Cell analysis system based on virtual fluorescence generation technology and method therefor

Abstract

The present invention relates to a cell analysis system based on virtual fluorescence generation technology, and a method therefor. According to the present invention, the system comprises: an input unit that receives a quantitative phase image of a sample including at least one cell through digital holography microscopy; and a cell identification unit that identifies at least one living cell by applying, to the received quantitative phase image, a cell segmentation mask generation model and a nuclei segmentation mask generation model constructed previously, and that masks a position of the nucleus of the living cell to derive an identification result. The system may further comprise a post-processing unit that derives a final result by applying a preset post-processing method to the derived identification result.

Description

Cell analysis system and method based on virtual fluorescence generation technology

[0001] The present invention relates to a cell analysis system and method based on virtual fluorescence generation technology. More specifically, it relates to a cell analysis system and method based on virtual fluorescence generation technology that analyzes cells by generating virtual fluorescence in living cells in a phase image using a cell segmentation mask generation model and a nuclear segmentation mask generation model learned based on information obtained from a digital holographic microscope and a fluorescence microscope.

[0002] Cell analysis plays a crucial role in understanding the mechanisms that cause disease when individual cells respond to the environment or when mutations occur. Performing single-cell viability and nuclear analysis allows for the accurate assessment of cell health and function, which is necessary for understanding various stimuli or cellular responses.

[0003] Sophisticated methods are required to analyze such single cells. In the case of typical fluorescence imaging methods, long-term studies are difficult due to limitations such as extrinsic cell boundaries, long culture periods for staining, and stain toxicity.

[0004] In addition, when using digital holography microscopy (DHM), quantitative phase images can be provided without fluorescent labels that may alter the biological characteristics of cells, but it is difficult to accurately identify cell organelles or cell states (e.g., living or dead).

[0005] Therefore, there is a need for technology that automatically segments individual cells from phase images acquired through a digital holographic microscope to identify organelles and cell states.

[0006] The technology forming the background of the present invention is described in Korean Published Patent No. 10-2024-0134622 (published September 10, 2024).

[0007] As such, according to the present invention, the purpose is to provide a cell analysis system and method based on virtual fluorescence generation technology that analyzes cells by generating virtual fluorescence in living cells in a phase image using a cell segmentation mask generation model and a nuclear segmentation mask generation model learned based on information obtained from a digital holographic microscope and a fluorescence microscope.

[0008] According to an embodiment of the present invention for achieving such technical challenges, a cell analysis system based on virtual fluorescence generation technology may further include: an input unit that receives a quantitative phase image of a sample containing at least one cell through a digital holographic microscope; a cell identification unit that identifies at least one living cell by applying a pre-established cell segmentation mask generation model and a nuclear segmentation mask generation model to the input quantitative phase image and masks the location of the nucleus of the living cell to derive an identification result, and a post-processing unit that derives a final result by applying a pre-established post-processing method to the derived identification result.

[0009] The cell segmentation mask generation model described above can distinguish the boundaries of all cells when receiving the quantitative phase image, and can mask living cells with a first fluorescence and dead cells with a second fluorescence.

[0010] The above nuclear splitting mask generation model can receive a quantitative phase image of a living cell masked with a first fluorescence and represent the location of the nucleus of the living cell masked with the first fluorescence by masking it with a third fluorescence.

[0011] The cell segmentation mask generation model described above can receive the quantitative phase image as input, distinguish the boundaries of all cells, mask the living cells with a first fluorescence, and mask the dead cells with a second fluorescence to generate a segmentation mask and a distance map according to cell viability.

[0012] The above nuclear segmentation mask generation model can generate a segmentation mask of a living cell nucleus by receiving a segmentation mask based on the cell viability, masking the location of the cell nucleus masked by the first fluorescence with the third fluorescence, and comparing the mask marked with the third fluorescence with the correct image to remove the location of the nucleus that is not in an overlapping position.

[0013] The above post-processing unit can separate the background, the region of the living cell, and the region of the nucleus of the living cell, and apply a pre-set post-processing method to correct the boundaries of the cells based on the distance map of the identification result, thereby separating cells that are in contact or overlapping.

