Cell segmentation method based on multi-channel fluorescence characteristic fusion

Abstract

The invention discloses a cell segmentation method based on multi-channel fluorescence feature fusion, and relates to the technical field of cell segmentation. Comprising the following steps of data preprocessing and channel synchronization, network architecture design, cell nucleus positioning, mask generation and the like. Through a deep learning technology and in combination with multi-channel image data such as DAPI and SpGreen, accurate automatic segmentation of cell nucleus and cytoplasm regions is realized, challenges of intercellular adhesion and overlapping are effectively solved through introduction of an attention mechanism and multi-scale feature fusion, meanwhile, cell boundaries are optimized through distance transformation and a conditional random field CRF, the segmentation precision is improved, and the segmentation efficiency is improved. The subsequent cell counting and co-expression analysis function enables a user to accurately evaluate cell distribution and co-expression conditions of different fluorescence channels, and a segmentation result and a data visualization report provided by the method provide an efficient and accurate analysis tool for cell biology research, and promote automation and refinement of cell image analysis.

Description

Technical Field

[0001] This invention relates to the field of cell segmentation technology, and more specifically to a cell segmentation method based on multi-channel fluorescence feature fusion. Background Technology

[0002] Cell segmentation is a crucial foundational technology in cell biology, pathological analysis, drug screening, and precision medicine research. Its primary purpose is to accurately identify and separate individual cell regions from microscopic imaging data to enable cell counting, morphological analysis, and quantitative studies of multi-channel fluorescence signals. With the development of multi-channel fluorescence microscopy, researchers can simultaneously acquire spatial distribution information of the cell nucleus, cytoplasm, and various biomarkers, such as DAPI, SpGreen, SpGold, and CY5 fluorescence channels, which are widely used in cell structure and function studies. Current cell segmentation methods mainly include threshold-based methods, traditional image processing methods, and some deep learning-based segmentation methods. These methods have achieved, to some extent, automated segmentation of cell nuclei or cell regions and are applied in high-throughput imaging analysis and biomedical image processing.

[0003] However, in practical applications, existing technologies still suffer from several drawbacks. Traditional cell segmentation methods typically rely on a single fluorescence channel or simple feature extraction strategies, making it difficult to fully utilize the complementary information contained in multi-channel fluorescence images. When cells are densely packed, in complex growth states, or when cells are adhered or overlapped, incomplete segmentation, blurred cell boundaries, or even missegmentation can easily occur. Some existing deep learning segmentation methods lack effective constraint mechanisms when dealing with the relationship between the cell nucleus and cytoplasm, failing to establish a precise mapping relationship between the nucleus and cytoplasm, leading to inaccurate cell boundary expansion and affecting the reliability of subsequent cell counting and co-expression analysis. Furthermore, differences in brightness, noise interference, and spatial offset exist between different fluorescence channels. Existing technologies often lack systematic standardization, alignment, and boundary optimization processes, resulting in insufficient stability and generalization ability of segmentation results, thereby reducing the accuracy and practicality of cell image analysis. Therefore, we propose a cell segmentation method based on multi-channel fluorescence feature fusion. Summary of the Invention

[0004] In view of the above-mentioned shortcomings of the existing technology, the first objective of the present invention is to provide a cell segmentation method based on multi-channel fluorescence feature fusion, thereby solving the problems in the background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A cell segmentation method based on multi-channel fluorescence feature fusion includes the following steps:

[0007] S1. Standardization, noise removal, and spatial alignment of multi-channel image data;

[0008] S2. Design of a two-stream structure and attention mechanism model based on deep convolutional neural networks;

[0009] S3. Use the DAPI signal to generate a precise nuclear mask and perform thresholding.

[0010] S4. Use the SpGreen channel to extract cytoplasmic regions and perform probability map prediction;

[0011] S5. Combine the DAPI mask with the cytoplasmic region map to perform precise mask mapping;

[0012] S6. Cell boundary optimization and noise removal, restoring cell morphology and segmentation accuracy;

[0013] S7. Segmentation accuracy evaluation and model optimization based on indicators such as IoU and Dice coefficient;

[0014] S8. Perform cell counting and co-expression analysis to assess the interactions between different channels;

[0015] S9. Generate a visual report of the segmentation results and export the data to support subsequent analysis and sharing.

