Image analysis device, image analysis method, and recording medium

Abstract

An image analysis device, wherein a partial image generation means generates, from an input image, a partial image smaller than the input image. A feature space generation means converts a plurality of partial images into feature vectors including the positions thereof, and maps the feature vectors to a feature space. A reference image mapping means maps the feature vector of the reference image designated by a user to the feature space. A cluster selection means clusters the feature space and selects a cluster most similar to the reference image among a plurality of clusters. An important area extraction means extracts an important area from the input image on the basis of the positions of the plurality of partial images belonging to the selected cluster. The image analysis device can assist the decision making of the user.

Description

Image analysis device, image analysis method, and recording medium

[0001] The present disclosure relates to a technique for extracting a feature region included in an image.

[0002] There are known techniques for supporting histopathological examinations by analyzing pathological images. For example, Patent Literature 1 describes a pathological image diagnostic system that divides a pathological image into predetermined segmented regions and outputs the segmented regions in association with their importance.

[0003] JP 2010-281637 A

[0004] However, even with the technique of Patent Document 1, it is not always possible to accurately extract the desired important feature region from the pathological image.

[0005] One object of the present disclosure is to provide an image analysis device that is capable of accurately extracting a desired important feature region from an image.

[0006] In one aspect of the present disclosure, an image analysis device comprises: a partial image generation means for generating, from an input image, partial images smaller than the input image; a feature space generation means for converting a plurality of partial images into feature vectors including their positions and mapping the feature vectors into a feature space; a reference image mapping means for mapping a feature vector of a reference image designated by a user into the feature space; a cluster selection means for clustering the feature space and selecting, from a plurality of clusters, the cluster most similar to the reference image; and an important region extraction means for extracting an important region from the input image based on the positions of a plurality of partial images belonging to the selected cluster.

[0007] In another aspect of the present disclosure, an image analysis method includes performing partial image generation to generate, from an input image, partial images smaller than the input image; performing feature space generation to convert multiple partial images into feature vectors including their positions and map the feature vectors into a feature space; performing reference image mapping to map the feature vector of a reference image specified by a user into the feature space; clustering the feature space and performing cluster selection to select, from multiple clusters, the cluster most similar to the reference image; and performing important region extraction to extract important regions from the input image based on the positions of multiple partial images belonging to the selected cluster.

[0008] In yet another aspect of the present disclosure, a recording medium records a program that causes a computer to execute the following processes: partial image generation, which generates, from an input image, partial images that are smaller than the input image; feature space generation, which converts multiple partial images into feature vectors including their positions and maps the feature vectors into a feature space; reference image mapping, which maps the feature vector of a reference image specified by a user into the feature space; clustering the feature space and selecting a cluster from multiple clusters that is most similar to the reference image; and important region extraction, which extracts important regions from the input image based on the positions of multiple partial images that belong to the selected cluster.

[0009] According to the present disclosure, it is possible to provide an image analysis device that is capable of accurately extracting a desired important feature region from an image.

[0010] FIG. 1 is a diagram conceptually illustrating an image analysis device according to the present disclosure; FIG. 2 is a block diagram illustrating a hardware configuration of an image analysis device according to the present disclosure; FIG. 3 is a block diagram illustrating a functional configuration of an image analysis device according to the present disclosure; FIG. 4 is an explanatory diagram illustrating processing of an image division unit and a learning data generation unit; FIG. 5 is an explanatory diagram illustrating processing of a tissue region extraction unit; FIG. 6 shows an extraction result of a tissue region; FIG. 7 is a flowchart of a feature region extraction process; FIG. 8 is a block diagram illustrating a functional configuration of another image analysis device according to the present disclosure; and FIG. 9 is a flowchart of processing by another image analysis device according to the present disclosure.

[0011] Preferred embodiments of the present disclosure will now be described with reference to the drawings. First Embodiment Overall Configuration FIG. 1 is a conceptual diagram of an image analysis device. The image analysis device 10 extracts a target important tissue region from input image data (hereinafter also referred to as an "input image") based on the input image and a reference image. The input image in this embodiment is a pathological tissue image of a patient. The reference image in this embodiment is a reference image of the target important tissue region. The reference image is specified by the user. The extracted important tissue region is used for tasks such as pathological diagnosis, treatment selection, and prognosis prediction.

