Tissue slice thickness estimation device, tissue slice thickness evaluation device, tissue slice thickness estimation method, tissue slice thickness estimation program, and recording medium

Abstract

A tissue slice thickness estimation device 1 comprises: a differential image creation unit 10 for creating a differential image of a microscope image captured under a shallow depth of focus condition, and a microscope image captured under a deep depth of focus condition; and an estimation unit 20 for estimating the thickness of a tissue slice from the differential image. This tissue slice thickness estimation method includes: a differential image creation step for creating a differential image of a microscope image captured under a shallow depth of focus condition, and a microscope image captured under a deep depth of focus condition; and an estimation step for estimating the thickness of a tissue slice from the differential image.

Description

[Technical Field]

[0001] The present invention relates to a tissue slice thickness estimation device, a tissue slice thickness evaluation device, a tissue slice thickness estimation method, a tissue slice thickness estimation program, and a recording medium. [Background technology]

[0002] It is known that the thickness of a tissue section is related to its absorbance (brightness). For example, Non-Patent Document 1 discloses the absorbance and thickness of hematoxylin-eosin stained (hereinafter referred to as "HE stained") tissue sections. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] “Optical density-based image analysis method for the evaluation of hematoxylin and eosin staining”, Elizabeth Chlipala, Christine M. Bendzinski, Kevin Chu, Joshua I. Johnson, Miles Brous, Karen Copeland & Brad Bolon, JOURNAL OF HISTOTECHNOLOGY 2020, VOL. 43, NO. 1, 29.37 https: / / doi.org / 10.1080 / 01478885.2019.1708611 Summary of the Invention [Problem to be solved by the invention]

[0004] Tissue sections, such as pathological tissue sections, are prepared for the pathological diagnosis of various diseases occurring in organs. Generally, organs removed during surgery are fixed in formalin, hardened in paraffin, thinly sliced, and then mounted on glass slides for use. In the field of pathological diagnosis, which deals with images, the use of AI is expected to assist in diagnosis, but its adoption has been slower than in other fields such as radiology and endoscopy. This is mainly due to the large number of manual, analog processes involved in the process of preparing the slides that serve as the source of the images. Because of these manual processes, standardization of analog processes is extremely difficult, and when the environment changes, for example, when changing hospitals, sufficient reproducibility cannot be achieved, and the developed AI for pathological diagnosis cannot demonstrate its full potential.

[0005] In the process of creating tissue sections, the "thin section" is a step that exhibits particularly large variations. Typically, medical technicians at each hospital manually shave paraffin blocks to a thickness of 4 μm. However, the variations that occur during this process cannot be completely eliminated. Furthermore, hospitals generally do not have the means to measure the thickness of the sliced ​​paraffin, so the current situation is that technicians rely on their own intuition to perform their work. However, differences in the thickness of tissue sections have a significant impact on the images observed under a microscope, which not only affects pathological diagnosis but also serves as a major obstacle to the development of AI for pathological diagnosis.

[0006] Figure 1 shows the cutting of tissue sections (top, side view of the tissue block), the cut tissue sections (middle, top view of the sections), and microscopic images of these tissue sections (bottom, left: cancerous tissue cut at 4 μm, right: normal tissue cut at 8 μm). The left side shows a 4 μm-thick section, and the right side shows a 8 μm-thick section. When cut at 4 μm, cells do not overlap, resulting in a focused image. On the other hand, when cut at 8 μm, cells overlap, resulting in an out-of-focus image. As shown in the bottom image, in the microscopic image of the normal tissue cut at 8 μm, cells appear double, making them appear identical to the cells in the cancerous tissue cut at 4 μm. Furthermore, some cells appear crushed due to being out of focus. To identify cancer cells using tissue, it is necessary to accurately determine the shape, size, density, and arrangement of nuclei. However, if the tissue section is thick, such accurate judgment is not possible, which may lead to misdiagnosis.