[0016] The present invention is further configured such that: in step S1, the standardization, noise removal, and spatial alignment processing of multi-channel image data:

[0017] S1.1 Load all four fluorescence channel data (DAPI, SpGreen, SpGold, CY5) from the imaging device, ensuring that the image data format of each channel is consistent, perform noise reduction and standardization processing, and unify the image size and resolution;

[0018] S1.2. Normalize the image for each channel to eliminate brightness differences between channels. A commonly used normalization method is Z-score normalization:

[0019]

[0020] in, The grayscale value for each pixel, The average value of the pixels. The standard deviation of the pixels;

[0021] S1.3 To reduce processing complexity, the image is cropped into regions of interest and spatially aligned to ensure that all image channels are spatially consistent.

[0022] S1.4. Perform local contrast enhancement on the image of each channel, and use the CLAHE (Contrast Limiting Adaptive Histogram Equalization) algorithm to increase the details of the image;

[0023] S1.5. Combine the information from all channels to form a multi-channel composite image, ready for input into the network.

[0024] The present invention is further configured such that: in step S2, the design of the two-stream structure and attention mechanism model based on a deep convolutional neural network:

[0025] S2.1 The network adopts a dual-stream structure: one stream processes DAPI channel data (nuclear information), and the other stream processes SpGreen channel data (cytoplasmic information).

[0026] S2.2 Each stream uses a convolutional neural network (CNN) to extract high-level features of its respective channel, typically using ResNet or VGG architecture as the base network;

[0027] S2.3. Introduce an attention mechanism into the network to enhance its focus on important regions. Use a channel attention mechanism to improve the focus on DAPI high-signal regions:

[0028]

[0029] in, The weight matrix to be learned. For input features, For bias terms, Use the Sigmoid activation function;

[0030] S2.4. Combining multi-scale features, contextual information is fused through Feature Pyramid Networks (FPN), enabling the network to capture cell information at different scales.

[0031] S2.5 The network output is a nuclear mask, representing the spatial location of the cell nucleus.

[0032] The present invention is further configured such that in step S3, generating a precise nuclear mask using the DAPI signal and performing thresholding processing:

[0033] S3.1. The spatial location of the cell nucleus is learned through a deep network. The DAPI channel is used as the main input to generate a cell nucleus mask. An adaptive thresholding method is used to separate the cell nucleus region.

[0034] S3.2 Extract the core region from the DAPI channel image. These regions correspond to the center of the cell nucleus. Further refine the boundary of the nucleus region using the watershed algorithm.

[0035] S3.3 Optimize the threshold setting based on the local brightness information of the image to dynamically adjust the extraction range of cell nuclei;

[0036] S3.4. Remove small noises in the nuclear mask through morphological operations (such as opening and closing operations) to further enhance the accuracy of the nuclear region;

[0037] S3.5. The processed nuclear mask is passed to the subsequent steps for cytoplasmic boundary extraction.

[0038] The present invention is further configured such that: in step S4, extracting cytoplasmic regions using the SpGreen channel and performing probability map prediction:

[0039] S4.1. Based on the SpGreen channel, a convolutional neural network is used to extract the probability map of the cytoplasmic region;

[0040] S4.2 The convolutional layer extracts deep features from the SpGreen channel to obtain a multi-layer feature representation of the cytoplasm;

[0041] S4.3 Obtain the predicted probability map of the cytoplasm using the Sigmoid activation function:

[0042]

[0043] in, It is a convolution kernel used for cytoplasmic region prediction. It is the input SpGreen feature. It is the Sigmoid activation function;

[0044] S4.4. Further optimize the boundaries of the cytoplasmic region using Conditional Random Field (CRF) to correct ambiguous regions between cells;

[0045] S4.5 Use the CRF-optimized cytoplasmic probability map as input for subsequent steps.

[0046] The present invention is further configured such that: in step S5, combining the DAPI mask with the cytoplasmic region map, precise mask mapping is performed.