[0012] 2 is a block diagram showing the hardware configuration of the image analysis device 10. As shown in the figure, the image analysis device 10 includes a processor 11, an interface (IF) 12, a read-only memory (ROM) 13, a random access memory (RAM) 14, a storage medium 15, and an input unit 16. The components are connected to each other via a bus 18, for example.

[0013] The processor 11 is a computer such as a CPU (Central Processing Unit), and executes a prepared program to control the entire image analysis device 10. Specifically, the processor 11 may be a CPU, a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), an FPU (Floating Point number Processing Unit), a PPU (Physics Processing Unit), a TPU (Tensor Processing Unit), a quantum processor, a microcontroller, or a combination thereof.

[0014] The processor 11 also loads programs stored in the ROM 13, the storage medium 15, etc., and executes each process coded in the program. The processor 11 also functions as part or all of the image analysis device 10. The processor 11 then executes the feature region extraction process described below.

[0015] IF 12 inputs and outputs data to and from external devices. Specifically, the pathological tissue image and the reference image are input to image analysis device 10 via IF 12.

[0016] The ROM 13 stores various programs executed by the processor 11. The RAM 14 stores pathological tissue images and reference images input from external devices. The RAM 14 may also store a segmentation model, which will be described later. The RAM 14 is also used as a working memory while the processor 11 is executing various processes.

[0017] The storage medium 15 is a non-volatile, non-temporary storage device such as a disk-shaped recording medium or a semiconductor memory. The storage medium 15 may be configured to be detachable from the image analysis device 10. The storage medium 15 stores various programs executed by the processor 11.

[0018] The input unit 16 is, for example, a mouse, a keyboard, etc., and is used by the user to input data.

[0019] 3 is a block diagram showing the functional configuration of the image analysis device 10 according to the first embodiment. Functionally, the image analysis device 10 includes an image dividing unit 111, an image embedding unit 112, a tissue region extraction unit 113, a segmentation unit 114, and a feature calculation unit 115.

[0020] A pathological tissue image is input to image analyzing device 10 via IF 12. The pathological tissue image is input to image dividing unit 111 and tissue region extraction unit 113. A reference image is also input to image analyzing device 10 via IF 12. The reference image is input to image embedding unit 112.

[0021] The image dividing unit 111 divides the pathological tissue image into partial images smaller than the pathological tissue image. Hereinafter, the entire pathological tissue image will also be referred to as the "whole image." FIG. 4 is an explanatory diagram of the processing performed by the image dividing unit 111 and the image embedding unit 112. As shown in FIG. 4, the image dividing unit 111 divides the whole image WI into multiple partial images PI and outputs them to the image embedding unit 112. In the example of FIG. 4, the whole image WI is an image of a patient's pathological tissue. However, the image dividing unit 111 only needs to generate partial images for the pathological tissue region, and does not need to generate partial images for the background region. In the example of FIG. 4, the image dividing unit 111 divides the whole image WI into multiple partial images PI using grid division. Alternatively, the whole image WI may be divided into multiple partial images PI centered on characteristic points.

[0022] The image embedding unit 112 converts each of the input partial images into a feature vector, performs dimensionality reduction, and maps the resulting image onto a low-dimensional feature space. The image embedding unit 112 also converts the input reference image into a feature vector, performs dimensionality reduction, and maps the resulting image onto the low-dimensional feature space. FIG. 4 shows an example in which each partial image PI and the reference image are mapped onto a two-dimensional feature space. In reality, the dimension of the feature space corresponds to the dimension of the feature vector. In FIG. 4, one partial image PI is mapped as one point on the two-dimensional feature space. Also, in FIG. 4, the reference image is mapped as one point on the two-dimensional feature space. Examples of dimensionality reduction algorithms that can be used include t-SNE (t-distributed stochastic neighbor embedding) and UMAP (uniform manifold approximation and projection).

[0023] Next, the image embedding unit 112 combines the feature vector of the dimension-reduced partial image with the feature vector of the position of the partial image to generate a new feature vector of the partial image. Note that the position of the partial image is, for example, the center position of the partial image in the entire image. The image embedding unit 112 outputs the new feature vector of the partial image and the feature vector of the dimension-reduced reference image to the tissue region extraction unit 113.

[0024] The tissue region extraction unit 113 performs clustering on the set of new feature vectors of the partial image. For the clustering, a known clustering method such as kmeans can be used. FIG. 5 is an explanatory diagram of the processing of the tissue region extraction unit 113. In the example of FIG. 5, the tissue region extraction unit 113 classifies the set of new feature vectors of the partial image into four clusters C1 to C4 through clustering. Note that although the feature space in FIG. 5 shows a two-dimensional feature space, in reality, the dimensions of the feature space correspond to the dimensions of the feature vectors.

