JP5547597B2 - Pathological diagnosis support device, pathological diagnosis support method, control program for pathological diagnosis support, and recording medium recording the control program - Google Patents

Description

  The present invention relates to a technique for supporting pathological diagnosis by a computer system or the like.

  One pathological diagnosis is biopsy tissue diagnosis in which a part of an organ is collected from a patient as a specimen, and this is made into a tissue slice glass slide (preparation) and observed under a microscope for diagnosis. Such a diagnosis itself must be performed by a doctor, but the amount of work for that is enormous and the burden on the doctor is heavy. Therefore, in order to efficiently carry out observation work for diagnosis and reduce the burden on the doctor, various diagnosis support apparatuses using image processing technology by a computer system have been proposed.

  For example, in the technique described in Patent Document 1, a digital image is generated by enlarging and photographing a mammary tissue specimen with a microscope or the like, and a histopathological feature amount (for example, a nuclear region) found in a mammary tumor is detected from the image. Number, area, shape, presence / absence of papillary pattern, etc.). Based on the feature amount, a diagnosis category set in advance based on histopathological features is determined as a fitness level calculation category, and the degree to which the feature value of the image matches the fitness level calculation category is calculated. Diagnosis support is performed by displaying a fitness level calculation category name having a high fitness level.

Japanese Patent Laid-Open No. 10-197522 JP 2003-126045 A

  In general, a pathologist performs pathological diagnosis on a plurality of types of organs. As diagnostic criteria, there are diagnostic criteria common to organs and diagnostic criteria specialized for each organ. The development of diagnostic support devices that support this is specialized for each target organ or lesion. In the diagnostic support system that has only a support function specialized for a specific organ like the above-mentioned conventional technology, There is a problem that a doctor's pathological diagnosis cannot be sufficiently supported, for example, the operation method of the system changes every time the organ to be changed.

  On the other hand, as described in Patent Document 2, for example, a technique for providing a diagnosis support function specialized for each modality (for example, a computer tomography apparatus, an endoscope apparatus, etc.) in a plug-in format has been proposed. However, with this technique, there is a problem that it is impossible to grasp the tendency of the characteristic features of lesions that are commonly used as diagnostic criteria among organs that are important in pathological diagnosis.

  Thus, so far, diagnostic support systems that can handle different types of organs have not yet been put into practical use, and effectively support the diagnosis of pathologists who diagnose multiple types of organs. The technology that can be done has not been established.

  The present invention has been made in view of the above problems, and in the technical field of supporting pathological diagnosis by a doctor by a computer system or the like, it is possible to support pathological diagnosis more effectively by supporting multiple types of organs. The purpose is to provide.

In order to achieve the above object, a pathological diagnosis support apparatus according to the present invention obtains an image obtained by capturing an image of a specimen tissue, and a common feature quantity that is a feature quantity that does not depend on the type of organ based on the image. A common feature amount calculating means for calculating the image, a common classification means for classifying the image into a plurality of classification categories by a classifier machine-learned using a sample image given in advance based on the common feature amount, Based on the image, specialized feature quantity calculating means for calculating a specialized feature quantity that is a feature quantity specialized for the organ to be examined, and in advance according to the value of the specialized feature quantity and the specialized feature quantity Based on a comparison result with a predetermined threshold, specialized detection means for detecting a special region peculiar to an organ from the image, classification result by the common classification means, and detection result by the specialized detection means are integrated. Out And output means for outputting an image, the specialized feature quantity calculating means, is characterized in that it is interchangeable depending on the type of organ.

Further, the pathological diagnosis support method according to the present invention calculates an image acquisition step of acquiring an image obtained by imaging a sample tissue, a common feature amount that is a feature amount independent of the type of organ, based on the image, A common classification step of classifying the image into a plurality of classification categories by a classifier that has been machine-learned using a sample image given in advance based on the common feature amount, and an organ selection step of selecting and setting an organ to be examined And a specialized feature value that is a feature value specialized for the organ set as the examination target is calculated based on the image, and predetermined according to the value of the specialized feature value and the specialized feature value. a specialized detection step of detecting a specific special area in organs from the image based on the obtained and compared with the threshold value results, by integrating the classification result at the common classification step, and a detection result of said specialized detection step And an output step of outputting a force image, and in the specialized detection process is characterized by varying the type of the specific feature amount according to the type of the set organs.

