Information processing device, information processing method, and computer program

Abstract

To determine a cell nucleus on the basis of a unified criterion with good accuracy, and without significant variances.SOLUTION: This information processing device is a device for determining whether or not a cell nucleus in a biological tissue image is a specific type of cell nucleus, the device comprising a model acquisition unit and a determination execution unit. The model acquisition unit acquires a cell nucleus determination model. The cell nucleus determination model is generated by being trained, for each of a plurality of cell nuclei, by machine learning using training data that is associated with form information that indicates the form of a cell nucleus and score information that indicates the level of probability of being a specific type of cell nucleus. The determination execution unit determines, using form information regarding the cell nucleus to be determined and the cell nucleus determination model, whether or not the cell nucleus to be determined is the specific type of cell nucleus and outputs the determination result.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The technology disclosed in the present specification relates to information processing for determining whether a cell nucleus in a biological tissue image is a specific type of cell nucleus. Furthermore, the technology disclosed in the present specification relates to information processing for supporting a physician's understanding and diagnosis based on the result of the determination.

Background Art

[0002] Hirschsprung's disease is a congenital rare disease in which ganglion cells responsible for motility (peristalsis) are deficient in the portion of the intestinal tract near the anus, resulting in intestinal motility disorder. In addition, Hirschsprung's disease-related disorders are rare diseases in which ganglion cells are present in the intestinal tract, but their number is insufficient, or the ganglion cells are immature, leading to insufficient intestinal motility.

[0003] In the pathological diagnosis of Hirschsprung's disease or related disorders, an image obtained by imaging a piece of intestinal tissue collected from a patient is used to determine whether each cell nucleus present in the plexus region of the intestinal tract is a ganglion cell nucleus. After the pathological diagnosis, for example, a procedure is performed in which the site in the intestinal tract where no ganglion cells are present or the site where the number of ganglion cells is insufficient is resected, and the intestinal tract without motility disorder is anastomosed to the anus.

[0004] Conventionally, the determination of ganglion cell nuclei in the pathological diagnosis as described above has been carried out by a skilled pathologist visually observing images (for example, see Non-Patent Document 1).

Prior Art Documents

Non-Patent Documents

[0005]

Non-Patent Document 1

[0006] For appropriate treatment of Hirschsprung's disease or related disorders, accurate identification of ganglion cell nuclei in pathological diagnosis is crucial. Traditionally, subjective identification relies on the individual experience of each pathologist, resulting in insufficient accuracy and significant variability in results among physicians. In particular, because Hirschsprung's disease and related disorders are rare diseases, the number of facilities and physicians capable of performing accurate pathological diagnoses is limited. Therefore, there is a need for a technology that can perform accurate and consistent identification based on unified criteria. Furthermore, even when such a technology for consistent identification is realized, it does not provide new insights into how it differs from conventional identification or how to accumulate additional data. This creates a challenge in determining whether the identification can be used in actual medical practice, and if so, whether the results can be trusted. A technology that can solve these problems is desired.

[0007] Furthermore, this challenge is not limited to the identification of ganglion cell nuclei, but is a common challenge when determining whether or not cell nuclei in biological tissue images belong to a specific type of cell nucleus.

[0008] This specification discloses a technology capable of solving the above-mentioned problems. [Means for solving the problem]

[0009] The technologies disclosed herein can be implemented, for example, in the following forms:

[0010] (1) The information processing device disclosed herein is a device for determining whether a cell nucleus in a biological tissue image is of a specific type, and comprises a model acquisition unit and a determination execution unit. The model acquisition unit acquires a cell nucleus determination model. The cell nucleus determination model is a model generated by machine learning using training data, which is data in which morphological information indicating the shape of the cell nucleus and score information indicating the probability that each of a plurality of cell nuclei is of the specific type are associated. The determination execution unit uses the morphological information of the cell nucleus to be determined and the cell nucleus determination model to determine whether the cell nucleus to be determined is of the specific type, and outputs the result of the determination.

[0011] This information processing device uses morphological information of the target cell nucleus and a cell nucleus discrimination model to determine whether the target cell nucleus is of a specific type and outputs the discrimination result. Therefore, for each individual cell nucleus, the cell nucleus can be identified using the cell nucleus discrimination model based on the cell nucleus's morphology. Consequently, this information processing device can perform cell nucleus discrimination based on a unified standard, with high accuracy and minimal variability.

[0012] (2) The above-mentioned information processing device may further include a training data acquisition unit for acquiring the training data. Alternatively, the model acquisition unit may acquire the cell nucleus discrimination model by creating the cell nucleus discrimination model using the training data through machine learning. By adopting this configuration, the cell nucleus discrimination model can be acquired without using other devices, and the cell nucleus discrimination can be performed using the model.

[0013] (3) In the above-mentioned information processing device, the morphological information may be configured to represent the morphology of the cell nucleus using multiple morphological indicators. By adopting this configuration, it is possible to perform cell nucleus discrimination based on information that captures the morphology of the cell nucleus from multiple perspectives, and the accuracy of cell nucleus discrimination can be effectively improved. Furthermore, by adopting this configuration, since indicators that are understandable to humans are used, an expert can verify whether the structure of the cell nucleus discrimination model is appropriate or not.

[0014] (4) In the above-mentioned information processing device, the morphological information may be configured to include at least one of the following: information indicating the contour features of the cell nucleus, information indicating the degree of staining of the cell nucleus, and information indicating the pattern-like features of the cell nucleus. By adopting this configuration, it is possible to perform cell nucleus discrimination based on information that more accurately captures the morphology of the cell nucleus, and the accuracy of cell nucleus discrimination can be further effectively improved.