[0047] S5.1 Mapping the nuclear mask of the DAPI channel with the cytoplasmic probability map of the SpGreen channel;

[0048] S5.2. Based on the nuclear mask, perform distance transformation to generate a probability graph of the cytoplasmic boundary:

[0049]

[0050] in, It is a point on the nuclear mask. It is the result of distance transformation. It refers to each pixel in space. Represents Euclidean distance;

[0051] S5.3. By using constraint conditions, ensure that the cytoplasmic boundary does not exceed the SpGreen fluorescence region;

[0052] S5.4 Optimize the mask based on the distance transformation results, mapping the nuclear mask to the cytoplasm region to form a more accurate cell boundary;

[0053] S5.5 The final cell boundary represents the accurate cell location.

[0054] The present invention is further configured such that: in step S6, cell boundary optimization and noise removal, cell morphology is repaired and segmentation accuracy is improved;

[0055] S6.1 Denoise the segmentation results, remove areas with unclear boundaries, and further repair cell edges;

[0056] S6.2 Correcting the shape of cell boundaries through morphological repair techniques (expansion and corrosion operations);

[0057] S6.3 Apply multi-level image restoration methods, including artifact removal and enhancing the accuracy of true boundaries;

[0058] S6.4. By adjusting the morphological information of each cell, ensure that the cell structure conforms to biological characteristics;

[0059] S6.5 Output the corrected cell boundary image as the segmentation result.

[0060] The present invention is further configured such that: in step S7, segmentation accuracy evaluation and model optimization based on indicators such as IoU and Dice coefficient:

[0061] S7.1 Select multiple indicators (IoU, Dice coefficient) to evaluate the accuracy of the segmentation results;

[0062] S7.2 Optimize the hyperparameters (learning rate, batch size) of the network model to improve segmentation accuracy;

[0063] S7.3 Adjust the network structure based on the evaluation results, adding or removing certain layers to improve performance;

[0064] S7.4 Test the adjusted model on the validation set to ensure its performance on unseen data;

[0065] S7.5 Output the final segmentation accuracy evaluation results to provide a reference for subsequent applications.

[0066] The present invention is further configured such that, in step S8, cell counting and co-expression analysis are performed to evaluate the interactions between different channels:

[0067] S8.1 Count cells based on the segmentation results of cell boundaries and count the number of individual cells;

[0068] S8.2 Calculate the co-expression rate among different cells by analyzing the overlapping regions of different fluorescence channels;

[0069] S8.3 Use multi-component analysis to assess the intersection and differences between different channels;

[0070] S8.4. Apply statistical methods (chi-square test or T-test) to perform cell co-expression analysis;

[0071] S8.5 Generate the final cell count and co-expression analysis report.

[0072] The present invention is further configured such that: in step S9, generating a visualization report of the segmentation results and exporting the data, supporting subsequent analysis and sharing:

[0073] S9.1 Visualize the segmented image to display each cell and its region;

[0074] S9.2 Output the cell count and co-expression data for each cell to facilitate subsequent analysis;

[0075] S9.3 Archive the segmentation results and statistical data, and prepare them for sharing;

[0076] S9.4 Generate interactive visualization reports to help researchers understand the segmentation results;

[0077] S9.5. Fine-tune and improve the method according to application requirements.

[0078] Beneficial effects

[0079] Compared with known public technologies, the technical solution provided by this invention has the following beneficial effects:

[0080] This invention utilizes deep learning technology, combined with multi-channel image data such as DAPI and SpGreen, to achieve precise automatic segmentation of cell nuclei and cytoplasm regions. By introducing attention mechanisms and multi-scale feature fusion, it effectively addresses the challenges of cell adhesion and overlap. Furthermore, by optimizing cell boundaries through distance transformation and conditional random fields (CRF), it improves segmentation accuracy. Subsequent cell counting and co-expression analysis functions enable users to accurately assess cell distribution and co-expression of different fluorescence channels. The segmentation results and data visualization reports provided by this method offer an efficient and precise analytical tool for cell biology research, promoting the automation and refinement of cell image analysis. Attached Figure Description

[0081] Figure 1 This is a schematic flowchart of a cell segmentation method based on multi-channel fluorescence feature fusion according to the present invention. Detailed Implementation

[0082] To make the technical problems, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the described embodiments are only a part of the embodiments of this application, not all of them. The specific embodiments described herein are only used to explain the invention and are not intended to limit the invention. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0083] It should be further noted that the accompanying drawings and embodiments of the present invention mainly describe the concept of the present invention. Based on this concept, some specific forms and arrangements of connection relationships, positional relationships, power mechanisms, power supply systems, hydraulic systems and control systems may not be fully described. However, under the premise that those skilled in the art understand the concept of the present invention, they can implement the above-mentioned specific forms and arrangements in a well-known manner.