[0014] In a cell analysis method based on virtual fluorescence generation technology according to another embodiment of the present invention, the method may further include: a step in which an input unit receives a quantitative phase image of a sample containing at least one cell through a digital holographic microscope; a step in which a cell identification unit applies a pre-established cell segmentation mask generation model and a nuclear segmentation mask generation model to the input quantitative phase image to identify at least one living cell and masks the location of the nucleus of the living cell to derive an identification result, and a step in which a post-processing unit applies a pre-established post-processing method to the derived identification result to derive a final result.

[0015] As such, according to the present invention, observation is possible at the single-cell level, and the health and behavior of the cell can be determined through quantitative analysis such as dry mass.

[0016] In addition, it is possible to monitor changes in cell state over time, dynamically observe cell responses to various stimuli, and utilize it for biological research by applying it to various types of cancer cells.

[0017] In addition, virtual fluorescence images can be generated from phase images of a digital holographic microscope, which prevents problems caused by fluorescent markers and enables various analyses of cells.

[0018] FIG. 1 is a configuration diagram of a cell analysis system based on virtual fluorescence generation technology according to one embodiment of the present invention.

[0019] FIG. 2 is a flowchart of a cell analysis method based on virtual fluorescence generation technology according to another embodiment of the present invention.

[0020] FIG. 3 is a drawing illustrating an example of a dual imaging system according to another embodiment of the present invention.

[0021] FIG. 4 is a diagram illustrating images recorded through a dual imaging system according to another embodiment of the present invention and images for learning a cell segmentation mask generation model and a nuclear segmentation mask generation model.

[0022] FIG. 5 is a diagram illustrating the process of a cell splitting mask generation model and a nuclear splitting mask generation model according to another embodiment of the present invention.

[0023] FIG. 6 is a diagram illustrating an example of applying a post-processing method to a derived identification result according to another embodiment of the present invention.

[0024] FIG. 7 is a drawing illustrating the correct answer image and the identification result of the cell identification unit according to another embodiment of the present invention.

[0025] FIG. 8 is a diagram illustrating the results of applying a cell analysis method based on virtual fluorescence generation technology according to another embodiment of the present invention to liver cancer cells, bladder cancer cells, and lung cancer cells.

[0026] Preferred embodiments according to the present invention will be described in detail below with reference to the attached drawings. In this process, the thickness of lines or the size of components shown in the drawings may be exaggerated for clarity and convenience of explanation.

[0027] Furthermore, the terms described below are defined in consideration of their functions within the present invention, and these may vary depending on the intent or practice of the user or operator. Therefore, the definitions of these terms should be based on the content throughout this specification.

[0028] In the embodiments described below, a cell analysis system (100) based on virtual fluorescence generation technology is specifically described as being performed by a computing device comprising one or more processors and one or more memories capable of performing the following processes.

[0029] FIG. 1 is a configuration diagram of a cell analysis system based on virtual fluorescence generation technology according to one embodiment of the present invention.

[0030] As illustrated in FIG. 1, a virtual fluorescence generation technology-based cell analysis system (100) may include an input unit (110), a cell identification unit (120), and a post-processing unit (130).

[0031] First, the input unit (110) can receive a quantitative phase image of a sample containing at least one cell through a digital holography microscope (DHM).

[0032] Next, the cell identification unit (120) can identify at least one living cell by applying a pre-established cell segmentation mask generation model and a nuclear segmentation mask generation model to the input quantitative phase image, and can derive an identification result by masking the location of the nucleus of the living cell.

[0033] Here, the cell segmentation mask generation model is a model trained using quantitative phase images and fluorescence images of multiple cells captured through a dual imaging system as training data, and when a quantitative phase image is input, it distinguishes the boundaries of all cells and, based on the fluorescence image, masks living cells with a first fluorescence (e.g., blue) and dead cells with a second fluorescence (e.g., green). Additionally, the nuclear segmentation mask generation model may be a model trained using a quantitative phase image in which the nuclei of living cells are masked with a third fluorescence and a quantitative phase image in which living cells are masked with a first fluorescence as training data, and is trained to mark the location of the nuclei of cells masked with the first fluorescence (e.g., blue) by masking them with a third fluorescence (e.g., white).

