Classification device, learning device, classification method, learning method, and program

Abstract

In order to improve the accuracy of benignancy / malignancy classification of a specimen cell in pathological diagnosis, this classification device (1) comprises an acquisition unit (11) which acquires a pathological image, and a classification unit (12) which uses a classification model to classify whether the specimen cell is a benign cell or a malignant cell, the classification model having inputs of: a feature amount of a first weighted input image processed with first weighted information for emphasizing a first area of interest; and a feature amount of a second weighted input image processed with second weighted information for emphasizing a second area of interest that is different from the first area of interest.

Description

[Technical field]

[0001] The present invention relates to a classification device, a learning device, a classification method, a learning method, and a program. [Background technology]

[0002] A technology is disclosed relating to pathological diagnosis using a trained model that inputs a pathological image and performs image recognition on the pathological image.

[0003] Patent document 1 describes a configuration that outputs, based on an image of an organ to be examined, a classification result based on features that are independent of the type of organ, and a detection result of a special region based on features that are specialized for the organ to be examined. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2012-73179 Summary of the Invention [Problem to be solved by the invention]

[0005] In diagnosis, such as in classifying specimen cells contained as subjects in pathology images as benign or malignant cells, there is a demand for technology that uses a trained model that recognizes pathology images to improve the accuracy of classifying specimen cells as benign or malignant in pathology diagnosis.

[0006] One aspect of the present invention has been made in consideration of the above problems, and one example of an objective of the present invention is to provide a technique for improving the accuracy of classifying specimen cells into benign or malignant in pathological diagnosis. [Means for solving the problem]

[0007] A classification device according to one aspect of the present invention includes: an acquisition means for acquiring a pathology image including a specimen cell as a subject; The method uses a first generative model trained to generate first weighting information for emphasizing a first region of interest in the pathological image as an input, and generates a first weighted input image in which the pathological image is processed with the first weighting information; uses a second generative model trained to generate second weighting information for emphasizing a second region of interest in the pathological image that is different from the first region of interest, and generates a second weighted input image in which the pathological image is processed with the second weighting information; the first weighted input image is input, and the first weighted input image is generated. a classification means for generating features of the first weighted input image using a first feature analysis model trained to generate features of the first weighted input image, receiving the second weighted input image and generating features of the second weighted input image using a second feature analysis model trained to generate features of the second weighted input image, and for classifying the specimen cells as benign cells or malignant cells using a classification model trained to receive the features of the first weighted input image and the features of the second weighted input image, and classifying the specimen cells as benign cells or malignant cells.

[0008] A learning device according to one aspect of the present invention includes an acquisition means for acquiring a pathology image including specimen cells as subjects and classification information indicating whether the specimen cells are benign or malignant cells, and a first generative model that receives the pathology image as an input and generates first weighting information for highlighting a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information, and a second generative model that receives the pathology image as an input and generates second weighting information for highlighting a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information, and the first weighted input image is input and a feature of the first weighted input image is acquired. a learning means for receiving the second weighted input image and generating features of the second weighted input image using a first feature analysis model that generates features of the first weighted input image, receiving the second weighted input image and generating features of the second weighted input image using a second feature analysis model that generates features of the second weighted input image, receiving the features of the first weighted input image and the features of the second weighted input image and classifying the specimen cell as a benign cell or a malignant cell using a classification model that classifies the specimen cell as a benign cell or a malignant cell, and updating parameters of the first generation model, the second generation model, the first feature analysis model, the second feature analysis model and the classification model based on a comparison between a result of the classification and the classification information.

[0009] A classification method according to one aspect of the present invention includes a classification device that acquires a pathology image including specimen cells as a subject, and uses a first generative model that has been trained to generate first weighting information for emphasizing a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information, and uses a second generative model that has been trained to generate second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information, and The method includes: receiving an input image and generating features of the first weighted input image using a first feature analysis model trained to generate features of the first weighted input image; receiving the second weighted input image and generating features of the second weighted input image using a second feature analysis model trained to generate features of the second weighted input image; and receiving the features of the first weighted input image and the features of the second weighted input image and using a classification model trained to classify the specimen cells as benign cells or malignant cells.

[0010] A learning method according to one aspect of the present invention includes a learning device that acquires a pathology image including specimen cells as objects and classification information indicating whether the specimen cells are benign or malignant cells, and generates a first weighted input image in which the pathology image is processed by the first weighting information using a first generative model that receives the pathology image as an input and generates first weighting information for highlighting a first region of interest in the pathology image, and generates a second weighted input image in which the pathology image is processed by the second weighting information using a second generative model that receives the pathology image as an input and generates second weighting information for highlighting a second region of interest different from the first region of interest in the pathology image, and the first weighted input image is input, generating features of the first weighted input image using a first feature analysis model that generates features of the first weighted input image, the second weighted input image is input, and generating features of the second weighted input image using a second feature analysis model that generates features of the second weighted input image, the features of the first weighted input image and the features of the second weighted input image are input, and classifying the specimen cell as whether it is a benign cell or a malignant cell using a classification model that classifies the specimen cell as whether it is a benign cell or a malignant cell, and updating parameters of the first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model based on a comparison between a result of the classification and the classification information.

[0011] A program according to one aspect of the present invention is a program for causing a computer to function as a classification device, the program including: an acquisition means for acquiring a pathology image including specimen cells as a subject; and a first generative model that has been trained to receive the pathology image as an input and generate first weighting information for highlighting a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information; and a second generative model that has been trained to receive the pathology image as an input and generate second weighting information for highlighting a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information. a classification means for generating an image, receiving the first weighted input image and generating features of the first weighted input image using a first feature analysis model trained to generate features of the first weighted input image, receiving the second weighted input image and generating features of the second weighted input image using a second feature analysis model trained to generate features of the second weighted input image, and classifying the specimen cells as benign cells or malignant cells using a classification model trained to classify the specimen cells as benign cells or malignant cells.

