JP2011118540A - Case retrieval device, method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To comprehensively list up diseases in case retrieval. <P>SOLUTION: For disease classifications in which case images are not registered in a case table, the corresponding typical feature quantity is registered in a disease table. The disease classifications corresponding to the feature quantity of the case images similar to the feature quantity of an optional query image and the disease classifications corresponding to the typical feature quantity are respectively retrieved from the case table and the disease table, and thus the disease classifications regarding the query image can be comprehensively retrieved. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a technique for searching for a case related to a diagnosis target image based on a feature amount extracted from the diagnosis target image.

  In Patent Document 1, the image data input means inputs ROI image data P ′ set in the subject image, and the search means has a portion whose image characteristics are similar to the input ROI image data P ′. The similar image data having and the diagnostic data related to the similar image data are retrieved from the case database and read. The search means determines the similarity for the similarity determination item selected by the selection means and performs a search. The output means outputs search result information representing the search results by the search means to the display means, and the display means displays the search results on the monitor.

  Patent Document 2 is an example of a database search system that stores an electronic medical book. In general, an electronic medical book is content that can be viewed on a computer by digitizing the medical book. Each case data in the case database may be associated with data such as hyperlinks that can refer to information in related electronic medical books.

  Patent Documents 3 and 4 are examples of a hierarchical medical database.

  Patent Document 5 is an example in which medical information is classified and registered in a database.

  Patent Document 6 is an example of a technique for extracting a lesion area based on position information of an attention area.

  Patent Document 7 generates a control command for a capsule endoscope based on case information including image data of a lesion having a characteristic amount such as a typical shape, color, and size. It is assumed that the case information is representative data selected from a huge amount of data cultivated by past diagnosis.

  According to Patent Document 8, the reference image preparation support device includes a database that stores a plurality of medical images, an image collection device that inputs examination medical images to be examined, and each of the examination medical images and the plurality of medical images. A workstation that detects a type of abnormality for the medical image, selects a medical image of the same type of abnormality as the examination medical image, and displays the selected medical image together with the examination medical image. Assume that past images and typical case images to be referred to can be prepared and prepared for interpretation.

  Non-Patent Document 1 is an example of a technique for calculating an image feature amount in a region of interest.

JP 2004-5364 A JP 2004-118566 A JP 2007-105461 A JP 2004-185330 A JP 2004-334647 A JP 2008-245719 A JP 2008-237639 A JP 2001-187044 A

Couts, Edwards, Taylor, "Dynamic Appearance Model", 5th Computer Vision European Conference Report, Germany, Springer, 1998, Vol. 2, 484-498 (TF Coootes, GJ Edwards, CJ Taylor “Active Appearance Models” , In Proc. 5th European Conference on Computer Vision, Springer)

  When there are many cases that are not registered in the case DB, or there are diseases that do not have images in the case DB, such as rare cases that do not have images even though many cases are registered, the diseases that should exist comprehensively are similar. There is a problem that it is not listed in image search. When a diagnosis is performed with reference to the result of the similar image search, if there is a missing disease, the diagnosis result may be output without considering the possibility of the disease missing from the search result. An object of the present invention is to comprehensively list diseases by case search.

  The case search apparatus according to the present invention includes a query image input unit that inputs a query image, a query image feature amount calculation unit that calculates a feature amount of the query image input by the query image input unit, and a feature of the case image and the case image A first table for storing the amount and the disease classification in association with each other, a non-image disease classification for which the case image is not stored in the first table, and a typical feature amount corresponding to the disease classification The feature amount of the first table similar to the feature amount of the query image by comparing the feature amount of the query image calculated by the query image feature amount calculating unit with the second table and the feature amount of the first table. And a comparison between the feature quantity of the query image and the typical feature quantity of the second table to search for a second table similar to the feature quantity of the query image. A first search unit that searches for a second similar feature value that is a typical feature value of the message, a first output unit that outputs a case image and a disease classification corresponding to the first similar feature value in the first table, And a second output unit that outputs an imageless disease classification corresponding to the second similar feature amount in the table of 2.

  Preferably, the calculation unit that calculates the feature amount of the input case image by inputting the case image, the case image, the feature amount of the case image calculated by the calculation unit, and the disease classification of the case image are associated with each other. A first registration unit for registering in the table.