[0025] Next, the tissue region extraction unit 113 selects the cluster that is most similar to the reference image in a feature space that represents the relationship between the classified clusters and the reference image. For example, the tissue region extraction unit 113 calculates the center of gravity of each cluster. Then, the tissue region extraction unit 113 calculates the Euclidean distance between the center of gravity of each cluster and the reference image, and selects the cluster with the shortest Euclidean distance. In Figure 5, the tissue region extraction unit 113 selects C4 as the cluster that is most similar to the reference image.

[0026] Next, the tissue region extraction unit 113 selects a predetermined number of feature vectors from the feature vectors belonging to the selected cluster. Then, the tissue region extraction unit 113 maps positions included in the selected feature vectors onto the entire image. In FIG. 5 , the tissue region extraction unit 113 selects a predetermined number of points from all points belonging to C4. At this time, the tissue region extraction unit 113 selects the predetermined number of points randomly and so that the points are spaced apart by at least a predetermined distance. Then, the tissue region extraction unit 113 maps positions of the partial images corresponding to the selected points onto the entire image, as indicated by arrow 51.

[0027] The entire image onto which the positions of the partial images belonging to the selected cluster are mapped will be hereinafter referred to as a “mapping image.” The mapping image is used as information for identifying the segmentation target in the prompt described below.

[0028] Next, the tissue region extraction unit 113 generates a prompt for segmentation. The prompt instructs which region of the image to segment. The tissue region extraction unit 113 generates a prompt including an instruction statement and a mapping image. The tissue region extraction unit 113 then outputs the prompt and the entire image to the segmentation unit 114.

[0029] The segmentation unit 114 inputs the prompt and the entire image into a segmentation model and obtains a mask corresponding to the prompt from the segmentation model. This segmentation model is a machine learning model that segments a predetermined region from an image. The segmentation model is constructed by learning using a pair of an image and a mask image as training data. For example, the Segment Anything Model (SAM) published by Meta, Inc. can be used as the segmentation model. Figure 6 shows the output result of the segmentation model. In Figure 6, a mask 52 is generated on the entire image by the segmentation model.

[0030] The segmentation unit 114 extracts the tissue region masked by the segmentation model, and outputs the extracted tissue region to the feature amount calculation unit 115.

[0031] The feature calculation unit 115 converts the tissue region input from the segmentation unit 114 into tissue feature amounts. Examples of tissue feature amounts include Rayleigh flow index, divergence, curl, self-correlation, texture, cell density, and degree of variation in cell position. The feature calculation unit 115 outputs the tissue feature amounts to downstream tasks. Examples of downstream tasks include tasks such as pathological diagnosis, treatment selection, and prognosis prediction.

[0032] As described above, the image analysis device 10 of this embodiment can generate appropriate prompts according to the purpose and use them in a segmentation model. In particular, the image analysis device 10 of this embodiment can generate prompts to segment areas in an image whose boundaries are unclear. For example, the image analysis device 10 can generate prompts to segment elongated cells (elongated areas) in the interstitium or cells arranged in a spiral or radial pattern. This allows the image analysis device 10 to extract tissue regions important for pathological diagnosis, treatment selection, prognosis prediction, and the like.

[0033] In the above configuration, the image dividing unit 111 is an example of a partial image generating means, the image embedding unit is an example of a feature space generating means and a reference image mapping means, the tissue region extraction unit 113 is an example of a cluster selecting means, the segmentation unit 114 is an example of an important region extracting means, and the feature calculation unit 115 is an example of an important region feature calculation means.

[0034] [Feature Region Extraction Processing] Next, the feature region extraction processing will be described. Fig. 7 is a flowchart of the feature region extraction processing performed by the image analysis device 10. This processing is realized by the processor 11 shown in Fig. 2 executing a program prepared in advance and operating as the elements shown in Fig. 3.

[0035] First, a pathological tissue image and a reference image are input to image analysis device 10 via IF 12 (step S111). The pathological tissue image is input to image segmentation unit 111. The reference image is input to image embedding unit 112. Next, image segmentation unit 111 segments the pathological tissue image into a plurality of partial images (step S112). Image segmentation unit 111 outputs the plurality of partial images to image embedding unit 112.