[0012] A program according to one aspect of the present invention is a program for causing a computer to function as a learning device, the program including: an acquisition means for acquiring a pathology image including a specimen cell as a subject and classification information indicating whether the specimen cell is a benign cell or a malignant cell; a first generative model for receiving the pathology image as an input and generating first weighting information for highlighting a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information; a second generative model for receiving the pathology image as an input and generating second weighting information for highlighting a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information; and the second weighted input image is input, and a feature of the first weighted input image is generated using a first feature analysis model that generates a feature of the first weighted input image; the second weighted input image is input, and a feature of the second weighted input image is generated using a second feature analysis model that generates a feature of the second weighted input image; the features of the first weighted input image and the features of the second weighted input image are input, and the specimen cell is classified as being a benign cell or a malignant cell using a classification model that classifies the specimen cell as being a benign cell or a malignant cell; and the first generation model, the second generation model, the first feature analysis model, the second feature analysis model, and the classification model are updated based on a comparison between a result of the classification and the classification information. Effect of the Invention

[0013] According to one aspect of the present invention, it is possible to improve the accuracy of classifying specimen cells into benign or malignant in pathological diagnosis. [Brief description of the drawings]

[0014] [Figure 1] FIG. 1 is a block diagram showing a configuration of a classification device according to a first exemplary embodiment of the present invention. [Diagram 2] FIG. 2 is a flow chart showing the flow of a classification method according to the first exemplary embodiment of the present invention. [Diagram 3] FIG. 1 is a block diagram showing a configuration of a learning device according to a first exemplary embodiment of the present invention. [Figure 4] FIG. 2 is a flow chart showing the flow of a learning method according to the first exemplary embodiment of the present invention. [Diagram 5] FIG. 11 is a block diagram showing a configuration of a classification device according to an exemplary embodiment 2 of the present invention. [Figure 6] FIG. 11 is a schematic diagram showing a flow of processing executed by a classification unit according to the second exemplary embodiment of the present invention. [Figure 7] FIG. 2 is a block diagram showing an example of a hardware configuration of a classification device and a learning device according to each exemplary embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Example Embodiment 1 A first exemplary embodiment of the present invention will be described in detail with reference to the drawings. This exemplary embodiment is a basic form of the exemplary embodiments described later.

[0016] (Overview of Classification Device 1) The classification device 1 according to this exemplary embodiment is a device that classifies specimen cells as to whether they are benign cells or malignant cells.

[0017] As an example, the classification device 1 first receives as input a pathology image containing specimen cells as a subject, and uses a generative model that has been trained to generate weighting information for highlighting areas of interest in the pathology image to generate an image in which the pathology image has been weighted using the weighting information (hereinafter referred to as an "input image in which the pathology image has been processed using the weighting information").

[0018] Next, the classification device 1 receives the weighted input image and generates features of the weighted input image using a feature analysis model that has been trained to generate features of the weighted input image.

[0019] The classification device 1 receives the weighted feature amount of the input image and classifies the specimen cells as benign or malignant cells using a classification model that has been trained to classify the specimen cells as benign or malignant cells. As an example, the classification device 1 can be used for cytology in rapid on-site evaluation (ROSE).

[0020] An example of weighting information is an attention map that indicates an attention area by setting a weight for each area. The following describes the case where the weighting information is the attention map.

[0021] In addition, the number of generative models and feature analysis models is not particularly limited, but as an example, the following describes the use of two generative models (hereinafter, a first generative model and a second generative model) and feature analysis models (a first feature analysis model and a second feature analysis model).

[0022] In addition, in the following, the attention ups generated by the first and second generative models are referred to as the first and second attention maps, respectively. Also, the input images processed by the first and second attention maps are referred to as the first and second weighted input images, respectively. Also, the attention regions in the first and second weighted input images, i.e., the regions highlighted in the attention maps, are referred to as the first and second attention regions, respectively.

[0023] Although the specific configurations of the generative model, the feature analysis model, and the classification model do not limit this exemplary embodiment, as an example, a deep neural network having a convolutional layer such as a CNN (Convolution Neural Network) can be used. As the convolutional layer, a network having the following configuration can be used, although it is not limited thereto. · Batch normalization layer Activation layers (e.g., ReLU (Rectified Linear Unit) layers) · 2D convolutional layer Pooling Layer (Configuration of Classification Device 1) The configuration of a classification device 1 according to this exemplary embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of a classification device 1 according to this exemplary embodiment.

[0024] 1, the classification device 1 includes an acquisition unit 11 and a classification unit 12. The acquisition unit 11 and the classification unit 12 are components that respectively realize an acquisition means and a classification means in this exemplary embodiment.

[0025] The acquiring unit 11 acquires a pathology image including a specimen cell as a subject. The acquiring unit 11 supplies the acquired pathology image to the classifying unit 12.

[0026] The classifier 12 uses a first generative model to generate a first weighted input image in which the pathological image is processed by the first attention map.

[0027] Furthermore, the classification unit 12 uses a second generative model to generate a second weighted input image in which the pathological image is processed by the second attention map.

[0028] Moreover, the classification unit 12 generates a first weighted feature amount of the input image using the first feature analysis model.

[0029] Moreover, the classification unit 12 generates second weighted features of the input image using a second feature analysis model.

[0030] Furthermore, the classification unit 12 classifies the specimen cells as benign cells or malignant cells using a classification model.

[0031] As described above, the classification device 1 according to this exemplary embodiment includes an acquisition unit 11 that acquires a pathology image including specimen cells as a subject, and a first generative model that has been trained to receive the pathology image as an input and generate a first attention map for highlighting a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed by the first attention map, a second generative model that has been trained to receive the pathology image as an input and generate a second attention map for highlighting a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed by the second attention map, and a classification unit (12) that receives an input image having a weighted component and generates features of the first weighted input image using a first feature analysis model that has been trained to generate features of the first weighted input image, a classification unit (12) that receives an input image having a weighted component and generates features of the second weighted input image using a second feature analysis model that has been trained to generate features of the second weighted input image, and a classification unit (12) that receives an input image having a weighted component and generates features of the second weighted input image using a classification model that has been trained to classify the specimen cells as benign cells or malignant cells.