  Preferably, the image processing apparatus includes a second registration unit that inputs an imageless disease classification and a typical feature amount and registers them in the second table.

  Preferably, the image processing apparatus includes an imageless disease classification detection unit that detects a disease classification in which a case image is not registered, and the second registration unit selects a disease classification in which the case image detected by the imageless disease classification detection unit is not registered. The corresponding typical feature value is input, and the input typical feature value and the disease classification are associated with each other and registered in the second table.

  Preferably, the imageless disease classification detection unit registers a flag indicating a disease classification in which it is detected that a case image is not registered in association with the disease classification.

  Preferably, the imageless disease classification detection unit updates the flag corresponding to the case image every time the case image in the first table is added or deleted by the first registration unit.

  Preferably, the search unit compares the feature amount of the query image with the feature amount of the first table in accordance with a selection instruction input from the operation device, so that the first table is similar to the feature amount of the query image. Similar to the feature amount of the query image by comparing the feature amount of the first search pattern and / or query image for searching for the first similar feature amount that is the feature amount of the second table with the typical feature amount of the second table A second search pattern for searching for a second similar feature amount that is a feature amount of the second table is executed.

  Preferably, the search unit compares the feature amount of the case image of the predetermined electronic medical book database with the feature amount of the query image, thereby comparing the feature amount of the case image of the electronic medical book database similar to the feature amount of the query image. The first output unit outputs a case image and a disease classification corresponding to the third similar feature in the electronic medical book database, and the second output unit The disease classification corresponding to the third similar feature is output in the medical book database.

  The first output unit starts or stops the output of the case image in accordance with an instruction to the operating device.

  The case search method according to the present invention includes a step in which a computer inputs a query image, a step of calculating a feature amount of the input query image, a case image, the feature amount of the case image, and a disease classification in association with each other. A step of storing in the table of 1, a step of storing in the second table the image-less disease classification that is a disease classification for which no case image is stored in the first table and a typical feature amount corresponding to the disease classification; By comparing the calculated feature amount of the query image with the feature amount of the first table, a first similar feature amount that is a feature amount of the first table similar to the feature amount of the query image is searched, and By comparing the feature quantity of the query image with the typical feature quantity of the second table, the second similar feature that is the typical feature quantity of the second table similar to the feature quantity of the query image. A step of outputting a case image and disease classification corresponding to the first similar feature amount in the first table, and an image-less disease classification corresponding to the second similar feature amount in the second table. And a step of outputting.

  A case search program for causing a computer to execute this case search method is also included in the present invention.

  According to the present invention, even for a disease classification without a case image, the disease classification related to the query image can be searched based on the corresponding typical feature amount, and the disease classification search range and result presentation range become comprehensive.

Schematic configuration diagram of case image retrieval system The figure which shows an example of a case table The figure which shows an example of a disease table The figure which shows an example of the classification structure of case DB Figure showing output example of search results

  FIG. 1 is a schematic configuration diagram of a case image search system 100 according to a preferred embodiment of the present invention. This system includes a search system 101 and a registration system 102. The search system 101 includes a feature amount calculation unit 1, a similarity search unit 2, an image presence / absence determination unit 3, an image display unit 4, a disease display unit 5, and a case database (DB) 10. The registration system 102 includes a case registration unit 7, a feature amount calculation unit 8, an imageless disease registration unit 9, a case database (DB) 10, and a flag control unit 11.

  The case image retrieval system 100 may be configured by a computer (a circuit including a CPU, a RAM, a ROM, and other circuits necessary for arithmetic processing, a data storage medium, a data input / output circuit, a display circuit, an operation device, a communication circuit, etc.). it can. Each block may be configured integrally with one computer, but each block is configured by a single computer or logic circuit, and these blocks are connected by data communication means such as a network or a system bus. A system may be configured. The feature amount calculation unit 1, the similarity search unit 2, the image presence / absence determination unit 3, the case registration unit 7, the feature amount calculation unit 8, the no-image disease registration unit 9, and the flag control unit 11 are executed by a computer and the computer. A program module (stored in a computer-readable recording medium such as a ROM) can be substituted, and does not necessarily require a specific hardware configuration. The image display unit 4 and the disease display unit 5 can be configured by a display display, and the case DB 10 can be configured by a large-capacity storage medium such as a hard disk.

