Microscopic layered image data analysis device and data analysis method

Abstract

Provided is a microscopic layered image data analysis device for analyzing microscopic z-stack images for use in cytologic diagnosis. This microscopic layered image data analysis device for a specimen S containing a large number of cells comprises: a microscope (20) that acquires z-stack images of the specimen S; an imaging unit (30) that images the z-stack images acquired by the microscope (20); a storage unit (70) that stores the z-stack images captured by the imaging unit (30); and a control unit (50) that detects cells from the z-stack images stored in the storage unit (70), acquires a cell image group corresponding to each of the detected cells, crops a cell image from images forming the cell image group for each of the cells, acquires the confidence level of each of a plurality of class classifications for each of the cell images of the cells, and visualizes the confidence level acquired for each of the class classifications.

Description

Data analysis device and data analysis method for microscopic stacked images

[0001] The present invention relates to a data analysis device and a data analysis method for microscopic stacked images.

[0002] To digitize specimens with depth, such as cytology samples, techniques have been developed to capture Z-stack images of the entire slide. However, capturing Z-stack images of the entire slide results in a huge amount of data. Therefore, techniques such as extended depth-of-field (EDF) have been developed to generate a focused composite image from the Z-stack images (see, for example, U.S. Pat. No. 9,332,190).

[0003] In addition, a technology has been developed that compresses captured Z-stack image data to reduce and record the data size while retaining the information on each Z layer of the Z-stack image (e.g., International Publication No. 2023 / 233922). Furthermore, artificial intelligence is used to analyze large amounts of digitized cell images and detect and classify objects. When analyzing large amounts of cell images, it is sometimes necessary to analyze the type, state, and proportion of cells contained in the entire cell population. In this case, a technology has also been developed that classifies the type and state of individual cells using artificial intelligence, obtains the confidence level for each class classification as the output of the artificial intelligence, and depicts the confidence level for each class classification of a large number of cells using a histogram, two-dimensional plot, or the like (e.g., International Publication No. 2020 / 004101).

[0004] FIG. 1 is a schematic diagram of a portion of a microscope Z-stack image containing a large number of cells. In the example shown in FIG. 1, six Z-layers (L1 to L6) are shown. The distance between Z-layers in the Z-stack image is, for example, 1 μm or less. As shown in FIG. 1, when the distance between Z-layers is narrower than the depth of the cell, a single cell is observed across multiple Z-layers.

[0005] When such Z-stack images are processed using the EDF method, the images of cells C and D, which are overlapping in the depth direction, are linked, as shown in Figure 2. For example, in cytological diagnosis, it is necessary to observe cell clumps where many cells have accumulated. Therefore, using data processed using the EDF method, it is difficult to observe and analyze cell clumps where many cells have accumulated.

[0006] Conventionally, there has been no established analytical method for the entire cell population, taking into account the three-dimensional arrangement of the cell population, for a large number of cells contained in a Z-stack image. The present invention aims to provide a data analysis device and data analysis method for microscopic stacked images that analyze microscopic Z-stack images used in cytological diagnosis.

[0007] A first aspect of the present invention is a data analysis device for microscopic stacked images of a specimen containing a large number of cells, comprising: a microscope that acquires Z-stack images of the specimen; an imaging unit that captures the Z-stack images acquired by the microscope; a memory unit that stores the Z-stack images captured by the imaging unit; and a control unit that detects cells from the Z-stack images stored in the memory unit; acquires a cell image group corresponding to each detected cell; cuts out a cell image from images that constitute the cell image group for each cell; acquires a confidence level for each of a plurality of class classifications for the cell image of each cell; and visualizes the confidence level acquired for each class classification.

[0008] A second aspect of the present invention is a data analysis method for microscopic stacked images of a specimen containing a large number of cells, comprising: acquiring Z-stack images of the specimen with a microscope; capturing the Z-stack images acquired by the microscope with an imaging unit; storing the Z-stack images captured by the imaging unit in a memory unit; detecting cells from the Z-stack images stored in the memory unit with a control unit; acquiring a cell image group corresponding to each detected cell with the control unit; extracting a cell image from images constituting the cell image group with respect to each of the detected cells; acquiring confidence levels for each of a plurality of class classifications with respect to the cell image of each of the cells; and visualizing the confidence levels acquired for each of the class classifications.

[0009] According to the present invention, it is possible to analyze microscopic Z-stack images used in cytological diagnosis.