[0034] At this time, the fluorescent image may include a first mask divided into three classes that distinguish living cells, dead cells, and the background, and a second mask that indicates the location of the nucleus based on the nucleus of the stained cell.

[0035] Next, the post-processing unit (130) can derive a final result by applying a pre-set post-processing method (e.g., a watershed algorithm) to the derived identification result. At this time, since the description of the watershed algorithm is the same as previously disclosed, a redundant description is omitted.

[0036] Specifically, the post-processing unit (130) can derive a final result by applying the derived identification result and cell viability to a pre-set post-processing method (e.g., a watershed algorithm).

[0037] Below, a cell analysis method based on virtual fluorescence generation technology will be explained in more detail using FIGS. 2 to 8.

[0038] FIG. 2 is a flowchart of a cell analysis method based on virtual fluorescence generation technology according to another embodiment of the present invention.

[0039] As illustrated in FIG. 2, the input unit (110) can receive a quantitative phase image of a sample containing at least one cell through a digital holographic microscope (S210).

[0040] Specifically, the input unit (110) receives a hologram of a sample through a digital holographic microscope and can acquire a quantitative phase image.

[0041] Here, the digital holographic microscope irradiates the sample with an object beam to receive amplitude and phase information of the sample, combines the reference beam and the object beam to generate a hologram, and then reconstructs the complex plane of the object beam in the form of intensity changes.

[0042] At this time, the input unit (110) can extract amplitude and phase information from the complex plane of the reconstructed object beam and reconstruct the phase using the phase value of each pixel through the [Equation 1] below.

[0043]

[0044] Here, λ is the wavelength of the coherent source, and h(x,y) is the cell thickness at position (x,y), and The average intracellular refractive index at position (x,y) is , and is the refractive index of the surrounding culture medium.

[0045] In addition, the input unit (110) can obtain a quantitative phase image by calculating the optical path difference of light passing (OPD) through the following [Equation 2] on the quantitative phase image obtained through hologram reconstruction.

[0046]

[0047] Here, λ is the wavelength of the coherent source, h(x,y) is the cell thickness at position (x,y), and is the average intracellular refractive index at position (x,y), and is the refractive index of the surrounding culture medium.

[0048] Next, the cell identification unit (120) can identify at least one living cell by applying a pre-established cell segmentation mask generation model and a nuclear segmentation mask generation model to the input quantitative phase image, and derive an identification result by masking the location of the nucleus of the living cell (S220).

[0049] Here, the cell segmentation mask generation model is a model trained using quantitative phase images and fluorescence images of multiple cells captured through a dual imaging system as training data, and when a quantitative phase image is input, it distinguishes the boundaries of all cells and, based on the fluorescence image, masks living cells with a first fluorescence and dead cells with a second fluorescence. Additionally, the nuclear segmentation mask generation model may be a model trained using a quantitative phase image in which the nucleus of a living cell is masked with a third fluorescence and a quantitative phase image in which a living cell is masked with a first fluorescence as training data, and is trained to mark the location of the nucleus of the cell masked with the first fluorescence by masking it with the third fluorescence.

[0050] At this time, the fluorescent image may include a first mask divided into three classes that distinguish living cells, dead cells, and the background, and a second mask that indicates the location of the nucleus based on the nucleus of the stained cell.

[0051] FIG. 3 is a drawing illustrating an example of a dual imaging system according to another embodiment of the present invention.

[0052] As illustrated in FIG. 3, the dual imaging system can acquire quantitative phase images and fluorescence images of a sample (S) containing liver cancer cells using a digital holographic microscope (red) and a fluorescence microscope (blue). In FIG. 3, BS (Beam splitter) is a beam splitter, LS (Light source) is a light source, S (Sample) is a sample, MO (Microscope objective) is an objective lens, DF (Dichromic mirror) is a dichromic mirror, ExF (Excitation filter) is an absorption filter that filters only light of wavelengths that the sample can absorb, EmF (Emission filter) is an emission filter that filters only light of wavelengths that the sample can emit, and M (Mirror) is a mirror.