In these inventions, when classifying images according to their feature values, elements common to organs and elements different for each organ are distinguished and handled. In addition, regarding elements that are different for each organ, it is possible to replace the types of feature amounts calculated according to the organ to be examined. Therefore, according to the present invention, it is possible to handle a plurality of types of organs in a single diagnosis support apparatus and method, and it is possible to support pathological diagnosis by a pathologist more effectively. In addition, it is possible to share functions, operation procedures, and the like that do not depend on the type of organ among organs, and it is possible to reduce an increase in the burden on pathologists due to the change or addition of an examination target organ. Here, the common classification means classifies images using a classifier by machine learning using sample images. Since the size and shape of cells vary greatly among organs and individuals, classification based on feature quantities can be performed with high accuracy by performing classification using a learning algorithm using a typical sample image. On the other hand, when detecting a special region peculiar to an organ, detection is performed using a detector not based on machine learning. In the field of pathology, features on images appearing in biopsy tissues have been studied in detail for each organ or its lesion, and many organs have established clear diagnostic criteria such as the presence or absence of certain features and the number of detections. . By providing the situation according to the diagnostic criteria for such an organ, the pathologist can connect to a highly accurate diagnosis by combining experience and quantitative data results.

  In the above-described pathological diagnosis support apparatus, for example, the specialized detection unit may be interchangeable with the specialized feature amount calculation unit. By doing so, the calculation of the feature amount and the detection of the special region based on the feature amount can be optimized for each organ, and the diagnosis support can be performed more efficiently.

  Further, the common feature amount calculating unit and the specialized feature amount calculating unit calculate the feature amount for each of a plurality of grid images obtained by dividing the image, and the common classification unit and the specialized detection unit classify the grid image. You may do it. By classifying the acquired images into a plurality of grid images and classifying them in units of grid images, more effective diagnosis support can be performed, for example, by clarifying the distribution status of lesion sites in the images.

  Here, the common features include the nuclear density in the image, the concentration of the nucleus, the arrangement of the nucleus, the size of the nucleus, the shape of the nucleus, the area ratio between the cytoplasm and the nucleus in the image, the nuclear variant, and the size of the nucleus. At least one of the feature amounts corresponding to each of the above can be used. These elements are effective for determining the presence or absence of a lesion in an image regardless of the type of organ. Therefore, by making the feature quantities corresponding to these common feature quantities, it becomes possible to efficiently support diagnosis based on a common diagnostic criterion between organs.

  The output unit may be configured to include a display unit that displays the classification result by the common classification unit and the detection result by the specialized detection unit in an integrated manner. In this way, the user can perform a comprehensive diagnosis based on a diagnosis standard common to organs and a diagnosis standard unique to the organ, and the effect of diagnosis support by the apparatus can be further improved.

Further, in order to achieve the above object, the control program for pathological diagnosis support according to the present invention calculates a common feature amount that is a feature amount independent of the type of organ based on an image obtained by imaging a specimen tissue, Based on the common feature amount, a common classification step of classifying the image into a plurality of classification categories by a classifier that has been machine-learned using a sample image given in advance , and an organ to be examined based on the image A special feature amount that is a special feature amount is calculated from the image to the organ based on a result of comparing the value of the special feature amount and a threshold value determined in advance according to the special feature amount. a specialized detection step of detecting a specific special area, the classification result at the common classification step, to execute an output step of outputting an output image obtained by integrating the detection results of the specific detection step in a computer, Duck, and in the specialized detection process is characterized by varying the type of the specific feature amount according to the type of the set organs. Furthermore, a recording medium according to the present invention is a computer-readable recording medium, and is characterized by recording a control program for supporting pathological diagnosis described above. According to these inventions, the pathological diagnosis support method according to the above invention can be executed by a computer, and diagnosis by a pathologist can be effectively supported.

  In this case, the program for executing the specialized detection step may be configured as a plug-in file for the control program body. In this way, it is possible to cope with a plurality of types of organs by appropriately applying a plug-in file corresponding to the organ to be examined and executing the control program. Even in this case, since many functions can be made the same regardless of the selected organ, it is possible to reduce an increase in the burden on the pathologist accompanying the change of the organ to be examined. In addition, a new function can be added by updating or adding a plug-in file, and a diagnostic support system with excellent extensibility can be constructed.

  According to the pathological diagnosis support technique according to the present invention, it is possible to handle a plurality of types of organs in a single diagnosis support apparatus and method, thereby providing statistical data of diagnostic criteria used in common between organs. Therefore, pathological diagnosis by a pathologist can be more effectively supported. In addition, by sharing functions and operation procedures that do not depend on the type of organ among organs, it is possible to reduce an increase in the burden on the pathologist accompanying the change or addition of the organ to be examined.

It is a figure which shows one Embodiment of the pathological diagnosis assistance apparatus concerning this invention. It is a block diagram which shows the structure of this pathological diagnosis assistance apparatus. It is a flowchart which shows the diagnostic assistance process by a diagnostic assistance apparatus. It is a flowchart which shows the selection process of the assistance function in the process of FIG. It is a flowchart which shows the outline | summary of a process of a common assistance function and a specialization assistance function. It is a figure which shows an example of the plug-in file specialized in the organ. It is a flowchart which shows the other example of the plug-in file specialized in the organ. It is a figure which shows an example of a display of a classification / integration result. It is a figure which shows an example of the result of having statistically processed the past support result.