[0015] (5) In the above-mentioned information processing device, the morphological information may be configured to include information indicating the contour features of the cell nucleus, information indicating the degree of staining of the cell nucleus, and information indicating the pattern-like features of the cell nucleus. By adopting this configuration, it is possible to perform cell nucleus discrimination based on information that captures the morphology of the cell nucleus more comprehensively and more accurately, thereby greatly improving the accuracy of cell nucleus discrimination.

[0016] (6) The above-mentioned information processing device may further include a discrimination result analysis unit that analyzes the discrimination of multiple cell nuclei and outputs information showing the combination rules of the morphological information accumulated in the structure of the cell nucleus discrimination model in a visually identifiable manner. Adopting this configuration makes it possible to perform mapping for each morphological feature of the cell nucleus and can support an objective understanding of the morphological categories of the cell nucleus. This makes it possible to determine whether the discrimination results of the cell nucleus discrimination model can be trusted, to recognize which of the multiple discrimination results given by the cell nucleus discrimination model should be trusted and which should not, and to consider whether the data used for training is appropriate in the first place, thereby supporting the verification of medical diagnostic technology in comparison with past knowledge.

[0017] (7) In the above-mentioned information processing device, the multidimensional clustering may be configured to be performed using some of the morphological indicators that contribute highly to the cell nucleus discrimination model from among the multiple morphological indicators included in the morphological information. By adopting this configuration, mapping can be performed for each morphological feature that contributes highly to the discrimination of the cell nucleus, and an objective understanding of the morphological categories of the cell nucleus can be effectively supported.

[0018] (8) In the above-described information processing device, the biological tissue image may be an image including the nerve plexus region of an intestinal tissue fragment, and the specific type of cell nucleus may be configured to be ganglion cell nuclei. By adopting this configuration, it is possible to determine whether or not the cell nuclei included in the nerve plexus region of an image including the nerve plexus region of an intestinal tissue fragment are ganglion cell nuclei based on a unified standard, with high accuracy and low variability. This makes it possible to provide useful information for appropriate treatment of Hirschsprung's disease or related disorders.

[0019] Furthermore, the technologies disclosed herein can be implemented in various forms, for example, in the form of an information processing device, an information processing method, a computer program that implements such a method, or a non-temporary recording medium on which such a computer program is stored. [Brief explanation of the drawing]

[0020] [Figure 1] Conceptual diagram illustrating the cell nucleus discrimination model MO in this embodiment. [Figure 2] Diagram illustrating the schematic configuration of the information processing device 100. [Figure 3] Flowchart showing the cell nucleus discrimination model acquisition process in this embodiment [Figure 4] An explanatory diagram showing an example of morphological information (FI) of the cell nucleus (CN). [Figure 5] Flowchart showing the cell nucleus discrimination process in this embodiment [Figure 6] Diagram illustrating the predictive accuracy of the cell nucleus discrimination model MO in the example. [Figure 7] An explanatory diagram showing morphological indices with high contribution in the cell nucleus discrimination model MO of the example [Figure 8] A flowchart showing the discrimination result analysis processing in the present embodiment [Figure 9] An explanatory diagram showing an example of the result of multidimensional clustering [Figure 10] An explanatory diagram showing an example of the result of re-performed multidimensional clustering targeting only data belonging to the first group Gr1 and the second group Gr2 in the multidimensional clustering shown in FIG. 9 [Figure 11] An explanatory diagram showing an example of a morphological index heatmap based on the multidimensional clustering result shown in FIG. 9 [Mode for Carrying Out the Invention]

[0021] A. Embodiment: A-1. Overview of the cell nucleus discrimination model MO: First, an overview of the cell nucleus discrimination model MO in the present embodiment will be described. FIG. 1 is an explanatory diagram conceptually showing the cell nucleus discrimination model MO in the present embodiment. The cell nucleus discrimination model MO is a model used to discriminate whether or not a cell nucleus CN on the biological tissue image Ib is a specific type of cell nucleus. As will be described in detail later, the cell nucleus discrimination model MO is a machine learning model that receives morphological information FI indicating the morphology of the cell nucleus CN as input, and outputs score information SI indicating the level of probability that the cell nucleus is the specific type of cell nucleus.

[0022] In the present embodiment, the cell nucleus discrimination model MO is used for discriminating whether or not a cell nucleus CN is a ganglion cell nucleus CN(Ne). More specifically, in the pathological diagnosis of Hirschsprung's disease or a related disease thereof, using an image of an intestinal tissue piece as the biological tissue image Ib, the cell nucleus discrimination model MO is used to discriminate whether each cell nucleus CN included in the nerve plexus region NPA on the image is a ganglion cell nucleus CN(Ne) or a cell nucleus other than the ganglion cell nucleus CN(Ne) (hereinafter referred to as "general cell nucleus CN(Ge)").

[0023] The score information SI, which indicates the probability that a cell is a ganglion cell nucleus (CN(Ne)) and is output by the cell nucleus discrimination model MO, may be, for example, a zero / one discrimination result indicating whether or not it is a ganglion cell nucleus (CN(Ne)), or a numerical value indicating the probability of it being a ganglion cell nucleus (CN(Ne)) in steps (for example, a numerical value indicating how many points out of 10), or both may be used. In the example shown in Figure 1, a zero / one discrimination to indicate whether or not it is a ganglion cell nucleus (CN(Ne)) is performed, and each cell nucleus (CN) on the displayed biological tissue image Ib is shown with an identifier indicating that it is a ganglion cell nucleus (CN(Ne)) and an identifier indicating that it is a general cell nucleus (CN(Ge)). Ganglion cell nucleus (CN(Ne)) is an example of a specific type of cell nucleus in the claims.

