Method, system, device and medium for recognizing punctate fluorescent signals in a cell nucleus

Abstract

The application discloses a method, system, device and medium for recognizing point fluorescence signals in cell nuclei. The method comprises inputting a pre-processed tissue slice scanning image into a cell nucleus segmentation network model to obtain a binary segmentation mask image; determining a target cell nucleus mask based on the binary segmentation mask image; multiplying a point fluorescence signal channel with the target cell nucleus mask to obtain a multiplication feature result image; inputting the multiplication feature result image into a point fluorescence signal recognition model to obtain a signal point heat map and a preliminary segmentation mask; fusing the signal point heat map, the preliminary segmentation mask and the point fluorescence signal channel to obtain a fusion feature image; inputting the fusion feature image into a signal point instance segmentation model to obtain a plurality of single signal point instances; and determining a target ACD score of the tissue slice scanning image based on the plurality of single signal point instances. The application can improve the accuracy, stability and efficiency of recognizing point fluorescence signals in cell nuclei.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to a method, system, device and medium for recognizing dotted fluorescence signals in the cell nucleus. Background Technology

[0002] With the rapid development of fluorescence labeling and microscopic imaging technologies, fluorescence in situ labeling and imaging of specific proteins or nucleic acid sequences within the cell nucleus have become key technologies in cell biology, pathology, and precision medicine. In particular, the identification and quantitative analysis of discrete, discontinuously distributed point-like fluorescence signals within the cell nucleus (such as fluorescence in situ hybridization signals, aggregation foci of certain proteins, transcriptional active sites, etc.) are of great significance for gene status assessment (such as HER2 gene amplification), viral infection detection, assessment of cellular functional status, and disease prognosis analysis.

[0003] Currently, methods for identifying and analyzing punctate fluorescence signals within cell nuclei mainly fall into two categories. One category is manual visual interpretation and semi-quantitative scoring methods, but these methods heavily rely on the professional experience and subjective judgment of the interpreters, resulting in problems such as strong subjectivity, poor repeatability, low efficiency, limited throughput, and insufficient quantitative accuracy. The other category is automatic analysis methods based on traditional image processing algorithms, but these methods suffer from insufficient environmental adaptability and robustness, limited feature expression capabilities, and weak processing capabilities for overlapping and dense signals when dealing with complex real-world biological samples.

[0004] In summary, existing technologies suffer from low accuracy, stability, and efficiency in identifying point-like fluorescent signals within the cell nucleus. Summary of the Invention

[0005] This application aims to provide a method, system, device, and medium for identifying dotted fluorescent signals in the cell nucleus, which can improve the accuracy, stability, and efficiency of identifying dotted fluorescent signals in the cell nucleus.

[0006] In a first aspect, embodiments of this application provide a method for identifying punctate fluorescence signals within the cell nucleus, the method comprising: Acquire preprocessed tissue section scan images containing dotted fluorescence signal channels and nuclear marker channels; The preprocessed tissue slice scan image is input into the nuclear segmentation network model, and the nuclear marker channel is processed by the nuclear segmentation network model to obtain a binary segmentation mask image. Based on the binary segmentation mask image, the target cell nucleus mask is determined; The dotted fluorescence signal channel is multiplied with the target cell nuclear mask to obtain a multiplication feature result map; The multiplication feature result map is input into the dotted fluorescence signal recognition model to obtain a signal point heatmap and a preliminary segmentation mask; The signal point heatmap, the preliminary segmentation mask, and the dotted fluorescence signal channel are fused to obtain a fused feature map; The fused feature map is input into the signal point instance segmentation model to obtain multiple single signal point instances, and the target ACD score of the tissue slice scan image is determined based on the multiple single signal point instances.

[0007] In some embodiments, the step of inputting the preprocessed tissue slice scan image into a nuclear segmentation network model, and processing the nuclear marker channels through the nuclear segmentation network model to obtain a binary segmentation mask image includes: A semantic segmentation network model is constructed using an encoder-decoder structure. The Dice loss and classification loss are weighted and summed to construct a composite loss function; Based on the composite loss function, the segmentation network model is trained using a training dataset to obtain a cell nucleus segmentation network model. The training dataset contains labels that indicate whether each cell nucleus is valid or invalid. The preprocessed tissue slice scan image is input into the nuclear segmentation network model, and the nuclear marker channels are processed by the nuclear segmentation network model to obtain a binary segmentation mask image.

[0008] In some implementations, determining the target cell nucleus mask based on the binary segmentation mask image includes: Connectivity analysis is performed on the binary segmentation mask to retain the nucleus regions that are classified as valid. Based on the effective cell nucleus region, a target cell nucleus mask is generated.

[0009] In some embodiments, inputting the multiplication feature result map into a point-like fluorescence signal recognition model to obtain a signal point heatmap and a preliminary segmentation mask includes: Two parallel decoding heads are connected after the shared feature encoder to construct a dotted fluorescence signal recognition model; After the multiplication feature result map is input into the shared feature encoder for data processing, the output of the shared feature encoder is decoded by two parallel decoding heads to obtain a signal point heatmap and a preliminary segmentation mask.

