Intelligent cerebral apoplexy cell state recognition method based on Cellpose and fluorescence image analysis and application thereof in drug effect screening

Abstract

The invention discloses an intelligent cerebral apoplexy cell state recognition method based on Cellpose and fluorescence image analysis and application of the intelligent cerebral apoplexy cell state recognition method in drug effect screening. The intelligent cerebral apoplexy cell state recognition method comprises the following steps: S1, establishing a cerebral apoplexy ischemia reperfusion cell model; s2, adding a nitrite fluorescent probe into the cells subjected to ischemia reperfusion treatment to obtain a fluorescent microscopic image of the cells; s3, importing the fluorescence microscopic image into a Cellpose model for cell segmentation, and obtaining an ROI region of each cell; s4, extracting fluorescence intensity characteristics and cell morphological characteristics in the ROI region of each cell; and S5, inputting the extracted multi-dimensional features into a classification model for cell state recognition, so as to distinguish a single cell into three states of health, mild injury or serious injury. According to the method, a complete link of image acquisition-Cellpose analysis-feature extraction-state discrimination-drug effect evaluation is constructed, and the practical application value of a deep learning image algorithm in disease model research and drug screening scenes is expanded.

Description

TECHNICAL FIELD

[0001] The application belongs to the cross field of artificial intelligence and biological image processing, and particularly relates to an intelligent cerebral apoplexy cell state recognition method based on Cellpose and fluorescence image analysis and application thereof in drug efficacy screening. BACKGROUND

[0002] Cerebral apoplexy (also known as stroke) is a kind of acute cerebral vascular disease, and ischemic cerebral apoplexy is the most common one. Its pathogenesis includes ischemia, hypoxia of brain tissue and secondary injury caused by reperfusion process.

[0003] In recent years, fluorescence probes have been developed for visualizing detection of nitrite ions in cells, providing a new technical approach for cell damage evaluation. However, traditional image analysis methods rely on manual identification, segmentation and fluorescence intensity reading, which are low in efficiency and strong in subjectivity, and are not suitable for high-throughput experiments.

[0004] Deep learning algorithms, especially general cell segmentation tools such as Cellpose, have shown extremely high segmentation accuracy and adaptability in the field of biological image processing, but there is still a lack of research and application of applying them to cerebral apoplexy cell models, combining cell morphology and nitrite fluorescence signals, to construct an integrated system of cell state recognition and drug screening. SUMMARY

[0005] In view of the problems in the current cerebral apoplexy cell model research that manual analysis of nitrite ion fluorescence images is complicated, segmentation is inaccurate, state recognition is not intelligent, and drug screening process is inefficient, an intelligent cerebral apoplexy cell state recognition method based on Cellpose and fluorescence image analysis and application thereof in drug efficacy screening are proposed. The application realizes automatic, quantitative and batch recognition of cerebral apoplexy cell state for the first time by using AI method, constructs a complete link of "image acquisition-Cellpose analysis-feature extraction-state discrimination-drug efficacy evaluation", and expands the practical application value of deep learning image algorithms in disease model research and drug screening scenarios.

[0006] To achieve the above purpose, the following technical solutions are adopted in the present application: In a first aspect, the present application provides an intelligent cerebral apoplexy cell state recognition method based on Cellpose and fluorescence image analysis, comprising the following steps: S1, establishing a cerebral apoplexy ischemia-reperfusion cell model; S2, adding a nitrite fluorescence probe to the cells treated by ischemia-reperfusion to obtain fluorescence microscopic images of the cells; S3, importing the fluorescence microscopic images into a Cellpose model for cell segmentation to obtain the ROI region of each cell; S4, extracting a fluorescence intensity feature and a cell morphology feature in the ROI region of each cell, the fluorescence intensity feature including average fluorescence intensity and total fluorescence intensity, and the cell morphology feature including cell area; S5, inputting the extracted multi-dimensional features into a classification model for cell state recognition, so as to distinguish a single cell into three states of healthy, mild injury and severe injury.