[0053] In a dual imaging system, the digital holographic microscope acquires a quantitative phase image through the above [Equation 1] and [Equation 2], and the fluorescence microscope separates the fluorescence light emitted by irradiating with light of a specific wavelength using an excitation filter through a dichroic mirror and an emission filter to produce a fluorescence image in which only the fluorescence-labeled components of the cells in the sample are visible.

[0054] FIG. 4 is a diagram illustrating an image recorded through a dual imaging system according to another embodiment of the present invention and an image for learning a cell segmentation mask generation model and a nuclear segmentation mask generation model.

[0055] Figure 4(a) is a quantitative phase image obtained through a digital holographic microscope, Figure 4(b) is an image showing living cells stained blue, Figure 4(c) is an image showing dead cells stained green, Figure 4(d) is a segmentation mask for living cells in Figure 4(b), Figure 4(e) is a segmentation mask for dead cells in Figure 4(c), Figure 4(f) is a mask combining the segmentation mask for living cells and the segmentation mask for dead cells, Figure 4(g) is a nuclear segmentation mask for living cells generated based on Figure 4(b), and Figure 4(h) is a distance map of the cell segmentation mask.

[0056] As shown in Figure 4, the fluorescence image and quantitative phase image, which are the training data of the cell segmentation mask generation model, are matched in pairs.

[0057] According to one embodiment of the present invention, 25 pairs of 675 pairs of quantitative phase images and fluorescence images were used as test data, and the remaining 650 pairs were used as training data by applying data augmentation technology.

[0058] FIG. 5 is a diagram illustrating the process of a cell splitting mask generation model and a nuclear splitting mask generation model according to another embodiment of the present invention.

[0059] As illustrated in Fig. 5, the cell viability mask generation model masks living cells with a first fluorescence and dead cells with a second fluorescence to generate a segmentation mask and a distance map based on cell viability, and the nuclei mask generation model receives the segmentation masks based on cell viability (Live cell mask, Dead cell mask) as input, masks the location of the nucleus of the cell masked by the first fluorescence with a third fluorescence to denote it, and generates a segmentation mask of the nucleus of a living cell by comparing the generated nuclei mask indicated by the third fluorescence with the ground truth image to remove the locations of nuclei at non-overlapping positions. At this time, the distance map indicates proximity to the cell boundary.

[0060] In addition, the cell segmentation mask generation model was constructed to identify regions of living cells and dead cells, respectively, using a first feature pyramid network (FPN) structure, and the nuclear segmentation mask generation model was constructed to locate cell nuclei by segmenting the identified regions of living cells using a second feature pyramid network structure. At this time, the first and second feature pyramid networks were constructed using an attention mechanism and ASPP (Atrous spatial pyramid pooling) to extract features and improve segmentation accuracy.

[0061] According to one embodiment of the present invention, the cell segmentation mask generation model and the nuclear segmentation mask generation model have a structure that extracts features at multiple scales through a series of convolutional layers (top-down path), reduces the number of channels through convolutional blocks, and merges with lateral connections (bottom-up path) to facilitate the smooth fusion of features of different resolutions. In this case, the top-down path is constructed using bottleneck blocks with varying numbers of layers and channels, the convolutional blocks apply spatial and attention mechanisms at each feature level centered on features contributing to cell segmentation, and the bottom-up path integrates instance normalization and Rectified Linear Units (ReLU) to expand resolution while preserving feature integrity.

[0062] In addition, the cell segmentation mask generation model and the nuclear segmentation mask generation model can identify and segment living cells and cell nuclei in quantitative phase images through a combination of adversarial loss functions and conventional loss functions.

[0063] Here, the cell splitting mask generation model and the nuclear splitting mask generation model can use the adversarial loss calculated using [Equation 3] below.