[0010] Schematic diagram of a cut-out portion of a Z-stack image Schematic diagram of an image obtained by processing a Z-stack image using the EDF method Schematic diagram of a data analysis device according to an embodiment Diagram showing a method for acquiring Z-stack images Flowchart showing a data analysis method according to an embodiment Diagram showing an image of a cell image group Diagram showing a cell image of each cell Schematic diagram of a cell image of each cell Diagram showing class classification of the cell image of each cell (a) and (b) are histograms of confidence (a), (b), and (c) are two-dimensional plots of confidence First two-dimensional plot of multi-stage gating Second two-dimensional plot of multi-stage gating

[0011] The data analysis device and method for microscopic stacked images will be described below with reference to the drawings. The data analysis device and method for microscopic stacked images of the present invention analyze microscopic Z-stack images containing a large number of cells to obtain information about the entire cell population contained in the Z-stack image. FIG. 3 shows a schematic configuration diagram of the data analysis device of this embodiment. The data analysis device of this embodiment includes a microscope 20, an imaging unit 30, a focal plane moving mechanism 40, and a control unit 50. The data analysis device may further include a stage 10, a stage moving mechanism 42, an image buffer 55, a memory unit 70, and a transmission unit 80. The data analysis device may also include an illumination unit 22.

[0012] A specimen S, which is an object to be imaged, is placed on a stage 10. In the example shown in FIG. 3 , the stage 10 is parallel to the XY plane. A microscope 20 acquires an image of the specimen S at a focal plane. The microscope 20 has an optical system such as a lens. An imaging unit 30 captures the image acquired by the microscope 20 based on an imaging timing signal. The imaging unit 30 is, for example, a digital camera.

[0013] The focal plane moving mechanism 40 moves the focal plane of the microscope 20 in the optical axis direction based on the focal plane control signal. The optical axis direction is perpendicular to the stage 10. In the example shown in FIG. 3 , the optical axis direction is the Z direction. In this embodiment, the optical axis direction (Z direction) is the vertical direction, but the optical axis direction is not limited to the vertical direction. The stage moving mechanism 42 moves the microscope 20 and the stage 10 relatively in directions (X and Y directions) perpendicular to the optical axis direction based on the stage movement control signal.

[0014] The image buffer 55 may be disposed between the imaging unit 30 and the storage unit 70. The image buffer 55 may be configured as part of the output side of the imaging unit 30. The image buffer 55 temporarily stores image data of multiple images captured by the imaging unit 30. The storage unit 70 stores the image data sent from the image buffer 55. The transmission unit 80 transmits the compressed image data to the outside.

[0015] The control unit 50 controls the imaging unit 30, the focal plane moving mechanism 40, the stage moving mechanism 42, the storage unit 70, and the transmission unit 80. Specifically, the control unit 50 transmits a focal plane control signal to the focal plane moving mechanism 40 to move the focal plane to a predetermined focal coordinate (Z coordinate). The control unit 50 transmits an imaging timing signal to the imaging unit 30 to instruct the timing of capturing an image.

[0016] The control unit 50 transmits a stage movement control signal to the stage movement mechanism 42 to relatively move the microscope 20 and the stage 10 in directions (X and Y directions) perpendicular to the optical axis direction. Based on the focal plane control signal and the stage movement control signal, the control unit 50 transmits an image capturing timing signal to the image capturing unit 30 at the timing when the focal coordinates become predetermined coordinates. This allows the image capturing unit 30 to capture an image of the specimen S at the desired focal plane.

[0017] The control unit 50 may control the transmission unit 80 to transmit the image data to the outside. The control unit 50 may add additional information such as imaging conditions to the image data stored in the storage unit 70 or to the image data transmitted from the transmission unit 80 to the outside.

[0018] Here, a Z-stack image will be described with reference to FIG. 4 . FIG. 4 shows an example of a method for acquiring a Z-stack image. A Z-stack image is obtained by stacking planar images of multiple Z layers (L1 to L5) having different focal planes in the depth direction in a microscope 20. For example, using an imaging unit 30 equipped with an area sensor such as a CCD or CMOS, imaging is repeated in an optical system that forms an image of the focal plane on the specimen S on the area sensor, while gradually changing the distance between the objective lens of the microscope 20 and the specimen S. In this way, planar images of multiple Z layers (L1 to L5) are continuously acquired, thereby obtaining a Z-stack image.