[0007] In the above technical solution, in S1, the specific steps of establishing a cerebral apoplexy ischemia-reperfusion cell model are: S11, cell culture and processing: selecting SH-SY5Y human neuroblastoma cell lines, culturing the cells in complete culture medium containing 10% fetal bovine serum, and culturing under the condition of 37°C, 5% CO2 and 95% O2 environment; S12, ischemia-reperfusion processing: processing the cells under the condition of oxygen-glucose deprivation to simulate ischemia; and then restoring to the conventional culture condition to simulate the reperfusion process.

[0008] In the above technical solution, in S3, the specific steps of importing the fluorescence microscopic image into the Cellpose model for cell segmentation to obtain the ROI region of each cell are: the Cellpose model automatically identifies the cell contour, generates a segmentation mask image for each cell, accurately identifies the boundary of each cell in each image, and outputs the ROI mask image of each cell, wherein the nucleus region is marked as 1 and the background region is marked as 0.

[0009] In the above technical solution, the specific method of the Cellpose model for cell segmentation of the fluorescence microscopic image is: (1) image preprocessing: converting the collected fluorescence microscopic image into an RGB format, and extracting the red channel as the main signal source of the nucleus fluorescence; (2) threshold segmentation: applying the Otsu method to the extracted channel image for adaptive threshold segmentation to separate the nucleus region from the background to obtain a binary image, wherein the nucleus region is marked as 1 and the background region is marked as 0; (3) morphological image denoising and repair: applying morphological operations to the binary image, including open operation to remove small noise and close operation to fill the holes in the nucleus region to ensure the accuracy of the segmentation result; (4) region labeling and attribute extraction: using connected domain analysis method to process the binary image, identifying all independent nucleus regions, and extracting basic attributes of each region, the basic attributes including area and centroid position.

[0010] In the above technical solution, in S4, the specific method of extracting the fluorescence intensity feature and the cell morphology feature is: in the ROI region of each cell, the sum of all red pixel values in the region is calculated using the summation function of NumPy, representing the total red light intensity of the cell, and the calculation formula is: total red light intensity = Σ (red channel pixel value), and the average red light intensity in the cell region is calculated, and the calculation formula is: average red light intensity = total red light intensity / number of pixels in the cell region. The number of pixels in the statistical mask is counted by counting the number of pixels with a pixel value of 1 in the ROI mask image generated by Cellpose.

[0011] In the above technical solution, in S5, the standard for cell state recognition is: according to the extracted red light intensity and area features, combined with the set threshold standard, the state of each cell is classified: Healthy cells: average red light intensity is lower than threshold A, and area is normal; Mildly damaged cells: average red light intensity is between threshold A and threshold B, and area is slightly reduced; Severely damaged cells: average red light intensity is higher than threshold B, and area is significantly reduced.

[0012] In the above technical solution, the following steps are further included: S6, the state of each experimental group of cells is counted, and the number of different state cells in each group and the average fluorescence intensity are output.

[0013] In a second aspect, the application provides the application of the above cell state recognition method in pharmacodynamic screening.

[0014] In the above technical solution, in the process of pharmacodynamic screening, the ischemia-reperfusion cell model with different candidate drugs is processed and imaged, the imaging image is segmented and features are extracted, the average fluorescence intensity and the proportion of severely damaged cells of the cells in each drug treatment group are compared, and the protective effect of the drug is evaluated; the candidate drugs are sorted according to the fluorescence attenuation degree and the proportion of severely damaged cells, and the pharmacodynamic sorting result is output, and the automatic state recognition and intelligent pharmacodynamic screening of the cerebral apoplexy cell model are completed.

[0015] In a third aspect, the application provides a cerebral apoplexy drug screening system, which comprises a cell state recognition module, and the cell state recognition module adopts the above cell state recognition method for cell state recognition.