[0010] In some embodiments, fusing the signal point heatmap, the preliminary segmentation mask, and the dotted fluorescence signal channel to obtain a fused feature map includes: Upsample the signal point heatmap and the preliminary segmentation mask to obtain a first upsampled result corresponding to the signal point heatmap and a second upsampled result corresponding to the preliminary segmentation mask; The first upsampling result, the second upsampling result, and the dotted fluorescence signal channel are spliced ​​together to obtain the spliced ​​feature. The concatenated features are then fused and dimensionality reduced using a lightweight convolutional module to obtain a fused feature map.

[0011] In some implementations, inputting the fused feature map into a signal point instance segmentation model to obtain multiple single signal point instances includes: Construct a signal point instance segmentation model that includes a self-attention mechanism, a spatial pyramid pooling module, and a label-controlled watershed algorithm; The fused feature map is input into the spatial pyramid pooling module to obtain the output result of the spatial pyramid pooling module; The output of the spatial pyramid pooling module is input into the self-attention mechanism to obtain the signal instance offset vector field and the signal semantic enhancement map. Based on the signal instance offset vector field and the signal semantic enhancement map, the attribution metric matrix is ​​determined; Based on the attribution metric matrix, the label-controlled watershed algorithm is used to perform instance segmentation on the fused feature map to obtain a single signal point instance segmentation mask; Multiple single-signal-point instances are obtained by segmenting the mask based on the single-signal-point instance.

[0012] In some implementations, determining the target ACD score of the tissue slice scan image based on the plurality of single signal point instances includes: Based on the multiple single signal point instances, the target point density and the target cluster ratio are calculated. The target point density and the target cluster ratio are input into the target ACD scoring model to obtain the target ACD score of the tissue slice scan image.

[0013] Secondly, embodiments of this application also provide a system for recognizing dotted fluorescence signals within the cell nucleus, the system comprising: The data acquisition unit is used to acquire preprocessed tissue section scan images containing dotted fluorescence signal channels and nuclear marker channels; The data processing unit is used to input the preprocessed tissue slice scan image into the cell nuclear segmentation network model, and process the cell nuclear marker channels through the cell nuclear segmentation network model to obtain a binary segmentation mask image. The data determination unit is used to determine the target cell nucleus mask based on the binary segmentation mask image; The data multiplication unit is used to multiply the dotted fluorescence signal channel with the target cell nuclear mask to obtain a multiplication feature result map; The signal recognition unit is used to input the multiplication feature result map into the dotted fluorescence signal recognition model to obtain a signal point heatmap and a preliminary segmentation mask. The data fusion unit is used to fuse the signal point heatmap, the preliminary segmentation mask, and the dotted fluorescence signal channel to obtain a fused feature map; The instance segmentation unit is used to input the fused feature map into the signal point instance segmentation model to obtain multiple single signal point instances, and to determine the target ACD score of the tissue slice scan image based on the multiple single signal point instances.

[0014] Thirdly, embodiments of this application also provide an electronic device, including at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, the instructions being executed by the at least one control processor to enable the at least one control processor to perform a method for identifying dotted fluorescent signals in the cell nucleus as described above.

[0015] Fourthly, embodiments of this application also provide a computer-readable storage medium storing computer-executable instructions for causing a computer to perform a method for identifying dotted fluorescent signals within a cell nucleus as described above.

[0016] Compared with the prior art, this application has the following beneficial effects: This application involves acquiring preprocessed tissue section scan images containing dotted fluorescence signal channels and nuclear marker channels; inputting the preprocessed tissue section scan images into a nuclear segmentation network model, which processes the nuclear marker channels to obtain a binary segmentation mask image; determining the target nuclear mask based on the binary segmentation mask image; multiplying the dotted fluorescence signal channels with the target nuclear mask to obtain a multiplication feature result image; inputting the multiplication feature result image into a dotted fluorescence signal recognition model to obtain a signal point heatmap and a preliminary segmentation mask; fusing the signal point heatmap, the preliminary segmentation mask, and the dotted fluorescence signal channels to obtain a fused feature image; inputting the fused feature image into a signal point instance segmentation model to obtain multiple single signal point instances, and determining the target ACD score of the tissue section scan image based on these multiple single signal point instances. Thus, the nuclear segmentation network model can preserve regularly shaped nuclei and suppress segmentation responses to excessively small, dark, irregular, or overly overlapping nuclei, thereby obtaining an accurate and effective target nuclear mask. Based on this accurate and effective target nuclear mask, a dotted fluorescence signal recognition model is used to obtain a signal point heatmap and a preliminary segmentation mask, resulting in relatively accurate signal point heatmaps and preliminary segmentation masks. Finally, the signal point heatmap, preliminary segmentation mask, and dotted fluorescence signal channels are fused, and multiple single signal point instances are obtained through a signal point instance segmentation model, which improves the accuracy, stability, and efficiency of dotted fluorescence signal recognition within the cell nucleus. Based on multiple single signal point instances, the target ACD score of the tissue section scan image is determined, resulting in an accurate ACD score, which can effectively assist in gene status assessment (such as HER2 gene amplification), viral infection detection, cell functional status assessment, and disease prognosis analysis. Attached Figure Description