  The case DB 10 stores a case table and a disease classification table. 2 shows information on the case table, and FIG. 3 shows an example of information on the disease classification table. In the figure, the case table and the disease classification table are shown separately, but both may be merged into the same table.

  The case table stores a case image to be a search target image and accompanying information corresponding to each case image. Here, the case images are classified according to disease (for example, according to disease name code corresponding to ICD10) and findings (characteristics of lesions) according to disease classification information included in corresponding incidental information, and case DB10. Stored in

  In the case DB 10, statistical information, disease information, and medical information of electronic medical books (such as disease information of a disease that is easily mistaken for a specific disease and a case image related to the disease) are linked by hyperlinks or the like. ing. The link information can be included in the accompanying information. One or more pieces of medical information may be linked to one case image. The electronic medical book may be stored in the case DB 10 or may be stored in a database different from the case DB 10 (see, for example, Patent Document 2).

  The accompanying information can include case information. Case information includes imaging diagnosis name, pathological diagnosis name, findings (infiltrative shadow, tumor, honeycomb, etc.), imaging modality information such as “CR” and “CT”, patient name, age, sex, etc. Includes patient-related information indicating personal information, registrant (name of doctor in charge), date of case conference, imaging part information indicating imaging part, text-based diagnostic information such as diagnostic report created by interpretation doctor .

The statistical information for each disease includes, for example, the following information.
・ Similarity in representative cases (highest similarity in disease)
-Average similarity-Total number of registrations of relevant diseases in the case DB-Number of cases (case pattern) registered near the representative case-Characteristics in the case pattern ... Patient characteristics in the main symptoms and case patterns ... Average age, medical history, Smoking history, etc., national / regional morbidity, etc. In the case table, the feature amount (second feature amount) extracted from the site of each case image by the feature amount calculation unit 8 is stored. However, if the case image itself is stored in the case DB 10, the second feature amount can be obtained by analyzing the case image later by the feature amount calculation unit 8 or the like. It is not always necessary to register the feature amount of the present invention when implementing the present invention. The case image is input from the case registration unit 7 to the feature amount calculation unit 8. The feature amount calculation method of the feature amount calculation unit 8 is the same as that of the feature amount calculation unit 1.

  The case registration unit 7 registers information in the case table. The case registration unit 7 inputs a case image from a desired medical imaging device. Case images registered in the case table by the case registration unit 7 are various medical imaging apparatuses (CT apparatus, MRI apparatus, PET apparatus, X-ray apparatus, US apparatus, endoscope, mammography apparatus, microscope, and the like as the query image. For example). The accompanying information registered in the case table by the case registration unit 7 can be input from a medical imaging device, an operation device, an information terminal connected via a network, or the like.

  As illustrated in FIG. 3, the disease table stores the feature amount input by the imageless disease registration unit 9 and the accompanying information in association with each other. This accompanying information includes disease classification. The disease table includes an image presence / absence flag indicating a disease classification not associated with any case image in the case table or a disease classification associated with at least one case image in the case table.

  In the case table illustrated in FIG. 2, the disease classification “classification B” and “classification D” are not associated with any case image. For this reason, the disease table stores an image presence flag “none” corresponding to “classification B” and “classification D”. Alternatively, in the case table, the disease classifications “class A”, “class C”, and “class E” are associated with some case image. For this reason, image presence / absence flags “present” corresponding to “class A”, “class C”, and “class E” are stored.

  The flag control unit 11 displays the contents of the image presence / absence flag in the disease table in conjunction with the registration of the case image corresponding to the desired disease classification by the case registration unit 7 and the deletion from the case table. Update.

  The disease table stores the typical feature quantity of the case image in the disease classification in association with the disease classification of the image presence flag “none”. This typical feature amount is input from the no-image disease registration unit 9. This typical feature amount is included in the second feature amount that the similarity search unit 2 compares with the first feature amount of the query image. Note that the disease table may store a typical feature amount of a case image in the disease classification in association with the disease classification of the image presence flag “present”. This is because even if the image presence / absence flag is “present”, it may change to “absent” due to the deletion of the case image.