[0016] The beneficial effects of the present application are that the cell image segmentation of the present application adopts the Cellpose model, extracts two types of features of fluorescence intensity and cell morphology for state discrimination, uses an AI model to classify the cells as healthy / injured, and constructs a complete process: image acquisition → segmentation → feature extraction → state recognition → drug screening output, which has medical research and high-throughput drug development value. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The operation flow chart for cell state recognition and drug screening; Figure 2 The cell number chart for cell state recognition and drug screening; Figure 3 The cell recognition chart for cell state recognition and drug screening, the left chart is after staining, and the right chart is without staining; Figure 4 The drug efficacy ranking chart, the real-time imaging of ONOO⁻ (a) Confocal fluorescence microscope imaging results using TJO staining (10 μM, 30 minutes) show the distribution of ONOO⁻ in SY5Y cells treated with drugs No. 26, 27, 29, and 36 (10 μM) after 0 or 4 ischemia treatment and 2 hours of reperfusion after 4 hours of ischemia. (b) Average fluorescence intensity data of chart (a). The mean values of the remaining four groups were compared with the mean value of the control group. DETAILED DESCRIPTION

[0018] In order to better illustrate the purpose, technical scheme and advantages of the present application, the present application will be further described below in combination with specific embodiments. The present application can be implemented in many different forms, and should not be understood as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure will be thorough and complete, and will fully convey the concept of the present application to those skilled in the art, and the present application will be limited only by the claims.

[0019] The present application provides an intelligent cerebral apoplexy cell state recognition method based on Cellpose and fluorescence image analysis and its application in drug screening, comprising the following steps: Establishing a cerebral apoplexy ischemia-reperfusion cell model; Adding a nitrite fluorescent probe to the cells treated by ischemia-reperfusion to obtain fluorescence microscopic images of the cells.

[0020] Importing the fluorescence microscopic images into the Cellpose model for cell segmentation to obtain the ROI area of each cell.

[0021] Extracting fluorescence intensity features and cell morphology features in the ROI area of each cell, the fluorescence intensity features including average fluorescence intensity and total fluorescence intensity, and the morphology features including cell area.

[0022] The extracted multi-dimensional features are input into a classification model for cell state recognition. Preferably, the classification model is a cell state discrimination model constructed based on a shallow neural network or a random forest, which is used to discriminate individual cells into three states: healthy, mild injury, or severe injury.

[0023] The cell states of each experimental group are statistically analyzed, and the proportion of the number of cells in each group in different states and the average fluorescence level are output.

[0024] In the drug efficacy screening step, the ischemia-reperfusion cell model with different candidate drugs is processed and imaged, and the imaging images are segmented and feature extracted according to the above steps. The average fluorescence intensity and the proportion of severely injured cells of the cells in each drug treatment group are compared to evaluate the protective effect of the drug; the candidate drugs are sorted according to the fluorescence attenuation degree and the proportion of severe injury, and the drug efficacy sorting result is output, completing the automatic state recognition and intelligent drug efficacy screening of the cerebral apoplexy cell model.

[0025] Further, a cerebral apoplexy ischemia-reperfusion cell model is established and a fluorescence microscopic image is obtained, and the specific steps are as follows: Cell culture and processing: SH-SY5Y (human neuroblastoma) cell line is selected, and the cells are cultured in complete culture medium containing 10% fetal bovine serum (FBS), and the culture conditions are 37°C, 5% CO2, and 95% O2 environment.

[0026] Ischemia-reperfusion processing: the cells are treated under oxygen-glucose deprivation (OGD) conditions for 4 hours to simulate ischemia; then, they are restored to normal culture conditions for 2 hours to simulate the reperfusion process.

[0027] Nitrite fluorescent probe staining: to detect the concentration of nitrite ions in the cells, a specific concentration of fluorescent probe (for example, 10 μM NO2⁻ probe) is added for staining. The cells are incubated in the probe solution for 30 minutes, avoiding light exposure.

[0028] Microscopic image acquisition: a laser confocal microscope (such as Leica TCS SP8) is used to acquire images of the stained cells, and the acquisition channel is the red light channel. A specific wavelength of excitation light source (such as 590 nm) and emission spectrum (such as 700-780 nm) are used for fluorescence imaging. The fluorescence microscopic images produced in this step will be used for subsequent cell segmentation and state analysis.