[0017] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a schematic flowchart of an embodiment of the method for recognizing dotted fluorescence signals in the cell nucleus provided in this application; Figure 2 This is a schematic diagram of multichannel scanning and image registration in the preferred embodiment of the method for recognizing dotted fluorescence signals in the cell nucleus provided in this application; Figure 3 This is a schematic diagram of the effective cell nucleus recognition effect in the preferred embodiment of the method for recognizing dotted fluorescence signals in the cell nucleus provided in this application; Figure 4 This is a schematic diagram illustrating the effect of dotted signal detection and segmentation in the preferred embodiment of the method for identifying dotted fluorescent signals in the cell nucleus provided in this application; Figure 5This is a schematic diagram of an embodiment of the cell nucleus spot fluorescence signal recognition system provided in this application; Figure 6 This is a schematic diagram of the structure of an embodiment of the electronic device provided in this application. Detailed Implementation

[0018] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0019] In the description of this application, the use of terms such as "first," "second," etc., is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or the order of the technical features indicated.

[0020] In the description of this application, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application 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, and therefore should not be construed as a limitation of this application.

[0021] In the description of this application, it should be noted that, unless otherwise explicitly defined, terms such as "setup," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this application in conjunction with the specific content of the technical solution.

[0022] To address the issues of low accuracy, stability, and efficiency in identifying dotted fluorescence signals within the cell nucleus in related technologies, this application proposes a method, system, device, and medium for identifying dotted fluorescence signals within the cell nucleus.

[0023] Reference Figure 1 This application provides a schematic flowchart of a method for identifying punctate fluorescence signals within the cell nucleus. This method is applied to an electronic device, such as a server or a mobile terminal. Figure 1 As shown, the method for identifying the punctate fluorescence signal in the cell nucleus may include the following steps S101 to S107.

[0024] Step S101: Obtain a preprocessed tissue section scan image containing dotted fluorescence signal channels and nuclear marker channels.

[0025] Specifically, input a scanned image of a tissue section labeled with multiplex immunofluorescence (mIHC) (i.e., a pre-processed scanned image of a tissue section). This image can simultaneously contain the nucleus-specific protein or nucleic acid sequence (discrete, non-continuously distributed dotted fluorescent signals in the nucleus, i.e., dotted fluorescent signal channels) and the spectral information of nuclear markers (DAPIs) (i.e., nuclear marker channels).

[0026] Step S102: Input the preprocessed tissue slice scan image into the cell nuclear segmentation network model, and process the cell nuclear marker channels through the cell nuclear segmentation network model to obtain a binary segmentation mask image.

[0027] In some implementations, a segmentation network model can be constructed using a semantic segmentation network with an encoder-decoder structure. A composite loss function is constructed by weighted summing of the Dice loss and classification loss. Based on this composite loss function, the segmentation network model is trained using a training dataset containing valid or invalid labels for each cell nucleus. Preprocessed tissue slice scan images are input into the cell nucleus segmentation network model, which processes the cell nucleus marker channels to obtain a binary segmentation mask. In this way, the cell nucleus segmentation network model can preserve regularly shaped cell nuclei and suppress segmentation responses to excessively small, dark, irregular, or excessively overlapping cell nuclei, thereby obtaining an accurate binary segmentation mask.

[0028] Specifically, a semantic segmentation network with an encoder-decoder structure (such as U-Net++ or its variants) is used as the backbone network model (i.e., the segmentation network model). The encoder (such as ResNet, EfficientNet, etc.) is used to extract multi-scale features, and the decoder gradually recovers spatial details through skip connections to achieve pixel-level cell kernel segmentation.

[0029] A weighted sum of the Dice loss and classification loss is constructed to create a composite loss function. Based on this composite loss function, the segmentation network model is trained using the training dataset to obtain a nucleus segmentation network model. Tissue slice scan images (i.e., nucleus marker channels) are input into the trained nucleus segmentation network model, which processes individual nucleus channels (DAPI) in the tissue slice scan images. The nucleus segmentation network model outputs a binary segmentation mask of the same size as the input, where foreground pixels (value 1) represent valid nucleus regions, and background pixels (value 0) represent non-nucleus regions or invalid nucleus regions.

[0030] Step S103: Determine the target cell nucleus mask based on the binary segmentation mask image.

[0031] In some implementations, connected component analysis can be performed on the binary segmentation mask image to retain the nucleus regions classified as valid. Based on these valid nucleus regions, a target nucleus mask is generated. This retains only the nucleus regions classified as valid, laying a solid data foundation for subsequent point fluorescence signal recognition model processing, thereby improving the efficiency and accuracy of the model's output.