[0084] When a component is referred to as "fixed to" or "set on" another component, it can be directly on or indirectly on that other component. When a component is referred to as "connected to" another component, it can be directly connected to or indirectly connected to that other component.

[0085] The directional terms "inner" and "outer" refer to the inner and outer contours of each component itself. The terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0086] For ease of description, spatial relative terms such as "above," "over," "on the upper surface of," "above," etc., are used herein to describe the spatial positional relationship of a device or feature as shown in the figures to other devices or features. It should be understood that spatial relative terms are intended to encompass different orientations in use or operation beyond the orientation of the device as described in the figures. For example, if the device in the figures were inverted, a device described as "above" or "above" other devices or structures would subsequently be positioned as "below" or "under" other devices or structures. Thus, the exemplary term "above" can include both "above" and "below." The device may also be positioned in other different ways, and the spatial relative descriptions used herein will be interpreted accordingly.

[0087] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, and "several" means one or more, unless otherwise explicitly specified.

[0088] The present invention will now describe a cell segmentation method based on multi-channel fluorescence feature fusion.

[0089] Example 1

[0090] like Figure 1 As shown, the present invention provides a technical solution: a cell segmentation method based on multi-channel fluorescence feature fusion, comprising the following steps:

[0091] S1. Standardization, noise removal, and spatial alignment of multi-channel image data;

[0092] S2. Design of a two-stream structure and attention mechanism model based on deep convolutional neural networks;

[0093] S3. Use the DAPI signal to generate a precise nuclear mask and perform thresholding.

[0094] S4. Use the SpGreen channel to extract cytoplasmic regions and perform probability map prediction;

[0095] S5. Combine the DAPI mask with the cytoplasmic region map to perform precise mask mapping;

[0096] S6. Cell boundary optimization and noise removal, restoring cell morphology and segmentation accuracy;

[0097] S7. Segmentation accuracy evaluation and model optimization based on indicators such as IoU and Dice coefficient;

[0098] S8. Perform cell counting and co-expression analysis to assess the interactions between different channels;

[0099] S9. Generate a visual report of the segmentation results and export the data to support subsequent analysis and sharing;

[0100] In step S1, the standardization, noise removal, and spatial alignment of multi-channel image data:

[0101] S1.1 Load all four fluorescence channel data (DAPI, SpGreen, SpGold, CY5) from the imaging device, ensuring that the image data format of each channel is consistent, perform noise reduction and standardization processing, and unify the image size and resolution;

[0102] S1.2. Normalize the image for each channel to eliminate brightness differences between channels. A commonly used normalization method is Z-score normalization:

[0103]

[0104] in, The grayscale value for each pixel, The average value of the pixels. The standard deviation of the pixels;

[0105] S1.3 To reduce processing complexity, the image is cropped into regions of interest and spatially aligned to ensure that all image channels are spatially consistent.

[0106] S1.4. Perform local contrast enhancement on the image of each channel, and use the CLAHE (Contrast Limiting Adaptive Histogram Equalization) algorithm to increase the details of the image;

[0107] S1.5. Combine the information from all channels to form a multi-channel composite image, ready for input into the network;

[0108] In step S2, the design of the two-stream structure and attention mechanism model based on deep convolutional neural networks:

[0109] S2.1 The network adopts a dual-stream structure: one stream processes DAPI channel data (nuclear information), and the other stream processes SpGreen channel data (cytoplasmic information).