  FIG. 1 is a diagram showing an embodiment of a pathological diagnosis support apparatus according to the present invention. The pathological diagnosis support apparatus 1 includes a host computer 10 and an imaging unit 20. The host computer 10 has a configuration and functions equivalent to, for example, a known personal computer or workstation terminal, and includes a processor unit 100 that executes various control programs, an input unit 150 that receives operation input from a user, A display unit 160 that displays various types of information, a disk drive 170 that reads out data by accessing an optical disk OD that is an external recording medium on which various types of data such as a control program and image data are recorded, and an interface 180 are provided. . The host computer 10 can be connected to a telecommunication line 30 such as a local LAN or the Internet via an interface 180.

  As the display unit 160, for example, a liquid crystal display can be used. Further, as the input unit 150, for example, a touch panel integrated with the display unit 160 may be used in addition to a pointing device such as a mouse or a keyboard. Further, the recording medium on which the control program and the image data are recorded is not limited to the optical disk, and any hard disk, memory card, USB memory, or the like can be used.

  The imaging unit 20 includes a microscope 202 to which a CCD camera 201 is attached. The microscope 202 forms an enlarged optical image of a preparation prepared from a specimen tissue collected from a patient on a light receiving unit of the CCD camera 201, and the CCD camera 201 captures the image and converts it into digital data to create a so-called virtual slide. . The specimen tissue is stained by an appropriate staining method according to the purpose of observation (for example, hematoxylin / eosin (HE) staining).

  The host computer 10 performs various kinds of image processing on the image of the specimen tissue imaged by the imaging unit 20 and converted into digital data (RGB data) according to the control program read from the optical disk OD, and displays the image on the display unit 160. Thus, the pathological diagnosis by the user (diagnostic doctor) is supported.

  FIG. 2 is a block diagram showing the configuration of the pathological diagnosis support apparatus 1. More specifically, each functional block of FIG. 2 is realized by software in the processor unit 100 by executing a control program read from the optical disk OD by the host computer 10 or received via the telecommunication line 30. . The contents of the diagnosis support processing realized by these functional blocks will be described in detail later. Here, the outline of each functional block will be briefly described. In FIG. 2, the contents of information exchanged between the functional blocks are shown in parentheses.

  An image (target image) to be processed in the pathological diagnosis support apparatus 1 is, for example, an entire virtual slide obtained by imaging a sample tissue and converting it into data, or an image obtained by dividing the virtual slide into a plurality of images. Also, a plurality of images obtained by individually capturing a plurality of sample tissues collected from the same or a plurality of patients may be set as target images. When one virtual slide is divided into a plurality of target images, information that can specify the position of each target image in the original virtual slide so that the virtual slide can be restored in a later process (for example, coordinates Information) is desirably attached to each target image data.

  The RGB image data corresponding to the target image imaged by the imaging unit 20 is given to the grid image generation unit 101 and a specialization support processing unit 120 described later. The grid image generation unit 101 divides the target image input from the imaging unit 20 into a plurality of grid images with a grid of an arbitrary size that is determined in advance or specified by the user. At this time, information (for example, coordinate information) indicating where each grid image is located in the original target image is given to the grid image data.

  The support function selection unit 102 functions as an interface that receives an operation input from the user via the input unit 150. When the user selects and inputs a desired function from the support functions that can be executed by the device, the support function selection unit 102 supports the input. The function selection unit 102 transmits selection information corresponding to the selected function to each unit of the apparatus.

  The image information including the grid image data generated by the grid image generation unit 101 and the selection information output from the support function selection unit 102 are given to the common support processing unit 110 that performs processing independent of the type of the examination target organ. It is done. In addition, the target image captured by the imaging unit 20 and the selection information output from the support function selection unit 102 are given to the specialization support processing unit 120 that performs processing specialized for the organ to be examined.

  The common support processing unit 110 executes a process that does not depend on the type of the organ to be examined, for example, a process that extracts a part corresponding to a cell nucleus or cytoplasm in an image and obtains its position and number. More specifically, among the feature quantities that represent the features of the image, a feature quantity that is less dependent on the organ, that is, can be applied in common without limiting the organ, is calculated as a “common feature quantity”. Classify images based on quantity.