[0029] Further, in order to realize automatic counting and spatial positioning of the cell nucleus, the present application uses computer vision methods to perform image segmentation and centroid extraction on the cell nucleus in the fluorescence image, and the specific method is as follows: Image Preprocessing: First, convert the collected fluorescence image to RGB format. Then, extract the red channel as the main source of nuclear fluorescence signal.

[0030] Threshold Segmentation: Apply Otsu's method to the extracted channel image for adaptive threshold segmentation, separating the nuclear region from the background, resulting in a binary image where the nuclear region is marked as 1 and the background region is marked as 0.

[0031] Morphological Image Denoising and Repair: Apply morphological operations to the binary image, including opening operation to remove small noise and closing operation to fill the holes in the nuclear region, ensuring the accuracy of the segmentation result.

[0032] Region Labeling and Attribute Extraction: Use connected component analysis method to process the binary image, identify all independent nuclear regions, and extract the basic attributes of each region, such as area, centroid position, etc.

[0033] Further, use the Cellpose deep learning model to segment the cell image, accurately identify the outline and position of each cell, and the specific operation is as follows: Load Cellpose Model: First, we use the pre-trained Cellpose deep learning model to process single-cell fluorescence images. The Cellpose model can automatically identify cell contours and generate a segmentation mask image for each cell (i.e., the identification of each cell region). The image is the previously collected cell fluorescence image, where the nucleus is labeled by fluorescent probes. The segmentation process is that Cellpose uses deep learning algorithms to accurately identify the boundaries of each image, outputting the outline of each cell. The model output is the ROI region of each cell (the pixel value of the region is 1, and the other regions are 0). Through the previous segmentation result, we can generate an ROI mask, using the segmentation mask generated by the model, we get a binary mask image, where each cell region is marked as 1 and the other background region is marked as 0. This mask image will provide the basis for subsequent feature extraction and analysis. Next, based on the ROI mask image, use OpenCV and NumPy libraries to extract the red light intensity and area of each cell region from the original image.

[0034] In the fluorescence image, the red light signal of the cell is used to reflect the damage degree of the cell. Therefore, we extract the pixel value of the red channel from the original image as the intensity basis of each cell region.

[0035] For each cell's ROI region, we use the sum function of NumPy to calculate the sum of all red pixel values in this region. This represents the total red light intensity of the cell, which is used to reflect the degree of cell damage (the higher the red light intensity, the greater the damage the cell may have suffered). The formula used is: total red light intensity = Σ (red channel pixel value), and the average red light intensity in this cell region is calculated. The formula used is: red light mean = total red light intensity / number of pixels in this cell region. Extracting the maximum pixel value of the red light channel in the cell as the maximum red light intensity of the cell helps to capture the brightest fluorescent signal. The formula is: red light maximum = max (red channel pixel value).

[0036] The number of pixels in the statistical mask can be counted by counting all pixels with a value of 1 in the ROI mask image generated by Cellpose, which is equivalent to calculating the area of the cell. If necessary, the area of the cell can be converted from the number of pixels to the actual area unit (e.g., microns squared) according to the magnification of the microscope and the size of the pixels.

[0037] The cell features extracted by the above steps (including total red light intensity, average red light intensity, and area) can be used for cell damage state determination and drug screening.

[0038] State determination criteria: Based on the extracted red light intensity and area characteristics, combined with the set threshold criteria, we can classify the state of each cell. Generally, the red light intensity of the cell is proportional to the degree of damage, so: Healthy cells: average red light intensity below threshold A, normal area; Mildly damaged cells: average red light intensity between threshold A and threshold B, slightly smaller area; Severely damaged cells: average red light intensity above threshold B, significantly smaller area.

[0039] Determination method: By training a classification model (such as random forest or support vector machine) or setting a manual threshold (e.g., red light intensity above a certain threshold is determined as a damaged cell), we classify the state of the cell.

[0040] Further, the drug efficacy screening step can be performed after simulating the ischemic microenvironment, and then re-oxygenation and drug administration: replace the normal DMEM complete culture medium with 100 μM of the candidate natural medicine, and re-oxygenate for 2 hours under normal culture conditions (5% CO2, 37°C). Probe detection and confocal imaging: after removing the culture medium, add complete culture medium containing 10 μM TJO, and incubate in the dark for 30 minutes. Then, add 0.8ul Hoechest to stain the nucleus per dish. Before imaging, wash with PBS for 3 times to remove unbound probes. Finally, use a Leica confocal microscope to capture fluorescence images, and use ImageJ software to quantitatively analyze the fluorescence intensity (excitation wavelength 590 nm, emission wavelength 700-780 nm).