[0019] Here, the planar image is repeatedly captured at a Z-layer distance of, for example, 250 nm or 500 nm per Z layer. The Z-layer distance is typically about 200 nm to 2 μm. The depth of field near the focal plane is determined by the optical system. Therefore, it is preferable to set the optical system so that the Z-layer distance and the depth of field are approximately equal. For example, the depth of field is determined by the numerical aperture of the objective lens and the numerical aperture of the illumination light. When capturing images using transmitted light, the aperture of the transmitted illumination light incident on the specimen S can be adjusted by an aperture diaphragm.

[0020] The distance between the specimen S and the objective lens is changed by moving the entire optical system of the microscope 20, or a part of the optical system including the objective lens, in the optical axis direction using a vertical stage (Z stage). Alternatively, the distance between the specimen S and the objective lens may be changed by placing the stage 10 carrying the specimen S on the vertical stage and moving it in the optical axis direction. The vertical stage may be a single stage. Alternatively, the vertical stage may be configured as a combination of two or more types of stages, such as a coarse movement stage that retracts the optical system and a fine movement stage that finely adjusts the distance between the objective lens and the specimen. The vertical stage can perform fine movement with sufficient accuracy relative to the distance between the Z layers of the stacked image.

[0021] The imaging range of the stacked image during imaging is determined according to the characteristics of the specimen S. For example, in the case of a cytological specimen, it is assumed that many cells are fixed directly on the glass slide, with other cells stacked on top of that. In this case, the imaging range is set to a height sufficient to accommodate the height of the stacked cells from the surface of the glass slide.

[0022] It is preferable to use autofocus processing to determine the imaging range of stacked images. Specifically, the degree of focusing of each Z layer image in a Z stack image temporarily acquired for autofocusing is quantified, and the range including the Z layer whose degree of focus meets certain conditions is determined as the imaging range. A method for quantifying the degree of focusing is to use the method described below to detect cells or cell nuclei contained in the Z stack image, select the Z layer most suitable for observation of each cell or cell nucleus, and then use the number of images of each selected cell or cell nucleus as an indicator of the degree of focusing. In this case, it is possible to set the range of Z layers in which a certain number or more of the selected cells or cell nuclei are present as the imaging range.

[0023] The control unit 50 acquires in advance the parameters of the imaging range and the distance between Z layers depending on the type of specimen S. The control unit 50 switches the parameters of the imaging range and the distance between Z layers depending on the type of slide or the content of the image data acquired by focus processing and the position of the vertical stage. This makes it possible to efficiently acquire images of sufficient quality to reconstruct an appropriate three-dimensional image.

[0024] In order to integrate the acquired stacked images from multiple locations, the control unit 50 records data on the vertical stage position during imaging. For example, the control unit 50 detects the vertical stage position using an encoder mounted on the stage 10. The control unit 50 may also detect the vertical stage position by integrating the amount of movement of the stage 10 from the origin position. The control unit 50 may also detect the vertical stage position by monitoring, using a digital circuit, pulses from a stepping motor that operates the stage 10 and control pulses input to a controller of the stepping motor.

[0025] In the example shown in FIG. 4 , an image may be captured after the focal plane has completed its movement and stopped. Alternatively, the focal plane may be continuously moved, and an image may be captured when the focal plane reaches a predetermined position. The latter imaging method can capture images faster than the former imaging method. Note that in the latter imaging method, if the focal plane movement distance during the exposure time of the imaging unit 30 is sufficiently large compared to the depth of field of the optical system, the image may become blurred. To avoid this, it is preferable to set the exposure time of the imaging unit 30 and the focal plane movement speed to match the depth of field of the optical system. Furthermore, if a faster focal plane movement speed is desired, using a flash lamp for the illumination 22 can reduce image blurring caused by focal plane movement.

[0026] The control unit 50 transmits a stage movement control signal that relatively moves the microscope 20 and the stage 10 in a direction perpendicular to the optical axis direction. The control unit 50 also transmits a focal plane control signal that moves the focal plane to a predetermined focal coordinate. When the vertical stage moves to a specific position where imaging is desired, the control unit 50 transmits an imaging timing signal to the imaging unit 30. The imaging unit 30 captures an image at the timing when it receives the imaging timing signal.

[0027] The data analysis method of this embodiment will be described below with reference to Fig. 5. Fig. 5 is a flowchart showing the data analysis method of this embodiment. First, the imaging unit 30 captures Z stack images acquired by the microscope 20 based on an imaging timing signal (step S1).