  A plurality of types of processing algorithms are stored and stored in the common support function storage unit 112, and a processing algorithm corresponding to the selection information given from the support function selection unit 102 is selected from these, and the common support function execution unit 111 performs the selection. This is executed to calculate the feature amount and classify the image based on the feature amount. More specifically, it is realized by a combination of a dynamic link library corresponding to the feature amount calculator 113 for calculating each feature amount as described above and a dynamic link library corresponding to the classifier 114 for performing classification based on the feature amount. Are stored in the common support function storage unit 112, and a plug-in file is appropriately added to a processing routine corresponding to the common support function execution unit 111. By incorporating and executing, the above-described various processing algorithms are selectively executed. The result of the processing executed in this way is stored in the common support result storage unit 115.

  Selection of a plug-in file can be performed by a user, that is, a pathologist, via the input unit 150 and the support function selection unit 102. In addition, the function of the apparatus can be expanded by enabling updating or addition of plug-in files as necessary.

  On the other hand, the specialization support processing unit 120 performs processing specialized for the organ to be examined. That is, the specialization support function storage unit 122 calculates a feature quantity (special feature quantity) specialized to detect a region having a feature of a lesion site in the organ (special area). A plug-in file combining 123 and a detector 124 that detects a special region based on the organ is prepared for each organ or for each type of lesion in each organ, and a plug corresponding to the type of organ designated by the user. IN is incorporated in a processing routine corresponding to the specialization support function execution unit 121 and executed. The result is stored in the specialization support result storage unit 125.

  The type and number of feature amounts calculated by each plug-in stored in the specialization support function storage unit 122 and the configuration of the detector used for determining the lesion in each organ are naturally different for each organ. However, an example thereof will be described later.

  The support processing results stored in the common support result storage unit 115 and the specialized support result storage unit 125 are integrated by the support result integration unit 131. The support result integration unit 131 performs a set operation such as taking a union or a product set of the common support result and the special support result in each support process, thereby the common support processing unit 110 and the special support processing unit 120. The processing results at are integrated into a comprehensive processing result. For example, composite image data is created by superimposing the color coding of the image based on the classification result by the common support processing unit 110 and the color coding of the image based on the classification result by the specialization support processing unit 120. The image is appropriately processed by the display processing unit 132 and displayed on the display unit 160. The user (pathologist) can make a diagnosis by viewing the displayed image.

  FIG. 3 is a flowchart showing a diagnosis support process by the diagnosis support apparatus 1 configured as described above. FIG. 4 is a flowchart showing the support function selection process during the process of FIG. In the diagnosis support process using the pathological diagnosis support apparatus 1, first, a support function selection process for causing the user to select a desired function from a plurality of support functions that can be executed is executed (step S101).

  As shown in FIG. 4, in the support function selection process, a list of support functions that can be executed first is displayed on the display unit 160 (step S201). Here, for the convenience of the user, it is desirable that the common support processing that can be executed by the common support processing unit 110 and the specialized support processing that can be executed by the specialization support processing unit 120 are displayed separately. The support function selection unit 102 receives a function selection input from the user for each of the common support function and the specialized support function from the displayed functions (steps S202 and S203). Note that the common support function and the special support function may be selected in any order, and selection inputs may be received in any order for each function.

  As the common support function, for example, the type of classifier used for classification, that is, the classification algorithm and the feature quantity to be used can be selected. As a specialization support function, the type of organ to be examined and the type of lesion to be detected can be selected in the first sense, but the feature amount to be used can be further selected. Also good.

  When the support function is selected by the user, the selection information is sent from the support function selection unit 102 to each of the common support processing unit 110 and the specialized support processing unit 120, and information on the plug-in file corresponding to the selected function Is stored in each of the common support function execution unit 111 and the specialized support function execution unit 121 (step S204).

  When the support function is selected in this way, a target image to be inspected next is acquired (step S102). The target image is mainly a virtual slide image obtained by imaging the entire specimen tissue. However, when the size of the image is large, the virtual slide image may be divided into a plurality of partial images, and each partial image may be used as the target image. Further, the virtual slide image can be received from the imaging unit 20 or may be received from another imaging device or storage via the electric communication line 30 or the like. The target image is divided into a plurality of grid images by the grid image generation unit 101 and sent to the common support processing unit 110 to execute the common support function (step S103). Also, the special support function is executed by sending the target image to the special support processor 120 (step S104).

  FIG. 5 is a flowchart showing an outline of processing of the common support function and the special support function. In the processing of the common support function shown in FIG. 5A, calculation of a feature amount (common feature amount) that can be applied generally to a living tissue that does not depend on the type of organ (step S301), and an image based thereon Classification (step S302).

  More specifically, for example, a color image is decomposed into RGB components on a pixel-by-pixel basis, and a certain threshold value is provided for those luminance levels or color difference components, so that a nuclear region (basophilic and B luminance level is compared from the image) 2) and cytoplasmic region (acidophilic and relatively high R luminance level).