[0032] Therefore, the classification device 1 according to this exemplary embodiment focuses on a first region of interest and a second region of interest, which are different regions in a pathological image, to classify whether a specimen cell is a benign cell or a malignant cell. For example, the classification device 1 according to this exemplary embodiment can differentiate a region of interest depending on the type of cells contained in a pathological image in a pathological diagnosis for classifying whether a specimen cell is a benign cell or a malignant cell. Therefore, the classification device 1 according to this exemplary embodiment can provide an effect of improving the accuracy of classification of specimen cells into benign and malignant cells in a pathological diagnosis.

[0033] (Classification method S1 flow) The flow of the classification method S1 according to this exemplary embodiment will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of the classification method S1 according to this exemplary embodiment.

[0034] (Step S11) In step S11, the acquiring unit 11 acquires a pathology image including a specimen cell as a subject. The acquiring unit 11 supplies the acquired pathology image to the classifying unit 12.

[0035] (Step S12) In step S12, the classification unit 12 uses the first generative model to generate a first weighted input image in which the pathological image is processed by the first attention map.

[0036] Furthermore, the classification unit 12 uses a second generative model to generate a second weighted input image in which the pathological image is processed by the second attention map.

[0037] Moreover, the classification unit 12 generates a first weighted feature amount of the input image using the first feature analysis model.

[0038] Moreover, the classification unit 12 generates second weighted features of the input image using a second feature analysis model.

[0039] Furthermore, the classification unit 12 classifies the specimen cells as benign cells or malignant cells using a classification model.

[0040] As described above, the classification method S1 according to this exemplary embodiment includes acquiring a pathology image including specimen cells as a subject, using a first generative model trained to use the pathology image as an input and generate a first attention map for highlighting a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed by the first attention map, using a second generative model trained to use the pathology image as an input and generate a second attention map for highlighting a second region of interest different from the first region of interest in the pathology image to generate a second weighted image in which the pathology image is processed by the second attention map, and The configuration includes: inputting an input image having a weighted image, generating a feature of the first weighted input image using a first feature analysis model trained to generate a feature of the first weighted input image; inputting a second weighted input image, generating a feature of the second weighted input image using a second feature analysis model trained to generate a feature of the second weighted input image; inputting the feature of the first weighted input image and the feature of the second weighted input image, and classifying the specimen cells as benign or malignant using a classification model trained to classify the specimen cells as benign or malignant. Therefore, the classification method S1 according to this exemplary embodiment can achieve the same effect as the above-mentioned classification device 1.

[0041] (Overview of learning device 2) The learning device 2 according to this exemplary embodiment is a device that updates parameters of a first generative model, a second generative model, a first feature analysis model, a second feature analysis model, and a classification model.

[0042] As an example, the learning device 2 acquires a pathology image including a specimen cell as a subject, and classification information (correct label) indicating whether the specimen cell is a benign cell or a malignant cell, and updates the parameters of each model based on a comparison between the classification result obtained by the classification model and the classification information.

[0043] The first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model are as described above.

[0044] (Configuration of learning device 2) The configuration of the learning device 2 according to this exemplary embodiment will be described with reference to Fig. 3. Fig. 3 is a block diagram showing the configuration of the learning device 2 according to this exemplary embodiment.

[0045] 3, the learning device 2 includes an acquisition unit 21 and a learning unit 22. The acquisition unit 21 and the learning unit 22 are components that respectively realize an acquisition means and a learning means in this exemplary embodiment.

[0046] The acquiring unit 21 acquires a pathology image including a specimen cell as a subject, and classification information indicating whether the specimen cell is a benign cell or a malignant cell. The acquiring unit 11 supplies the acquired pathology image and classification information to the learning unit 22.

[0047] The learning unit 22 updates parameters of the first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model based on a comparison between the result of classification by the classification model and the classification information, thereby learning the first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model. As an example, the learning device 2 can be used to learn a classification model used for cytology in rapid on-site evaluation (ROSE).

[0048] As described above, the learning device 2 according to this exemplary embodiment includes an acquisition unit 21 that acquires a pathology image including specimen cells as subjects and classification information indicating whether the specimen cells are benign or malignant cells, an acquisition unit 22 that receives the pathology image as an input and uses a first generation model that generates first weighting information for highlighting a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information, and a second generation model that receives the pathology image as an input and generates second weighting information for highlighting a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information, and the first weighted input image is input and the first weighted input image is processed. a learning unit 22 that generates features of a first weighted input image using a first feature analysis model that generates features of a force image, a second weighted input image is input, a second feature analysis model that generates features of the second weighted input image is used to generate features of the second weighted input image, the features of the first weighted input image and the features of the second weighted input image are input, and the specimen cells are classified as benign cells or malignant cells using a classification model that classifies the specimen cells as benign cells or malignant cells, and updates parameters of the first generation model, the second generation model, the first feature analysis model, the second feature analysis model and the classification model based on a comparison between the classification result and the classification information.

[0049] Therefore, the learning device 2 according to this exemplary embodiment can accurately learn a first generation model that generates a first attention map, a second generation model that generates a second attention map, a first feature analysis model that generates a feature amount of a first weighted input image, a second feature analysis model that generates a feature amount of a second weighted input image, and a classification model that classifies whether a specimen cell is a benign cell or a malignant cell. Therefore, the learning device 2 according to this exemplary embodiment can provide an effect of improving the accuracy of classification of specimen cells into benign and malignant cells in pathological diagnosis.

[0050] (Flow of learning method S2) The flow of the learning method S2 according to this exemplary embodiment will be described with reference to FIG. 4 1 is a flow diagram showing the flow of a learning method S2 according to this exemplary embodiment.

[0051] (Step S21) In step S21, the acquisition unit 21 acquires a pathology image including a specimen cell as a subject, and classification information (correct answer label) indicating whether the specimen cell is a benign cell or a malignant cell. The acquisition unit 11 supplies the acquired pathology image and classification information to the learning unit 22.

[0052] (Step S22) Learning section 22 , minutes Based on a comparison between the result of classification by the class model and the classification information, parameters of the first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model are updated.