[0110] S2.2 Each stream uses a convolutional neural network (CNN) to extract high-level features of its respective channel, typically using ResNet or VGG architecture as the base network;

[0111] S2.3. Introduce an attention mechanism into the network to enhance its focus on important regions. Use a channel attention mechanism to improve the focus on DAPI high-signal regions:

[0112]

[0113] in, The weight matrix to be learned. For input features, For bias terms, Use the Sigmoid activation function;

[0114] S2.4. Combining multi-scale features, contextual information is fused through Feature Pyramid Networks (FPN), enabling the network to capture cell information at different scales.

[0115] S2.5 The network output is a nuclear mask, representing the spatial location of the cell nucleus;

[0116] In step S3, which involves generating a precise nuclear mask using the DAPI signal and then performing thresholding:

[0117] S3.1. The spatial location of the cell nucleus is learned through a deep network. The DAPI channel is used as the main input to generate a cell nucleus mask. An adaptive thresholding method is used to separate the cell nucleus region.

[0118] S3.2 Extract the core region from the DAPI channel image. These regions correspond to the center of the cell nucleus. Further refine the boundary of the nucleus region using the watershed algorithm.

[0119] S3.3 Optimize the threshold setting based on the local brightness information of the image to dynamically adjust the extraction range of cell nuclei;

[0120] S3.4. Remove small noises in the nuclear mask through morphological operations (such as opening and closing operations) to further enhance the accuracy of the nuclear region;

[0121] S3.5. The processed nuclear mask is passed to the subsequent steps for cytoplasmic boundary extraction;

[0122] In step S4, extracting cytoplasmic regions using the SpGreen channel and performing probability map prediction:

[0123] S4.1. Based on the SpGreen channel, a convolutional neural network is used to extract the probability map of the cytoplasmic region;

[0124] S4.2 The convolutional layer extracts deep features from the SpGreen channel to obtain a multi-layer feature representation of the cytoplasm;

[0125] S4.3 Obtain the predicted probability map of the cytoplasm using the Sigmoid activation function:

[0126]

[0127] in, It is a convolution kernel used for cytoplasmic region prediction. It is the input SpGreen feature. It is the Sigmoid activation function;

[0128] S4.4. Further optimize the boundaries of the cytoplasmic region using Conditional Random Field (CRF) to correct ambiguous regions between cells;

[0129] S4.5 Use the CRF-optimized cytoplasmic probability map as input for subsequent steps;

[0130] In step S5, combining the DAPI mask with the cytoplasmic region map, precise mask mapping is performed:

[0131] S5.1 Mapping the nuclear mask of the DAPI channel with the cytoplasmic probability map of the SpGreen channel;

[0132] S5.2. Based on the nuclear mask, perform distance transformation to generate a probability graph of the cytoplasmic boundary:

[0133]

[0134] in, It is a point on the nuclear mask. It is the result of distance transformation. It refers to each pixel in space. Represents Euclidean distance;

[0135] S5.3. By using constraint conditions, ensure that the cytoplasmic boundary does not exceed the SpGreen fluorescence region;

[0136] S5.4 Optimize the mask based on the distance transformation results, mapping the nuclear mask to the cytoplasm region to form a more accurate cell boundary;

[0137] S5.5 The final cell boundary represents the accurate cell location;

[0138] In step S6, cell boundary optimization and noise removal, cell morphology and segmentation accuracy are restored:

[0139] S6.1 Denoise the segmentation results, remove areas with unclear boundaries, and further repair cell edges;

[0140] S6.2 Correcting the shape of cell boundaries through morphological repair techniques (expansion and corrosion operations);

[0141] S6.3 Apply multi-level image restoration methods, including artifact removal and enhancing the accuracy of true boundaries;

[0142] S6.4. By adjusting the morphological information of each cell, ensure that the cell structure conforms to biological characteristics;

[0143] S6.5 Output the corrected cell boundary image as the segmentation result;

[0144] In step S7, segmentation accuracy evaluation and model optimization based on metrics such as IoU and Dice coefficient:

[0145] S7.1 Select multiple indicators (IoU, Dice coefficient) to evaluate the accuracy of the segmentation results;

[0146] S7.2 Optimize the hyperparameters (learning rate, batch size) of the network model to improve segmentation accuracy;