[0007] For these reasons, the development of AI for pathological diagnosis requires the standardization of tissue sections. In recent years, efforts have been made to fully standardize pathological diagnosis, aiming to fully automate the preparation of pathological specimens and image acquisition. Since standardization is based on mechanization, fully automated tissue sectioning devices and automatic staining equipment are used. However, these devices are expensive, costing tens of millions of yen, making their widespread adoption in general hospitals extremely difficult. Therefore, even if a mechanized standardization process were completed, it would be limited to a limited number of large-scale facilities. Furthermore, general hospitals, which produce many pathological specimens, have difficulty producing standardized tissue sections. This means that many hospitals and medical facilities cannot benefit from AI for pathological diagnosis. Therefore, there is a need to develop technology that can produce standardized tissue sections without introducing the expensive equipment mentioned above.

[0008] The thickness of tissue sections can be measured using, for example, a confocal laser microscope or a white light interference microscope. Therefore, such a device would allow us to know the thickness of the tissue sections prepared by laboratory technicians, and thus, for example, to determine the degree to which the thickness deviates from the target thickness. This would likely lead to improved technician skills and the standardization of tissue sections.

[0009] However, microscopes like those mentioned above are for industrial use and are not usually available in general hospitals, etc. However, if it becomes possible to estimate the thickness of prepared tissue sections using ordinary optical microscopes, etc., which are available in general hospitals, etc., it is expected that this will become a fundamental technology for developing AI for pathological diagnosis.

[0010] The present invention has been made in consideration of these problems, and its purpose is to provide a technique that can estimate the thickness of a prepared tissue section from an image obtained by a normal optical microscope or the like. [Means for solving the problem]

[0011] In order to solve the above problem, a tissue slice thickness estimation device according to one embodiment of the present invention includes a differential image creation unit that creates a differential image between a microscopic image of a tissue slice taken under shallow focal depth conditions and a microscopic image of the tissue slice taken under deep focal depth conditions, and an estimation unit that estimates the thickness of the tissue slice from the differential image.

[0012] In one embodiment, the change in depth of focus may be produced by adjusting the condenser of the optical microscope.

[0013] In one embodiment, the tissue slice thickness estimation device may include a model generation unit that performs machine learning using a differential image between a microscopic image of the tissue slice captured under a shallow depth of focus and a microscopic image of the tissue slice captured under a deep depth of focus, and an actual measured value of the thickness of the tissue slice, as training data, to generate an estimation model that estimates the thickness of the tissue slice when an image of the tissue slice is input. In this case, the estimation unit estimates the thickness of the tissue slice using the estimation model generated by the model generation unit.

[0014] In certain embodiments, the tissue section may be HE-stained tissue, unstained deparaffinized tissue, or unstained undeparaffinized tissue.

[0015] Another aspect of the present invention is a tissue section thickness evaluation device comprising any one of the tissue section thickness estimation devices described above and a tissue section thickness evaluation unit, which evaluates the usability of the tissue section for pathological diagnosis based on the thickness and color of the tissue section estimated by the tissue section thickness estimation device.

[0016] Yet another aspect of the present invention is a method for estimating the thickness of a tissue section, comprising: a step of creating a difference image between a microscope image captured under a shallow depth of focus condition and a microscope image captured under a deep depth of focus condition; and a step of estimating the thickness of the tissue section from the difference image.

[0017] Yet another aspect of the present invention is a tissue section thickness estimation program that causes a computer to execute a difference image creation step of creating a difference image between a microscopic image of a tissue section captured under a shallow depth of focus condition and a microscopic image of the tissue section captured under a deep depth of focus condition, and an estimation step of estimating the thickness of the tissue section from the difference image.

[0018] Yet another aspect of the present invention is a recording medium having recorded thereon a program for causing a computer to execute a differential image creation step of creating a differential image between a microscopic image of a tissue slice captured under a shallow depth of focus condition and a microscopic image of the tissue slice captured under a deep depth of focus condition, and an estimation step of estimating the thickness of the tissue slice from the differential image.

[0019] Another aspect of the present invention is a tissue section thickness estimation device. This device includes an image data generation unit that generates microscopic image data of a focused tissue section, an estimation unit that estimates the thickness of the tissue section from the microscopic image of the tissue section, and a model generation unit that performs machine learning using the microscopic image data of the focused tissue section and actual measurements of the thickness of the tissue section as training data to generate an estimation model that estimates the thickness of the tissue section when a microscopic image of the tissue section is input. The estimation unit estimates the thickness of the tissue section using the estimation model generated by the model generation unit.