In addition, for each grid image, features corresponding to the following elements:
Nuclear density ... (total number of pixels in the nucleus region) / (total number of pixels in tissues other than the nucleus);
N / C ratio (Nuclear to Cytoplasmic Ratio)
... (total number of pixels in the nuclear region) / (total number of pixels in the cytoplasmic region);
Nuclear variant: Variance or standard deviation of roundness of each nuclear region;
Nuclear size difference: variance or standard deviation of the area of each nuclear region;
Are calculated as a common feature amount. In general, cancerous cells tend to have a deformed nucleus or vary in size, so these feature values are suitable for detecting a cancerous tissue regardless of the type of organ.

  In addition to these, the stroma ratio, that is, the area ratio of the stroma in the tissue may be added to the common feature amount. However, the distinction between parenchyma and stroma in cells is handled differently depending on the type of organ, and it is not necessarily a uniform definition. You may make it treat as quantity.

  Then, these common feature quantities are input to a classifier, and the image is classified into a plurality of classification categories, for example, “normal tissue”, “tissue suspected of becoming cancerous”, and the like. The image classification based on the common feature amount is preferably performed by a learning algorithm such as a neural network, a support vector machine (SVM), or discriminant analysis. These classifiers input feature amounts calculated from images to be classified, and output classification results based on previously learned results. Since the size and shape of cells vary greatly among organs and individuals, classification based on feature quantities can be performed with high accuracy by performing classification using a learning algorithm using a typical sample image. It is desirable that a plurality of types of learning algorithms are plugged in and stored in the common support function storage unit 112 so that the user can select them appropriately.

  On the other hand, in the processing of the specialization support function shown in FIG. 5B, a feature amount (specialized for detecting a lesion in the examination target organ is executed by executing a plug-in file corresponding to the function selected by the user. Calculation (step S401), and detection of a special region based on the calculation (step S402). For example, a configuration example of a plug-in file for cancer detection in the prostate is as follows.

  An image feature that is widely used in the diagnosis of prostate cancer is eosinophilic crystalloid that exists in glandular tissue. Eosinophilic crystalloid in prostate tissue is a needle-like substance mainly present in the gland duct region, and its size is constant to some extent, and it has a characteristic that it is easy to stain red with HE staining but difficult to stain blue. . No other tissue or substance in the prostate that has all these characteristics is known, and crystalloids can be detected with high accuracy from these properties. Therefore, as the specialized feature amount corresponding to prostate cancer, for example, those corresponding to elements such as R and B luminance levels, area and shape can be used. By setting a predetermined threshold value for these feature value values and comparing the threshold value with the calculated value, it is possible to detect crystalloid, which is a special region used for the diagnosis of prostate cancer.

  FIG. 6 is a diagram showing an example of a plug-in file specialized for an organ. More specifically, FIG. 6A is a flowchart showing the contents of processing of a plug-in (PI) file specialized for detection of prostate cancer, and FIG. 6B is a decision tree embodied by this flowchart. It is a figure which shows the example of. In this process, a target image to be processed is acquired (step S502), and a specialized feature amount for the above-described crystalloid detection is calculated from the target image (step S502).

  Next, the special area is detected based on the calculated feature amount. When the organ or even the type of lesion is specified as described above, in many cases, the special area becomes extremely clear. For example, in this example, the presence / absence of the special region can be determined based on whether or not the acidophilic crystalloid is present in the image. That is, it is determined whether or not a part having the above characteristics, that is, a crystalloid is detected in the target image (step S503). If not detected, the target image is determined using the common support result. .

  In FIG. 6A, the process surrounded by the alternate long and short dash line, that is, the process of step S503 is equivalent to the decision tree shown in FIG. In this way, if the criteria of the special area are clear, in other words, if the diagnostic criteria are clear, the learning type classifier is not used, and the diagnostic criteria that do not depend on learning like the decision tree in this figure are used. It is possible to process with a rule-based detector along. This is possible because the feature quantity to be used and the special region detection algorithm based on the feature quantity are specialized only for the function of detecting a lesion of a specific organ. Such a feature amount and a special region detection algorithm based on the feature amount may not be applied if the target organ is different, but the feature amount and detection algorithm specialized for the organ are plugged in this way. If the internal organs are prepared and prepared for each organ in advance, diagnosis support for a plurality of organs can be performed with the same device by selecting and executing them according to the type of organ to be examined.

  Here, when a special region that is considered to be a crystalloid is not detected, the user determines whether there is a suspicion of cancer using the common support result. For example, in this case, a classifier based on a learning algorithm is applied. Thus, it may be determined whether there is a suspicion of cancer comprehensively from various feature amounts.

