjects 27 and the subject 29 after the brain form standardization processing are stored in the standard image data storage unit 9 and the standard subject image data storage unit 13, respectively. It is assumed that the standard image data storage unit 9 stores an average image and a standard deviation image after brain morphology standardization processing.

  The genetic testing device 31 obtains genetic information of the subject 29. The genetic information of the subject 29 is stored in the subject genetic information storage unit 15. A disease to be diagnosed is input to the disease input unit 17. The combination storage unit 19 stores the correspondence between the gene type corresponding to the disease and the region of interest. For example, when the disease to be diagnosed is AD, one of the corresponding genes is the APOEε4 gene. The combination storage unit 19 stores a hippocampus as a region of interest corresponding to a type having the APOEε4 gene, and a parietal leaf as a region of interest corresponding to a type having no APOEε4 gene.

  When a specific disease is input to the disease input unit 17, the region-of-interest setting unit 21 refers to the combination storage unit 19 and specifies a specific gene corresponding to this specific disease. Then, the type of the specific gene of the subject is specified with reference to the subject gene information storage unit 15. Then, the region of interest corresponding to the type of the specific gene of the subject is set with reference to the combination storage unit 19.

  The analysis unit 23 uses the average image and the standard deviation image of the standard image data storage unit 9 to store the brain morphology stored in the standard subject image data storage unit 13 in the region of interest set by the region of interest setting unit 21. The brain image data of the subject 29 after the standardization process is analyzed. The diagnosis result display unit 25 displays the diagnosis result. For example, if atrophy is observed in the region of interest in the hippocampus or parietal lobe, it is displayed that AD may have occurred.

  With reference to FIG. 3, operation | movement of the medical assistance apparatus 1 of FIG. 2 is demonstrated. The standardization process for the normal cases in steps S1 and S2 and the genetic test of the subject are performed before photographing the subject 29 in step S3, but before the setting of the region of interest in step S5. That's fine.

  As preprocessing, the standardization unit 7 performs brain morphology standardization processing on the brain image data of the healthy individuals 27 (step S1). The standard image data storage unit 9 stores brain image data after brain morphology standardization processing. The brain morphology standardization process will be specifically described later.

  Subsequently, the genetic testing device 31 performs genetic testing of the subject (step S2). The subject genetic information storage unit 15 stores genetic information obtained by genetic testing.

  The imaging unit 3 captures the subject 29 and obtains brain image data (step S3). The subject image data storage unit 11 stores the brain image data of the subject 29.

  The standardization unit 7 performs brain form standardization processing on the subject's brain image data (step S4). The standard subject image data storage unit 13 stores brain image data after brain morphology standardization processing.

  The region-of-interest setting unit 21 identifies a specific gene corresponding to the specific disease input to the disease input unit 17 with reference to the combination storage unit 19, and refers to the subject gene information storage unit 15 to examinee 29 The specific gene type in the gene information is specified, and the region of interest corresponding to the specific gene type is set with reference to the combination storage unit 19 (step S5). Thus, since the region of interest is set according to the individual characteristic of genetic information, the medical support device of the present invention can realize more advanced personalized medicine.

  The analysis unit 23 refers to the standard image data storage unit 9 and the standard subject image data storage unit 13 to create a Z score map (step S6), and determines the degree of atrophy in the region of interest based on the Z score map. Quantify (step S7). The Z score map will be specifically described later. Then, the diagnosis result display unit 25 displays the diagnosis result (step S8).

  The brain morphology standardization process by the standardization unit 7 in FIG. 2 will be described. In the brain morphology standardization process, first, a Montreal Neurological Institute (MNI) standard brain (T1 template image) is selected as a reference image. The enlargement / reduction ratio of the processed image is calculated and the enlargement / reduction process is added so that the brain size of the processed image matches the brain size of the reference image. Next, the processed image is rigidly transformed by translation and rotation in the X-axis, Y-axis, and Z-axis directions so that the brain position of the processed image matches the brain position of the reference image. By performing this brain morphology standardization processing, the brain sizes and anatomical positions of all the processed images become the same. Therefore, morphological changes can be analyzed by comparing voxel values at the same position.