[0028] Next, the control unit 50 loads the Z stack image into the computer memory (step S2). Preferably, the Z stack image is loaded into a GPU (Graphical Processing Unit) memory. If the Z stack image captured in step S1 has been compressed and saved, the compressed data is decoded. If the compressed data is decoded using a GPU or a device connected to the GPU, the decoded data is output to the GPU memory. All Z layers of the compressed Z stack image may be decoded, or only some of the Z layers may be decoded.

[0029] Next, the control unit 50 detects cells from each Z layer of the Z-stack image (step S3). Specifically, the control unit 50 detects cells by detecting cell nuclei as shown in FIG. 1. For example, the control unit 50 can detect cell nuclei by binarizing the image based on the color tone, density, contrast, etc. of the cell nuclei and then analyzing the size, shape, etc. from the binarized image. Alternatively, the control unit 50 may detect cell nuclei using an object detection algorithm based on deep learning, such as Yolo (You Only Look Once). As a result, for example, in the example shown in FIG. 1, it is possible to detect that the cell nucleus of cell A spans three Z layers L2 to L4. It is also possible to detect that the cell nucleus of cell B spans two Z layers L4 and L5.

[0030] Next, the control unit 50 acquires cell image groups (step S4). For example, in the example shown in FIG. 1 , cell A is detected across five Z layers L1 to L5. Therefore, the control unit 50 acquires cell image groups corresponding to cell A from the five Z layers L1 to L5. Similarly, the control unit 50 acquires cell image groups corresponding to cell B from the five Z layers L2 to L6. The total number of cell image groups acquired in this manner corresponds to the total number of cells included in the Z stack image of the specimen S.

[0031] Fig. 6 is a diagram showing an example of images in a cell image group. In the example shown in Fig. 6, cell A spans Z layers L1 to L9. These nine images make up the cell image group corresponding to cell A. Similarly, cell B spans Z layers L1 to L5. These five images make up the cell image group corresponding to cell B.

[0032] In step S4, a method may be adopted in which cell images detected in adjacent Z layers are grouped together if the difference in XY position of the cell nuclei is 25% or less of the width and height of the region of interest (ROI). Alternatively, the area of ​​the overlapping region in adjacent Z layers may be used as an index, or cell image groups may be acquired by taking into account shape and color information in addition to position.

[0033] Next, the control unit 50 extracts cell images from each cell image group (step S5). Specifically, the control unit 50 determines, for each cell, one or more Z layers suitable for observing and analyzing the cell from the images that make up the cell image group. Then, the control unit 50 extracts, for each cell, a cell image from the determined Z layer.

[0034] More specifically, it is preferable to select as the cell image, from among the images constituting the cell image group, an image in which the cell nucleus is in focus. In step S3, if a cell is detected by detecting the cell nucleus, the Z-stack image in which the cell nucleus is detected may be selected as the cell image. Furthermore, for the multiple images constituting the cell image group, cell images may be selected using statistics such as standard deviation or variance as an index. In the example shown in FIG. 6 , the numbers below each image indicate the standard deviation of the pixel brightness values ​​of each image. The control unit 50 may select as the cell image, for each cell, the image in which the standard deviation of the pixel brightness values ​​is the largest. Alternatively, cell images may be selected from the multiple images constituting the cell image group by image processing such as edge detection.

[0035] Here, the control unit 50 preferably crops the cell image so that the cell nucleus of the target cell is positioned at the center. The size of the cell image may be any size required for subsequent analysis. FIG. 7 shows an example of a cell image of each cell. In this way, as shown in FIG. 7, a cell image suitable for cell observation and analysis is selected for each cell. FIG. 8 is a schematic diagram of a cell image of each cell corresponding to FIG. 1. As described with reference to FIG. 2, when Z-stack images such as those in FIG. 1 are processed using the EDF method, images of cells C and D, which are overlapping in the depth direction, are linked as shown in FIG. 2. In contrast, according to the present invention, different images can be obtained for cells C and D, which are overlapping in the depth direction, as shown in FIG. 8.

[0036] In step S3, the control unit 50 does not have to detect cell nuclei for all Z layers. The control unit 50 may detect cell nuclei every predetermined Z layer (for example, every four Z layers). This reduces the time required to detect cell nuclei.

[0037] Next, the control unit 50 analyzes the cell images of each cell obtained in step S5 and performs class classification (step S6). The control unit 50 then obtains a confidence factor for each class classification of each cell. Here, the confidence factor can be obtained using, for example, a softmax function. In this case, the sum of the confidence factors for each class is 1. Alternatively, the confidence factor can be obtained using a sigmoid function. In this case, the sum of the confidence factors for each class does not necessarily equal 1. When drawing a histogram or two-dimensional plot of confidence factors, the confidence factor obtained using the softmax function may result in a discontinuous cell population at both ends of confidence factors 0 and 1, resulting in an unnatural distribution. Therefore, using a sigmoid function may be preferable because it results in a more natural distribution. The following example shows confidence factors obtained using a sigmoid function.