  In FIG. 3, typical features corresponding to each of “class A”, “class C”, and “class E” corresponding to the image presence flag “present”, and a disease classification “class B” corresponding to the image presence flag “absent”. And the typical feature amount corresponding to each of “class D” is stored, but corresponds to each of “class A”, “class C”, and “class E” corresponding to the image presence flag “present”. There may be no typical feature amount.

  Further, incidental information other than disease classification, for example, link information to an electronic medical book may be associated with the typical feature amount. The disease table is used to provide information related to a disease classification in which case images are not registered in the case table. However, not only the case table but also a feature quantity related to a disease classification in which case images are not registered in the database of electronic medical books may be stored. That is, the disease table can be applied to an electronic medical book without a case image.

  Since the typical feature amount is merely a numerical value and does not include the personal information of the patient, it is considered that the degree of restriction of disclosure by other hospitals and research facilities is low, and is easier to obtain than the case image.

  Individual disease classifications of individual accompanying information included in the case table and the disease table correspond to logical storage units provided in the case DB 10 such as directories, folders, and partitions. This logical storage unit can take a hierarchical structure (tree structure) (see FIG. 4).

  Classification may be performed by storing a data set corresponding to each disease classification in a physical storage unit. An individual logical storage unit may be provided for each physical storage unit. For example, a database of electronic medical books is constructed in a physical or logical storage unit different from the case DB 10, and a hierarchical structure different from the case DB 10 can be constructed therein.

  The timing for registering data in the case table by the case registration unit 7 and the timing for registering the typical feature amount and accompanying information in the disease table by the disease registration unit 9 are arbitrary. For example, the flag control unit 11 first sets all the image presence / absence flags corresponding to each disease classification in the disease table to a default value of “none”, and then permits the case registration unit 7 to register data in the case table. To do. In response to this, the user registers a desired case image and accompanying information in the case registration unit 7. This registration is performed based on a user instruction, but the source of data to be registered is arbitrary. When a predetermined time has elapsed since the registration of the case image and accompanying information in the case table by the case registration unit 7 but the image presence flag corresponding to a certain disease classification is still “none”, the disease registration unit 9 causes the disease table Registration of typical cases and associated information corresponding to the disease classification is started.

  Alternatively, the disease registration unit 9 first registers the comprehensive disease classification and the typical feature amount in the disease table, and the image presence / absence flag corresponding to them registers the default value “none”. Thereafter, data registration in the case table by the case registration unit 7 may be sequentially performed, and the image presence flag of the disease classification corresponding to the case image registered in the case table may be updated from the default value “none” to “yes”.

  The start of data registration in the disease table may be fully automatic or semi-automatic. For example, the flag control unit 11 detects the number of registered case images corresponding to each disease category from the case DB 10 periodically (for example, every other week, every other month, etc.), and the disease category with the zero registered number is detected. If a case image of the disease classification is zero, a warning that the collection of typical features is desirable is output with a display message or a beep sound. In response to this, the no-image disease registration unit 9 acquires a typical feature amount from an arbitrary source designated by the user using the operation device, associates the disease category with the acquired typical feature amount, and stores it in the disease table. Store. The source of the accompanying information registered in association with the typical feature amount is arbitrary, and examples thereof include an operation device, a recording medium reading device, and a server on a network.

  When the case registration unit 7 registers a new case image and accompanying information in a disease category that has never been registered, the flag control unit 11 changes the image presence / absence flag corresponding to the disease category from “None” to “Yes” Update to When the disease registration unit 9 registers a new typical feature amount in a disease category that has never been registered, the flag control unit 11 keeps the image presence flag corresponding to the disease category “None”. The case registration unit 7 may delete a desired case image and accompanying information from the case table in response to an instruction from the user. If all the case images corresponding to a certain disease classification are deleted, the flag control unit 11 updates the image presence / absence flag corresponding to the disease classification from “present” to “absent”. In response to the image presence / absence flag corresponding to a specific disease classification being updated from “Yes” to “None”, the disease registration unit 9 starts data registration corresponding to the disease classification or a warning prompting the start To confirm the user's intention, and if there is an instruction, data registration may be started.