[0024] By using the cell nucleus discrimination model MO, it is possible to accurately and consistently determine whether each cell nucleus CN in an image of an intestinal tissue fragment (as a biological tissue image Ib) is a ganglion cell nucleus CN(Ne) or a general cell nucleus CN(Ge), without requiring visual inspection by a pathologist. Of course, the discrimination result obtained by the cell nucleus discrimination model MO may be confirmed by a pathologist as the final discrimination result. In this way, by using the cell nucleus discrimination model MO, it is possible to accurately and consistently determine whether or not a cell is a ganglion cell nucleus CN(Ne), which in turn can lead to appropriate treatment and improved functional prognosis for Hirschsprung's disease or related disorders.

[0025] A-2. Configuration of the information processing device 100: Next, we will explain the configuration of the information processing device 100 for obtaining a cell nucleus discrimination model MO and for performing cell nucleus discrimination using the cell nucleus discrimination model MO. Figure 2 is an explanatory diagram showing the schematic configuration of the information processing device 100. The information processing device 100 is composed of a computer (PC, server, etc.).

[0026] The information processing device 100 comprises a control unit 110, a storage unit 120, a display unit 130, an operation input unit 140, and an interface unit 150. These units are connected to each other via a bus 190 so that they can communicate with one another. The information processing device 100 may also include a speaker as an output means.

[0027] The display unit 130 of the information processing device 100 is configured, for example, as a liquid crystal display, and displays various images and information. The operation input unit 140 is configured, for example, as a keyboard, mouse, buttons, microphone, trackpad, etc., and receives operations and instructions from the administrator. The display unit 130 may also function as the operation input unit 140 by being equipped with a touch panel. The interface unit 150 is configured, for example, as a LAN interface or USB interface, and communicates with other devices via wired or wireless connection.

[0028] The storage unit 120 of the information processing device 100 is composed of, for example, ROM, RAM, a hard disk drive (HDD), and stores various programs and data, and is used as a workspace and temporary storage area for data when executing various programs. For example, the storage unit 120 stores a cell nucleus discrimination program CP, which is a computer program for executing the cell nucleus discrimination model acquisition process described later. The cell nucleus discrimination program CP is provided in a state where it can be stored on a computer-readable recording medium (not shown), such as a CD-ROM, DVD-ROM, or USB memory, or it is provided in a state where it can be obtained from an external device (for example, a server on the cloud or other terminal device) via the interface unit 150, and is stored in the storage unit 120 in a state where it can be operated on the information processing device 100.

[0029] Furthermore, the storage unit 120 of the information processing device 100 stores the training data TD, the cell nucleus discrimination model MO, the target image Ibo, and the discrimination result data RD in advance or during the execution of the cell nucleus discrimination model acquisition process described later. The contents of this information and data will be explained in conjunction with the description of the cell nucleus discrimination model acquisition process described later.

[0030] The control unit 110 of the information processing device 100 is composed of, for example, a CPU, and controls the operation of the information processing device 100 by executing a computer program read from the storage unit 120. For example, the control unit 110 reads and executes a cell nucleus discrimination program CP from the storage unit 120, and functions as a cell nucleus discrimination processing unit 111 for executing the cell nucleus discrimination model acquisition process described later. The cell nucleus discrimination processing unit 111 includes a training data acquisition unit 112, a model acquisition unit 113, a target image acquisition unit 114, a region identification unit 115, a cell nucleus identification unit 116, a morphological information acquisition unit 117, a discrimination execution unit 118, and a discrimination result analysis unit 119. The functions of each of these units will be explained in accordance with the description of the cell nucleus discrimination model acquisition process described later.

[0031] A3. Cell nuclear discrimination model acquisition process: Next, the cell nucleus discrimination model acquisition process performed by the information processing device 100 of this embodiment will be described. Figure 3 is a flowchart of the cell nucleus discrimination model acquisition process in this embodiment. The cell nucleus discrimination model acquisition process is a process for acquiring the cell nucleus discrimination model MO described above. In this embodiment, the information processing device 100 acquires the cell nucleus discrimination model MO by creating the cell nucleus discrimination model MO itself using predetermined machine learning. The cell nucleus discrimination model acquisition process is started when a user operates the operation input unit 140 of the information processing device 100 and inputs a start command.

[0032] First, the training data acquisition unit 112 (Figure 2) of the information processing device 100 acquires a training image It, which is a biological tissue image Ib used to create a cell nucleus discrimination model MO (S110). In this embodiment, the training image It is an image of an intestinal tissue sample taken from a patient with Hirschsprung's disease or a related disease, and includes the nerve plexus region (NPA). It is preferable for the training data acquisition unit 112 to acquire multiple training images It for multiple patients.

[0033] Next, the region identification unit 115 (Figure 2) of the information processing device 100 identifies the nerve plexus region NPA in each training image It (S120). The identification of the nerve plexus region NPA can be achieved by the region identification unit 115 performing predetermined image processing. Alternatively, for example, a pathologist may identify the nerve plexus region NPA by visually inspecting each training image It and input information specifying the location of the nerve plexus region NPA via the operation input unit 140, which the region identification unit 115 then uses to identify the nerve plexus region NPA.