[0064]

[0065] The first discriminator ( It is a hostile loss of ), and is the first characteristic pyramid network, and The second discriminator ( It is a hostile loss of ), and It is a pyramid network with the second characteristic, and is the first feature of the pyramid network's adversarial loss, and is the second feature of a pyramid network, which is adversarial loss, and is in the distribution of y variables It is the average of, and y is the correct cell splitting mask, and is the discrimination result of the first discriminator for the correct answer cell splitting mask, and is in the distribution of x and y variables It is the average of, is the discrimination result of the first discriminator for the cell splitting mask generated by passing the input quantitative phase image through the first feature pyramid network, and z is the actual correct nuclear splitting mask for the nucleus, and is in the distribution of z variables It is the average of, is the discrimination result of the second discriminator for the actual correct answer core splitting mask, and is in the distribution of x, y, and z variables It is the average of, is the discrimination result of the second discriminator for the cell division mask masked by the first fluorescence among the cell division masks generated by passing the input quantitative phase image through the first feature pyramid network, and the nuclear division mask generated by passing the quantitative phase image through the second feature pyramid network, and x is the input quantitative phase image, and is in the distribution of x variables or It is the average of.

[0066] In addition, the cell splitting mask generation model and the nuclear splitting mask generation model can handle class imbalance using the dice coefficient loss as shown in [Equation 4] below.

[0067]

[0068] Here, P is the predicted partition map and T is the actual partition map.

[0069] At this time, insertion loss ( ) is to ensure that the nucleus region of a living cell is included within the region of the living cell, and exclusion loss( ) is to ensure that the region of the nucleus of a living cell is not included within the region of a dead cell.

[0070]

[0071] Here, is the number of pixels in the region of the nucleus of a living cell that is not included in the region of the living cell, and is the number of pixels in the region of the nucleus of a living cell included in the region of a living cell, and is the number of pixels of the region of the nucleus of a living cell included in the region of a dead cell, and is the number of pixels in the area of ​​dead cells.

[0072] In addition, the cell segmentation mask generation model can predict the distance between living cells and dead cells more accurately by using distance loss through [Equation 6] below.

[0073]

[0074] Here, d is the actual distance map, and is a predicted distance map.

[0075] Additionally, the total loss of the cell identification unit (120) is the sum of each loss (adversarial loss of the first and second feature pyramid networks, dice count loss of the first and second feature pyramid networks, insertion loss, exclusion loss, and distance loss).

[0076] At this time, the first and second discriminators can better distinguish between real data and fake data by performing updates using the adversarial loss of the first discriminator and the adversarial loss of the second discriminator, and the first and second feature pyramid networks can maintain the accuracy and realism of segmenting the regions of living cells and nuclei in the derived identification results above a certain level by minimizing the total sum of the combined losses.

[0077] Next, the post-processing unit (130) can derive a final result by applying a pre-set post-processing method (e.g., a watershed algorithm) to the derived identification result (S230). At this time, since the watershed algorithm is the same as that disclosed in the prior art, a redundant explanation is omitted.

[0078] Specifically, the post-processing unit (130) can derive a final result by applying the derived identification result and cell viability to a pre-set post-processing method (e.g., a watershed algorithm).

[0079] FIG. 6 is a diagram illustrating an example of applying a post-processing method to a derived identification result according to another embodiment of the present invention.

[0080] As illustrated in FIG. 6, the post-processing unit (130) can apply a calculated threshold value to the distance map of the derived identification result to precisely correct the cell boundary and apply it to the watershed algorithm to derive the final result.

[0081] Through this, the post-processing unit (130) can correct the boundaries of the cell and the boundaries of the nucleus region, and clearly target and correct the living cells and dead cells.

[0082] At this time, the post-processing unit (130) can apply a kernel of a preset size (e.g., 3×3) to the cell segmentation mask generated in S220, calculate the boundary between the background and the living cell through [Equation 7] below, and obtain a mask for the background.

[0083]

[0084] Here, δ is the threshold value (e.g., 0.4× is the maximum value of, and is a distance map on a grayscale image, and is a marker to be applied to the watershed algorithm, and is a mask to which a threshold value calculated from the distance map is applied.

[0085] At this time, silver If is greater than the threshold, it is 1, and It is a mask that is 0 when the threshold is lower than or equal to.

[0086] Additionally, the post-processing unit (130) can separate the background, the region of the living cell, and the region of the nucleus of the living cell, generate markers for applying a watershed algorithm, and apply the watershed algorithm to the generated markers to correct the boundaries of the cells based on the distance map of the identification results, thereby separating cells that are in contact or overlapping.