[0036] Next, the image embedding unit 112 converts each of the input partial images into a feature vector, performs dimensionality reduction, and maps the resulting vector onto a low-dimensional feature space. The image embedding unit 112 also converts the input reference image into a feature vector, performs dimensionality reduction, and maps the resulting vector onto the low-dimensional feature space (step S113). The image embedding unit 112 also combines the feature vector of the dimensionally reduced partial image with the feature vector of the position of the partial image to generate a new feature vector for the partial image. The image embedding unit 112 outputs the new feature vector for the partial image and the feature vector for the reference image to the tissue region extraction unit 113.

[0037] Next, the tissue region extraction unit 113 performs clustering on the new set of feature vectors of the partial images and selects the cluster most similar to the reference image (step S114). Next, the tissue region extraction unit 113 maps the positions of the partial images belonging to the selected cluster onto the entire image and generates a prompt for segmentation (step S115). The tissue region extraction unit 113 outputs the prompt and the entire image to the segmentation unit 114. Next, the segmentation unit 114 inputs the prompt and the entire image into a segmentation model and obtains a mask corresponding to the prompt from the segmentation model. The segmentation unit 114 then extracts the tissue region masked by the segmentation model (step S116). The segmentation unit 114 outputs the extracted tissue region to the feature calculation unit 115. Next, the feature calculation unit 115 converts the tissue region into tissue features and outputs them to a downstream task (step S117). The process then ends.

[0038] [Variation] In the above embodiment, the image embedding unit 112 combines the feature vector of the dimensionally reduced partial image (hereinafter also referred to as the "feature vector of the partial image") with the feature vector of the position of the partial image to generate a new feature vector of the partial image.

[0039] Instead of generating a new feature vector for the partial image, the image embedding unit 112 may associate and store the feature vector of the partial image with the position of the partial image. In this case, the tissue region extraction unit 113 performs clustering on the set of feature vectors of the partial images and selects a cluster similar to the reference image. Then, using the association described above, the tissue region extraction unit 113 obtains the position of the partial image from the feature vector of the partial image belonging to the selected cluster and maps the position to the entire image.

[0040] 8 is a block diagram showing the functional configuration of an image analysis device according to Embodiment 2. The image analysis device 200 includes a partial image generation unit 201, a feature space generation unit 202, a reference image mapping unit 203, a cluster selection unit 204, and an important region extraction unit 205.

[0041] 9 is a flowchart of processing by the image analysis device of the second embodiment. The partial image generation means 201 generates partial images smaller than the input image from the input image (step S201). The feature space generation means 202 converts multiple partial images into feature vectors including their positions and maps the feature vectors to the feature space (step S202). The reference image mapping means 203 maps the feature vector of a reference image designated by a user to the feature space (step S203). The cluster selection means 204 clusters the feature space and selects the cluster most similar to the reference image from multiple clusters (step S204). The important region extraction means 205 extracts important regions from the input image based on the positions of multiple partial images belonging to the selected cluster (step S205).

[0042] According to the image analyzing device 200 of the second embodiment, it is possible to provide an image analyzing device that can extract important feature regions from an image. Furthermore, the image analyzing device 200 can support a user's decision-making.

[0043] A part or all of the above-described embodiments can be described as, but not limited to, the following supplementary notes.

[0044] (Supplementary Note 1) An image analysis device comprising: a partial image generation means for generating, from an input image, partial images smaller than the input image; a feature space generation means for converting a plurality of partial images into feature vectors including their positions and mapping the feature vectors into a feature space; a reference image mapping means for mapping a feature vector of a reference image designated by a user into the feature space; a cluster selection means for clustering the feature space and selecting, from a plurality of clusters, the cluster most similar to the reference image; and an important region extraction means for extracting an important region from the input image based on the positions of a plurality of partial images belonging to the selected cluster.

[0045] (Supplementary Note 2) The image analysis device according to Supplementary Note 1, further comprising: an important region feature amount calculation unit that calculates a feature amount from the important region.

[0046] (Supplementary Note 3) The image analysis device according to Supplementary Note 1, wherein the important region extraction means selects a predetermined number of partial images from a plurality of partial images belonging to a selected cluster, generates a prompt including the input image and positions of the selected partial images, and inputs the prompt to a machine learning model that segments a predetermined region from the input image, thereby extracting an important region from the input image.

[0047] (Supplementary Note 4) The image analysis device according to Supplementary Note 1, wherein the cluster selection means selects a cluster that is most similar to the reference image based on a distance between the center of gravity of each cluster in the feature space and the reference image.