[0053] As described above, the learning method S2 according to this exemplary embodiment includes obtaining a pathology image including specimen cells as subjects and classification information indicating whether the specimen cells are benign or malignant cells; using a first generative model that takes the pathology image as input and generates first weighting information for highlighting a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information; using a second generative model that takes the pathology image as input and generates second weighting information for highlighting a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information; The configuration includes: generating a feature quantity of the first weighted input image using a first feature analysis model that generates a feature quantity of the input image; inputting a second weighted input image; generating a feature quantity of the second weighted input image using a second feature analysis model that generates a feature quantity of the second weighted input image; inputting the feature quantity of the first weighted input image and the feature quantity of the second weighted input image; classifying the specimen cell as a benign cell or a malignant cell using a classification model that classifies the specimen cell as a benign cell or a malignant cell; and updating parameters of the first generation model, the second generation model, the first feature analysis model, the second feature analysis model, and the classification model based on a comparison between the classification result and the classification information. Therefore, the learning method S2 according to this exemplary embodiment can achieve the same effect as the learning device 2 described above.

[0054] Exemplary embodiment 2 A second exemplary embodiment of the present invention will be described in detail with reference to the drawings. Note that components having the same functions as those described in the first exemplary embodiment are given the same reference numerals, and the description thereof will be omitted as appropriate.

[0055] (Overview of Classification Device 1A) The classification device 1A according to this exemplary embodiment is a device that classifies specimen cells as to whether they are benign cells or malignant cells.

[0056] As an example, the classification device 1A first receives as input a pathology image containing specimen cells as a subject, and uses a generative model that has been trained to generate weighting information for highlighting areas of interest in the pathology image to generate an image in which the pathology image has been weighted using the weighting information (hereinafter referred to as an "input image in which the pathology image has been processed using the weighting information").

[0057] Next, the classification device 1A receives the weighted input image and generates features of the weighted input image using a feature analysis model that has been trained to generate features of the weighted input image.

[0058] Then, the classification device 1A receives the features of the weighted input image and classifies the specimen cells as benign cells or malignant cells using a classification model that has been trained to classify the specimen cells as benign cells or malignant cells.

[0059] The specimen cells included as subjects in the pathology image are cells that have been stained in advance by a staining method that stains at least the cytoplasm, such as Diff-Quik staining or Papanicolaou staining.

[0060] Also in this exemplary embodiment, a case will be described in which the weighting information is an attention map.

[0061] In addition, in this exemplary embodiment, the number of generative models and feature analysis models is not particularly limited, but as an example, the following describes the case where two generative models (hereinafter, a first generative model and a second generative model) and feature analysis models (a first feature analysis model and a second feature analysis model) are used.

[0062] The first attention map, the second attention map, the first weighted input image, the second weighted input image, the first attention region, the second attention region, the generative model, the feature analysis model, and the classification model are as described in the above-mentioned embodiment.

[0063] Here, the first region of interest in the first weighted input image and the second region of interest in the second weighted input image are not particularly limited as long as they are different regions from each other, but as an example, at least the cytoplasm of the specimen cells is stained, and the first region of interest and the second region of interest are regions having different staining intensities from each other.

[0064] As an example of the first and second regions of interest, at least the cytoplasm of a specimen cell may be stained, and in the weighted input image, a region where the staining intensity is equal to or greater than a first threshold may be set as the first region of interest, and a region where the staining intensity is equal to or greater than a second threshold that is lower than the first threshold and is lower than the first threshold may be set as the second region of interest.

[0065] As another example of the first region of interest and the second region of interest, the first region of interest is a region having a higher staining intensity than the second region of interest, and the second region of interest is a region containing stained cilia.

[0066] As yet another example of the first and second regions of interest, the first region of interest is a region whose distance from the pixel with the highest staining intensity in the cell is shorter than a first predetermined length (in other words, a region close to the center of the cell or the center of the nucleolus), and the second region of interest is a region whose distance from the center of the cell or the center of the nucleolus is shorter than a second predetermined length that is longer than the first predetermined length, and longer than the first predetermined length.

[0067] The classification device 1A is also a device that updates parameters of the first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model. In other words, the classification device 1A also has the configuration of the learning device 2 described above.

[0068] As an example, the classification device 1A acquires a pathology image including a specimen cell as a subject and classification information indicating whether the specimen cell is a benign cell or a malignant cell, and updates the parameters of each model based on a comparison between the classification results obtained by the classification model and the classification information.

[0069] (Configuration of Classification Device 1A) The classification device 1A according to this exemplary embodiment includes a control unit 10, a storage unit 30, and an input / output unit 31, as shown in FIG.

[0070] The storage unit 30 stores data to be referenced by the control unit 10, which will be described later. Examples of the data stored in the storage unit 30 include pathological images PP and classification information CI.

[0071] The input / output unit 31 is an interface that acquires data from other connected devices or outputs data to other connected devices. The input / output unit 31 supplies data acquired from other devices to the control unit 10 and outputs data supplied from the control unit 10 to other devices.

[0072] The input / output unit 31 may be a communication module that communicates with other devices via a network. Although the specific configuration of the network does not limit the present exemplary embodiment, as an example, a wireless LAN (Local Area Network), a wired LAN, a WAN (Wide Area Network), a public line network, a mobile data communication network, or a combination of these networks can be used.

[0073] (Control unit 10) The control unit 10 controls each component included in the classification device 1A. As shown in Fig. 5, the control unit 10 also functions as an acquisition unit 11A, a classification unit 12, and a learning unit 22. The acquisition unit 11A, the classification unit 12, and the learning unit 22 are components that respectively realize an acquisition means, a classification means, and a learning means in this exemplary embodiment.

[0074] The acquisition unit 11A acquires data supplied from the input / output unit 31. The acquisition unit 11A stores the acquired data in the storage unit 30. The acquisition unit 11A further includes a first acquisition unit 110 and a second acquisition unit 120, as shown in FIG.

[0075] The first acquisition section 110 acquires a pathology image PP including a specimen cell as a subject. In other words, the first acquisition section 110 has the functions of the acquisition section 11 in the above-described exemplary embodiment.

[0076] The second acquisition unit 120 acquires a pathology image including a specimen cell as a subject, and classification information indicating whether the specimen cell is a benign cell or a malignant cell. In other words, the second acquisition unit 120 has the function of the acquisition unit 21 in the above-mentioned exemplary embodiment.