[0034] Next, the cell nucleus identification unit 116 (Figure 2) of the information processing device 100 identifies the cell nucleus CN targeting the nerve plexus region NPA identified in S120 (S130). The identification of the cell nucleus CN can be achieved by the cell nucleus identification unit 116 performing predetermined image processing. Alternatively, for example, a pathologist may identify the cell nucleus CN by visually inspecting the nerve plexus region NPA of each training image It, and input information specifying the position of the cell nucleus CN via the operation input unit 140, which the cell nucleus identification unit 116 then uses to identify the cell nucleus CN.

[0035] Next, the morphological information acquisition unit 117 (Figure 2) of the information processing device 100 acquires (measures) morphological information FI for each cell nucleus CN identified in S130 (S140).

[0036] Figure 4 is an explanatory diagram showing an example of morphological information FI for the cell nucleus (CN). In this embodiment, morphological information FI includes information indicating the contour characteristics of the cell nucleus (size, roundness, etc.), information indicating the staining characteristics of the cell nucleus (color intensity, etc.), and information indicating the pattern characteristics of the cell nucleus (texture, etc.). In the example in Figure 4, 11 morphological indicators, including area and length, are used to indicate the contour characteristics of the cell nucleus; 27 morphological indicators, including the mean blue intensity, standard deviation blue intensity, and coefficient of variation blue intensity, are used to indicate the staining characteristics of the cell nucleus; and 9 morphological indicators, including blue correlation, energy, and homogeneity blue, are used to indicate the pattern characteristics of the cell nucleus. In other words, in this embodiment, a total of 47 morphological indicators are used as morphological information FI.

[0037] The acquisition (measurement) of morphological information FI of cell nuclei (CN) can be achieved by the morphological information acquisition unit 117 performing predetermined image processing. Alternatively, for example, a pathologist may manually measure the morphological information FI of each cell nucleus (CN) and input the measured values ​​via the operation input unit 140, which the morphological information acquisition unit 117 then uses to acquire the morphological information FI.

[0038] Next, the training data acquisition unit 112 (Figure 2) of the information processing device 100 acquires score information SI indicating the probability that each cell nucleus CN identified in the nerve plexus region NPA of each training image It is a ganglion cell nucleus CN(Ne) (S150). As described above, the score information SI may be a zero / one judgment result indicating whether or not it is a ganglion cell nucleus CN(Ne), or it may be a numerical value indicating the probability of it being a ganglion cell nucleus CN(Ne) in steps, or both may be used.

[0039] The acquisition of score information SI is achieved, for example, by a pathologist visually inspecting each cell nucleus CN in the image, assigning a score indicating the likelihood that it is a ganglion cell nucleus CN(Ne), and inputting this score via the operation input unit 140.

[0040] The training data acquisition unit 112 of the information processing device 100 stores the data obtained through processing S110 to S150, that is, data in which morphological information FI indicating the shape of the cell nucleus CN and score information SI indicating the probability that it is a ganglion cell nucleus CN(Ne) are associated for each of the multiple cell nuclei CN, as training data TD in the storage unit 120.

[0041] Next, the model acquisition unit 113 (Figure 2) of the information processing device 100 creates a cell nucleus discrimination model MO using machine learning with the training data TD (S160). The model acquisition unit 113 takes morphological information FI, which shows the morphology of cell nuclei CN included in the training data TD, as input (explanatory variable), and outputs score information SI, which shows the probability that the cells included in the training data TD are ganglion cell nuclei CN(Ne) (target variable). The model acquisition unit 113 creates the cell nucleus discrimination model MO by performing machine learning that increases a predetermined evaluation index that shows the accuracy of discrimination of whether or not a cell is a ganglion cell nucleus CN(Ne).

[0042] For machine learning to create a cell nucleus discrimination model (MO), various known machine learning algorithms can be used, but in this embodiment, the LASSO regression model is used. The LASSO regression model is a relatively simple machine learning model, is easy to tune, fully automated, and is a model that can be interpreted.

[0043] The cell nucleus discrimination model MO created in this manner is stored in the storage unit 120 of the information processing device 100. With this, the acquisition process of the cell nucleus discrimination model MO is completed.

[0044] A4. Cell nuclear discrimination processing: Next, the cell nucleus discrimination process performed by the information processing device 100 of this embodiment will be described. Figure 5 is a flowchart of the cell nucleus discrimination process in this embodiment. The cell nucleus discrimination process is a process that uses the cell nucleus discrimination model MO to determine whether or not the cell nucleus CN to be discriminated is a ganglion cell nucleus CN(Ne). The cell nucleus discrimination process is started when the user operates the operation input unit 140 of the information processing device 100 and inputs a start command.

[0045] First, the target image acquisition unit 114 (Figure 2) of the information processing device 100 acquires the target image Ibo (S310). The target image Ibo is a biological tissue image Ib that is the subject of cell nucleus discrimination processing, and more specifically, it is an image of an intestinal tissue sample taken from a patient with Hirschsprung's disease or a related disease. The biological tissue image Ib may be acquired in the form of a virtual slide, which is a high-resolution, wide-area stitched image.

[0046] Next, the region identification unit 115 (Figure 2) of the information processing device 100 identifies the nerve plexus region NPA in the target image Ibo (S320). The identification of the nerve plexus region NPA can be achieved by the region identification unit 115 performing predetermined image processing. Alternatively, for example, a pathologist may identify the nerve plexus region NPA by visually inspecting the target image Ibo and input information specifying the location of the nerve plexus region NPA via the operation input unit 140, which the region identification unit 115 then uses to identify the nerve plexus region NPA.