[0032] Specifically, connected component analysis is performed on the binary segmentation mask map output by the network, and only the cell nucleus regions classified as "valid" are retained to generate the final valid cell nucleus mask (i.e., the target cell nucleus mask), which is used for region constraints in subsequent steps.

[0033] It should be noted that the above connectivity component analysis can use four-connected regions or eight-connected regions. Regenerating the final effective nucleus mask from the "effective" nucleus region can be done using techniques known to those skilled in the art, which will not be specifically described in this embodiment.

[0034] Step S104: Multiply the dotted fluorescence signal channel with the target cell nuclear mask to obtain the multiplication feature result map.

[0035] Specifically, by multiplying the target protein fluorescence channel image (i.e., the dotted fluorescence signal channel) with the effective cell nuclear mask, the analysis is strictly limited to the effective nuclear region, greatly reducing background interference and laying a good data foundation for the subsequent dotted fluorescence signal recognition model processing.

[0036] Step S105: Input the multiplication feature result map into the point fluorescence signal recognition model to obtain the signal point heatmap and preliminary segmentation mask.

[0037] In some implementations, a point-like fluorescence signal recognition model can be constructed by connecting two parallel decoding heads after a shared feature encoder. The multiplied feature result image is input to the shared feature encoder for data processing, and then the output of the shared feature encoder is decoded by the two parallel decoding heads to obtain a signal point heatmap and a preliminary segmentation mask. In this way, obtaining a signal point heatmap and a preliminary segmentation mask through the point-like fluorescence signal recognition model can yield relatively accurate signal point heatmaps and preliminary segmentation masks.

[0038] Specifically, a shared feature encoder is used, followed by two parallel decoders, forming a dual-task learning network (i.e., a point fluorescence signal recognition model). The multiplied feature result image is input into the shared feature encoder for data processing. The output of the shared feature encoder is then decoded by the two parallel decoders. One decoder is a signal point heatmap regression head, which outputs a signal point heatmap where the value of each pixel represents the probability (between 0 and 1) that the location belongs to the center of a signal point. This solves the problem of distinguishing weak signals from the background. The other decoder is a preliminary signal point instance segmentation head, which performs coarse pixel-level segmentation on each potential signal point, generating a preliminary mask. However, at this stage, overlapping signals may still be segmented as a single entity.

[0039] It should be noted that the aforementioned shared feature encoder may employ an encoder structure known to those skilled in the art, and this embodiment will not describe it in detail. Similarly, the aforementioned decoding head may employ a decoder known to those skilled in the art, and this embodiment will not describe or limit it in detail.

[0040] Step S106: Fuse the signal point heatmap, the preliminary segmentation mask, and the dotted fluorescence signal channel to obtain a fused feature map.

[0041] In some implementations, the signal point heatmap and the preliminary segmentation mask can be upsampled to obtain a first upsampled result corresponding to the signal point heatmap and a second upsampled result corresponding to the preliminary segmentation mask. The first upsampled result, the second upsampled result, and the point-like fluorescence signal channels are then concatenated to obtain the concatenated features. These concatenated features are then fused and dimensionality-reduced using a lightweight convolutional module to obtain a fused feature map. Thus, instance segmentation using the fused feature map enables accurate separation of individual signal points.

[0042] Specifically, firstly, the signal point heatmap and the preliminary segmentation mask are upsampled to the same spatial size as the original target protein fluorescence channel image (i.e., the dotted fluorescence signal channel); then, the three (including the first upsampling result corresponding to the signal point heatmap, the second upsampling result corresponding to the preliminary segmentation mask, and the original target protein fluorescence channel image) are concatenated along the channel dimension; finally, a lightweight convolutional module is used to fuse and reduce the dimensionality of the concatenated features to generate a fused feature map containing multi-source information, which serves as the input for the subsequent attention mechanism and spatial pyramid pooling module.

[0043] It should be noted that the lightweight convolution module in this embodiment can be a lightweight convolution module known to those skilled in the art, such as depthwise separable convolution, which will not be specifically described in this embodiment.

[0044] Step S107: Input the fused feature map into the signal point instance segmentation model to obtain multiple single signal point instances, and determine the target ACD score of the tissue slice scan image based on the multiple single signal point instances.

[0045] In some implementations, a signal point instance segmentation model can be constructed by including a self-attention mechanism, a spatial pyramid pooling module, and a label-controlled watershed algorithm. The fused feature map is input to the spatial pyramid pooling module to obtain its output. The output of the spatial pyramid pooling module is then input to the self-attention mechanism to obtain a signal instance offset vector field and a signal semantic enhancement map. Based on the signal instance offset vector field and the signal semantic enhancement map, an attribution metric matrix is ​​determined. According to the attribution metric matrix, the label-controlled watershed algorithm is used to segment the fused feature map, resulting in a single-signal-point instance segmentation mask. Multiple single-signal-point instances are obtained based on the single-signal-point instance segmentation mask. Thus, by using the label-controlled watershed algorithm to segment the fused feature map based on the attribution metric matrix, an accurate single-signal-point instance segmentation mask can be obtained, thereby improving the accuracy, stability, and efficiency of identifying point-like fluorescence signals within the cell nucleus.