[0147] S7.3 Adjust the network structure based on the evaluation results, adding or removing certain layers to improve performance;

[0148] S7.4 Test the adjusted model on the validation set to ensure its performance on unseen data;

[0149] S7.5 Output the final segmentation accuracy evaluation results to provide a reference for subsequent applications;

[0150] In step S8, cell counting and co-expression analysis are performed to assess the interactions between different channels:

[0151] S8.1 Count cells based on the segmentation results of cell boundaries and count the number of individual cells;

[0152] S8.2 Calculate the co-expression rate among different cells by analyzing the overlapping regions of different fluorescence channels;

[0153] S8.3 Use multi-component analysis to assess the intersection and differences between different channels;

[0154] S8.4. Apply statistical methods (chi-square test or T-test) to perform cell co-expression analysis;

[0155] S8.5 Generate the final cell count and co-expression analysis report;

[0156] Step S9, generating a visualization report of the segmentation results and exporting the data, supports subsequent analysis and sharing:

[0157] S9.1 Visualize the segmented image to display each cell and its region;

[0158] S9.2 Output the cell count and co-expression data for each cell to facilitate subsequent analysis;

[0159] S9.3 Archive the segmentation results and statistical data, and prepare them for sharing;

[0160] S9.4 Generate interactive visualization reports to help researchers understand the segmentation results;

[0161] S9.5. Fine-tune and improve the method according to application requirements.

[0162] In this embodiment, deep learning technology, combined with multi-channel image data such as DAPI and SpGreen, enables precise automatic segmentation of cell nuclei and cytoplasm regions. By introducing attention mechanisms and multi-scale feature fusion, the challenges of cell adhesion and overlap are effectively addressed. Simultaneously, cell boundaries are optimized through distance transformation and conditional random field (CRF), improving segmentation accuracy. Subsequent cell counting and co-expression analysis functions allow users to accurately assess cell distribution and co-expression of different fluorescence channels. The segmentation results and data visualization reports provided by this method offer efficient and accurate analytical tools for cell biology research, promoting the automation and refinement of cell image analysis.

[0163] Working principle:

[0164] like Figure 1As shown, this cell segmentation method based on multi-channel fluorescence feature fusion ensures the consistency of images across channels through data preprocessing and image normalization. By loading data from four channels—DAPI, SpGreen, SpGold, and CY5—noise removal and uniform image size adjustment are performed to eliminate brightness differences between channels and spatial alignment is carried out. The images are further refined through local contrast enhancement and the CLAHE algorithm to improve the visibility of cell regions. Images from all channels are stitched together into a composite image, which serves as the input for subsequent deep learning models, laying the foundation for accurate cell segmentation.

[0165] Using a deep convolutional neural network (CNN) and a two-stream structure, the model processes image data from the DAPI channel (nucleus) and the SpGreen channel (cytoplasm) respectively. The DAPI channel mainly provides spatial information of the nucleus, while the SpGreen channel provides information of the cytoplasmic region. Each stream uses a convolutional neural network (such as ResNet or VGG) to extract deep features, and a channel attention mechanism is introduced to focus on the high signal region of DAPI to enhance attention to the nucleus. Combined with multi-scale feature fusion of pyramid network (FPN), the spatial information of the nucleus and cytoplasm is further optimized, enabling the model to capture the morphological features of cells at different scales.

[0166] In the segmentation of cell nucleus and cytoplasm, the method utilizes the DAPI signal to generate an accurate nuclear mask and refines the boundary of the nuclear region through adaptive thresholding and watershed algorithms. In the SpGreen channel, the network predicts the probability map of the cytoplasmic region and optimizes the cytoplasmic boundary through conditional random field (CRF) to repair blurred areas. Combining the DAPI mask and the cytoplasmic region map, the cytoplasmic boundary is optimized through distance transformation and ensures that the cytoplasmic region does not exceed the fluorescence range of SpGreen, thereby obtaining accurate single-cell boundaries. Morphological repair and noise removal operations further improve cell morphology and ensure segmentation accuracy.