[0020] In one embodiment, the microscope image of the tissue section in focus may be the image when the microscope's condenser adjustment is turned on.

[0021] Yet another aspect of the present invention is a method for estimating the thickness of a tissue section. This method includes an image data generation step of generating microscopic image data of a tissue section in focus, an estimation step of estimating the thickness of the tissue section from the microscopic image of the tissue section, and a model generation step of performing machine learning using the microscopic image data of the in-focus tissue section and actual measurements of the thickness of the tissue section as training data to generate an estimation model that estimates the thickness of the tissue section when an image of the tissue section is input. The estimation step estimates the thickness of the tissue section using the estimation model generated in the model generation step.

[0022] Yet another aspect of the present invention is a tissue section thickness estimation program. This program causes a computer to execute an image data generation step of generating microscopic image data of a focused tissue section, an estimation step of estimating the thickness of the tissue section from the microscopic image of the tissue section, and a model generation step of performing machine learning using the microscopic image data of the focused tissue section and an actual measured thickness of the tissue section as training data to generate an estimation model for estimating the thickness of the tissue section when an image of the tissue section is input. The estimation step estimates the thickness of the tissue section using the estimation model generated in the model generation step.

[0023] Yet another aspect of the present invention is a recording medium. The recording medium stores a program for causing a computer to execute the following steps: an image data generation step for generating microscopic image data of a tissue section in focus; an estimation step for estimating the thickness of the tissue section from the microscopic image of the tissue section; and a model generation step for performing machine learning using the microscopic image data of the in-focus tissue section and an actual measurement of the thickness of the tissue section as training data to generate an estimation model for estimating the thickness of the tissue section when an image of the tissue section is input. The estimation step stores a tissue section thickness estimation program, characterized in that the estimation step estimates the thickness of the tissue section using the estimation model generated in the model generation step.

[0024] In addition, any combination of the above components, or mutual substitution of the components or expressions of the present invention between methods, devices, programs, temporary or non-temporary storage media on which programs are recorded, systems, etc., are also valid aspects of the present invention. [Effects of the Invention]

[0025] According to the present invention, the thickness of a prepared tissue section can be estimated from an image taken with a normal optical microscope or the like. [Brief explanation of the drawings]

[0026] [Figure 1] The images show the process of cutting tissue sections (top), the cut tissue sections (middle), and microscopic images of these tissue sections (bottom). [Figure 2] 1 is a functional block diagram of a tissue slice thickness estimation device according to a first embodiment. [Figure 3] 3 is a schematic diagram showing the operation of the tissue slice thickness estimation device of FIG. 2. FIG. [Figure 4] FIG. 10 is a functional block diagram of a tissue slice thickness estimation device according to a second embodiment. [Figure 5] FIG. 10 is a functional block diagram of a tissue section thickness evaluation device according to a third embodiment. [Figure 6]10 is a flowchart showing the process of a tissue section thickness estimation method according to a fourth embodiment. [Figure 7] 10 is a graph showing the results of a verification experiment. [Figure 8] 10 is a graph showing the results of a verification experiment. [Figure 9] 10 is a graph showing the results of a verification experiment. [Figure 10] 10 is a graph showing the results of a verification experiment. [Figure 11] FIG. 13 is a functional block diagram of a tissue slice thickness estimation device according to a seventh embodiment. [Figure 12] 13 is a flowchart showing the process of a tissue section thickness estimation method according to an eighth embodiment. [Figure 13] This figure shows the experimental results of using a generative AI model to estimate the thickness of a single tissue section pixel by pixel from a focused microscopic image. The left image is the input image. The right image is a heat map of the thickness data of the pathological tissue section generated by the generative AI model. DETAILED DESCRIPTION OF THE INVENTION

[0027] The present invention will be described below based on preferred embodiments with reference to the drawings. In the embodiments and modifications, identical or equivalent components and members are designated by the same reference numerals, and redundant descriptions will be omitted where appropriate. The dimensions of the components in the drawings are enlarged or reduced as appropriate for ease of understanding. Some components that are not important for explaining the embodiments are omitted from the drawings. Terms including ordinal numbers such as "first" and "second" are used to describe various components, but these terms are used only to distinguish one component from another and do not limit the components.