[0087] In addition, the post-processing unit (130) can correct the cell boundaries to be more clearly visible through a watershed algorithm.

[0088] FIG. 7 is a drawing illustrating the correct answer image and the identification result of the cell identification unit according to another embodiment of the present invention.

[0089] FIG. 7(a) is an input quantitative phase image, FIG. 7(b) is a segmentation mask for living cells, FIG. 7(c) is a segmentation mask for dead cells, FIG. 7(d) is a distance map, FIG. 7(e) is the final result with the basin algorithm applied, FIG. 7(f) is a segmentation mask for the nucleus of a living cell, and FIG. 7(g) is an image combining FIG. 7(b) to FIG. 7(f).

[0090] As shown in Figure 7, the cell segmentation mask generation model segments the regions of living cells and dead cells almost similarly to the ground truth image, but when the basin algorithm is applied, it can be seen that the boundaries between living cells and dead cells are segmented more clearly.

[0091] In addition, [Table 1] and [Table 2] below are evaluation indicators of identification results for living cells, dead cells, and nuclei generated using various models.

[0092]

[0093]

[0094] Referring to [Table 1] and [Table 2] above, it can be seen that the cell division mask generation model (Proposed model) according to another embodiment of the present invention in [Table 1] clearly distinguishes living cells and dead cells compared to other models, and the nuclear division mask generation model (Proposed model) in [Table 2] correctly distinguishes the nuclei of living cells.

[0095] FIG. 8 is a diagram illustrating the results of applying a cell analysis method based on virtual fluorescence generation technology according to another embodiment of the present invention to liver cancer cells, bladder cancer cells, and lung cancer cells.

[0096] FIG. 8(a) shows quantitative phase images (left image) and identification results (right image) of a cell identification unit (120) for liver cancer, bladder cancer, and lung cancer cells, respectively, FIG. 8(b) is a quantitative result calculated by a cell splitting mask generation model using a splitting mask for living cells, and FIG. 8(c) is a quantitative result calculated by a nuclear splitting mask generation model using a splitting mask for the nucleus of a living cell.

[0097] As shown in Fig. 8(b), the present invention can be seen that the difference in the optical path of light is similarly low for liver cancer cells (green) and bladder cancer cells (blue), and the difference in the optical path of light is high for lung cancer cells (yellow), and the projection area of ​​lung cancer cells is low so that the dry mass is distinguished from other cell types, but it is difficult to distinguish for liver cancer cells and bladder cancer cells.

[0098] As shown in Fig. 8(c), for the quantitative results regarding the nuclei of living cells, the difference in the optical path of light is lowest for liver cancer cells (green) and highest for lung cancer cells (yellow), and the projected area of ​​the nuclei is similar for liver cancer cells, bladder cancer cells (blue) and lung cancer cells, and liver cancer cells, bladder cancer cells and lung cancer cells can be distinguished by dry mass.

[0099] According to the embodiments of the present invention described above, observation is possible at the single-cell level, and the health and behavior of the cell can be determined through quantitative analysis such as dry mass.

[0100] In addition, it is possible to monitor changes in cell state over time, dynamically observe cell responses to various stimuli, and utilize it for biological research by applying it to various types of cancer cells.

[0101] In addition, virtual fluorescence images can be generated from phase images of a digital holographic microscope, which prevents problems caused by fluorescent markers and allows for various analyses of cells.

[0102] Although the present invention has been described with reference to the embodiments illustrated in the drawings, this is merely illustrative, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom. Accordingly, the true technical scope of protection of the present invention should be determined by the technical spirit of the following claims.

[0103] [Explanation of the symbol]

[0104] 100: Cell analysis system based on virtual fluorescence generation technology

[0105] 110: Input section

[0106] 120: Cell identification unit

[0107] 130: Post-processing unit

Claims

1. An input unit that receives a quantitative phase image of a sample containing at least one cell through a digital holographic microscope; and A cell analysis system based on virtual fluorescence generation technology comprising a cell identification unit that identifies at least one living cell by applying a pre-established cell segmentation mask generation model and a nuclear segmentation mask generation model to the input quantitative phase image, and derives an identification result by masking the location of the nucleus of the living cell.