[0077] The classifier 12 uses a first generative model to generate a first weighted input image in which the pathological image is processed by the first attention map.

[0078] Furthermore, the classification unit 12 uses a second generative model to generate a second weighted input image in which the pathological image is processed by the second attention map.

[0079] Moreover, the classification unit 12 generates a first weighted feature amount of the input image using the first feature analysis model.

[0080] Moreover, the classification unit 12 generates second weighted features of the input image using a second feature analysis model.

[0081] Furthermore, the classification unit 12 classifies the specimen cells as benign cells or malignant cells using a classification model.

[0082] The learning unit 22 updates parameters of the first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model based on a comparison between the result of classification by the classification model and the classification information.

[0083] (Overview of Processing Executed by Classification Unit 12) The process executed by the classification unit 12 will be described with reference to Fig. 6. Fig. 6 is a schematic diagram showing the flow of the process executed by the classification unit 12 according to this exemplary embodiment.

[0084] First, the classification unit 12 acquires a pathological image PP including a specimen cell as a subject from the storage unit 30. The classification unit 12 supplies the acquired pathological image PP to the first generative model GM1 and the second generative model GM2.

[0085] Next, the classifier 12 generates a weighted input image that has been processed by the attention map AM generated by the generative model GM.

[0086] As an example, in the case of an attention map in which the area of ​​interest is represented using weights of 0 to 1, the classification unit 12 takes a pathological image PP as input and multiplies the first attention map AM1 generated by the first generative model GM1 and each corresponding pixel value in the pathological image PP to generate a first weighted input image processed by the first attention map AM1.

[0087] Similarly, the classification unit 12 takes the pathological image PP as input and generates a second weighted input image processed by the second attention map AM2 by multiplying each corresponding pixel value in the pathological image PP with the second attention map AM2 generated by the second generative model GM2.

[0088] Next, the classification unit 12 generates a feature F3 by combining a feature F1 of the first weighted input image generated by a first feature analysis model using the generated first weighted input image as an input and a feature F2 of the second weighted input image generated by a second feature analysis model using the generated second weighted input image as an input. As an example, the features F1 and F2 are usually vectors of 1024 elements.

[0089] The classification unit 12 then inputs the feature amount F3 into the classification model CM, and refers to the classification result CR of the classification model to classify the specimen cell as either a benign cell or a malignant cell.

[0090] As an example, when the classification result CR indicates that the specimen cells included as the subject in the pathological image PP are benign cells, the classification unit 12 classifies the specimen cells included as the subject in the pathological image PP as benign cells. On the other hand, when the classification result CR indicates that the specimen cells included as the subject in the pathological image PP are malignant cells, the classification unit 12 classifies the specimen cells included as the subject in the pathological image PP as malignant cells.

[0091] Here, the classification unit 12 may skip the processes by the first generative model GM1 and the second generative model GM2 that generate the attention map.

[0092] As an example, in the case of an attention map in which areas of interest are represented using weights between 0 and 1, the attention maps generated by the first generative model GM1 and the second generative model GM2 may be set so that the weight is 1 in all areas of the pathological image PP.

[0093] The classification unit 12 generates a weighted input image by multiplying the corresponding pixel values ​​in the attention map AM and the pathological image PP. Therefore, when the attention map is set so that the weight in all areas of the pathological image PP is 1, the classification unit 12 skips the processing by the first generation model GM1 and the second generation model GM2, and inputs the pathological image PP to the feature analysis model as a weighted input image.

[0094] (Processing Executed by Learning Unit 22) The process executed by the learning unit 22 will be described below.

[0095] As an example, the learning unit 22 learns to generate a first attention map AM1 for highlighting a first region of interest having high staining intensity, and a second attention map AM2 for highlighting a second region of interest having a lower staining intensity than the first region of interest and a medium staining intensity, so that it is possible to accurately determine whether a sample cell is a malignant cell or a benign cell.

[0096] As another example, the learning unit 22 learns to generate a second attention map AM2 for highlighting a second region including stained cilia and a first attention map AM1 for highlighting a first region having a higher staining intensity than the second region, so that it is possible to accurately determine whether a sample cell is a malignant cell or a benign cell.

[0097] As an example, the following describes a configuration in which a first attention map AM1 is generated by using an area where the staining intensity is equal to or greater than a first threshold as the first attention area, and a second generative model GM2 generates a second attention map AM2 by using an area where the staining intensity is equal to or greater than a second threshold that is lower than the first threshold and is lower than the first threshold as the second attention area.

[0098] The learning unit 22 first To set the initial values ​​of the weights, The first generative model GM1 and the second generative model GM 2 As an example, the learning unit 22 sets As an initial value, The weight of the first region of interest, which is the region where the staining intensity is equal to or greater than the first threshold (for example, in a 256-level image, the region where the pixel value is equal to or greater than 180), is set to 1, and the weight of the other regions is set to 0. Hello , the first generative model GM1.

[0099] In addition, the learning unit 22 As an initial value, The weight of the second region of interest, which is a region in which the staining intensity is equal to or greater than a second threshold lower than the first threshold and is lower than the first threshold (for example, in a 256-level image, the region with pixel values ​​between 64 and 179), is set to 1, and the weight of the other regions is set to 0. Hello , and set it to the second generation model GM2. In this way, the learning unit 22 may set the initial values ​​of the first attention region and the second attention region in the first attention map AM1 and the second attention map AM2 to be relatively higher than the other regions.

[0100] Here, the learning unit 22 may further divide the above-mentioned regions other than the first region of interest and the second region of interest, and set weights. As an example, the learning unit 22 may set an initial value for the second generative model GM2 such that the weight of a region where the color intensity is equal to or greater than a first threshold (for example, in the case of an image with 256 gradations, a region where the pixel value is equal to or greater than 180) is 0.5.

[0101] The learning unit 22 may also set initial values ​​of weights differently depending on pixel values ​​in the first and second attention regions. In other words, the learning unit 22 may set initial values ​​of the first attention map AM1 and the second attention map AM2 so that the first and second attention regions are regions with different staining intensities. As an example, the learning unit 22 may set the initial values ​​of the first generative model GM1 to 1 for a region with pixel values ​​of 180 to 209, 0.9 for a region with pixel values ​​of 210 to 229, and 0.8 for a region with pixel values ​​of 230 to 255. The learning unit 22 may also set initial values ​​of weights differently depending on pixel values ​​in regions other than the first and second attention regions.