[0028] Before describing the embodiments of the present invention in detail, we will first explain the underlying findings. Generally, machine learning in the medical field is often based on small data, rather than the big data typically used in machine learning, and on data with unstable accuracy. After extensive research, the inventors discovered that, for images of the same tissue section captured using an optical microscope, by creating a differential image between an image with a shallow depth of focus, high resolution, and low contrast (e.g., an image captured when the condenser adjustment of the optical microscope is turned on) and an image with a deep depth of focus, low resolution, and high contrast (e.g., an image captured when the condenser adjustment of the optical microscope is turned off), the thickness of the tissue section can be estimated relatively accurately, even if the data is small. The following embodiments are based on this finding.

[0029] [First embodiment] 2 is a functional block diagram of a tissue section thickness estimation device 1 according to a first embodiment of the present invention. The tissue section thickness estimation device 1 includes a subtraction image creation unit 10 and an estimation unit 20.

[0030] A microscope image of a tissue slice taken under shallow depth of focus conditions and a microscope image of a tissue slice taken under deep depth of focus conditions are input to the differential image creation unit 10. The differential image creation unit 10 creates a differential image between these two optical microscope images.

[0031] The change in focal depth may be generated, for example, by adjusting the condenser of the optical microscope (more specifically, by adjusting the condenser aperture or optical axis).

[0032] As will be shown in the verification experiments described below, the above tissue sections may be HE-stained tissues, unstained tissues that have been deparaffinized, or unstained tissues that have not been deparaffinized.

[0033] The estimation unit 20 estimates the thickness of the tissue section from the difference image created by the difference image creation unit 10. For example, the thicker the tissue section, the greater the change in the difference image. By utilizing such properties of the difference image, the thickness of the tissue section can be estimated relatively accurately even if the number of difference images is not very large (i.e., even if the data is small). Any suitable method can be used for the estimation, such as a statistical method using multiple regression analysis or decision tree analysis, machine learning, AI, or deep learning.

[0034] Figure 3 shows a schematic diagram of the operation of the tissue section thickness estimation device 1. The two leftmost images (a, a') were taken of the same thin section with the optical microscope's condenser adjustment on and off. There is little difference in the resulting optical microscope images, and the signal in the subtraction image is weak. The two rightmost images (b, b') were taken of the same thick section with the optical microscope's condenser adjustment on and off. There is a significant difference in the resulting optical microscope images, and the signal in the subtraction image is strong. The estimation unit 20 estimates the thickness of the tissue section from this subtraction image. Typical optical microscope images vary greatly depending on the tissue section's location (organ) and staining intensity, and this variety far exceeds the differences due to the section's thickness. However, the subtraction image offsets the effects of the tissue section's location (organ) and staining intensity, strongly reflecting the section's thickness.

[0035] [Second embodiment] In a particularly useful embodiment, the estimation unit 20 estimates the thickness of the tissue section using an estimation model generated by supervised machine learning, as described below. FIG. 4 is a functional block diagram of a tissue section thickness estimation device 2 according to a second embodiment of the present invention. The tissue section thickness estimation device 2 includes a subtraction image creation unit 10, an estimation unit 20, and a model generation unit 30. That is, the tissue section thickness estimation device 2 includes the model generation unit 30 in addition to the configuration of the tissue section thickness estimation device 1 in FIG. 2. The other configuration of the tissue section thickness estimation device 2 is common to the configuration of the tissue section thickness estimation device 1. Below, the tissue section thickness estimation device 2 will be described, focusing on the differences from the tissue section thickness estimation device 1.

[0036] The difference image created by the difference image creation unit 10 and the actual measured thickness of the tissue slice on which the difference image is based are input to the model generation unit 30. The actual measured thickness may be measured using a confocal microscope, an industrial laser microscope, or the like.

[0037] The model generation unit 30 performs machine learning using, as training data, microscopic images of tissue slices taken with a shallow depth of focus, microscopic images of tissue slices taken with a deep depth of focus, and actual measured values ​​of the thickness of the tissue slices. Through this machine learning, when an image of a tissue slice (which may be a microscopic image taken with a shallow depth of focus or a deep depth of focus) is input, the model generation unit 30 generates an estimation model that estimates and outputs the thickness of the tissue slice.