[0041] Different drug groups: through the treatment of different drug groups (such as drugs A, B, and C), we can observe whether the drugs can effectively weaken the red light intensity in the cells. Generally, the effectiveness of the drugs is manifested as a decrease in the fluorescence intensity in the cells or the recovery of the normal morphology of the cells.

[0042] Screening result output: compare the red light intensity changes and the proportion of damaged cells of each group of cells, and output the screening results of the drugs. For example, the average fluorescence intensity of the cells in the drug A group decreases by 40%, which is determined to have a significant protective effect on the cells.

[0043] As one of the embodiments, the intelligent cerebral apoplexy cell state recognition method based on Cellpose and fluorescence image analysis and its application in the drug efficacy screening system includes the following steps: First step: input information verification and task initialization The user first inputs the following key information through the system interface: the file path of the original fluorescence image; the saving path of the analysis results; the microscope parameters (including the magnification and the pixel physical size: such as 0.65 μm / pixel); and whether to enable the drug screening analysis module. The system performs legality verification on the paths and parameters input by the user: if the image file does not exist or the format is not supported (non-TIFF / JPG / PNG), the system pops up a prompt "invalid image path" and requires the user to re-input; if the saving path has no permission to write or does not exist, the system guides the user to re-specify the saving directory; if the parameters are missing or the format is not correct, the system prompts and requires completion. After the verification is passed, the system loads the image and initializes the task state, and prepares to enter the image processing process.

[0044] Second step: image preprocessing and red channel extraction The system performs the following preprocessing operations on the imported fluorescence image in the image processing module: image reading and format conversion: convert the input image to standard RGB format to ensure consistent color space; image denoising: apply Gaussian Blur to the image for low-intensity smoothing to suppress high-frequency noise; contrast enhancement: apply histogram equalization to enhance image contrast and improve cell region edge definition; color channel extraction: extract the red channel from the RGB image to reflect the intracellular nitrite fluorescence signal intensity, save it as a grayscale image for subsequent analysis; image size recording: the system records image width, height, pixel size, etc. information for area conversion.

[0045] Step 3: Cell recognition and mask generation (Cellpose segmentation) After image preprocessing, the system enters the cell recognition module and performs the following operations in sequence: default call Cellpose official pre-trained model (cyto mode), or user-specified local fine-tuned model; set segmentation parameters such as target diameter (diameter) and channel type (channels = [0, 0] for grayscale image).

[0046] Input the red channel image into the Cellpose model; the model outputs a single-cell mask image, where each cell region is labeled with a unique number and other regions are 0; the mask image is in two-dimensional matrix format, with each pixel value representing the corresponding cell number. The mask image is saved in NumPy format and a color segmentation image is generated overlaid on the original image for user verification; the system records the pixel coordinate set of each cell mask for structured input for subsequent calculations.

[0047] Step 4: Red light intensity and morphological parameter extraction After generating the cell mask image, the system iterates through each cell ROI region in sequence to extract the following feature information: Red light intensity extraction: for each cell mask corresponding region, extract the pixel value set of the red channel original image; after extraction, calculate the following indicators: 1. Red light total intensity = sum of all pixel values in the ROI region; 2. Red light average intensity = mean of ROI region pixel values; 3. Red light maximum intensity = maximum pixel value in the ROI region.

[0048] The above indicators are used to evaluate the total amount, distribution balance and peak expression of intracellular nitrite concentration.