[0046] Specifically, the fused feature map is input into the signal point instance segmentation model. First, the fused feature map is processed by the spatial pyramid pooling module to obtain the output of the spatial pyramid pooling module. Then, the output of the spatial pyramid pooling module is processed by the self-attention mechanism to obtain the signal instance offset vector field and the signal semantic enhancement map. Alternatively, the fused feature map can be input into the self-attention mechanism and the spatial pyramid pooling module separately to obtain the signal instance offset vector field output by the self-attention mechanism and the signal semantic enhancement map output by the spatial pyramid pooling module, respectively.

[0047] The self-attention mechanism enables the signal point instance segmentation model to model long-range spatial relationships between signal points. For example, it can learn that "when two highlighted regions are very close and have an intensity dip in the middle, they are likely two independent signal points rather than a large irregular signal." The spatial pyramid pooling module captures contextual information at multiple scales through convolutions with different dilation rates, which is crucial for determining whether a signal point is a "single large point" or an "aggregation of multiple small points."

[0048] After learning the signal instance offset vector field and the signal semantic enhancement map, feature processing is performed by combining post-processing algorithms (such as the Marker-controlled Watershed algorithm) to accurately segment high-brightness points. Based on the learned shape and context prior, it predicts which point fluorescence signal center (i.e., candidate center) each pixel in the overlapping region should belong to, thereby achieving intelligent de-clustering of clusters and finally outputting a series of accurately labeled and separated single signal point instances.

[0049] In some implementations, the target point density and target cluster proportion can be calculated based on multiple single signal point instances; the target point density and target cluster proportion are then input into a target ACD scoring model to obtain the target ACD score of the tissue section scan image. This allows for accurate ACD scoring, effectively assisting in gene status assessment (such as HER2 gene amplification), viral infection detection, cell function status assessment, and disease prognosis analysis.

[0050] Specifically, the total number of signal points and the total number of effective cell nuclei are counted across multiple single-signal point instances. The target point density is obtained by dividing the total number of signal points by the total number of effective cell nuclei. When the centroid distance between two or more signal points is less than N times (e.g., 1.5 times) of their average radius, these signal points are determined to belong to the same cluster. The total number of signal points located in the cluster is counted, and then the proportion of the target cluster is calculated using the following formula: ; The target point density and target cluster ratio are then input into the target ACD scoring model to obtain the target ACD score of the tissue slice scan image.

[0051] To facilitate understanding by those skilled in the art, a set of preferred embodiments is provided below: Currently, methods for identifying and analyzing punctate fluorescence signals within the cell nucleus mainly fall into two categories: 1. Manual visual interpretation and semi-quantitative scoring method.

[0052] This is a traditional method long used in clinical and basic research. Operators observe samples using a fluorescence microscope, manually counting, comparing, and classifying punctate signals in multiple cell nuclei within the field of view according to established clinical or experimental standards (e.g., the HER2 FISH scoring criteria in the ASCO / CAP guidelines), ultimately providing a qualitative or semi-quantitative score (e.g., positive / negative, or 0, 1+, 2+, and 3+ grading). This method heavily relies on the professional experience and subjective judgment of the interpreter, and therefore has the following inherent drawbacks: (1) High subjectivity and poor repeatability: Different observers, or even the same observer at different times, have difficulty in completely unifying the judgment criteria for signal strength, boundary definition, overlapping signal segmentation and background fluorescence interference, resulting in significant intra-observer and inter-observer differences in the scoring results, which affects the reliability of the conclusions.

[0053] (2) Inefficient and limited throughput: The manual counting and scoring process is cumbersome and time-consuming, making it difficult to meet the needs of large-scale sample screening or high-throughput experiments, thus becoming a bottleneck for clinical diagnosis and scientific research analysis.

[0054] (3) Insufficient quantitative accuracy: Manual interpretation can usually only perform rough counting or intensity grading, and cannot accurately quantify the more complex morphological features of the signal (such as the spatial distribution, aggregation and shape heterogeneity of signal points, etc.), which may result in the loss of information containing important biological significance.

[0055] 2. Automatic analysis method based on traditional image processing algorithms.

[0056] To overcome the shortcomings of manual methods, several automated or semi-automated analysis software programs based on computer vision and traditional image processing algorithms have been developed. These solutions typically follow a workflow of "image preprocessing (e.g., noise reduction, background correction) → cell nucleus segmentation → signal point detection (e.g., intensity threshold-based segmentation, morphological manipulation) → feature extraction and counting → simple rule determination." However, these techniques have the following limitations when dealing with complex real-world biological samples: (1) Insufficient environmental adaptability and robustness: Its performance is highly dependent on pre-set fixed parameters (such as threshold, filter kernel size, etc.). Differences in biological sample preparation, uneven fluorescence staining, fluctuations in imaging conditions (such as exposure time, focal length, etc.) and complex background noise (such as non-specific staining, cell autoluminescence and impurity interference, etc.) can all cause the fixed parameters to fail, resulting in missed detection, false detection or segmentation errors.