[0038] Machine learning may be performed using known AI. The specific AI method is not particularly limited, but neural networks such as convolutional neural networks (CNN), recurrent neural networks (RNN), and long short-term memory (LSTM) networks may be used. In this case, different neural networks may be mixed for each computational model while sharing the input layer.

[0039] As will be shown in the verification experiments described below, a sufficient correlation is found between the microscopic images of tissue slices taken under shallow depth of focus conditions, the microscopic images of tissue slices taken under deep depth of focus conditions, and the measured thickness values ​​of the tissue slices. The model generation unit 30 of this embodiment can generate highly accurate estimation models of tissue slices.

[0040] An optical microscope image of a tissue slice is input to the estimation unit 20. The estimation unit 20 uses the estimation model generated by the model generation unit 30 to estimate the thickness of the tissue slice.

[0041] According to this embodiment, it is possible to provide an apparatus that estimates the thickness of a prepared tissue slice with high accuracy from an image obtained by a normal optical microscope or the like.

[0042] [Third embodiment] 5 is a functional block diagram of a tissue section thickness evaluation device 3 according to a third embodiment of the present invention. The tissue section thickness evaluation device 3 includes the tissue section thickness estimation device 1 of FIG. 1 and a tissue section thickness evaluation unit 40.

[0043] The thickness of the tissue section estimated by the tissue section thickness estimation device 1 and the color of the tissue section are input to the tissue section thickness evaluation unit 40. Based on the thickness and color of the tissue section, the tissue section thickness evaluation unit 40 evaluates whether the tissue section can be used for pathological diagnosis.

[0044] When considering usability in pathological diagnosis, for example, there are cases where a tissue section of a certain color requires very accurate thickness, while a tissue section of another color is relatively tolerant of thickness error. To address this situation, the tissue section thickness evaluation unit 40 comprehensively evaluates the thickness and color of the tissue section. This allows the tissue section thickness evaluation device 3 to be used as an application (medical device) that determines whether an acquired tissue section can be used for pathological diagnosis.

[0045] [Fourth embodiment] 6 is a flowchart showing the process of the tissue section thickness estimation method according to the fourth embodiment. This method includes a subtraction image creation step S1 and an estimation step S2.

[0046] In the differential image creating step S1, a differential image is created between a microscopic image of a tissue slice photographed under a shallow focal depth condition and a microscopic image of a tissue slice photographed under a deep focal depth condition.

[0047] In the estimation step S2, the thickness of the tissue slice is estimated from the difference image created in the difference image creation step S1.

[0048] According to this embodiment, the thickness of the prepared tissue slice can be estimated from an image taken by a normal optical microscope or the like using a computer or the like.

[0049] [Fifth embodiment] The fifth embodiment is a tissue section thickness estimation program (computer program). Fig. 6 is a flowchart showing the processing executed by a computer based on the tissue section thickness estimation program according to the fifth embodiment. This program causes the computer to execute a subtraction image creation step S1 and an estimation step S2.

[0050] In the differential image creating step S1, a differential image is created between a microscopic image of a tissue slice photographed under a shallow focal depth condition and a microscopic image of a tissue slice photographed under a deep focal depth condition.

[0051] In the estimation step S2, the thickness of the tissue slice is estimated from the difference image created in the difference image creation step S1.

[0052] According to this embodiment, a program for causing a computer to execute a method for estimating the thickness of a prepared tissue section from an image obtained by a normal optical microscope or the like can be implemented as software.

[0053] [Sixth embodiment] The sixth embodiment is a recording medium. This recording medium records a computer program that causes a computer to execute the difference image creation step S1 and the estimation step S2. Figure 6 is a flowchart showing the processing that the computer executes based on the program recorded on the recording medium according to the sixth embodiment.

[0054] In the differential image creation step S1, a microscopic image of the tissue slice photographed under a shallow focal depth condition and a microscopic image of the tissue slice photographed under a deep focal depth condition are created.

[0055] In the estimation step S2, the thickness of the tissue slice is estimated from the difference image created in the difference image creation step S1.