[0102] Next, the learning unit 22 updates parameters of the first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model using an optimization algorithm such as the steepest descent method based on a comparison between the classification result CR and the classification information CI.

[0103] As an example, if the classification result CR indicates that the specimen cells are benign cells and the classification information CI indicates that the specimen cells are malignant cells, the learning unit 22 updates the parameters of the first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model so that when a pathological image PP is input to the classification unit 12, the classification result CR outputs a classification result indicating that the specimen cells are malignant cells.

[0104] By repeating the above, the first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model can be trained with high accuracy.

[0105] (Effect 1 of Classification Device 1A) As described above, the classification device 1A according to this exemplary embodiment is configured to include an acquisition unit 11A that acquires a pathology image PP including specimen cells as a subject, and a classification unit 12 that classifies the specimen cells as benign or malignant based on a feature F1 of a first weighted input image generated using a first attention map AM1 for emphasizing a first region of interest, and a feature F2 of a second weighted input image generated using a second attention map AM2 for emphasizing a second region of interest.

[0106] For example, if a certain cell can be classified as either a benign cell or a malignant cell by focusing on areas with high staining intensity, and another cell different from the certain cell can be classified as either a benign cell or a malignant cell by focusing on areas with relatively low staining intensity, in the classification device 1A of this exemplary embodiment, the first and second regions of interest can be set to regions with different staining intensities.

[0107] Therefore, the classification device 1A according to this exemplary embodiment can improve the accuracy of classifying specimen cells into benign or malignant in pathological diagnosis.

[0108] As another example, a certain cell can be classified as a malignant cell by focusing on the vicinity of the nucleolus, and another cell different from the certain cell can be classified as a malignant cell by focusing on the vicinity of the cilia. other When a cell can be classified as a benign cell, in the classification device 1A according to this exemplary embodiment, a first region of interest can be set to a region having a higher staining intensity than a second region of interest, and the second region of interest can be set to a region containing stained cilia.

[0109] Therefore, the classification device 1A according to this exemplary embodiment can improve the accuracy of classifying specimen cells into benign or malignant in pathological diagnosis.

[0110] (Effect 2 of Classification Device 1A) Furthermore, in the classification device 1A according to this exemplary embodiment, parameters of the first generation model, the second generation model, the first feature analysis model, the second feature analysis model, and the classification model are updated based on a comparison between the classification result CR and the classification information CI. Therefore, according to the learning device 2 according to this exemplary embodiment, the first generation model, the second generation model, the first feature analysis model, the second feature analysis model, and the classification model can be trained with high accuracy, thereby obtaining an effect of improving the accuracy of classification of specimen cells into benign or malignant in pathological diagnosis.

[0111] [Software implementation example] Classifier 1, 1A Some or all of the functions of the learning device 2 may be realized by hardware such as an integrated circuit (IC chip), or may be realized by software.

[0112] In the latter case, classifier 1, 1A The learning device 2 is realized, for example, by a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter, referred to as computer C) is shown in FIG. 7. The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores the computer C as the classification device 1, 1A , and a program P for operating the learning device 2 are recorded in the memory C2. In the computer C, the processor C1 reads the program P from the memory C2 and executes it to operate the classification device 1, 1A , each function of the learning device 2 is realized.

[0113] The processor C1 may be, for example, a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a microcontroller, or a combination of these. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination of these.

[0114] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may further include a communication interface for transmitting and receiving data to and from other devices. The computer C may further include an input / output interface for connecting input / output devices such as a keyboard, a mouse, a display, and a printer.

[0115] Furthermore, the program P can be recorded on a non-transitory tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can obtain the program P via such a recording medium M. Furthermore, the program P can be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also obtain the program P via such a transmission medium.

[0116] [Additional Note 1] The present invention is not limited to the above-described embodiment, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining the technical means disclosed in the above-described embodiment are also included in the technical scope of the present invention.

[0117] [Additional Note 2] Some or all of the above-described embodiments can be described as follows. However, the present invention is not limited to the following described aspects.

[0118] (Appendix 1) An acquisition means for acquiring a pathology image including a specimen cell as a subject; The method uses a first generative model that has been trained to use the pathology image as an input and to generate first weighting information for emphasizing a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information, and uses a second generative model that has been trained to use the pathology image as an input and to generate second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information, and the first weighted input image is input and a feature of the first weighted input image is processed. a classification means for generating features of the first weighted input image using a first feature analysis model trained to generate features of the first weighted input image, for receiving the second weighted input image and generating features of the second weighted input image using a second feature analysis model trained to generate features of the second weighted input image, and for classifying the specimen cells as benign cells or malignant cells using a classification model trained to receive the features of the first weighted input image and the features of the second weighted input image,

[0119] (Appendix 2) 2. The classification device according to claim 1, wherein at least the cytoplasm of the specimen cell is stained, and the first region of interest and the second region of interest are regions having different staining intensities.

[0120] (Appendix 3) 3. The classification device of claim 2, wherein the second region of interest is a region including stained cilia, and the first region of interest is a region having a higher staining intensity than the second region of interest.

[0121] (Appendix 4) The present invention relates to a pathology image processing system, comprising: an acquisition means for acquiring a pathology image including specimen cells as a subject and classification information indicating whether the specimen cells are benign or malignant; a first generation model that receives the pathology image as an input and generates first weighting information for emphasizing a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information; a second generation model that receives the pathology image as an input and generates second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information; and a first feature generation system that receives the first weighted input image and generates a feature amount of the first weighted input image. and learning means for generating features of the first weighted input image using a feature analysis model, generating features of the second weighted input image using a second feature analysis model to which the second weighted input image is input and which generates features of the second weighted input image, classifying the specimen cell as whether it is a benign cell or a malignant cell using a classification model to which the features of the first weighted input image and the features of the second weighted input image are input and which classifies the specimen cell as whether it is a benign cell or a malignant cell, and updating parameters of the first generation model, the second generation model, the first feature analysis model, the second feature analysis model, and the classification model based on a comparison between a result of the classification and the classification information.