[0057] (2) Limited feature representation ability: Traditional algorithms mainly rely on low-level image features designed manually (such as intensity, size and roundness), which are difficult to effectively represent and learn the high-level abstract features of point signals in complex backgrounds (such as the association with specific subnuclear structures, the distinction between weak specific signals and strong background noise).

[0058] (3) Weak ability to process overlapping and dense signals: When signal points in the cell nucleus are dense or partially overlapping, traditional segmentation methods based on connected component analysis are difficult to accurately separate individual signal points, resulting in serious distortion of the count.

[0059] In summary, existing related technologies can be categorized as schemes that primarily rely on parameter-sensitive traditional image processing algorithms, combined with simple rules, for the automatic identification and scoring of punctate fluorescence signals within cell nuclei. Their fundamental shortcomings stem from the algorithms themselves: 1) They lack sufficient adaptability and generalization ability for complex and variable biological image data; 2) Their feature extraction and pattern recognition capabilities are insufficient, making it difficult to match the performance of experienced experts; 3) The various stages of the overall analysis process (segmentation, detection, and scoring) are relatively isolated, failing to achieve end-to-end optimization. These problems result in the accuracy, stability, and efficiency of related automated methods still needing improvement in practical applications, especially when facing scenarios with high sample heterogeneity and inconsistent image quality. Therefore, there is an urgent need for a novel automatic identification and scoring method that can adaptively learn complex image features, possess strong robustness to changes in imaging conditions, and more accurately simulate the comprehensive judgment logic of experts.

[0060] To address the problems existing in the aforementioned related technologies, this embodiment provides a method for identifying and scoring punctate fluorescence signals within the cell nucleus based on artificial intelligence deep learning. This method, by constructing a multi-level, modular deep learning analysis workflow, achieves fully automated, high-precision, and highly robust analysis from raw fluorescence micrographs to standardized scoring. The specific technical solution of this embodiment includes the following: I. Image Processing.

[0061] The input consists of scanned images of tissue sections labeled with multiplex immunofluorescence (mIHC) (i.e., pre-processed scanned images of tissue sections). These images can simultaneously contain nucleus-specific protein or nucleic acid sequences (presented as discrete, discontinuous punctate fluorescent signals within the nucleus, i.e., punctate fluorescent signal channels) and spectral information of nuclear markers (DAPIs) (i.e., nuclear marker channels). Figure 2 As shown, Figure 2 This is a schematic diagram of multi-channel scanning and image registration. Channel 1 is the red fluorescence channel: a dotted fluorescence signal channel, which uses H2AX to label dotted signals in the nucleus; Channel 2 is the blue fluorescence channel: a nuclear marker channel, which uses DAPI to label the cell nucleus.

[0062] II. Identification of intranuclear punctate fluorescence signals.

[0063] 1. Cell nucleus quality screening and segmentation based on deep learning.

[0064] The goal of this step is to accurately identify and screen out the effective cell nuclei suitable for analysis, while automatically excluding apoptotic cell nuclei, nuclear debris, cell nuclei that are stained too lightly or too darkly, and cell nuclei that overlap significantly.

[0065] (1) Network model structure: A semantic segmentation network with an encoder-decoder structure (such as U-Net++ or its variants) is used as the backbone network model (i.e., the segmentation network model). The encoder (such as ResNet, EfficientNet, etc.) is used to extract multi-scale features, and the decoder gradually recovers spatial details through skip connections to achieve pixel-level cell kernel segmentation.

[0066] (2) Input and Output: The input is a tissue slide scan image (i.e., the nuclear marker channel). The trained nuclear segmentation network model processes the individual nuclear channel (DAPI) image in the tissue slide scan image. The nuclear segmentation network model outputs a binary segmentation mask image of the same size as the input, where the foreground pixels (value 1) represent qualified nuclear regions and the background pixels (value 0) represent non-nuclear regions or invalid nuclear regions.

[0067] (3) AI definition and implementation of "effective cell nucleus": Traditional methods rely on fixed thresholds such as area and intensity for screening. This embodiment incorporates the "effectiveness" judgment into the learning objective of the network model.

[0068] Training data preparation: When annotating the training data, the annotator not only delineates the outline of the cell nucleus, but also labels each cell nucleus as "valid" or "invalid" according to morphological criteria (such as integrity, staining uniformity, and appropriate size).

[0069] Loss function design: A composite loss function is used. .in, Dice loss is used to optimize the overall segmentation accuracy of cell nuclei. An additional classification loss is applied, based on each connected component (i.e., each predicted nucleus instance). During segmentation, the network model outputs an instance validity probability for each predicted nucleus instance through a global pooling layer and a fully connected layer. This loss function forces the segmentation network model to learn to prioritize identifying and preserving regularly shaped nuclei during segmentation, while suppressing segmentation responses to excessively small, dark, irregular, or overly overlapping nuclei, thus training the nucleus segmentation network model.