[0056] According to this embodiment, a program for causing a computer to execute a method for estimating the thickness of a prepared tissue section from an image obtained by a normal optical microscope or the like can be recorded on a recording medium.

[0057] [Verification experiment] The present inventors conducted an experiment to verify the effectiveness of the above-described technique. The results of estimating the thickness of a tissue section using the tissue section thickness estimation device 2 of Fig. 4 are shown in Figs. 7 to 10. In Figs. 7 to 10, the horizontal axis represents the measured value of the thickness of the tissue section, and the vertical axis represents the estimated value of the thickness of the tissue section.

[0058] Figures 7 and 8 show the experimental results for 100 HE-stained slides of colonic mucosa tissue sections. This experiment used neural network regression model learning, with the created subtraction images and the actual thickness of the same area measured with a confocal laser microscope as training data. The optical microscope magnification was set to 400x. The subtraction images were created by photographing the HE-stained tissue sections with and without condenser adjustment using a microscope digital camera. The machine learning used a VGG16-based architecture with an Adam optimization function (learning rate 0.0001), an input resolution of 224 × 224 pixels, 100 epochs, and a batch size of 16.

[0059] When cross-validation was performed with the sample divided into three sections, the R2 scores (generalization accuracy) were 0.919, 0.813, and 0.845, respectively, with an average of 0.859 (mean cv). The RMSLE (logarithmic mean square error) was 0.067, 0.095, and 0.076, respectively, with an average of 0.079. The MAPE (mean absolute percentage error) was 0.063, 0.101, and 0.077, respectively, with an average of 0.081. Figure 7 shows the experimental results with an auxiliary line indicating a ±10% error. Figure 8 shows the experimental results with an auxiliary line indicating a ±20% error. As can be seen from these figures, we were able to verify that practical accuracy was achieved, with an error within approximately 20% for all section thicknesses.

[0060] Figures 9 and 10 show the experimental results for 102 unstained (undeparaffinized) colonic mucosa tissue sections. This experiment involved neural network regression model learning using the subtraction images and the actual thickness of the same area measured with an industrial laser microscope as training data. The optical microscope magnification was set to 200x. The subtraction images were created by photographing the unstained (undeparaffinized) tissue sections with and without condenser adjustment using a microscope digital camera. Machine learning was performed using a VGG16-based architecture with an Adam optimization function (learning rate 0.0001), an input resolution of 224 × 224 pixels, 300 epochs, and a batch size of 16.

[0061] When cross-validation was performed by dividing the specimen into three parts, the R2 scores (generalization accuracy) were 0.824, 0.619, and 0.772, respectively, with an average of 0.738 (mean cv). The RMSLE (logarithmic mean square error) was 0.141, 0.202, and 0.141, respectively, with an average of 0.161. The MAPE (mean absolute percentage error) was 0.122, 0.189, and 0.125, respectively, with an average of 0.1456. Figure 9 shows the experimental results with an auxiliary line indicating a ±10% error. Figure 10 shows the experimental results with an auxiliary line indicating a ±20% error. As can be seen from these figures, although the accuracy is lower than that of HE-stained images, it was verified that thickness estimation using AI is possible to a certain extent from unstained tissue sections.

[0062] [Seventh embodiment] In the above embodiment, the thickness of the tissue section is estimated using a differential image between a microscopic image of the tissue section taken under a shallow depth of focus and a microscopic image of the tissue section taken under a deep depth of focus. However, this is not limited to this, and the thickness of the tissue section can also be estimated using machine learning from a single microscopic image of the tissue section in focus.

[0063] 11 is a functional block diagram of a tissue section thickness estimation device 4 according to a seventh embodiment of the present invention. The tissue section thickness estimation device 4 includes an image data creation unit 12, an estimation unit 22, and a model generation unit 32.

[0064] A focused microscope image of a tissue slice is input to the image data creation unit 12. The image data creation unit 12 creates image data of this microscope image.

[0065] The model generation unit 32 performs machine learning using the microscopic image data of the in-focus tissue section and the actual measured thickness of the tissue section as training data, and generates an estimation model that estimates the thickness of the tissue section when an image of the tissue section is input.