  Next, the creation of the Z score map in step S6 of FIG. 3 will be described. The data stored in the standard image data storage unit 9 is referred to as “normal standard brain”. In this method, an average value and a standard deviation image are created by calculating an average value and a standard deviation for each voxel from an MR image of a normal case subjected to a brain morphology standardization process. The data stored in the standard subject image data storage unit 13 is referred to as “processed image”. The Z score map is created by substituting the values of the normal standard brain and the processed image into Expression (1). Here, Mean (x, y, z) and SD (x, y, z) represent the average image and standard deviation image of the normal standard brain, and Input (x, y, z) represents the processed image. By using the Z score map, it is possible to quantitatively analyze what change is occurring in which part of the processed image as compared with the normal standard brain. For example, a voxel with a negative Z score map indicates a region with atrophy as compared to a normal standard brain.

  Subsequently, the quantification of the degree of atrophy in step S7 in FIG. 3 will be described. A voxel with a negative Z score map indicates a region with atrophy compared to the normal standard brain. Therefore, for example, in order to evaluate atrophy, an average of voxel values having a negative Z score map may be obtained in the region of interest. For example, if regions of interest are set in the hippocampus and parietal lobe of the normal standard brain, and the average of the values of voxels with negative Z score maps is obtained, the degree of brain atrophy of the hippocampus and parietal lobe can be quantified. In current interpretation, when the atrophy occurs in the entire brain, the onset can be diagnosed, but the degree of atrophy focusing on a specific site has not been evaluated. The quantification of the atrophy of the site related to the gene according to the present invention enables interpretation in consideration of the individual constitution related to the gene, and the onset of the disease can be diagnosed earlier than at present.

  Note that if regions of interest are set in various parts of the brain, the degree of brain atrophy in each part can be quantified. Talairach Atlas is used to set regions of interest in various parts of the brain. FIG. 4 shows the Talairach Atlas. The brain morphology standardization process is performed so that this atlas matches the normal standard brain. Then, the Z score map of all the parts divided by the atlas can be calculated to quantify the degree of atrophy of various parts of the brain.

  The present invention can be applied not only to the medical support apparatus shown in the first embodiment, but also to an image data mining technique for automatically classifying and visualizing patterns of body tissue morphological changes. Currently, many hospitals operate hospital information systems and store big data such as images and examination data. In addition, information sharing with local hospitals is progressing with cloud technology, and a system is in place to store medical records over the lifetime of patients. From now on, it is important to conduct research and development on next-generation information and communication technology (ICT) medical technology to utilize big data sleeping in hospitals for the prediction and prevention of disease, ultra-early diagnosis of diseases, and personalized treatment.

  In this embodiment, an image data mining technique for automatically classifying and visualizing morphological change patterns will be described. In order to explain specifically, MR images of AD cases are targeted. This embodiment is realized by using data mining for feature analysis of anatomical spatial distribution in addition to brain morphology standardization processing. Data mining is a technique for discovering frequent patterns and meaningful structures in data. According to the present embodiment, regularity and anomaly included in large-scale image data can be found, and the result is expected to become one of the basic technologies supporting next-generation ICT medical care.

  In the following, we propose a method for classifying the regularity of morphological change patterns between image data using hierarchical clustering and a technique for visualizing the relationship between regularity and anomaly between image data and finding abnormal patterns.

  In the following, the Alzheimer ’s disease neuroimaging initiative (ADNI) database was used as an experimental sample for data mining. As described above, the age range was limited from 52 cases prepared as a normal set, and 30 cases aged 71 to 79 years were used as experimental subjects. In addition, 40 cases (56-91 years old) prepared as an AD set were the subjects of the experiment.

  FIG. 5 is a block diagram showing an example of the configuration of the image processing apparatus according to the embodiment of the present invention. FIG. 6 is a flowchart showing an example of the operation of the image processing apparatus 51 of FIG.

  5 includes an imaging unit 53, a reference image data storage unit 55, a standardization unit 57, a standard image data storage unit 59, a subject image data storage unit 61, a standardization unit 57, A standard subject image data storage unit 63, an onset patient gene information storage unit 67, an analysis unit 69, and a candidate display unit 71 are provided.

  The genetic testing device 77 performs genetic testing of the onset patients 75 (step SG1 in FIG. 6). The onset gene information storage unit 67 stores the result of the genetic test.

  The operations of the imaging unit 53 and the standardization unit 57 for the healthy individuals 73 and the onset patients 75 are the same as those of the imaging unit 3 and the standardization unit 7 in FIG. The reference image data storage unit 55 and the subject image data storage unit 61 store image data captured by the imaging unit 53. The standardization unit 57 performs the above-described brain form standardization process (step SG2 in FIG. 6). The standard image data storage unit 59 and the standard subject image data storage unit 63 store the image data after the brain morphology standardization process. The standard image data storage unit 59 stores an average image and a standard deviation image of the image data of healthy persons after the brain morphology standardization process. The standard subject image data storage unit 63 stores the image data after the brain form standardization processing of each subject.