[0070] (4) Post-processing: Connectivity analysis is performed on the binary segmentation mask map output by the network model. Only the cell nucleus regions classified as "valid" are retained to generate the final valid cell nucleus mask (i.e., the target cell nucleus mask), which is used for region constraints in subsequent steps. The "valid" cell nucleus regions are as follows: Figure 3 As shown, Figure 3The area within the white circle represents the identified valid nucleus region. Connectivity analysis can be performed using four-connected or eight-connected regions. The "valid" nucleus region is then used to regenerate the final valid nucleus mask, which can be achieved using techniques known to those skilled in the art; this embodiment will not describe these techniques in detail.

[0071] 2. Point signal detection and preliminary segmentation based on dual-task learning.

[0072] This step involves identifying and initially locating dotted fluorescence signals within the effective cell nucleus region provided in the previous step.

[0073] (1) Network structure: A shared feature encoder is used, followed by two parallel decoders to form a dual-task learning network (i.e., a dotted fluorescence signal recognition model). The shared feature encoder adopts an encoder structure known to those skilled in the art, and will not be described in detail in this embodiment.

[0074] The first decoding head is a signal point heatmap regression head. This head outputs a heatmap, where the value of each pixel represents the probability (between 0 and 1) that the location belongs to the center of the signal point, which solves the problem of distinguishing weak signals from the background.

[0075] The second decoding head is the initial segmentation head for signal point instances. This head performs coarse pixel-level segmentation on each potential signal point, generating an initial mask, but at this point, overlapping signals may still be segmented as a whole.

[0076] (2) Model input: The result of multiplying the target protein fluorescence channel image (i.e., the dotted fluorescence signal channel) with the effective cell nuclear mask is used as the model input. The dotted fluorescence signal analysis is strictly limited to the effective nuclear region, which greatly reduces background interference and improves the model processing efficiency.

[0077] (3) Training and Advantages: By learning two related tasks simultaneously through a shared encoder, the network model can extract more robust features. The heatmap regression task makes the network more sensitive to the localization of signal centers, while the segmentation task helps to clarify signal boundaries. The two complement each other, significantly improving the model's detection recall and accuracy even with low signal-to-noise ratios.

[0078] 3. Dense signal segmentation based on attention mechanism.

[0079] This step is key to overcoming technical bottlenecks and is specifically designed to process partially overlapping and dense clusters of fluorescent signals, enabling precise separation of individual signal points.

[0080] (1) Model input: The signal point heatmap and preliminary segmentation mask output in the previous step are used as guides, and the high-resolution detail features of the original target protein fluorescence channel image are fused again to obtain the fused feature map.

[0081] Specifically, firstly, the signal point heatmap and the preliminary segmentation mask are upsampled to the same spatial size as the original target protein fluorescence channel image; then, the three (including the first upsampling result corresponding to the signal point heatmap, the second upsampling result corresponding to the preliminary segmentation mask, and the original target protein fluorescence channel image) are concatenated along the channel dimension; finally, a lightweight convolutional module is used to fuse and reduce the dimensionality of the concatenated features to generate a fused feature map containing multi-source information, which serves as the input for the subsequent attention mechanism and spatial pyramid pooling module.

[0082] It should be noted that the lightweight convolution module in this embodiment can be a lightweight convolution module known to those skilled in the art, such as depthwise separable convolution, which will not be specifically described in this embodiment.

[0083] (2) Model Construction: The signal point instance segmentation model includes a self-attention mechanism, a spatial pyramid pooling (ASPP) module, and a label-controlled watershed algorithm. Specifically, the fused feature map is input into the signal point instance segmentation model. First, the fused feature map is processed by the spatial pyramid pooling module to obtain the output of the spatial pyramid pooling module. Then, the output of the spatial pyramid pooling module is processed by the self-attention mechanism to obtain the signal instance offset vector field and the signal semantic enhancement map. Alternatively, the fused feature map can be input into the self-attention mechanism and the spatial pyramid pooling module respectively to obtain the signal instance offset vector field output by the self-attention mechanism and the signal semantic enhancement map output by the spatial pyramid pooling module.

[0084] The self-attention mechanism enables the signal point instance segmentation model to model long-range spatial relationships between signal points. For example, it can learn that "when two highlighted regions are very close and have an intensity dip in the middle, they are likely two independent signal points rather than a large irregular signal." The ASPP module captures contextual information at multiple scales through convolutions with different dilation rates, which is crucial for determining whether a signal point is a "single large point" or an "aggregation of multiple small points."

[0085] After learning the signal instance offset vector field and the signal semantic enhancement map, feature processing is performed by combining post-processing algorithms (such as the Marker-controlled Watershed algorithm) to accurately segment high-spot points. Based on the previously learned shape and context priors, it predicts which point fluorescence signal center (i.e., candidate center) each pixel in the overlapping region should belong to, thereby achieving intelligent de-clustering of clusters and finally outputting a series of accurately labeled and separated single signal point instances.