[0038] For example, the control unit 50 performs image classification for each cell image using the Multi-Axis Vision Transformer (MaxViT) method. FIG. 9 is a diagram showing an example of class classification for cell images of each cell. In the example shown in FIG. 9, a model pre-trained to classify cell images into four classes: leukocytes (Leu), glandular cells (Glan), epithelial cells (Epi), and debris (Debris) is used. Note that the model was trained to classify foreign matter such as dust and images in which no cells are visible in the center of the cell image. In the example shown in FIG. 9, the confidence level for each class classification when classifying the cell image using this model is shown on the right side of the cell image.

[0039] The control unit 50 then visualizes the certainty of each class classification (step S7). Specifically, the control unit 50 outputs the certainty of each class classification as a histogram, a two-dimensional plot, or the like. An example in which the control unit 50 outputs the certainty of each class classification as a histogram will be described below with reference to FIG. 10. Two cytological specimens were analyzed, and a histogram of the certainty of debris was created from the obtained results. FIG. 10(a) shows the histogram for cytological specimen A. FIG. 10(b) shows the histogram for cytological specimen B. In FIGS. 10(a) and 10(b), the horizontal axis of the graph indicates the certainty of debris. The vertical axis of the graph indicates the frequency of the certainty of debris. An enlarged view of a portion of the graph with the vertical axis enlarged is shown in the upper right corner of the graph.

[0040] For cytological specimen A shown in FIG. 10( a), the total number of cells analyzed was 11,859. Cells with a debris confidence level of 0.8 or higher were identified as debris. Consequently, 349 cells were identified as debris. Similarly, for cytological specimen B shown in FIG. 10( b), the total number of cells analyzed was 133,136. Of these, 1,687 cells were identified as debris. The control unit 50 can then display cell images of cells identified as debris, i.e., cells included within the debris gate, on the output unit. In this way, using a histogram makes it possible to visualize the quantity and proportion of cells classified into a specific class (debris in the example shown in FIGS. 10( a) and 10(b)). Furthermore, by setting a gate on the histogram, specific classes can be easily extracted.

[0041] It should be noted that the visualization of the certainty of each class classification by the control unit 50 is not limited to a histogram or a two-dimensional plot. For example, t-distributed stochastic neighbor embedding (t-SNE), Self-Organizing Map (SOM), Sub-Image Anomaly Detection with Deep Pyramid Correspondences (SPADE), Uniform Manifold Approximation and Projection (UMAP), Isometric Mapping (Isomap), Locally Linear Embedding (LLE), Autoencoder, Variational Autoencoder (VAE), Principal Component Analysis (PCA), Independent Component Analysis (ICA), Linear Discriminant Analysis (LDA), Factor Analysis (FA), Multidimensional Scaling (MDS), Non-negative Matrix Factorization (NMF), Kernel PCA, tangent distance-based manifold learning, Maximum Variance Unfolding (MVU), Neighborhood Components Analysis (NCA), Laplacian Eigenmaps, Diffusion Maps, Random Projection (RP), Feature The data may be plotted after dimension reduction using techniques such as agglomeration, sparse PCA, tangent PCA (tPCA), and deep embedding clustering (DEC). When plotting after dimension reduction, for example, by color-coding the estimated class classification of each cell, it is possible to understand which types of cells are distributed at which positions in the plot.As a method for estimating the class classification of each cell, for example, a class classification with the maximum confidence level or a method for setting the class classification by analysis using gates set on a histogram or two-dimensional plot can be used.

[0042] An example in which the control unit 50 outputs the certainty of each class classification in a two-dimensional plot will be described below with reference to FIG. 11 . Three cytological specimens were analyzed, and two-dimensional plots of the certainty of debris and leukocytes were created from the obtained results. The results are shown in FIG. 11 . FIG. 11( a) shows the two-dimensional plot for cytological specimen A. FIG. 11( b) shows the two-dimensional plot for cytological specimen B. FIG. 11( c) shows the two-dimensional plot for cytological specimen C. In FIGS. 11( a), (b), and (c), the horizontal axis of the graph indicates the certainty of debris. The vertical axis of the graph indicates the certainty of leukocytes. The color tone indicates concentration. The square in the upper left corner of the graph indicates the gate set to identify leukocytes.