  The source of the typical feature data to be registered in the disease registration unit 9 is arbitrary, and includes an operation device, an information terminal connected via a network, a CDROM, a DVD or other storage medium reader, or a database of electronic medical books (however, case images The data can be input and registered from a record in which all or a part of a record storing only the feature quantity is included. The typical feature amount corresponding to a certain disease classification is a standard feature amount of an image corresponding to the disease classification, but what is used as the typical feature amount is arbitrary. For example, a value obtained by statistically processing image features (simple average, weighted average, median, etc.) of an image corresponding to a certain disease classification collected at academic societies or other hospitals is a typical feature corresponding to the disease classification. To do. Alternatively, a typical image is selected from the collected image samples by manual operation, and the feature amount extracted from the image by the feature amount calculation unit 8 is used as a typical feature amount corresponding to the disease classification as it is.

  The feature amount calculation unit 1 is a diagnosis target image (query image) such as a CT image, an MRI image, a PET image, an X-ray image (including a CR image), a US (ultrasound) image, an endoscopic image, a mammography, or a pathological image. Enter. The feature amount calculation unit 1 can also input these query images and incidental information such as the size of the query image and the imaging date and time and case information. The communication protocol between the feature quantity calculation unit 1 and the medical image diagnostic apparatus, and the format of the query image and incidental information comply with a predetermined standard such as DICOM (Digital Imaging and Communication in Medicine).

  The feature amount calculation unit 1 analyzes the input query image and creates a feature amount (first feature amount) necessary for searching for case images or disease classifications accumulated in the case DB 10.

  The feature amount calculated by the feature amount calculation unit 1 is a feature amount related to pixel values such as an average, variance, maximum value, minimum value, and luminance histogram of image data, a position of a region of interest, a circularity or moment of an outline, and a radius of a cross section. In addition to the feature amount related to the shape such as volume and area, the shape information and texture information of the lesioned part of the ROI (region of interest) obtained by the method of Non-Patent Document 1 may be used. Alternatively, a value obtained by processing information once registered in the case DB 10 with a predetermined mathematical formula (for example, principal component analysis, independent component analysis) may be used as the feature amount. In short, the outline / pixel value, position information, etc. of an image obtained by a known method can be used as the feature amount.

  Further, as a preprocessing for calculating the feature amount, a lesion area of the image may be specified, and the feature quantity may be calculated for the lesion area. As a specific example of a lesion area, there is a method of extracting an area based on position information of an attention area as in Patent Document 6.

  In addition, the identification of a lesion area is not necessarily fully automatic, and semi-automatic image processing technology using a feature amount (a feature amount related to a pixel value or a feature related to a shape such as a contour) in the vicinity of a region designated by the user on the operation device. May be extracted automatically. In this way, by using a feature amount specialized for a lesion site, a case image similar to the focused lesion site can be searched.

  The similarity search unit 2 compares the first feature amount calculated by the feature amount calculation unit 1 with the feature amount of the case table accumulated in the case DB 10 and the feature amount of the disease table (second feature amount). The similarity is calculated. The second feature amount is created and stored in advance according to the same technique as that for the query image.

  The similarity search unit 2 calculates the similarity between the first feature value and the second feature value of the case table according to the selection from the operation device (pattern 1), or the first feature value and the disease table The degree of similarity between the first feature amount and the second feature amount of the case table is calculated and the first feature amount is calculated (pattern 2). And the second feature quantity of the disease table are calculated (pattern 3), the first feature quantity and the second feature quantity of the case table, the second feature quantity of the disease table, and the electronic medical book Any one of extraction from the case images in the database or calculation of the similarity between the feature amounts registered in the database (pattern 4) may be executed. Pattern 1 requires a search for a case image and does not require a search for a typical case without an image. Pattern 2 does not require a search for a case image and requires a search for a typical case without an image. When a search for a typical case is necessary, the pattern 4 may be selected when a case image of a medical book is desired to be searched.

  As a method for calculating the similarity, a known method, for example, a difference between feature value values, a least square method on a feature amount space (a weighted space is also possible), or the like can be employed. For convenience of explanation, the similarity S is defined by the following mathematical formula. However, the standard for calculating the similarity S is not limited to this definition in order to implement the present invention.