  The analysis unit 65 performs analysis using the image data of the healthy individuals 73 and the image data of each group of the affected person 75 (step SG3 in FIG. 6). Then, the gene type related to one or a plurality of image data having the same characteristics is extracted with reference to the onset gene information storage unit 67 (step SG4 in FIG. 6). For example, among genes related to diseases, many types are common to genes related to image data with the same characteristics, but types that are rarely found in genes related to image data without characteristics are extracted. To do. And a candidate display part displays the candidate of the combination of the region of interest and gene type as a change part of the image data by a disease (step SG5). This is a combination candidate stored in the combination storage unit 19 of FIG. According to the image processing apparatus 51 of FIG. 5, such combination candidates can be obtained automatically.

  Hereinafter, classification by hierarchical clustering and analysis by a scatter diagram will be described. Classification by hierarchical clustering helps to understand the characteristics of several groups with the same tendency. Analysis with scatter charts helps to understand individual characteristics.

  First, classification by hierarchical clustering will be described. The grouping unit 65 applies hierarchical clustering that classifies similar images with high similarity into the same group using the similarity.

Hereinafter, the similarity is calculated by employing a cross-correlation coefficient. The cross-correlation coefficient outputs a value from −1.0 to 1.0. If the similarity between the two images is high, the cross-correlation coefficient shows a value close to 1.0, and conversely if the difference is large, it shows a value close to -1.0. A cross-correlation coefficient (CC) between the image A and the image B is calculated by the equation (2). Here, A _AVE and B _AVE represent average values of pixels, and σ _A and σ _B represent standard deviations, respectively. I, J, and K indicate the range of the region of interest for which the cross correlation coefficient is calculated. That is, in this method, the region of interest is a cube designated by I, J, and K in Expression (1).

  If it is desired to set a more complicated region of interest, the values of the pixels within the complex region of interest obtained by raster scanning the image may be arranged in order and expressed only by the variable I. By using the cross-correlation coefficient calculated by the variable I, a complex region of interest can be analyzed in the same manner as described below.

Although the degree of similarity between images can be calculated by the above-described method, there is a degree of freedom in how to set the region of interest, and the classification result changes depending on the method of giving it. Therefore, in this experiment, the similarity was calculated by setting the regions of interest in the Axial direction, Coronal direction, and local region (part).
Condition 1: When similarity is calculated by setting a region of interest of 6 × 95 × 79 around an Axial slice including the hippocampus.
Condition 2: A similarity is calculated by setting a region of interest of 68 × 6 × 79 around a Coronal slice including the hippocampus.
Condition 3: A similarity is calculated by setting a local region of interest having a size of 6 × 25 × 50 so as to surround the hippocampus.

  Since the image size after brain morphology standardization processing is 68 × 95 × 79, the region of interest in condition 1 is the maximum size in the height (Y-axis) and width (X-axis) directions of the condition 2 The area was set as the maximum size in the depth (Z axis) and width (X axis) directions of the image. The width was set to 6 so as to include the target region.

  Hierarchical clustering was performed using the similarities calculated under the above three conditions to classify the image data. That is, three classification results were obtained: a classification result based on similarity in the Axial direction, a classification result based on similarity in the Coronal direction, and a classification result based on similarity in the hippocampal region. Here, hierarchical clustering is a technique for collecting the most similar image data sequentially and classifying similar cases like a tree diagram. In this study, the Ward method is adopted as a hierarchical clustering method. The Ward method is a method of combining the two clusters X and Y with the smallest amount of information loss. Hierarchical clustering was performed using the free software R.

  To execute hierarchical clustering, it is necessary to give a distance matrix composed of distances (closeness) between data. In the distance matrix, the distance between similar cases has a small value, but the degree of similarity based on the cross-correlation coefficient increases as the similarity is similar. Therefore, a distance matrix was created by using the following conversion formula (3). Here, Distance Matrix (m, n) represents the distance between the mth and nth cases. By using this conversion formula, when the value of the similarity matrix (Similarity Matrix) based on the cross-correlation coefficient is 1.0, the value of the distance matrix is 0.0.

  To visualize the features of AD groups classified by hierarchical clustering, a Z score map was created. First, brain morphology standardization processing was performed on the image data of 30 healthy subjects, and the average value and standard deviation image of the normal brain were created by calculating the average value and standard deviation for each voxel for 30 cases with standardized brain morphology. . Next, an average image was created for each AD group classified by hierarchical clustering. The difference from the normal brain was imaged by formula (1) so that the characteristics of the AD group could be visualized. By this processing, it is possible to visually analyze at which site the characteristic morphological change occurs in the automatically classified AD group.