[0049] Morphological parameter extraction: cell area calculation can be achieved by counting the number of non-zero pixels in the mask region, multiplying by the physical size of the pixel, and converting to microns 2Cell area per unit; Area (pm 2 ) = ROI pixel number x (pixel side length) 2 Cell contour and centroid extraction can use OpenCV to extract the edge contour of each cell for subsequent visualization; calculate the centroid coordinates (x, y) of each cell ROI for numbering and positioning. After calculation, the number, area, red light index, and centroid coordinates of all cells are output in CSV / JSON format; at the same time, the labeled image (with cell number and centroid visualization) is saved. The calculation of cell morphology is mainly to prevent the interference of nitrite ions free in the extracellular with the probe to emit fluorescence.

[0050] Step 5: Cell state determination and drug efficacy analysis The system determines the cell state according to the extracted fluorescence and morphological parameters, and executes the drug efficacy screening module (if enabled): State determination logic: According to the preset threshold model (or trained model), the system divides each cell into three categories: Healthy cells: average red light intensity below threshold A, normal area; Mildly damaged cells: average red light intensity between threshold A and threshold B, slightly reduced area; Severely damaged cells: average red light intensity above threshold B, significantly reduced area.

[0051] Drug efficacy analysis process (enabled when enabled): For different drug treatment groups, statistics: 1. The proportion of the number of cells in each state; 2. Average red light intensity; 3. The proportion of severely damaged cells decreased.

[0052] Take "severe cell reduction rate" and "average fluorescence drop rate" as weighted indicators to calculate drug efficacy score, and finally output drug efficacy ranking list and bar chart, as shown in Figure 4 .

[0053] Example 1 Cell state recognition Comprising the following steps: Step 1: Culture of cells in human neuroblastoma cells (SH-SY5Y), specifically including: SY5Y cells were prepared using complete culture medium, and the culture medium was composed as follows: 1% penicillin-streptomycin double antibody (Cytiva, SV30010), 10% fetal bovine serum (FBS, Gibco), 89% minimum essential medium (DMEM, Gibco, C11965500BT).

[0054] All culture plates and dishes were placed in a 37°C constant temperature incubator, with constant humidity in the box and a gas environment of 5% CO2 and 95% O2.

[0055] SY5Y cells (8x103cells / well) were seeded in 96-well plates containing 100 μL of culture medium and incubated at 37°C, 5% CO2for 24 h.

[0056] Then the cells were incubated in fresh culture medium containing different concentrations of fluorescent probe TJO (0.0, 2.0, 5.0, 10.0, 20.0 and 40.0 μM) for 24 h.

[0057] Next, MTT solution (5.0 mg / mL, PBS) was added to each well (10 μL / well, 0.5 mg / mL), and after 4 h, the residual MTT solution was removed.

[0058] Finally, 150 μL of DMSO was added to each well to dissolve the formazan crystals; after 10 min of shaking, the absorbance values of the wells were recorded at 570 nm using an enzyme-linked immunosorbent assay, and the cytotoxic effects of the fluorescent probe TJO were evaluated using the formula A / A0x 100% (where A and A0are the absorbance values of the experimental and control groups, respectively).

[0059] Step 2: Construction of ischemia-reperfusion model and drug screening The oxygen-glucose deprivation / reperfusion (OGD / R) method was used to construct a cerebral ischemia-reperfusion (CIRI) cell model.

[0060] Oxygen-glucose deprivation (OGD): SH-SY5Y cells were replaced with sugar-free DMEM medium and transferred to a hypoxic incubator (1% O2, 5% CO2, 37°C) for 4 hours to simulate an ischemic microenvironment.

[0061] Reperfusion and drug administration: normal DMEM complete medium was replaced, and 100 μM of the candidate natural drug was added, respectively, and reoxygenated for 2 hours under normal culture conditions (5% CO2, 37°C).

[0062] Step 3: Probe detection and confocal imaging After removing the culture medium, complete medium containing 10 μM TJO was added, and incubated in the dark for 30 minutes; then, 0.8 ul of Hoechest was added to each dish to stain the nucleus. Before imaging, the unbound probe was removed by washing with PBS three times, and finally, the fluorescence images were captured using a Leica confocal microscope, and the fluorescence intensity was quantitatively analyzed using ImageJ software (excitation wavelength 590 nm, emission wavelength 700-780 nm).