[0066] Machine learning may be performed using known AI. The specific AI method is not particularly limited, but neural networks such as convolutional neural networks (CNN), recurrent neural networks (RNN), and long short-term memory (LSTM) networks may be used. In this case, different neural networks may be mixed for each computational model while sharing the input layer.

[0067] The estimation unit 22 estimates the thickness of the tissue slice from the image data created by the image data creation unit 12 using the estimation model created by the model creation unit 32.

[0068] [Eighth embodiment] 12 is a flowchart showing the process of the tissue section thickness estimation method according to the eighth embodiment. This method includes an image data generation step S3, a model generation step S4, and an estimation step S5.

[0069] The image data generation step S3 generates microscopic image data of the in-focus tissue section.

[0070] In the model generation step S4, machine learning is performed using the microscopic image data of the in-focus tissue section and the actual measured thickness of the tissue section as training data, to generate an estimation model that estimates the thickness of the tissue section when an image of the tissue section is input.

[0071] In the estimation step S5, the thickness of the tissue section is estimated from the microscopic image.

[0072] According to this embodiment, the thickness of the prepared tissue slice can be estimated from an image taken by a normal optical microscope or the like using a computer or the like.

[0073] [Ninth embodiment] The ninth embodiment is a tissue section thickness estimation program (computer program). Fig. 12 is a flowchart showing the processing executed by a computer based on the tissue section thickness estimation program according to the ninth embodiment. This program causes the computer to execute a subtraction image creation step S1, an image data creation step S3, a model generation step S4, and an estimation step S5.

[0074] The image data generation step S3 generates microscopic image data of the in-focus tissue section.

[0075] In the model generation step S4, machine learning is performed using the microscopic image data of the in-focus tissue section and the actual measured thickness of the tissue section as training data, to generate an estimation model that estimates the thickness of the tissue section when an image of the tissue section is input.

[0076] In the estimation step S5, the thickness of the tissue section is estimated from the microscopic image using the estimation model generated in the model generation step S4.

[0077] According to this embodiment, a program for causing a computer to execute a method for estimating the thickness of a prepared tissue section from an image obtained by a normal optical microscope or the like can be implemented as software.

[0078] [Tenth embodiment] The tenth embodiment is a recording medium. This recording medium records a computer program that causes a computer to execute a difference image creation step S1, an image data creation step S3, a model generation step S4, and an estimation step S5. Figure 12 is a flowchart showing the processing that the computer executes based on the program recorded on the recording medium according to the tenth embodiment.

[0079] The image data generation step S3 generates microscopic image data of the in-focus tissue section.

[0080] In the model generation step S4, machine learning is performed using the microscopic image data of the in-focus tissue section and the actual measured thickness of the tissue section as training data, to generate an estimation model that estimates the thickness of the tissue section when an image of the tissue section is input.

[0081] In the estimation step S5, the thickness of the tissue section is estimated from the microscopic image using the estimation model generated in the model generation step S4.

[0082] According to this embodiment, a program for causing a computer to execute a method for estimating the thickness of a prepared tissue section from an image obtained by a normal optical microscope or the like can be recorded on a recording medium.

[0083] [Performance evaluation] The inventors evaluated the estimation results obtained by the above-described method of estimating the thickness of a tissue section using machine learning from a single in-focus microscopic image of the tissue section. Three evaluation indices were used: R2 score (coefficient of determination score), RMSLE (Root Mean Squared Logarithmic Error), and MAPE (Mean Absolute Percentage Error). The in-focus microscopic image of the tissue section used for machine learning was a single HE image taken when the microscope's condenser adjustment was turned on. The results are as follows: R2 score: 0.929 RMSLE: 0.093 MAPE:0.093 The evaluation results showed that by using this method for machine learning, it is possible to estimate the thickness of tissue sections with high accuracy.

[0084] In the seventh embodiment, CNN, RNN, LSTM, etc. are exemplified as specific machine learning techniques for generating an estimation model that estimates the thickness of a tissue slice when an image of the tissue slice is input. However, the machine learning technique is not limited to these, and a generative AI model, for example, can also be used.