  FIGS. 7A to 7E show the results of classifying the 40 AD patient images by hierarchical clustering using the similarity in the Axial direction (condition 1) as a Z score map. In the group (GA1) in (a), mild ventricular enlargement is observed. In the group (GA2) in (b), there is a slight enlargement of the sulci, but almost no enlargement of the ventricles. In the group (GA3) in (c), mild enlargement of the ventricles and sulci is observed. In the group (d4) (GA4), enlargement of the ventricles and sulci and significant atrophy of the hippocampus are observed. In the group (e5) (GA5), enlargement of the ventricle and sulcus is observed, but the degree of hippocampal atrophy is mild compared to the group (d).

  FIGS. 7F to 7J show the results of classifying the same case by hierarchical clustering using the Coronal direction similarity (Condition 2) on a Z score map. In the group (GC1) of (f), a slight enlargement of the sulcus is seen, but it is close to a normal case. In the group (GC2) of (g), there is a slight enlargement of the sulci, but there is almost no enlargement of the ventricles. Cases of group (GC3) in (h) were considered to have been classified into the same group because brain shape standardization processing did not operate normally because the positioning by SPM did not function well and the position shifted as a whole. It is done. When analyzing the regularity and abnormalities included in the image data, it is necessary to remove cases where the brain morphology standardization process used in the preprocessing does not operate normally. As in this example, if cases where the alignment by SPM did not operate normally are automatically classified by this method, it is easy to remove those cases. In the group (GC4) in (i), enlargement of the sulci and ventricles and significant atrophy of the hippocampus are observed. In the group (GC5) of (j), enlargement of the ventricle is observed, but the degree of sulcus enlargement and hippocampal atrophy are milder than in the group (i).

  As shown in FIG. 7, by displaying the Z score map for each group classified by hierarchical clustering, it is possible to visually understand how much morphological change occurs in which part in each group. be able to.

  Table 1 shows the relationship between groups classified according to the similarity in the Axial direction (condition 1) and the similarity in the coronal direction (condition 2). Table 2 shows the relationship between groups classified according to the Coronal direction similarity (condition 2) and the hippocampal region similarity (condition 3). If it is classified into the same group even if the setting of the region of interest is changed, the values should be collected on a diagonal line. As a result of the experiment, 17/40 cases gathered diagonally in Table 1, and 19/40 cases gathered diagonally in Table 2. From the result that only less than half of the cases were collected on the diagonal line, the classification result of hierarchical clustering differs depending on how the region of interest is set. In other words, when using this method, it is important to set the region of interest in the part of interest, and it is considered that it should be noted.

  Further, regularity and anomaly between image data may be displayed in a scatter diagram based on the degree of similarity. By displaying in a scatter diagram, it is possible to visualize the anomaly that certain image data has.

  FIGS. 8A and 8B respectively show similarity matrices from image data A to D relating to the similarity of (Condition 1) and the similarity of (Condition 2). In the similarity matrix, the similarity between the image data A and the image data B is given to the position of A row and B column. The similarity of the same image data is 1.0, which is a diagonal value. Since the similarity matrix is represented by a symmetric matrix, only the value of the lower triangular matrix is shown. First, looking at the similarity between cases A and B, the similarity value from (condition 1) to FIG. 8 is 0.2, and the similarity value from (condition 2) to FIG. The value is 0.9. By using these values, one point on the two-dimensional space shown in FIG. 8C is determined, and data can be plotted. That is, in FIG. 8C, if the horizontal axis is the similarity of (condition 1) and the vertical axis is the similarity of (condition 2), the relationship between the image data A and B is (0.2, 0.9). Can express. FIG. 8 shows that image data A and image data B have a low similarity in (condition 1) but a high similarity in (condition 2). Hereinafter, when the same operation is repeated, as shown in FIG. 8C, the relationship of the similarity to all combinations of all cases can be displayed in a scatter diagram. As in this example, when the similarity between case A and case B is one, it cannot be plotted in a two-dimensional space. However, if the similarity is calculated in two different parts of case A and case B, the relationship can be plotted in a two-dimensional space by the above method, and the atrophy pattern of different parts can be visualized and compared. By analyzing the scatter plot created by this method, regularity and anomaly included in each image data can be found.