The feature quantity _mi (i = 1, 2,..., N) is extracted from the feature area of the attention area of the registered case image, and M _i (i = 1, 2,..., N) is extracted from the attention area of the diagnostic image. Feature amount. w _i (i = 1, 2,..., n) is a weighting coefficient for each feature amount defined in advance for each disease. A subscript _i of m _i , M _i , and w _i indicates a coordinate axis on the feature amount space. The closer the two feature quantities are, the smaller the value of the similarity S is, and the farther the feature quantities are, the larger the value of the similarity S is.

  The similarity search unit 2 has a predetermined threshold value of the similarity S to the feature amount information of the query image among all the case images stored in the case DB 10 or the case images corresponding to the associated search items designated in advance from the operation device. A higher second feature amount (similar feature amount) is specified. The accompanying search items are information that can be arbitrarily specified by the user. For example, the imaging region (chest, breast, etc.), anatomical position information (liver right lobe, lung S1, etc.), imaging modality information, patient related information, image diagnosis Information that can narrow down case images by matching with incidental information in the case DB 10 such as name, pathological diagnosis name, findings (infiltrating shadow, tumor, honeycomb cell, etc.), imaging modality information, doctor name in charge, and the like.

  The image presence / absence determination unit 3 outputs a similar image, which is a case image corresponding to the similar feature amount specified by the similarity search unit 2 in the case table, to the image display unit 4 and accompanying information (disease classification) corresponding to the similar image. And the like are output to the disease display unit 5.

  Alternatively, with reference to the image presence / absence flag included in the accompanying information corresponding to the similar feature amount specified by the case similarity search unit 2, if the image presence / absence flag indicates “Yes”, the similar feature amount specified from the case table A similar image, which is a case image corresponding to, may be output to the image display unit 4, and accompanying information corresponding to the similar feature amount may be output to the disease display unit 5.

  The image display unit 4 and the disease display unit 5 may be configured by the same display device, but their display areas are distinguished.

  The disease display unit 5 includes disease information included in the accompanying information, for example, case information (image diagnosis name, pathological diagnosis name, findings, etc.) and medical information (electronic medical book) included in the electronic medical book linked by the case information. Disease name, pathogen, symptom, classification by affected area, etc.). The display item may be selected with an operation device such as a keyboard or a mouse.

  The disease display unit 5 can be replaced with output means other than the video display means. For example, the output unit may be a network I / F, and information may be output to a desired communication terminal (such as a personal computer, a mobile phone, or a PDA). Alternatively, the information may be output by the printing unit together with the video display unit. Further, information may be output by outputting a synthesized voice from the voice output means.

  The image presence / absence determination unit 3 refers to the image presence / absence flag included in the accompanying information corresponding to the similar feature amount specified by the similarity search unit 2 and is specified from the disease table if the image presence / absence flag indicates “none”. The accompanying information corresponding to the similar feature amount is output to the disease display unit 5.

  That is, the similar image is displayed on the image display unit 4 only when the similar image exists, while the accompanying information is displayed on the disease display unit 5 regardless of whether the similar image is searched or not ( However, except when pattern 1 is selected).

  FIG. 5 shows an example of a similar image corresponding to the similar feature amount displayed by the image display unit 4 and its accompanying information (disease classification), and accompanying information (disease classification) corresponding to the similar feature amount displayed by the disease display unit 5. Show. This is output as a result screen of a similar case search by the similar search unit 2. A similar image and disease classification (image diagnosis name, pathological diagnosis name, findings, etc.) are displayed for a disease classification with a case image, and only a disease classification is displayed for a disease classification without a case image.

  The display / non-display of the similar image may be switched according to an instruction from the operation device. If non-display of a similar image is instructed, only the disease name is displayed regardless of the presence or absence of the similar image.

  Further, information other than the disease name included in the accompanying information, for example, a link to medical information may be displayed, and the corresponding medical information may be displayed by designating the link.

  As described above, even for disease classifications in which case images are not registered in the case table in the first place, typical feature amounts corresponding to the disease classifications are registered in the disease table, and case images similar to the feature amounts of any query image The disease classification corresponding to the feature amount and the disease classification corresponding to the typical feature amount are searched from the case table and the disease table, respectively, and the disease classification related to the query image can be comprehensively searched.