  FIG. 7 is a flowchart showing another example of a plug-in file specialized for an organ. More specifically, it is an example of a plug-in specialized for detection of gastric cancer. In the plug-in file for gastric cancer detection, after acquiring the target image (step S601), a feature amount suitable for detection of signet ring cells, which are special regions that are specifically found in the cancerous part of the stomach tissue, is calculated. (Step S602). The signet ring cells are characterized by a shape that is relatively circular and has a flat nucleus at the periphery. Therefore, for example, a feature quantity corresponding to the circularity of the cytoplasm, the position of the nucleus in the cell, the shape of the nucleus, etc. can be used as the “specialized feature quantity” in this case. The presence or absence of signet ring cells can be determined by comparing these values with a preset threshold value.

  The detector in this case is configured as follows, for example. First, when a special region that is a signet ring cell is detected in the target image (step S603) and the detected number is equal to or greater than a predetermined threshold (step S604), the tissue of the image is suspected to be cancerous. The special area is displayed on the display unit 160 as a specialization support result. Although this is not the case, if the number of signet ring cells detected within a certain distance from the tissue of the image is greater than or equal to the threshold value, that is, there are many special areas that are regarded as signet ring cells in the surrounding area. If it is determined that the image is cancerous, the region is displayed on the display unit 160 as a specialization support result (step S605). Also in this example, a portion surrounded by a one-dot chain line functions as a detector. The detector may have only a function of detecting signet ring cells. Even in such a configuration, the diagnosis of the user can be supported by displaying the area of the signet ring cells on the display unit 160 as a specialized support result.

  As described above, in the pathological diagnosis support apparatus 1 according to this embodiment, the contents of the executed processes are different when, for example, the prostate is selected as the organ to be examined and when the stomach is selected. Yes. However, from the user's point of view, there is no change in that the function of finding a part suspected of having cancerous in the organ to be examined is being executed. In addition, functions other than those specialized for organs are naturally the same. This means that even if the target organ changes, there is no change in the operability and usability of the device, and the user is provided with a completely different system for each organ as in the prior art. Can be greatly reduced.

  In addition, when a new diagnostic criterion is established for a certain organ, a classification function based on the new diagnostic criterion can be immediately added to the apparatus by creating and incorporating a corresponding plug-in file. In this way, by making a function specialized for an organ into a plug-in, a diagnosis support apparatus excellent in application and expandability can be obtained.

  Returning to FIG. 3, the description of the diagnosis support process will be continued. The classification results and special area detection results executed by the common support function execution unit 111 and the special support function execution unit 121 as described above are stored in the common support result storage unit 115 and the special support result storage unit 125, respectively. Each is stored (step S105). These results are appropriately integrated by the support result integration unit 131 and sent to the display processing unit 132 (step S106), and the display processing unit 132 generates display image data for displaying the integrated classification result on the screen. It is sent to the display unit 160. As a result, the final classification result and detection result are displayed on the display unit 160 and presented to the user (step S107). In this way, diagnosis by the user (pathologist) is supported. Note that the result of rule-based detection by the specialization support function execution unit 121 may be stored in the specialization support result storage unit 125 and used for integration of support results.

  FIG. 8 is a diagram showing an example of the display of the classification / integration result. Here, an example in the case of diagnosing cancer from a stomach tissue image is shown. From the execution result of the common support function, for example, the coordinates of the cancerous region in the tissue image specified from the nuclear density, the N / C ratio, the nuclear atypia, the size of the nucleus, and the like can be provided as display information. Further, from the execution result of the specialization support function, the presence / absence and number of signet ring cells in the image, the color display of a special region corresponding to the signet ring cells, and the like can be provided as display information.

  In the example of FIG. 8, the virtual slide image Iv to be inspected or a partial image thereof is displayed on the display screen 161 of the display unit 160. Further, among the grid images Ig obtained by dividing the virtual slide image Iv, those that are determined to be normal are displayed as they are, while the common support function execution unit 111 determines that the cancer is suspected of being cancerous. The grid image region R1 is darkly hatched, and the region R2 of the signet ring cell, which is judged to be cancerous by the specialization support function execution unit 121, is visually highlighted with a lighter hatching. As a result, the difference in the classification result for each region can be visually recognized at a glance.

  In addition, a text area It is provided on the left side of the screen for displaying a list of functions selected by the user, various calculation results, and the like so that the user can check classification conditions and more detailed classification results as necessary. It can be done. With these displays, the pathologist as the user can efficiently perform pathological diagnosis.

  For example, as shown in FIG. 9, the past support results accumulated in the common support result storage unit 115 and the specialized support result storage unit 125 are statistically processed by the support result integration unit 131 and displayed on the display unit 160. By doing so, it is possible to extract the tendency of features on an image common to each organ or between each organ, and to support work for establishing academic research and new diagnostic criteria.