[0047] Next, the cell nucleus identification unit 116 (Figure 2) of the information processing device 100 identifies the cell nucleus CN (CN) targeting the nerve plexus region NPA identified in S320 (S330). The identification of the cell nucleus CN can be achieved by the cell nucleus identification unit 116 performing predetermined image processing. Alternatively, for example, a pathologist may identify the cell nucleus CN by visually inspecting the nerve plexus region NPA of the target image Ibo, and input information specifying the position of the cell nucleus CN via the operation input unit 140, which the cell nucleus identification unit 116 then uses to identify the cell nucleus CN.

[0048] Next, the morphological information acquisition unit 117 (Figure 2) of the information processing device 100 acquires (measures) morphological information FI for each cell nucleus CN identified in S330 (S340). If multiple cell nuclei CN are identified in S330, morphological information FI is acquired for each cell nucleus CN. In this embodiment, morphological information FI consisting of the 47 morphological indicators described above is acquired for each cell nucleus CN. The identification of morphological information FI can be achieved by the morphological information acquisition unit 117 performing predetermined image processing.

[0049] Next, the discrimination execution unit 118 (Figure 2) of the information processing device 100 performs discrimination on each cell nucleus CN using morphological information FI and the cell nucleus discrimination model MO to determine whether or not it is a ganglion cell nucleus CN(Ne) (S350). That is, the discrimination execution unit 118 inputs the morphological information FI of the cell nucleus CN to the cell nucleus discrimination model MO, obtains score information SI which indicates the probability that it is a ganglion cell nucleus CN(Ne) output from the cell nucleus discrimination model MO, and uses the score information SI to determine whether it is a ganglion cell nucleus CN(Ne) or a general cell nucleus CN(Ge). If multiple cell nuclei CN are identified in S330, score information SI is obtained for each cell nucleus CN. Based on the score information SI, the discrimination execution unit 118 generates discrimination result data RD which indicates the discrimination result and stores it in the storage unit 120 of the information processing device 100.

[0050] Next, the discrimination execution unit 118 outputs the discrimination result based on the discrimination result data RD (S360). For example, as shown in Figure 1, the discrimination execution unit 118 displays the biological tissue image Ib (target image Ibo) on the display unit 130, and displays an identifier (such as a mark or letter) indicating that it is a ganglion cell nucleus CN(Ne) and an identifier indicating that it is a general cell nucleus CN(Ge) for each cell nucleus CN on the image. This allows the user to understand the discrimination result. The output of the discrimination result is not limited to display on the display unit 130, but may also be printed via a printer or output as sound via a speaker. With this, the cell nucleus discrimination process is completed.

[0051] A-5. Examples: The following describes an example of the cell nucleus discrimination model MO mentioned above. In creating the cell nucleus discrimination model MO in this example, approximately 350 images of normal areas were obtained from resected intestinal specimens of multiple Hirschsprung's disease patients who underwent surgery at the Department of Pediatric Surgery, Nagoya University Hospital. In each image, the nerve plexus region (NPA) and cell nucleus (CN) were manually identified, and morphological information (FI) was measured for each cell nucleus (CN), and it was labeled whether or not it was a ganglion cell nucleus (CN(Ne)). The cell nucleus discrimination model MO was created by machine learning using a portion of the data thus created as training data (TD). The remaining portion of the data was used as test data to validate the cell nucleus discrimination model MO.

[0052] Figure 6 is an explanatory diagram showing the predictive accuracy of the cell nucleus discrimination model MO in the example. As shown in Figure 6, the cell nucleus discrimination model MO in the example had an agreement rate of 98.7%, a sensitivity (percentage of cases where general cell nuclei CN(Ge) were correctly identified) of 98.6% (=(1784 / 1810)×100), and a specificity (percentage of cases where ganglion cell nuclei CN(Ne) were correctly identified) of 99.3% (=(434 / 437)×100), all of which were extremely high values. Thus, the cell nucleus discrimination model MO in the example showed extremely high discrimination accuracy.

[0053] Figure 7 is an explanatory diagram showing the morphological indices that contribute most to the cell nucleus discrimination model MO of the example. In Figure 7, morphological indices with large coefficient values ​​contribute more to the characteristics of ganglion cell nuclei (CN(Ne)), while morphological indices with small coefficient values ​​contribute more to the characteristics of general cell nuclei (CN(Ge)).

[0054] As shown in Figure 7, characteristics of ganglion cell nuclei (CN(Ne)) include, for example, being round, large, and having strong green brightness unevenness. Characteristics of general cell nuclei (CN(Ge)) include, for example, having strong brightness unevenness in the outline and a dark reddish outline. Of the 47 morphological indicators used to create the cell nucleus discrimination model MO, the 16 morphological indicators shown in Figure 7 can be said to be important indicator values ​​adopted by the cell nucleus discrimination model MO.

[0055] A-6. Analysis of discrimination results: Next, the discrimination result analysis process performed by the information processing device 100 of this embodiment will be described. Figure 8 is a flowchart of the discrimination result analysis process in this embodiment. The discrimination result analysis process is a process for analyzing the discrimination results using the cell nucleus discrimination model MO, which targets multiple cell nuclei CN. The discrimination result analysis process is started when a user operates the operation input unit 140 of the information processing device 100 and inputs a start command.

[0056] First, the discrimination result analysis unit 119 (Figure 2) of the information processing device 100 performs multidimensional clustering using morphological information FI on the discrimination results of multiple cell nuclei CN identified by the discrimination result data RD (S410). In this embodiment, analysis using UMAP (Uniform Manifold Approximation and Projection) is performed as the multidimensional clustering. UMAP is a type of unsupervised learning that performs dimensionality reduction and is a method for converting high-dimensional data into 2D or 3D for visualization. In this embodiment, UMAP analysis is performed using 16 morphological indices that are effective (i.e., used for learning) in the cell nucleus discrimination model MO, as shown in Figure 7, out of the 47 morphological indices that constitute the morphological information FI, and clustering is performed by compressing the 16 dimensions to 2D.