[0048] (Supplementary Note 5) The image analysis device according to Supplementary Note 2, wherein the input image is a pathological tissue image, and the important region feature amount calculation means calculates tissue feature amounts representing a state of cells from the important region.

[0049] (Supplementary Note 6) The image analysis device according to Supplementary Note 5, further comprising: a task execution means for executing a downstream task using the tissue feature, wherein the downstream task is a pathological diagnosis task.

[0050] (Supplementary Note 7) The image analysis device according to Supplementary Note 1, wherein the position of the partial image is a coordinate position of the partial image in the input image.

[0051] (Supplementary Note 8) An image analysis method comprising: performing partial image generation to generate, from an input image, partial images smaller than the input image; performing feature space generation to convert a plurality of partial images into feature vectors including their positions and map the feature vectors into a feature space; performing reference image mapping to map a feature vector of a reference image designated by a user into the feature space; clustering the feature space and selecting a cluster from a plurality of clusters that is most similar to the reference image; and performing important region extraction to extract important regions from the input image based on the positions of a plurality of partial images that belong to the selected cluster.

[0052] (Supplementary Note 9) The image analysis method according to Supplementary Note 8, further comprising: calculating an important region feature amount by calculating a feature amount from the important region.

[0053] (Supplementary Note 10) The image analysis method according to Supplementary Note 8, wherein the important region extraction comprises selecting a predetermined number of partial images from a plurality of partial images belonging to a selected cluster, generating a prompt including the input image and the positions of the selected partial images, and inputting the prompt into a machine learning model that segments a predetermined region from the input image, thereby extracting the important region from the input image.

[0054] (Supplementary Note 11) The image analysis method according to Supplementary Note 8, wherein the cluster selection comprises selecting a cluster that is most similar to the reference image based on a distance between the center of gravity of each cluster in the feature space and the reference image.

[0055] (Supplementary Note 12) The image analysis method according to Supplementary Note 9, wherein the input image is a pathological tissue image, and the important region feature calculation calculates a tissue feature representing a state of a cell from the important region.

[0056] (Supplementary Note 13) The image analysis device according to Supplementary Note 12, further performing task execution to execute a downstream task using the tissue feature, wherein the downstream task is a pathological diagnosis task.

[0057] (Supplementary Note 14) The image analysis device according to Supplementary Note 8, wherein the position of the partial image is a coordinate position of the partial image in the input image.

[0058] (Supplementary Note 15) A recording medium having recorded thereon a program that causes a computer to execute the following processes: partial image generation for generating, from an input image, partial images that are smaller than the input image; feature space generation for converting a plurality of partial images into feature vectors including their positions and mapping the feature vectors into a feature space; reference image mapping for mapping a feature vector of a reference image designated by a user into the feature space; clustering the feature space and selecting a cluster that is most similar to the reference image from a plurality of clusters; and important region extraction for extracting an important region from the input image based on the positions of a plurality of partial images that belong to the selected cluster.

[0059] (Supplementary Note 16) The recording medium according to Supplementary Note 15, further comprising: an important region feature amount calculation for calculating a feature amount from the important region.

[0060] (Appendix 17) The recording medium according to Appendix 15, wherein the important region extraction comprises selecting a predetermined number of partial images from a plurality of partial images belonging to a selected cluster, generating a prompt including the input image and the positions of the selected partial images, and inputting the prompt into a machine learning model that segments a predetermined region from the input image, thereby extracting the important region from the input image.

[0061] (Supplementary Note 18) The recording medium according to Supplementary Note 15, wherein the cluster selection comprises selecting a cluster that is most similar to the reference image based on a distance between the center of gravity of each cluster in the feature space and the reference image.

[0062] (Supplementary Note 19) The recording medium according to Supplementary Note 16, wherein the input image is a pathological tissue image, and the important region feature calculation calculates a tissue feature representing a state of a cell from the important region.

[0063] (Supplementary Note 20) The recording medium according to Supplementary Note 19, further comprising: performing task execution to execute a downstream task using the tissue feature; and the downstream task being a pathology diagnosis task.

[0064] (Supplementary Note 21) The recording medium according to Supplementary Note 15, wherein the position of the partial image is a coordinate position of the partial image in the input image.

[0065] Although the present disclosure has been described above with reference to the embodiments and examples, the present disclosure is not limited to the above-described embodiments and examples. Various modifications that can be understood by a person skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure.