[0167] By evaluating and optimizing the segmentation results, the method uses indicators such as IoU (Intersection over Union) and Dice coefficient to ensure the accuracy of cell segmentation. Through hyperparameter tuning and model optimization, the model's performance on unknown data is improved. Based on the segmentation results, cell counting is performed and the co-expression of fluorescence channels is analyzed. Statistical analysis (such as chi-square test) is used to evaluate the interaction of different channels. Visual reports and data export functions are generated, enabling researchers to easily view cell segmentation results and share and use the data in subsequent research and analysis.

[0168] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0169] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0170] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

Claims

1. A cell segmentation method based on multi-channel fluorescence feature fusion, characterized in that, Includes the following steps: S1. Standardization, noise removal, and spatial alignment of multi-channel image data; S2. Design of a two-stream structure and attention mechanism model based on deep convolutional neural networks; S3. Use the DAPI signal to generate a precise nuclear mask and perform thresholding. S4. Use the SpGreen channel to extract cytoplasmic regions and perform probability map prediction; S5. Combine the DAPI mask with the cytoplasmic region map to perform precise mask mapping; S6. Cell boundary optimization and noise removal, restoring cell morphology and segmentation accuracy; S7. Segmentation accuracy evaluation and model optimization based on indicators such as IoU and Dice coefficient; S8. Perform cell counting and co-expression analysis to assess the interactions between different channels; S9. Generate a visual report of the segmentation results and export the data to support subsequent analysis and sharing.

2. The cell segmentation method based on multi-channel fluorescence feature fusion according to claim 1, characterized in that: In step S1, the standardization, noise removal, and spatial alignment of multi-channel image data: S1.1 Load all four fluorescence channel data from the imaging device, ensuring that the image data format of each channel is consistent, perform noise reduction and standardization processing, and unify the image size and resolution; S1.

2. Normalize the image for each channel to eliminate brightness differences between different channels; a commonly used normalization method is Z-score standardization. ； in, The grayscale value for each pixel, The average value of the pixels. The standard deviation of the pixels; S1.3 To reduce processing complexity, the image is cropped into regions of interest and spatially aligned to ensure that all image channels are spatially consistent. S1.

4. Perform local contrast enhancement on the image of each channel, using the CLAHE (Contrast Limiting Adaptive Histogram Equalization) algorithm to increase image details. S1.

5. Combine the information from all channels to form a multi-channel composite image, ready for input into the network.

3. The cell segmentation method based on multi-channel fluorescence feature fusion according to claim 1, characterized in that: In step S2, the design of the two-stream structure and attention mechanism model based on deep convolutional neural networks: S2.1 The network adopts a dual-stream structure: one stream processes DAPI channel data, i.e., cell nuclear information, and the other stream processes SpGreen channel data, i.e., cytoplasmic information; S2.2 Each stream uses a convolutional neural network (CNN) to extract high-level features for its respective channel, typically using ResNet or VGG architecture as the base network; S2.

3. Introduce an attention mechanism into the network to enhance its focus on important regions. Use a channel attention mechanism to improve the focus on DAPI high-signal regions: ； in, The weight matrix to be learned. As input features, For bias terms, Use the Sigmoid activation function; S2.

4. Combining multi-scale features, contextual information is fused through a pyramid network, enabling the network to capture cell information at different scales; S2.5 The network output is a nuclear mask, representing the spatial location of the cell nucleus.

4. The cell segmentation method based on multi-channel fluorescence feature fusion according to claim 1, characterized in that: In step S3, which involves generating a precise nuclear mask using the DAPI signal and then performing thresholding: S3.

1. The spatial location of the cell nucleus is learned through a deep network. The DAPI channel is used as the main input to generate a cell nucleus mask. An adaptive thresholding method is used to separate the cell nucleus region. S3.2 Extract the core region from the DAPI channel image. These regions correspond to the center of the cell nucleus. Further refine the boundary of the nucleus region using the watershed algorithm. S3.3 Optimize the threshold setting based on the local brightness information of the image to dynamically adjust the extraction range of cell nuclei; S3.

4. Morphological manipulation is used to remove small noises in the nuclear mask, further enhancing the accuracy of the nuclear region. S3.