[0063] Step 4: Intelligent information extraction of cells: Segmentation of cell images using the Cellpose deep learning model to accurately identify the outline and location of each cell. The specific operations are as follows: Load the Cellpose model: First, we use the pre-trained Cellpose deep learning model to process single-cell fluorescence images. The Cellpose model can automatically identify cell contours and generate a segmentation mask image for each cell (i.e., the identification of each cell region). The image is the previously collected cell fluorescence image, in which the cell nucleus is labeled by a fluorescent probe. The segmentation process is that Cellpose accurately identifies the boundary of each cell in each image based on deep learning algorithms, and outputs the outline of each cell. The output of the model is the ROI region of each cell (the pixel value of the region is 1, and the other regions are 0). Through the segmentation results in the early stage, the ROI mask can be generated, and using the segmentation mask generated by the model, we get a binary mask image, in which each cell region is marked as 1, and the other background regions are marked as 0. This mask image will provide the basis for subsequent feature extraction and analysis. Next, based on the ROI mask image, use OpenCV and NumPy libraries to extract the red light intensity and area of each cell region from the original image.

[0064] In the fluorescence image, the red light signal of the cell is used to reflect the damage degree of the cell. Therefore, we extract the pixel value of the red channel from the original image as the intensity basis of each cell region.

[0065] Step 5: Calculate the damage degree of the cell: For each cell ROI region, we use the sum function of NumPy to calculate the sum of all red pixel values in the region. This represents the total intensity of the red light of the cell, which is used to reflect the degree of damage to the cell (the higher the red light intensity, the greater the damage the cell may have suffered). The formula used is: total red light intensity = Σ (red channel pixel value), and the average red light intensity in the cell region is calculated. The formula used is: red light mean = total red light intensity / number of pixels in the cell region. Extract the maximum pixel value of the red light channel in the cell as the maximum red light intensity of the cell, which helps to capture the brightest fluorescence signal. The formula is: red light maximum = max (red channel pixel value).

[0066] The number of pixels in the mask can be counted by counting the number of pixels with a value of 1 in the ROI mask image generated by Cellpose, which is equivalent to calculating the area of the cell. If necessary, the area of the cell can be converted from the number of pixels to the actual area unit (e.g., microns square) according to the magnification of the microscope and the size of the pixels.

[0067] The cell features (including red light total intensity, average red light intensity, area) extracted by the above steps can be used for cell damage state determination and drug screening.

[0068] State determination criteria: Based on the extracted red light intensity and area features, combined with the set threshold criteria, we can classify the state of each cell. Generally, the red light intensity of the cell is proportional to the degree of damage, so: Healthy cells: average red light intensity below threshold A, normal area; Mildly damaged cells: average red light intensity between threshold A and threshold B, slightly smaller area; Severely damaged cells: average red light intensity above threshold B, significantly smaller area.

[0069] Determination method: By training a classification model (such as random forest or support vector machine) or setting a manual threshold (for example, red light intensity exceeding a certain threshold is determined as a damaged cell), we classify the state of the cell.

[0070] Obviously, the above embodiments are only examples for the sake of clarity, and are not limiting of the embodiments. For those of ordinary skill in the art, other different forms of changes or variations can be made on the basis of the above description. Here, it is not necessary and impossible to exhaust all the embodiments. The obvious changes or variations derived therefrom are still within the protection scope of the present invention.

Claims

1. An intelligent stroke cell state recognition method based on Cellpose and fluorescence image analysis, characterized by: The method comprises the following steps: S1, establishing a cerebral apoplexy ischemia-reperfusion cell model; S2, adding a nitrite fluorescent probe to the cell after ischemia-reperfusion treatment to obtain a fluorescence microscopic image of the cell; S3, importing the fluorescence microscopic image into a Cellpose model for cell segmentation to obtain an ROI region of each cell; S4, extracting a fluorescence intensity feature and a cell morphology feature in the ROI region of each cell, wherein the fluorescence intensity feature comprises an average fluorescence intensity and a total fluorescence intensity, and the cell morphology feature comprises a cell area; S5, inputting the extracted multi-dimensional features into a classification model for cell state recognition to distinguish a single cell into three states of healthy, mild damage or severe damage.