  FIG. 9 is a diagram illustrating an example of a result of statistical processing of past support results. From the data accumulated in the common support result storage unit 115 and the specialized support result storage unit 125 as a result of analyzing the tissue collected from the patient, statistical data (average, variance, standard deviation, etc.) for each feature amount is obtained. As shown in FIG. 9, by displaying them together on the display screen 161, it becomes easier for the user to grasp the tendency on the image for each organ or lesion.

  For example, in an organ (for example, prostate) containing a large number of glandular tissues, even if it is a normal tissue, many nuclei constituting the gland are included. On the other hand, nuclear density is low in organs with a lot of muscles and fat (for example, the heart). In other words, the threshold value that should be determined to be normal differs even with the same feature amount depending on the organ. Therefore, it is possible to more effectively support a doctor's diagnosis by obtaining statistical data of common feature amounts in each organ in the support result integration unit 131 based on past support results.

  As described above, in this embodiment, in a pathological diagnosis support system that supports a pathological diagnosis by processing an image obtained by imaging a specimen tissue, a function specialized for each organ is configured as a plug-in file, which is necessary. In response to the above, it is possible to support diagnosis of a plurality of types of organs with a single system by being incorporated into a control program body and executed. For this reason, the user does not need to prepare a different system for each target organ, and can operate the apparatus with almost the same feeling of use even if the target organ changes. Thereby, the diagnosis support apparatus 1 of this embodiment can support the diagnosis with respect to multiple types of organs by a doctor very effectively.

  In addition, by updating or incorporating a plug-in file, it is possible to immediately cope with organs or diagnostic criteria that have not been supported so far, and it is possible to construct a diagnostic support system with excellent extensibility. It has become.

  As described above, in this embodiment, the imaging unit 20 functions as the “image acquisition unit” of the present invention, and the display unit 160 functions as the “output unit” of the present invention. The functions of the feature quantity calculator 113 and the classifier 114 realized by the common support processor 110 correspond to the “common feature quantity calculator” and the “common classifier” of the present invention, respectively, while the specialized support processor The function of the feature quantity calculator 123 realized by 120 and the function of the detector 124 for detecting a special area correspond to the “specialized feature quantity calculation means” and “specialization detection means” of the present invention, respectively.

  The present invention is not limited to the above-described embodiment, and various modifications other than those described above can be made without departing from the spirit of the present invention. For example, in the above embodiment, the prostate and stomach are taken as examples of organs, but the present invention can be applied to diagnosis support for other organs. In addition, the feature amounts listed in the description of the present embodiment are only some examples, and any feature amount not described here can be used.

  In the above embodiment, as a common support function, the image classification is classified into two classification categories of “normal” and “suspected of canceration”, but the number and type of classification categories are not limited to this and are arbitrary. is there. For example, the “normal” category is further subdivided to provide categories such as “normal tissue” and “normal tissue (gland duct)”, or the “cancerous” category is further subdivided into “highly differentiated adenocarcinoma”. A category such as “poorly differentiated adenocarcinoma” may be provided.

  In the above embodiment, the common support function processing algorithm stored in the common support function storage unit 112 is also plugged in so that future updates and additions of these functions can be supported. However, this is not an essential requirement in the present invention. That is, these may be fixed in advance incorporated in the control program body. In this case, for example, it is possible to update or add functions by updating the entire control program.

  In the above embodiment, the imaging unit 20 including the CCD camera 201 and the microscope 202 is provided as an “image acquisition unit”. However, the imaging unit is not essential in the present invention. As described above, it is also possible to receive an original image from the telecommunication line 30 via the interface 180 or from an external recording medium via an appropriate interface. In this case, these interfaces serve as “image acquisition means”. Will work. In this sense, the “step for acquiring an image” is not essential in the control program of the present invention.

  Moreover, in the said embodiment, what integrated the classification result by the common support function execution part 111 and the detection result by the specialization support function execution part 121 is displayed on the screen, However, The function which displays each of these results independently You may have.

  In this embodiment, processing by each functional block is realized by the host computer 10 executing the control program recorded on the optical disk OD. However, the recording medium for recording the control program is limited to the optical disk. Any other recording medium can be used as described above. The control program may be distributed through the telecommunication line 30.

  Moreover, although the pathological diagnosis support apparatus 1 of the said embodiment is comprised combining the general purpose host computer 10 and the imaging part 20, it is made to comprise a pathological diagnosis support apparatus combining an imaging part and a dedicated terminal. Also good. Further, by incorporating the control program according to the present invention into an existing virtual slide creation device, the virtual slide creation device may function as a pathological diagnosis support device.

  In the above embodiment, the display unit 160 functions as the “output unit” of the present invention. However, the output unit is not limited to the one that displays the result, and for example, prints the classification result on a recording sheet. Alternatively, the classification result may be transmitted to a recording medium or a telecommunication line. These may be used in combination.