5. The processed nuclear mask is passed to the subsequent steps for cytoplasmic boundary extraction.

5. The cell segmentation method based on multi-channel fluorescence feature fusion according to claim 1, characterized in that: In step S4, extracting cytoplasmic regions using the SpGreen channel and performing probability map prediction: S4.

1. Based on the SpGreen channel, a convolutional neural network is used to extract the probability map of the cytoplasmic region; S4.2 The convolutional layer extracts deep features from the SpGreen channel to obtain a multi-layer feature representation of the cytoplasm; S4.3 Obtain the predicted probability map of the cytoplasm using the Sigmoid activation function: ； in, It is a convolution kernel used for cytoplasmic region prediction. It is the input SpGreen feature. It is the Sigmoid activation function; S4.

4. Further optimize the boundaries of the cytoplasmic region using Conditional Random Field (CRF) to correct ambiguous regions between cells; S4.5 Use the CRF-optimized cytoplasmic probability map as input for subsequent steps.

6. The cell segmentation method based on multi-channel fluorescence feature fusion according to claim 1, characterized in that, In step S5, combining the DAPI mask with the cytoplasmic region map, precise mask mapping is performed: S5.1 Mapping the nuclear mask of the DAPI channel with the cytoplasmic probability map of the SpGreen channel; S5.

2. Based on the nuclear mask, perform distance transformation to generate a probability graph of the cytoplasmic boundary: ； in, It is a point on the nuclear mask. It is the result of distance transformation. It refers to each pixel in space. Represents Euclidean distance; S5.

3. By using constraint conditions, ensure that the cytoplasmic boundary does not exceed the SpGreen fluorescence region; S5.4 Optimize the mask based on the distance transformation results, mapping the nuclear mask to the cytoplasm region to form a more accurate cell boundary; S5.5 The final cell boundary represents the accurate cell location.

7. The cell segmentation method based on multi-channel fluorescence feature fusion according to claim 1, characterized in that, In step S6, cell boundary optimization and noise removal, cell morphology and segmentation accuracy are restored: S6.1 Denoise the segmentation results, remove areas with unclear boundaries, and further repair cell edges; S6.2 Correcting the shape of cell boundaries through morphological repair techniques, namely expansion and corrosion operations; S6.3 Apply multi-level image restoration methods, including artifact removal and enhancing the accuracy of true boundaries; S6.

4. By adjusting the morphological information of each cell, ensure that the cell structure conforms to biological characteristics; S6.5 Output the corrected cell boundary image as the segmentation result.

8. The cell segmentation method based on multi-channel fluorescence feature fusion according to claim 1, characterized in that, In step S7, segmentation accuracy evaluation and model optimization based on metrics such as IoU and Dice coefficient: S7.1 Select the IoU and Dice coefficient indicators to evaluate the accuracy of the segmentation results; S7.2 Optimize the hyperparameters of the network model to improve segmentation accuracy; S7.3 Adjust the network structure based on the evaluation results, adding or removing certain layers to improve performance; S7.4 Test the adjusted model on the validation set to ensure its performance on unseen data; S7.5 Output the final segmentation accuracy evaluation result.

9. The cell segmentation method based on multi-channel fluorescence feature fusion according to claim 1, characterized in that, In step S8, cell counting and co-expression analysis are performed to assess the interactions between different channels: S8.1 Count cells based on the segmentation results of cell boundaries and count the number of individual cells; S8.2 Calculate the co-expression rate among different cells by analyzing the overlapping regions of different fluorescence channels; S8.3 Use multi-component analysis to assess the intersection and differences between different channels; S8.4 Statistical methods were used to perform co-expression analysis of cells; S8.5 Generate the final cell count and co-expression analysis report.

10. The cell segmentation method based on multi-channel fluorescence feature fusion according to claim 1, characterized in that, Step S9, generating a visualization report of the segmentation results and exporting the data, supports subsequent analysis and sharing: S9.1 Visualize the segmented image to display each cell and its region; S9.2 Output the cell count and co-expression data for each cell to facilitate subsequent analysis; S9.3 Archive the segmentation results and statistical data, and prepare them for sharing; S9.4 Generate interactive visualization reports to facilitate understanding of the segmentation results; S9.

5. Fine-tune and improve the method according to application requirements.

Images