[0043] In cytology specimen A shown in Figure 11(a), the percentage of white blood cells was 5.1%. In cytology specimen B shown in Figure 11(b), the percentage of white blood cells was 14.0%. In cytology specimen C shown in Figure 11(c), the percentage of white blood cells was 69.3%. The control unit 50 can then cause the output unit to display cell images of cells identified as white blood cells, i.e., cells included within the gates in Figures 11(a), (b), and (c).

[0044] An example of multi-stage gating will be described below with reference to Figs. 12 and 13. First, as shown in Fig. 12, the control unit 50 outputs a first two-dimensional plot for a cytology specimen, with the horizontal axis of the graph representing the certainty of debris and the vertical axis of the graph representing the certainty of white blood cells. In Fig. 12, the square in the upper left of the graph represents a gate for identifying white blood cells. Furthermore, the square in the lower right of the graph represents a gate for identifying debris. The control unit 50 can cause the output unit to display cell images of cells identified as white blood cells and debris.

[0045] The control unit 50 performs the next stage of gating on cells within the gate indicated by the square at the bottom left of the graph in FIG. 12 . Specifically, as shown in FIG. 13 , the control unit 50 outputs a second two-dimensional plot for cells within the gate at the bottom left of the first two-dimensional plot, with the horizontal axis representing the certainty of epithelial cells and the vertical axis representing the certainty of glandular cells. In FIG. 13 , vertical and horizontal lines represent quadrant gates that classify the cell population into four. The lower right section of the graph, separated by the two lines of the quadrant gate, represents a gate for identifying epithelial cells. The upper left section of the graph represents a gate for identifying glandular cells. The lower left section of the graph represents a gate corresponding to other cells that do not fit into either of these categories.

[0046] The control unit 50 can display on the output unit cell images of the cells identified by each gate in Figures 12 and 13. In this way, by performing multi-stage gating, the cytological specimen can be analyzed in detail.

[0047] In addition, in step S5, if the control unit 50 determines, for each cell, multiple Z layers suitable for observing and analyzing the cell from the images that constitute the cell image group, the control unit 50 repeats steps S1 to S7 for each Z layer.

[0048] REFERENCE SIGNS LIST 10 Stage 20 Microscope 30 Imaging unit 40 Focal plane moving mechanism 42 Stage moving mechanism 50 Control unit 55 Image buffer 70 Storage unit 80 Transmission unit S Specimen

Claims

1. A data analysis device for microscopic stacked images of a specimen containing a large number of cells, comprising: a microscope that acquires Z-stack images of the specimen; an imaging unit that captures the Z-stack images acquired by the microscope; a memory unit that stores the Z-stack images acquired by the imaging unit; and a control unit that detects cells from the Z-stack images stored in the memory unit; acquires a cell image group corresponding to each detected cell; cuts out a cell image for each cell from images that constitute the cell image group; acquires a confidence level for each of the cell images of each cell for multiple class classifications; and visualizes the confidence level acquired for each class classification.

2. The microscopic stacked image data analysis device according to claim 1, wherein the control unit detects the cells by detecting cell nuclei from the Z-stack images stored in the memory unit.

3. A data analysis device for microscopic stacked images as described in claim 1 or 2, wherein the control unit acquires the cell image group from the multiple Z layers across which the cell is detected when the cell is detected across multiple Z layers that make up the Z stack image.

4. A data analysis method for microscopic stacked images of a specimen containing a large number of cells, comprising: acquiring Z-stack images of the specimen using a microscope; capturing the Z-stack images acquired by the microscope using an imaging unit; storing the Z-stack images captured by the imaging unit in a memory unit; a control unit detecting cells from the Z-stack images stored in the memory unit; the control unit acquiring a cell image group corresponding to each detected cell; the control unit extracting a cell image from images constituting the cell image group for each cell; the control unit acquiring confidence levels for each of a plurality of class classifications for the cell image of each cell; and the control unit visualizing the confidence levels acquired for each class classification.

5. The data analysis method for microscopic stacked images according to claim 4, wherein the control unit detects the cells by detecting cell nuclei from the Z-stack images stored in the memory unit.

6. A data analysis method for microscopic stacked images as described in claim 4 or 5, wherein the control unit, when the cell is detected across multiple Z layers that make up the Z stack image, acquires the cell image group from the multiple Z layers across which the cell is detected.

Images