2. The cell state identification method of claim 1, wherein: In S1, the specific steps of establishing the cerebral apoplexy ischemia-reperfusion cell model are as follows: S11, cell culture and treatment: a SH-SY5Y human neuroblastoma cell line is selected, and the cells are cultured in a complete culture medium containing 10% fetal bovine serum, and the culture conditions are 37°C, 5% CO2 and 95% O2 environment; S12, ischemia-reperfusion treatment: the cells are treated under oxygen-glucose deprivation conditions to simulate ischemia; then, the cells are restored to normal culture conditions to simulate the reperfusion process.

3. The cell state identification method of claim 1, wherein: In S3, the specific steps of importing the fluorescence microscopic image into the Cellpose model for cell segmentation to obtain the ROI region of each cell are as follows: the Cellpose model automatically identifies the cell contour and generates a segmentation mask image for each cell, accurately identifies the boundary of the cell in each image, and outputs the ROI mask image of each cell, wherein the cell nucleus region is marked as 1 and the background region is marked as 0.

4. The cell state identification method according to claim 3, characterized in that: The specific method of the Cellpose model for cell segmentation of the fluorescence microscopic image is as follows: (1) image preprocessing: the collected fluorescence microscopic image is converted into an RGB format, and the red channel is extracted as the main signal source of the cell nucleus fluorescence; (2) threshold segmentation: the extracted channel image is subjected to adaptive threshold segmentation by the Otsu method to separate the cell nucleus region from the background to obtain a binary image, wherein the cell nucleus region is marked as 1 and the background region is marked as 0; (3) morphological image denoising and repair: morphological operations are applied to the binary image, including open operation to remove small noise and close operation to fill the holes in the cell nucleus region to ensure the accuracy of the segmentation result; (4) region labeling and attribute extraction: the binary image is processed using a connected domain analysis method to identify all independent cell nucleus regions and extract the basic attributes of each region, wherein the basic attributes include area and centroid position.

5. The cell state identification method of claim 1, wherein: In S4, the specific method of extracting the fluorescence intensity feature and the cell morphology feature is as follows: in the ROI region of each cell, the sum of all red pixel values in the region is calculated using the summation function of NumPy to represent the total red light intensity of the cell, and the average red light intensity of the cell region is calculated, and the calculation formula is: total red light intensity = Σ (red channel pixel value), and the calculation formula is: average red light intensity = total red light intensity / number of pixels in the cell region. The number of pixels in the statistical mask is counted by counting the number of pixels with a value of 1 in the ROI mask image generated by Cellpose, and the area of the cell is calculated.

6. The cell state identification method of claim 5, wherein: In S5, the standard for cell state recognition is: according to the extracted red light intensity and area characteristics, combined with the set threshold standard, the state of each cell is classified: Healthy cells: average red light intensity is lower than threshold A, and area is normal; Mildly damaged cells: average red light intensity is between threshold A and threshold B, and area is slightly reduced; Severely damaged cells: average red light intensity is higher than threshold B, and area is significantly reduced.

7. The method of claim 1, wherein: Further comprising the following steps: S6, statistics of the state of each experimental group of cells, output the number of different state cells in each group and the average fluorescence intensity.

8. The cell state recognition method of any one of claims 1-7 for use in pharmacodynamic screening.

9. Use according to claim 8, characterized in that: In the process of pharmacodynamic screening, the ischemia-reperfusion cell model with different candidate drugs is processed and imaged, the imaging image is segmented and feature extracted, the average fluorescence intensity and the proportion of severely damaged cells of each drug treatment group are compared, and the protective effect of the drug is evaluated; according to the fluorescence attenuation degree and the proportion of severely damaged cells of each group, the candidate drugs are sorted, and the pharmacodynamic screening result is output, and the automatic state recognition and intelligent pharmacodynamic screening of the cerebral apoplexy cell model are completed.

10. A stroke drug screening system, characterized in that: The cerebral apoplexy drug screening system comprises a cell state recognition module, and the cell state recognition module adopts the cell state recognition method of any one of claims 1-7 for cell state recognition.

Images