[0057] Figure 9 is an explanatory diagram illustrating an example of the results of multidimensional clustering. As shown in Figure 9, UMAP analysis compresses the morphological indices for each cell nucleus (CN) into two-dimensional data. Furthermore, UMAP analysis enables clustering of the discrimination results for each cell nucleus (CN). In the example in Figure 9, the discrimination results for each cell nucleus (CN) are clustered into three groups (Gr1, Gr2, Gr3).

[0058] In the example shown in Figure 9, Group 1 (Gr1) consists of 234 ganglion cell nuclei (CN(Ne)) and 4 general cell nuclei (CN(Ge)). The ganglion cell nuclei (CN(Ne)) belonging to Group 1 (Gr1) were characterized by their dark color and visible nucleoli. The general cell nuclei (CN(Ge)) belonging to Group 1 (Gr1) were characterized by their dark color and patterned appearance.

[0059] Furthermore, Group 2 (Gr2) consists of 148 ganglion cell nuclei (CN(Ne)) and 139 general cell nuclei (CN(Ge)). The ganglion cell nuclei (CN(Ne)) belonging to Group 2 (Gr2) were lighter in color but, overall, similar to the ganglion cell nuclei (CN(Ne)) belonging to Group 1 (Gr1). The general cell nuclei (CN(Ge)) belonging to Group 2 (Gr2) were darker in color.

[0060] Furthermore, Group 3 (Gr3) consists of 55 ganglion cell nuclei (CN(Ne)) and 1677 general cell nuclei (CN(Ge)). The ganglion cell nuclei (CN(Ne)) belonging to Group 3 (Gr3) were characterized by their dark color and lack of pattern. The general cell nuclei (CN(Ge)) belonging to Group 3 (Gr3) were characterized by their dark color and almost complete lack of pattern.

[0061] Thus, multidimensional clustering allows for mapping based on the morphological (visual) features of cell nuclei (CN). Therefore, multidimensional clustering can support an objective understanding of morphological categories for cell nuclei. Such mapping results can be used, for example, in training on methods for identifying ganglion cell nuclei (CN) (Ne), or in selecting training data for updating or creating new cell nucleus identification models (MOs).

[0062] Figure 10 is an explanatory diagram showing an example of the results of a second multidimensional clustering, targeting only the data belonging to the first group Gr1 and the second group Gr2 in the multidimensional clustering shown in Figure 9. Columns A and B of Figure 10 show the results of the same multidimensional clustering, but column A clearly distinguishes between ganglion cell nuclei (CN(Ne)) and general cell nuclei (CN(Ge)), while column B clearly distinguishes between data belonging to the first group Gr1 and data belonging to the second group Gr2.

[0063] Referring to Figure 10, the analysis results shown in Figure 9 reveal that even in areas where data is concentrated and difficult to identify, a considerable degree of identification is actually possible by extracting specific groups and performing re-analysis. This re-analysis (morphological visualization) reveals what kind of images the difficult-to-identify data consisted of, allowing for the investigation of data deficiencies and the causes of misidentification. Furthermore, it is possible to extract specific groups and create a new cell nucleus discrimination model (MO).

[0064] Furthermore, the discrimination result analysis unit 119 (Figure 2) of the information processing device 100 creates a morphological index heatmap based on the multidimensional clustering results (S420). The morphological index heatmap is a map that represents the individual values ​​of the 2D data using color tones and shades. Note that the process in S420 may be executed in parallel with the process in S410. The processes in S410 and S420 can be collectively referred to as "unsupervised learning".

[0065] Figure 11 is an explanatory diagram showing an example of a morphological index heatmap based on the multidimensional clustering results shown in Figure 9. By referring to such a heatmap, the characteristics of each group identified by multidimensional clustering can be visualized. In Figure 11, the code "Gr1_Ne" represents ganglion cell nuclei (CN(Ne)) belonging to the first group Gr1, and the code "Gr1_Ge" represents general cell nuclei (CN(Ge)) belonging to the first group Gr1. The same applies to other codes. For example, referring to the heatmap shown in Figure 11, it can be said that cell nuclei (CN) belonging to the first group Gr1 are generally large in size and round (high "solidity"). Also, cell nuclei (CN) belonging to the second group Gr2 are characterized by a small variation in brightness values ​​(cv), but a high mean due to their whitish color. Also, cell nuclei (CN) belonging to the third group Gr3 are generally small in size and less round (low "solidity").

[0066] Thus, by referring to morphological index heatmaps based on multidimensional clustering results, it can be said that we can support our understanding of the differences between each morphological category for cell nuclei (CN).

[0067] The discrimination result analysis unit 119 adds information showing the multidimensional clustering results and morphological index heatmap to the discrimination result data RD stored in the storage unit 120.

[0068] Next, the discrimination result analysis unit 119 outputs (i.e., visualizes) the analysis results, more specifically, information showing the combination rules of morphological information FI accumulated in the structure of the cell nucleus discrimination model MO, based on the discrimination result data RD (S430). For example, the discrimination result analysis unit 119 displays the multidimensional clustering results shown in Figures 9 and 10, and the morphological index heatmap shown in Figure 11, on the display unit 130. This allows the user to understand the analysis results. The output of the analysis results is not limited to display on the display unit 130; it may also be printed via a printer or output as audio via a speaker. With this, the discrimination result analysis process is completed.