  1: feature amount calculation unit, 2: similarity search unit, 3: image presence / absence determination unit, 4: image display unit, 5: disease display unit, 7: case registration unit, 8: feature amount calculation unit, 9: disease without image Registration unit, 10: case database, 11: flag control unit

Claims (11)

A query image input unit for inputting a query image;
A query image feature amount calculation unit for calculating a feature amount of the query image input by the query image input unit;
A first table that stores a case image, a feature amount of the case image, and a disease classification in association with each other;
A second table that stores an imageless disease classification that is a disease classification in which no case image is stored in the first table and a typical feature amount corresponding to the disease classification;
The feature amount of the first table that is similar to the feature amount of the query image by comparing the feature amount of the query image calculated by the query image feature amount calculation unit with the feature amount of the first table. The second table representative similar to the feature amount of the query image is obtained by searching for the first similar feature amount and comparing the feature amount of the query image with the representative feature amount of the second table. A search unit for searching for a second similar feature quantity that is a feature quantity;
A first output unit that outputs a case image and a disease classification corresponding to the first similar feature amount in the first table;
A second output unit that outputs an imageless disease classification corresponding to the second similar feature amount in the second table;
A case retrieval apparatus comprising: A calculation unit that inputs the case image and calculates a feature amount of the input case image;
A first registration unit that registers the case image and the feature amount of the case image calculated by the calculation unit and the disease classification of the case image in association with each other in the first table;
The case search device according to claim 1, comprising:   The case search device according to claim 2, further comprising a second registration unit that inputs the image-less disease classification and the typical feature amount and registers them in the second table. With an imageless disease classification detection unit that detects a disease classification that is not registered in a case image,
The second registration unit inputs a typical feature amount corresponding to an unregistered disease category of the case image detected by the no-image disease category detection unit, and associates the input typical feature amount with the disease classification The case search device according to claim 3, wherein the case search device is registered in the second table.   5. The case retrieval apparatus according to claim 4, wherein the no-image disease classification detection unit registers a flag indicating a disease classification in which it is detected that the case image is not registered in association with the disease classification.   6. The case search according to claim 5, wherein the imageless disease classification detection unit updates a flag corresponding to the case image every time a case image in the first table is added or deleted by the first registration unit. apparatus.   The search unit compares the feature amount of the query image with the feature amount of the first table in accordance with a selection instruction input from an operating device, and thus the first similarity similar to the feature amount of the query image. By comparing the first search pattern for searching for the first similar feature quantity that is the feature quantity of the table and / or the feature quantity of the query image with the typical feature quantity of the second table, the query image The case search device according to claim 1, wherein a second search pattern for searching for a second similar feature amount that is a feature amount of the second table similar to the feature amount of the second table is implemented. The search unit compares a feature amount of a case image in a predetermined electronic medical book database with a feature amount of the query image, thereby resembling the case image feature of the electronic medical book database similar to the feature amount of the query image. Search for a third similar feature quantity,
The first output unit outputs a case image and a disease classification corresponding to the third similar feature amount in the electronic medical book database,
The case search device according to claim 1, wherein the second output unit outputs a disease classification corresponding to the third similar feature amount in the electronic medical book database.   The case search device according to any one of claims 1 to 8, wherein the first output unit starts or stops outputting the case image in response to an instruction to the operation device. Computer
Entering a query image;
Calculating a feature amount of the input query image;
Storing the case image, the feature amount of the case image, and the disease classification in association with each other in the first table;
Storing, in the second table, an imageless disease classification that is a disease classification in which no case image is stored in the first table, and a typical feature amount corresponding to the disease classification;
By comparing the calculated feature amount of the query image with the feature amount of the first table, a first similar feature amount that is a feature amount of the first table similar to the feature amount of the query image is obtained. A second characteristic value of the second table similar to the characteristic value of the query image is obtained by searching and comparing the characteristic value of the query image with the typical characteristic value of the second table. Searching for similar features,
Outputting a case image and a disease classification corresponding to the first similar feature amount in the first table;
Outputting an imageless disease classification corresponding to the second similar feature amount in the second table;
Search method for cases.   A case search program for causing a computer to execute the case search method according to claim 10.