[0122] (Appendix 5) The learning device described in Appendix 4, wherein at least the cytoplasm of the specimen cell is stained, and the learning means sets initial values ​​of the first weighting information and the second weighting information so that the first region of interest and the second region of interest are regions having different staining intensities.

[0123] (Appendix 6) The learning device described in Appendix 5, wherein the second region of interest is a region including stained cilia, the first region of interest is a region having a higher staining intensity than the second region of interest, and the learning means sets initial values ​​of the first region of interest and the second region of interest in the first weighting information and the second weighting information to be relatively higher than other regions.

[0124] (Appendix 7) a classification device acquires a pathology image including specimen cells as a subject, and generates a first weighted input image in which the pathology image is processed by the first weighting information using a first generative model that has been trained to generate first weighting information for emphasizing a first region of interest in the pathology image, and generates a second weighted input image in which the pathology image is processed by the second weighting information using a second generative model that has been trained to generate second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image, the first weighted input image being input; generating features of the first weighted input image using a first feature analysis model trained to generate features of the first weighted input image; a second weighted input image is input, and generating features of the second weighted input image using a second feature analysis model trained to generate features of the second weighted input image; and the features of the first weighted input image and the features of the second weighted input image are input, and classifying the specimen cell as whether it is a benign cell or a malignant cell using a classification model trained to classify the specimen cell as whether it is a benign cell or a malignant cell.

[0125] (Appendix 8) a learning device that acquires a pathology image including a specimen cell as a subject and classification information indicating whether the specimen cell is a benign cell or a malignant cell; a first generation model that receives the pathology image as an input and generates first weighting information for emphasizing a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information; a second generation model that receives the pathology image as an input and generates second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information; generating features of the first weighted input image using a first feature analysis model, generating features of the second weighted input image using a second feature analysis model to which the second weighted input image is input and generating features of the second weighted input image, classifying the specimen cell as a benign cell or a malignant cell using a classification model to which the features of the first weighted input image and the features of the second weighted input image are input and which classifies the specimen cell as a benign cell or a malignant cell, and updating parameters of the first generation model, the second generation model, the first feature analysis model, the second feature analysis model, and the classification model based on a comparison between a result of the classification and the classification information.

[0126] (Appendix 9) A program for causing a computer to function as a classification device, the program comprising: an acquisition means for acquiring a pathology image including a specimen cell as a subject; and a first generative model that has been trained to generate first weighting information for emphasizing a first region of interest in the pathology image, and generating a first weighted input image in which the pathology image is processed by the first weighting information, and a second generative model that has been trained to generate second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image, and generating a second weighted input image in which the pathology image is processed by the second weighting information, and a classification means for receiving an input of a weighted input image and generating features of the first weighted input image using a first feature analysis model trained to generate features of the first weighted input image, receiving an input of a second weighted input image and generating features of the second weighted input image using a second feature analysis model trained to generate features of the second weighted input image, and receiving an input of the features of the first weighted input image and the features of the second weighted input image and generating features of the second weighted input image using a classification model trained to classify the specimen cell as whether it is a benign cell or a malignant cell.

[0127] (Appendix 10) A program for causing a computer to function as a learning device, the program comprising: an acquisition means for acquiring a pathology image including a specimen cell as a subject and classification information indicating whether the specimen cell is a benign cell or a malignant cell, and a first generation model for receiving the pathology image as an input and generating first weighting information for highlighting a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information, and a second generation model for receiving the pathology image as an input and generating second weighting information for highlighting a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information, the first weighted input image being input and the first weighting information being used to generate a second weighted input image in which the pathology image is processed with the second weighting information, a learning means for receiving the second weighted input image and generating features of the second weighted input image using a first feature analysis model that generates features of the first weighted input image, receiving the second weighted input image and generating features of the second weighted input image using a second feature analysis model that generates features of the second weighted input image, receiving the features of the first weighted input image and the features of the second weighted input image and using a classification model that classifies the specimen cell as a benign cell or a malignant cell, and updating parameters of the first generation model, the second generation model, the first feature analysis model, the second feature analysis model, and the classification model based on a comparison between a result of the classification and the classification information.

[0128] [Additional Note 3] A part or all of the above-described embodiments can be further expressed as follows.

[0129] The system includes at least one processor, the processor performing an acquisition process of acquiring a pathology image including specimen cells as a subject, a process of taking the pathology image as an input and using a first generative model trained to generate first weighting information for emphasizing a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information, a process of taking the pathology image as an input and using a second generative model trained to generate second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information, and a process of extracting the first weighted input image from the first weighted input image. a classification process in which a first weighted input image is input, and features of the first weighted input image are generated using a first feature analysis model that has been trained to generate features of the first weighted input image; a second weighted input image is input, and features of the second weighted input image are generated using a second feature analysis model that has been trained to generate features of the second weighted input image; and a classification process in which the features of the first weighted input image and the features of the second weighted input image are input, and the specimen cells are classified as benign cells or malignant cells using a classification model that has been trained to classify the specimen cells as benign cells or malignant cells.

[0130] The classification device may further include a memory, and the memory may store a program for causing the processor to execute the acquisition process and the classification process. The program may be recorded in a computer-readable, non-transitory, tangible recording medium.

[0131] The system includes at least one processor, the processor performing an acquisition process of acquiring a pathology image including specimen cells as a subject and classification information indicating whether the specimen cells are benign cells or malignant cells, a process of taking the pathology image as an input and using a first generation model that generates first weighting information for emphasizing a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information, and a process of taking the pathology image as an input and using a second generation model that generates second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information, the first weighted input image being input and a learning process in which a first feature analysis model that generates a feature is used to generate a feature of the first weighted input image, the second weighted input image is input, and a second feature analysis model that generates a feature of the second weighted input image is used to generate a feature of the second weighted input image, the feature of the first weighted input image and the feature of the second weighted input image are input, and the specimen cell is classified as whether it is a benign cell or a malignant cell is used to classify the specimen cell as whether it is a benign cell or a malignant cell, and parameters of the first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model are updated based on a comparison between a result of the classification and the classification information.