[0066] 10 Image analysis device 111 Image division unit 112 Image embedding unit 113 Tissue region extraction unit 114 Segmentation unit 115 Feature amount calculation unit

Claims

1. A partial image generation means for generating a partial image smaller than the input image from the input image, a feature space generation means for converting a plurality of partial images into feature vectors including their positions and mapping the feature vectors into a feature space, a reference image mapping means for mapping the feature vector of a reference image specified by a user into the feature space, a cluster selection means for clustering the feature space and selecting the cluster most similar to the reference image among a plurality of clusters, and an important region extraction means for extracting an important region from the input image based on the positions of a plurality of partial images belonging to the selected cluster. An image analysis apparatus comprising:

2. The image analysis apparatus according to claim 1, further comprising an important region feature amount calculation means for calculating a feature amount from the important region.

3. The important region extraction means selects a predetermined number of partial images from a plurality of partial images belonging to the selected cluster, generates a prompt including the input image and the positions of the selected partial images, and inputs the prompt to a machine learning model for segmenting a predetermined region from the input image, thereby extracting an important region from the input image. The image analysis apparatus according to claim 1.

4. The cluster selection means selects the cluster most similar to the reference image based on the distance between the center of gravity of each cluster in the feature space and the reference image. The image analysis apparatus according to claim 1.

5. The input image is a pathological tissue image, and the important region feature amount calculation means calculates a tissue feature amount representing the state of cells from the important region. The image analysis apparatus according to claim 2.

6. A task execution means for executing a downstream task using the tissue feature amount, and the downstream task is a pathological diagnosis task. The image analysis apparatus according to claim 5.

7. The position of the partial image is the coordinate position of the partial image in the input image. The image analysis apparatus according to claim 1.

8. Perform partial image generation to generate a partial image smaller than the input image from the input image, convert a plurality of partial images into feature vectors including their positions, perform feature space generation to map the feature vectors into a feature space, perform reference image mapping to map the feature vector of a reference image specified by a user into the feature space, perform clustering on the feature space, perform cluster selection to select the cluster most similar to the reference image among a plurality of clusters, and perform important region extraction to extract an important region from the input image based on the positions of the plurality of partial images belonging to the selected cluster.

9. The image analysis method according to claim 8, further performing important region feature amount calculation to calculate a feature amount from the important region.

10. The important region extraction selects a predetermined number of partial images from the plurality of partial images belonging to the selected cluster, generates a prompt including the input image and the positions of the selected partial images, and inputs the prompt into a machine learning model that segments a predetermined region from the input image, thereby extracting an important region from the input image.

11. The cluster selection selects the cluster most similar to the reference image based on the distance between the center of gravity of each cluster in the feature space and the reference image.

12. The input image is a pathological tissue image, and the important region feature amount calculation calculates a tissue feature amount representing the state of cells from the important region.

13. Further perform task execution to execute a downstream task using the tissue feature amount, and the downstream task is a pathological diagnosis task.

14. The position of the partial image is the coordinate position of the partial image in the input image.

15. A recording medium storing a program that causes a computer to execute a process of generating a partial image that is smaller than the input image from the input image, converting a plurality of partial images into feature vectors including their positions, performing feature space generation to map the feature vectors into a feature space, performing reference image mapping to map the feature vector of a reference image specified by a user into the feature space, clustering the feature space, performing cluster selection to select the cluster most similar to the reference image among a plurality of clusters, and performing important region extraction to extract an important region from the input image based on the positions of the plurality of partial images belonging to the selected cluster.

16. The recording medium according to claim 15, further performing important region feature amount calculation for calculating a feature amount from the important region.

17. The recording medium according to claim 15, wherein the important region extraction includes selecting a predetermined number of partial images from the plurality of partial images belonging to the selected cluster, generating a prompt including the input image and the positions of the selected partial images, and inputting the prompt to a machine learning model that segments a predetermined region from the input image, thereby extracting an important region from the input image.

18. The recording medium according to claim 15, wherein the cluster selection selects the cluster most similar to the reference image based on the distance between the center of gravity of each cluster in the feature space and the reference image.

19. The recording medium according to claim 16, wherein the input image is a pathological tissue image, and the important region feature amount calculation calculates a tissue feature amount representing the state of cells from the important region.

20. The recording medium according to claim 19, further performing task execution for executing a downstream task using the tissue feature amount, wherein the downstream task is a pathological diagnosis task.

Images