  FIG. 9A shows the result of visualizing the relationship between regularity and anomaly between image data using a scatter diagram. Many cases show regularity with the same atrophy pattern, but A1 and A2 have abnormalities that are different from other atrophy patterns. The A1 case has significantly different characteristics from the other cases in both the Coronal direction similarity (condition 2) and the local region similarity (condition 3). A2 cases also tend to have different characteristics from other cases in local area similarity (condition 3). FIG. 9B shows an enlarged image of these cases. These two cases can show obvious atrophy in the ventricle and hippocampal regions.

  FIG. 10 shows the classification results of all 40 AD cases classified by the method of this example. Hierarchical clustering can automatically perform classification tasks that have been performed manually by doctors based on subjective judgments. Then, by examining the relationship between the gene and the image data, it becomes possible to detect the onset pattern corresponding to the gene type.

  DESCRIPTION OF SYMBOLS 1 Medical assistance apparatus, 3,53 Image pick-up part, 5,55 Reference image data storage part, 7,57 Standardization part, 9,59 Standard image data storage part, 11,61 Subject image data storage part, 13,63 Standard subject image Data storage unit, 15 Subject gene information storage unit, 17 Disease input unit, 19 Combination storage unit, 21 Region of interest setting unit, 23, 69 Analysis unit, 25 Diagnosis result display unit, 27, 73 Healthy subjects, 29 Subjects, 31 , 77 Gene testing device, 51 Image processing device, 61 Onset person image data storage part, 63 Standard onset person image data storage part, 67 Onset person gene information storage part, 71 Candidate display part, 75 Onset person

Claims (7)

A medical support device that sets a region of interest in subject image data obtained by photographing a body tissue of a subject,
A subject gene information storage means for storing information on the gene of the subject;
A region-of-interest setting means for setting the region of interest different corresponding to the type of the gene of a part of the subject;
A medical support apparatus comprising analysis means for analyzing the region of interest of the subject image data. The region of interest setting means, in the gene of the subject,
If there is a specific gene, set the first region of interest,
The medical support device according to claim 1, wherein when there is no specific gene, a second region of interest that is at least partially different from the first region of interest is set. The subject data is obtained by photographing the subject's brain,
The specific gene is APOEε4 gene,
The region of interest setting means includes
If the APOEε4 gene is present, the hippocampus is set as the first region of interest,
The medical support apparatus according to claim 2, wherein when there is no APOEε4 gene, a parietal lobe is set as the second region of interest. Reference image data storage means for storing a plurality of reference image data obtained by imaging a body tissue at the same location as the subject image data;
Standardization means for obtaining standard image data and standard subject image data by setting a plurality of portions common to each of the reference image data and the subject image data,
The medical analysis according to any one of claims 1 to 3, wherein the analysis unit analyzes the difference in the region of interest set by the region of interest setting unit by comparing the standard image data and the standard subject image data. Support device. An image analysis apparatus for obtaining a region of interest candidate set by the medical support apparatus according to claim 1,
Standardization means for obtaining standard onset image data by setting a plurality of sites for each of the onset image data obtained by photographing the body tissues of a plurality of onset individuals who have developed a specific disease,
Analysis target data storage means for storing each onset person and the onset person's genetic information correspondingly;
In the candidate region of interest that is a part of the set part, an extraction means for extracting one or a plurality of the standard onset image data that is not similar to the other standard onset image data;
A specifying means for identifying a gene type included in the gene information corresponding to the extracted standard onset image data and not included in the gene information corresponding to the standard onset image data not extracted as a candidate gene type; ,
An image analysis apparatus comprising candidate display means for displaying a combination of the candidate region of interest and the candidate gene type as a candidate for the combination of the region of interest and the gene type used by the region of interest setting means. A medical support method in a medical support device that analyzes image data obtained by photographing a body tissue,
The medical support device includes:
Reference image data storage means for storing a plurality of reference image data obtained by imaging the same part of the body tissue;
Subject image data storage means for storing subject image data obtained by imaging the same location as the reference image data of the subject's body tissue;
A subject gene information storage means for storing information of the subject's gene;
The standardization means provided in the medical support device sets a plurality of portions common to the respective reference image data to obtain standard image data, and the same plurality of the same as the standard image data for the subject image data A standardization step of setting the site of and obtaining standard subject image data;
A region-of-interest setting step in which the region-of-interest setting means included in the medical support device sets the region of interest that differs depending on the type of the gene of a part of the subject; and
The medical support method includes an analysis step in which the analysis unit included in the medical support device analyzes the difference in the region of interest set by the region of interest setting unit by comparing the standard image data and the standard subject image data.   The program for functioning a computer as a medical assistance apparatus in any one of Claim 1 to 4.