[0085] Figure 13 shows the experimental results of using a generative AI model to estimate section thickness for each pixel from a single focused microscopic image of a tissue section. The left image is the input image, i.e., an optical microscope image. The right image is a heat map of the pathological tissue section thickness data generated by the generative AI model, with the estimated thickness represented by gray levels. The generative AI model used an autoencoder as its base architecture. In this experiment, the model was trained using 49 microscopic images and the corresponding actual measurements of tissue section thickness taken with a confocal laser microscope. The experimental results demonstrate that it is possible not only to obtain a single representative value for section thickness within the range of the target image, but also to finely distinguish between areas with appropriate and inappropriate thicknesses for diagnosis across the entire image.

[0086] The present invention has been described above based on several embodiments. These embodiments are merely examples, and it will be understood by those skilled in the art that various modifications and changes are possible within the scope of the claims of the present invention, and that such modifications and changes also fall within the scope of the claims of the present invention. Therefore, the descriptions and drawings in this specification should be treated as illustrative rather than restrictive.

[0087] Any combination of the above-described embodiments and modifications is also useful as an embodiment of the present invention. A new embodiment resulting from the combination has the combined effects of each of the combined embodiments and modifications.

[0088] When understanding the abstract technical ideas of the embodiments and modifications, the technical ideas should not be interpreted as being limited to the contents of the embodiments and modifications. The above-described embodiments and modifications are merely illustrative examples, and many design modifications, such as changes, additions, and deletions of components, are possible. In the embodiments, the contents in which such design modifications are possible are emphasized by adding the notation "embodiment." However, it goes without saying that design modifications are also permitted even in contents not so notated. [Industrial Applicability]

[0089] The present invention relates to a tissue slice thickness estimation device, a tissue slice thickness evaluation device, a tissue slice thickness estimation method, a tissue slice thickness estimation program, and a recording medium. [Explanation of symbols]

[0090] 1. Tissue section thickness estimation device, 2. Tissue section thickness estimation device, 3. Tissue section thickness evaluation device, 4. Tissue section thickness estimation device 10. Difference image creation unit, 12. Image data creation unit, 20...Estimation section, 22... Estimation Department, 30··Model generation section, 32··Model generation part, 40··Tissue section thickness evaluation section, S1··Difference image creation step; S2··estimation step, S3: Image data creation step; S4··Model generation step, S5··Estimation step.

Claims

1. an image data generating unit that generates microscopic image data of the in-focus tissue section; an estimation unit that estimates the thickness of the tissue section from the microscopic image; a model generation unit that performs machine learning using the microscope image data of the in-focus tissue section and the actual measured thickness of the tissue section as training data, and generates an estimation model that estimates the thickness of the tissue section when an image of the tissue section is input; The tissue slice thickness estimation device, wherein the estimation unit estimates the thickness of the tissue slice using the estimation model generated by the model generation unit.

2. an image data generation step of generating microscopic image data of the in-focus tissue section; an estimation step of estimating a thickness of the tissue section from the microscopic image; a model generation step of performing machine learning using the microscopic image data of the focused tissue section and the actual measured thickness of the tissue section as training data to generate an estimation model that estimates the thickness of the tissue section when an image of the tissue section is input; The method for estimating a thickness of a tissue slice, wherein the estimation step estimates the thickness of the tissue slice using the estimation model generated in the model generation step.

3. an image data generation step of generating microscopic image data of the in-focus tissue section; an estimation step of estimating a thickness of the tissue section from the microscopic image; a model generation step of performing machine learning using the microscopic image data of the focused tissue section and the actual measured thickness value of the tissue section as training data to generate an estimation model that estimates the thickness of the tissue section when an image of the tissue section is input; The tissue slice thickness estimation program is characterized in that the estimation step estimates the thickness of the tissue slice using the estimation model generated in the model generation step.

4. an image data generation step of generating microscopic image data of the in-focus tissue section; an estimation step of estimating a thickness of the tissue section from the microscopic image; a model generation step of performing machine learning using the microscopic image data of the focused tissue section and the actual measured thickness value of the tissue section as training data to generate an estimation model that estimates the thickness of the tissue section when an image of the tissue section is input; A recording medium having a tissue slice thickness estimation program recorded thereon, wherein the estimation step estimates the thickness of the tissue slice using the estimation model generated in the model generation step.

Images