2. In Paragraph 1, A cell analysis system based on virtual fluorescence generation technology, further comprising a post-processing unit that derives a final result by applying a pre-set post-processing method to the derived identification result.

3. In Paragraph 1, The above cell division mask generation model is, When the above quantitative phase image is received as input, the boundaries of all cells are distinguished, and A cell analysis system based on virtual fluorescence generation technology that masks living cells with a first fluorescence and dead cells with a second fluorescence.

4. In Paragraph 3, The above nuclear splitting mask generation model is, A cell analysis system based on virtual fluorescence generation technology that receives a quantitative phase image of a living cell masked with a first fluorescence and displays the location of the nucleus of the living cell masked with the first fluorescence by masking it with a third fluorescence.

5. In Paragraph 3, The above cell division mask generation model is, The above quantitative phase image is received as input to distinguish the boundaries of all cells, and A cell analysis system based on virtual fluorescence generation technology that generates a segmentation mask and a distance map according to cell viability by masking the living cells with a first fluorescence and the dead cells with a second fluorescence.

6. In Paragraph 5, The above nuclear splitting mask generation model is, A segmentation mask based on the cell viability is received, and the location of the cell nucleus masked by the first fluorescence is masked by the third fluorescence and indicated. A cell analysis system based on virtual fluorescence generation technology that generates a segmentation mask of the nuclei of a living cell by comparing the mask labeled with the third fluorescence above with the correct image and removing the locations of nuclei at non-overlapping positions.

7. In Paragraph 2, The above post-processing unit is, A cell analysis system based on virtual fluorescence generation technology that separates the background, the region of living cells, and the region of the nucleus of living cells, and separates cells in contact or overlapping states by correcting the cell boundaries based on the distance map of the identification results by applying a preset post-processing method.

8. A step in which an input unit receives a quantitative phase image of a sample containing at least one cell through a digital holographic microscope; and A cell analysis method based on virtual fluorescence generation technology comprising the step of identifying at least one living cell by applying a pre-established cell segmentation mask generation model and a nuclear segmentation mask generation model to the quantitative phase image input by a cell identification unit, and deriving an identification result by masking the location of the nucleus of the living cell.

9. In Paragraph 8, A cell analysis method based on virtual fluorescence generation technology that further includes a step in which a post-processing unit applies a pre-set post-processing method to the derived identification result to derive a final result.

10. In Paragraph 8, The above cell division mask generation model is, When the above quantitative phase image is received as input, the boundaries of all cells are distinguished, and A cell analysis method based on virtual fluorescence generation technology that masks living cells with a first fluorescence and dead cells with a second fluorescence.

11. In Paragraph 10, The above nuclear splitting mask generation model is, A cell analysis method based on virtual fluorescence generation technology that receives a quantitative phase image of a living cell masked with a first fluorescence and displays the location of the nucleus of the living cell masked with the first fluorescence by masking it with a third fluorescence.

12. In Paragraph 10, The above cell division mask generation model is, The above quantitative phase image is received as input to distinguish the boundaries of all cells, and A cell analysis method based on virtual fluorescence generation technology that generates a segmented mask and a distance map according to cell viability by masking the living cells with a first fluorescence and the dead cells with a second fluorescence.

13. In Paragraph 12, The above nuclear splitting mask generation model is, A segmentation mask based on the cell viability is received, and the location of the cell nucleus masked by the first fluorescence is masked by the third fluorescence and indicated. A cell analysis method based on virtual fluorescence generation technology that generates a segmentation mask of the nuclei of a living cell by comparing the mask labeled with the third fluorescence above with the correct image and removing the locations of nuclei at non-overlapping positions.

14. In Paragraph 9, The step of deriving the above final result is, A cell analysis method based on virtual fluorescence generation technology that separates the background, the region of living cells, and the region of the nucleus of living cells, and corrects the boundaries of cells based on the distance map of the identification results by applying a pre-set post-processing method to separate cells in contact or overlapping states.

Images