[0132] The learning device may further include a memory, and the memory may store a program for causing the processor to execute the acquisition process and the learning process. The program may be recorded in a computer-readable, non-transitory, tangible recording medium. [Explanation of symbols]

[0133] 1, 1A classification device 2 Learning device 11, 11A, 21 Acquisition Department 110 First acquisition part 120 Second acquisition part 12 Classification section 22 Learning Department

Claims

1. An acquisition means for acquiring a pathology image including specimen cells as a subject; the pathology image is input, and a first generative model trained to generate first weighting information for emphasizing a first region of interest in the pathology image is used to generate a first weighted input image in which the pathology image is processed with the first weighting information; the pathology image is input, and a second generative model trained to generate second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image is used to generate a second weighted input image in which the pathology image is processed with the second weighting information; the first weighted input image is input, and a first feature analysis model trained to generate feature amounts for the first weighted input image is used to generate feature amounts for the first weighted input image; the second weighted input image is input, and a second feature analysis model trained to generate feature amounts for the second weighted input image is used to generate feature amounts for the second weighted input image; a classification means for receiving the feature quantities of the first weighted input image and the feature quantities of the second weighted input image and for classifying the specimen cells as benign cells or malignant cells using a classification model that has been trained to classify the specimen cells as benign cells or malignant cells.

2. The classification device according to claim 1, wherein at least the cytoplasm of the specimen cells is stained, and the first region of interest and the second region of interest are regions with different staining intensities.

3. The classification device according to claim 2, wherein the second region of interest is a region containing stained cilia, and the first region of interest is a region having a higher staining intensity than the second region of interest.

4. An acquisition means for acquiring a pathology image including specimen cells as a subject and classification information indicating whether the specimen cells are benign or malignant; a first generation model that receives the pathology image as an input and generates first weighting information for emphasizing a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information; a second generation model that receives the pathology image as an input and generates second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information; a first feature analysis model that receives the first weighted input image and generates feature quantities for the first weighted input image to generate feature quantities; a second feature analysis model that receives the second weighted input image and generates feature quantities for the second weighted input image to generate feature quantities; a learning means for receiving input features of the first weighted input image and features of the second weighted input image, and classifying the specimen cell as either a benign cell or a malignant cell using a classification model that classifies the specimen cell as either a benign cell or a malignant cell, and updating parameters of the first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model based on a comparison between the classification result and the classification information.

5. The learning device according to claim 4, wherein at least the cytoplasm of the specimen cells is stained, and the learning means sets initial values ​​of the first weighting information and the second weighting information so that the first region of interest and the second region of interest are regions with different staining intensities.

6. The learning device described in claim 5, wherein the second region of interest is a region containing stained cilia, the first region of interest is a region having a higher staining intensity than the second region of interest, and the learning means sets the initial values ​​of the first region of interest and the second region of interest in the first weighting information and the second weighting information to be relatively higher than those of other regions.

7. A classification device acquires a pathology image including specimen cells as a subject; and uses a first generative model trained to generate first weighting information for emphasizing a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information; uses a second generative model trained to generate second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information; receives the first weighted input image and uses a first feature analysis model trained to generate features of the first weighted input image to generate features of the first weighted input image; receives the second weighted input image and uses a second feature analysis model trained to generate features of the second weighted input image to generate features of the second weighted input image; a classification method comprising: receiving the features of the first weighted input image and the features of the second weighted input image, and classifying the specimen cells as benign cells or malignant cells using a classification model that has been trained to classify the specimen cells as benign cells or malignant cells.

8. A learning device acquires a pathology image including specimen cells as a subject and classification information indicating whether the specimen cells are benign or malignant cells; uses the pathology image as input and a first generative model that generates first weighting information for emphasizing a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information; uses the pathology image as input and a second generative model that generates second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information; uses a first feature analysis model that receives the first weighted input image and generates feature quantities for the first weighted input image; uses a second feature analysis model that receives the second weighted input image and generates feature quantities for the second weighted input image to generate feature quantities; a classification model that receives input features of the first weighted input image and features of the second weighted input image and classifies the specimen cell as either a benign cell or a malignant cell, and updates parameters of the first generative model, the second generative model, the first feature analysis model, the second feature analysis model, and the classification model based on a comparison between the classification result and the classification information.

9. A program that causes a computer to function as a classification device, the program comprising: an acquisition means that acquires a pathology image including specimen cells as a subject; and a first generative model that receives the pathology image as input and has been trained to generate first weighting information for emphasizing a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information; a second generative model that receives the pathology image as input and has been trained to generate second weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information; and a first feature analysis model that receives the first weighted input image and has been trained to generate feature quantities for the first weighted input image to generate feature quantities for the first weighted input image; a classification means for receiving the second weighted input image and generating features of the second weighted input image using a second feature analysis model that has been trained to generate features of the second weighted input image; and a classification means for receiving the features of the first weighted input image and the features of the second weighted input image and classifying the specimen cells as benign or malignant using a classification model that has been trained to classify the specimen cells as benign or malignant.

10. A program that causes a computer to function as a learning device, the program comprising: an acquisition means that acquires a pathology image including specimen cells as a subject and classification information indicating whether the specimen cells are benign or malignant; an acquisition means that receives the pathology image as input and uses a first generation model that generates first weighting information for emphasizing a first region of interest in the pathology image to generate a first weighted input image in which the pathology image is processed with the first weighting information; an acquisition means that receives the pathology image as input and uses a first generation model that generates first weighting information for emphasizing a second region of interest different from the first region of interest in the pathology image to generate a second weighted input image in which the pathology image is processed with the second weighting information; an acquisition means that receives the first weighted input image and uses a first feature analysis model that generates feature amounts of the first weighted input image to generate a feature amount of the first weighted input image; a learning means for receiving the second weighted input image and generating features of the second weighted input image using a second feature analysis model that generates features of the second weighted input image; receiving the features of the first weighted input image and the features of the second weighted input image and using a classification model that classifies the specimen cell as either a benign cell or a malignant cell, and updating parameters of the first generation model, the second generation model, the first feature analysis model, the second feature analysis model, and the classification model based on a comparison between the classification result and the classification information.

Images