[0069] A-7. Effects of this embodiment: As described above, the information processing device 100 of this embodiment is an information processing device for determining whether or not a cell nucleus in an image including the nerve plexus region of an intestinal tissue fragment as a biological tissue image is a ganglion cell nucleus CN(Ne), and comprises a model acquisition unit 113 and a determination execution unit 118. The model acquisition unit 113 acquires a cell nucleus determination model MO. The cell nucleus determination model MO is a model generated by machine learning using training data TD, which is data in which morphological information FI indicating the shape of the cell nucleus CN and score information SI indicating the probability that it is a ganglion cell nucleus CN(Ne) are associated for each of a plurality of cell nuclei CN. The determination execution unit 118 uses the morphological information FI for the cell nucleus CN to be determined and the cell nucleus determination model MO to perform determination of whether or not the cell nucleus CN to be determined is a ganglion cell nucleus CN(Ne), and outputs the determination result.

[0070] As described above, the information processing device 100 of this embodiment can use morphological information FI about the cell nucleus CN to be identified and the cell nucleus identification model MO to perform a determination of whether or not the cell nucleus CN to be identified is a ganglion cell nucleus CN(Ne), and output the determination result. Therefore, for each individual cell nucleus CN, it is possible to determine whether or not it is a ganglion cell nucleus CN(Ne) based on the morphology of the cell nucleus CN using the cell nucleus identification model MO. Accordingly, the information processing device 100 of this embodiment can perform the determination of ganglion cell nucleus CN(Ne) based on a unified criterion, with high accuracy and low variability. This makes it possible to provide useful information for appropriate treatment of Hirschsprung's disease or related disorders.

[0071] Furthermore, the information processing device 100 of this embodiment further includes a training data acquisition unit 112 that acquires training data TD. The model acquisition unit 113 acquires a cell nucleus discrimination model MO by creating a cell nucleus discrimination model MO using machine learning with the training data TD. Therefore, the cell nucleus discrimination model MO can be acquired without using other devices, and the discrimination of ganglion cell nuclei CN(Ne) can be performed using this model.

[0072] Furthermore, in the information processing device 100 of this embodiment, the morphological information FI is information that shows the morphology of the cell nucleus CN using multiple morphological indicators. Therefore, it is possible to perform discrimination of ganglion cell nuclei CN(Ne) based on information that captures the morphology of the cell nucleus CN from multiple perspectives, and the discrimination accuracy of ganglion cell nuclei CN(Ne) can be effectively improved.

[0073] Furthermore, in the information processing device 100 of this embodiment, the morphological information FI includes information indicating the contour features of the cell nucleus CN, information indicating the staining characteristics of the cell nucleus CN, and information indicating the pattern-like features of the cell nucleus CN. Therefore, it is possible to perform discrimination of ganglion cell nuclei CN(Ne) based on information that captures the morphology of the cell nucleus CN more comprehensively and accurately, thereby extremely effectively improving the discrimination accuracy of ganglion cell nuclei CN(Ne).

[0074] Furthermore, the information processing device 100 of this embodiment includes a discrimination result analysis unit 119 that performs multidimensional clustering using morphological information FI on the discrimination results for multiple cell nuclei CN and outputs the results of the multidimensional clustering. Therefore, mapping can be performed for each morphological feature of the cell nuclei CN, and it is possible to support an objective understanding of the morphological categories of the cell nuclei CN.

[0075] Furthermore, in the information processing device 100 of this embodiment, the multidimensional clustering is performed using some of the morphological indicators included in the morphological information FI that have a high contribution to the cell nucleus discrimination model MO. Therefore, mapping can be performed for each morphological feature that has a high contribution to the discrimination of the cell nucleus CN, and this can effectively support an objective understanding of the morphological categories of the cell nucleus CN.

[0076] B. Variations: The technologies disclosed herein are not limited to the embodiments described above and can be modified in various forms without departing from their essence, for example, the following modifications are possible.

[0077] The configuration of the information processing device 100 in the above embodiment is merely an example and can be modified in various ways. Furthermore, the contents of the cell nucleus discrimination model acquisition process, cell nucleus discrimination process, and discrimination result analysis process in the above embodiment are merely an example and can be modified in various ways. For example, in the above embodiment, the information processing device 100 acquires the cell nucleus discrimination model MO by creating a cell nucleus discrimination model MO, but the information processing device 100 may acquire the cell nucleus discrimination model MO generated by another device. In this case, the information processing device 100 does not need to have a training data acquisition unit 112.

[0078] The morphological information FI used to create the cell nucleus discrimination model MO in the above embodiment is merely an example and can be modified in various ways. For example, in the above embodiment, the morphological information FI is composed of 47 morphological indicators, but the number of morphological indicators constituting the morphological information FI can be arbitrarily changed. Also, in the above embodiment, the morphological information FI includes information indicating the contour features of the cell nucleus CN, information indicating the degree of staining of the cell nucleus CN, and information indicating the pattern-like features of the cell nucleus CN, but instead, the morphological information FI may include one or two of these. Furthermore, the morphological information FI may include other morphological indicators in addition to these morphological indicators.

[0079] In the above embodiment, the LASSO regression model is used as the machine learning algorithm for creating the cell nucleus discrimination model MO. However, the machine learning algorithm for creating the cell nucleus discrimination model MO can be arbitrarily changed, and other machine learning algorithms such as RIDGE regression, Support Vector Machine, Logistic regression, Random Forest, Neural Network, etc. may be used.