  Moreover, in the said embodiment, although the specialization assistance process part 120 becomes a structure which processes with respect to an object image, the grid image and image produced | generated by the grid image generation part 101 similarly to the common assistance process part 110 are shown. Processing may be performed on the information. Then, the region representing the grid image may be colored to show the detection result to the user.

  The present invention can be suitably applied to a technical field in which various pathological diagnoses by a doctor are supported by a computer system or the like.

10 Personal Computer 20 Imaging Unit (Image Acquisition Unit)
110 common support processing unit 113 feature quantity calculator (common feature quantity calculation means)
114 Classifier (common classification means)
120 Specialization Support Processing Unit 123 Feature Quantity Calculator (Specialized Feature Quantity Calculation Unit)
124 Detector (specialized detection means)
160 Display (output means)
Iv virtual slide image S101 organ selection process S102 image acquisition process S103 common classification process S104 specialized detection process S107 output process

Claims (11)

Image acquisition means for acquiring an image of the sample tissue;
A common feature amount calculating means for calculating a common feature amount that is a feature amount independent of the type of organ based on the image;
Based on the common feature amount, a common classification unit that classifies the image into a plurality of classification categories by a classifier machine-learned using a sample image given in advance ,
Based on the image, specialized feature quantity calculating means for calculating a specialized feature quantity that is a feature quantity specialized for an organ to be examined;
Specialized detection means for detecting a special region peculiar to an organ from the image based on a comparison result between a value of the specialized feature value and a threshold value predetermined according to the specialized feature value ;
An output means for outputting an output image obtained by integrating the classification result by the common classification means and the detection result by the specialized detection means;
The pathological diagnosis support apparatus characterized in that the specialized feature amount calculation means can be exchanged according to the type of organ.   The pathological diagnosis support apparatus according to claim 1, wherein the specialized detection unit is replaceable with the specialized feature amount calculating unit. The common feature amount calculating unit calculates a feature amount for each of a plurality of grid images obtained by dividing the image, and the common classification unit classifies the grid image,
The pathological diagnosis support apparatus according to claim 1, wherein the specialized feature amount calculating unit calculates a feature amount for the image, and the specialized detection unit detects the special region.   The common feature amount includes a feature amount corresponding to at least one of a nucleus density, a nucleus concentration, a nucleus arrangement, a nucleus size, and a nucleus shape in the image. Pathological diagnosis support device.   The pathological diagnosis support according to any one of claims 1 to 4, wherein the common feature amount includes a feature amount corresponding to at least one of an area ratio between a cytoplasm and a nucleus in the image, a nuclear atypia, and a size of the nucleus. apparatus. The pathological diagnosis support apparatus according to claim 1, wherein the output unit includes a display unit that displays the output image .   The pathological diagnosis support apparatus according to any one of claims 1 to 6, wherein the classification category includes a category corresponding to a normal tissue and a category corresponding to a tissue suspected of being cancerous. An image acquisition step of acquiring an image obtained by imaging a sample tissue;
Based on the image, a common feature value that is a feature value that does not depend on the type of organ is calculated, and based on the common feature value, the image is obtained by a classifier that is machine-learned using a sample image given in advance. A common classification process for classifying into multiple classification categories;
An organ selection process for selecting and setting an organ to be examined;
Based on the image, a specialized feature amount that is a feature amount specialized for the organ set as the examination target is calculated, and predetermined according to the value of the specialized feature amount and the specialized feature amount A specialized detection step of detecting a special region peculiar to an organ from the image based on a result of comparison with a threshold ;
An output step of outputting an output image obtained by integrating the classification result in the common classification step and the detection result in the specialized detection step;
The pathological diagnosis support method characterized in that, in the specialized detection step, the type of the specialized feature amount is varied according to the set organ type. A classifier that calculates a common feature quantity that is a feature quantity that does not depend on the type of organ based on an image obtained by imaging a specimen tissue, and performs machine learning using a sample image that is given in advance based on the common feature quantity a common classification step for classifying the image into a plurality of classification categories by,
Based on the image, a specialized feature value that is a feature value specialized for an organ to be examined is calculated, and a value of the specialized feature value and a threshold value that is predetermined according to the specialized feature value are calculated. A specialized detection step of detecting a special region specific to an organ from the image based on the comparison result ;
Causing the computer to execute an output step of outputting an output image obtained by integrating the classification result in the common classification step and the detection result in the specialized detection step;
A control program for pathological diagnosis support, characterized in that, in the specialized detection step, the type of the specialized feature is made different according to the set type of organ. The control program according to claim 9 , wherein the program for executing the specialized detection step is configured as a plug-in file for the control program main body. To claim 9 or 1 0 is recorded a control program according, the computer-readable recording medium.