[0080] In the discrimination result analysis process of the above embodiment, UMAP is used as the multidimensional clustering method, but the multidimensional clustering method can be arbitrarily changed, and for example, methods such as Principal component analysis or t-SNE may be used.

[0081] In the above embodiment, the information processing device 100 is equipped with a discrimination result analysis unit 119, and the discrimination result analysis unit 119 performs discrimination result analysis processing. However, the information processing device 100 does not necessarily need to be equipped with a discrimination result analysis unit 119, and the processing does not necessarily need to be performed.

[0082] The above embodiment illustrates information processing for identifying ganglion cell nuclei (CN(Ne)) in the pathological diagnosis of Hirschsprung's disease or related disorders. However, the techniques disclosed herein are not limited to identifying ganglion cell nuclei (CN(Ne)) but are also applicable to determining whether cell nuclei in biological tissue images belong to a specific type of cell nucleus. For example, the techniques disclosed herein are also applicable to identifying inflammatory cell nuclei (for understanding inflammatory responses), neuronal cell nuclei (for elucidating brain function), epithelial and endothelial cell nuclei (for elucidating skin function), cancer cell nuclei (for elucidating cancer invasiveness), and so on.

[0083] In the above embodiment, some of the configuration implemented by hardware may be replaced with software, and conversely, some of the configuration implemented by software may be replaced with hardware. [Explanation of Symbols]

[0084] 100: Information processing unit 110: Control unit 111: Cell nucleus discrimination processing unit 112: Training data acquisition unit 113: Model acquisition unit 114: Target image acquisition unit 115: Region identification unit 116: Cell nucleus identification unit 117: Morphological information acquisition unit 118: Discrimination execution unit 119: Discrimination result analysis unit 120: Storage unit 130: Display unit 140: Operation input unit 150: Interface unit 190: Bus CP: Cell nucleus discrimination program FI: Morphological information Ib: Biological tissue image Ibo: Target image It: Training image MO: Cell nucleus discrimination model NPA: Nerve plexus region RD: Discrimination result data SI: Score information TD: Training data

Claims

1. An information processing device for determining whether or not a cell nucleus in a biological tissue image is of a specific type of cell nucleus, A model acquisition unit for acquiring a cell nucleus discrimination model, wherein the cell nucleus discrimination model is a model generated by machine learning using training data, which is data in which morphological information indicating the shape of the cell nucleus and score information indicating the probability that each of a plurality of cell nuclei is of a specific type are associated. A discrimination execution unit that uses the morphological information of the cell nucleus to be discriminated and the cell nucleus discrimination model to perform a determination as to whether or not the cell nucleus to be discriminated is of the specified type and outputs the result of the determination, Equipped with, The morphological information is information that shows the morphology of a cell nucleus using multiple morphological indicators, and is an information processing device.

2. The information processing apparatus according to claim 1, further, The system includes a training data acquisition unit that acquires the aforementioned training data, The model acquisition unit is an information processing device that acquires the cell nucleus discrimination model by creating the cell nucleus discrimination model using the machine learning method with the training data.

3. An information processing apparatus according to claim 1 or claim 2, The morphological information includes at least one of the following: information indicating the contour features of the cell nucleus; information indicating the degree of staining of the cell nucleus; and information indicating the pattern-like features of the cell nucleus.

4. An information processing apparatus according to claim 3, The morphological information includes information indicating the contour features of the cell nucleus, information indicating the degree of staining of the cell nucleus, and information indicating the pattern-like features of the cell nucleus, in the information processing device.

5. An information processing apparatus according to any one of claims 1 to 4, further, An information processing device comprising a discrimination result analysis unit that analyzes the results of the discrimination for multiple cell nuclei and outputs information that shows the combination rules of the morphological information accumulated in the structure of the cell nucleus discrimination model in a visually identifiable manner.

6. An information processing device according to claim 5, The analysis is performed using a portion of the morphological indicators included in the morphological information that have a high contribution to the cell nucleus discrimination model, according to the information processing device.

7. An information processing device according to any one of claims 1 to 6, The aforementioned biological tissue image is an image of a fragment of intestinal tissue including the nerve plexus region. The aforementioned specific type of cell nucleus is a ganglion cell nucleus, an information processing device.

8. A method for processing information to determine whether or not a cell nucleus in a biological tissue image is of a specific type of cell nucleus, A step to obtain a cell nucleus discrimination model, wherein the cell nucleus discrimination model is a model generated by machine learning using training data, which is data in which morphological information indicating the shape of the cell nucleus and score information indicating the probability that each of a plurality of cell nuclei is of a specific type are associated. A step of using the morphological information of the cell nucleus to be identified and the cell nucleus identification model to determine whether the cell nucleus to be identified is of a specific type and outputting the result of the determination, Equipped with, The morphological information is information that shows the morphology of a cell nucleus using multiple morphological indicators, and is an information processing method.

9. A computer program for determining whether or not a cell nucleus in a biological tissue image is of a specific type of cell nucleus, On the computer, A process for obtaining a cell nucleus discrimination model, wherein the cell nucleus discrimination model is a model generated by machine learning using training data, which is data in which morphological information indicating the shape of the cell nucleus and score information indicating the probability that each of a plurality of cell nuclei is of a specific type are associated. A process that uses the morphological information of the cell nucleus to be identified and the cell nucleus identification model to determine whether the cell nucleus to be identified is of a specific type and outputs the result of the determination, Make it run, The morphological information is a computer program that describes the morphology of a cell nucleus using multiple morphological indicators.

Images