[0086] The specific implementation steps of the post-processing algorithm include: (1) Based on the signal instance offset vector field, the convergence center (i.e. candidate center) of all signal instances is identified by Hough Voting or clustering methods, and used as the markers / seeds of the watershed algorithm; at the same time, the predicted offset vector of each pixel is obtained from the signal instance offset vector field.

[0087] (2) For each non-background pixel in the image, a theoretical pointing vector is constructed based on its pixel coordinates and the coordinates of all (or neighboring) candidate centers. Constructing the pointing vector using coordinates is a technique known to those skilled in the art, and will not be described in detail in this embodiment.

[0088] (3) Using the cosine similarity formula, the consistency between the predicted offset vector of each pixel and the pointing vector pointing to each candidate center is calculated to obtain the geometric consistency score.

[0089] (4) Extract the high-dimensional feature vector of each pixel and the feature vector of each candidate center position from the signal semantic enhancement map; calculate the feature similarity between the two (such as cosine similarity or the inverse of Euclidean distance) to obtain the semantic similarity score.

[0090] (5) A weighted fusion method is used to combine the geometric consistency score and the semantic similarity score to generate an attribution metric matrix. This step actually defines a "generalized distance", that is, the higher the metric, the "closer" the distance.

[0091] (6) The label-controlled watershed algorithm based on the attribution metric matrix can accurately separate single signal point instances. Specifically, the fused feature map is used as an "energy topography map", and the "attribution metric" obtained in step (5) is inverted (or transformed by 1 - Metric) so that the candidate center is located at the bottom of the "basin", while the area with low attribution (i.e., uncertain which candidate center it belongs to) is located at the "ridge". The label points determined in step (1) are used as the water injection source, and the label-controlled watershed algorithm is executed on the above "energy topography map". The algorithm will automatically determine the best attribution of the pixel according to the guidance of the attribution gradient, thereby generating an accurate single signal point instance segmentation mask in the area with severe adhesion.

[0092] (7) Finally, based on the segmentation mask of single signal point instances, a series of precisely marked and separated single signal point instances can be obtained, each instance having its own contour, area, integrated optical density and other attributes.

[0093] It should be noted that the example obtained from the mask is processed using techniques known to those skilled in the art, and this embodiment will not describe it in detail.

[0094] 4. Multi-feature extraction and dynamic scoring engine.

[0095] This step calculates key parameters (including target point density and target cluster proportion) based on the separated single signal point instances, and then inputs the key parameters into the target ACD scoring model to predict the final ACD Score (i.e., target ACD score).

[0096] The key parameter calculation formulas include: (1) Point density: Total number of signal points / Total number of effective cell nuclei, unit: points / cell.

[0097] (2) Cluster proportion: First, define “cluster”: When the centroid distance between two or more signal points is less than N times (e.g., 1.5 times) of their average radius, these points are determined to belong to the same cluster.

[0098] .

[0099] Other auxiliary features, such as average point area and average point intensity, are used for quality control.

[0100] Dynamic scoring logic: The scoring is not a simple "if-else" rule, but rather uses a dynamic decision model based on fuzzy logic or a lightweight classifier to predict the score. The dynamic decision model can be a model known to those skilled in the art, such as a Bayesian classifier or a random forest, but this embodiment does not describe or limit it in detail.

[0101] The model input vector is: [point density, cluster proportion]. Model training involves using a large amount of expert-annotated sample data with the final ACD score (its point density and cluster proportion are obtained from the aforementioned key parameter calculation formulas) to train the dynamic decision model, resulting in the target ACD scoring model. Based on the target point density and target cluster proportion, the target ACD scoring model is used to classify cell nuclei, obtaining the results of point signal detection and segmentation. A schematic diagram of the point signal detection and segmentation effect can be found in [reference needed]. Figure 4 , Figure 4 The corresponding scoring results are shown in Table 1. Figure 4 Medium-dark green represents 0 points, green represents 1 point, light green represents 2 points, white represents 3 points, and light blue represents 4 points. The ACD scoring criteria are (0 to 4 points): 0 points: No staining, or less than 1 positive spot per 10 cells; 1 point: 1 to 3 positive spots per cell; 2 points: 4 to 9 positive spots per cell, with no positive spot clusters or very few positive spot clusters; 3 points: 10 to 15 positive spots per cell, and / or the percentage of clustered positive spots <10%; 4 points: more than 15 positive spots per cell, and / or the percentage of clustered positive spots >10%.

[0102] This objective ACD scoring model can learn the nonlinear decision boundaries and trade-offs in expert scoring (e.g., whether to favor a score of 2 or 3 when the point density is 9.5 points / cell and the cluster percentage is 8%). It simulates the comprehensive judgment of human experts better than fixed threshold rules, making the scoring more continuous and robust.

[0103] Table 1 shows the core scoring results.

[0104] Batch output: The system performs the above analysis on each field of view in a whole or batch of tissue slide scan images, automatically summarizes the data of all valid cell nuclei, calculates the global average point density and cluster ratio, and obtains the final ACD Score of the tissue slide scan image through a dynamic scoring engine, and generates a report containing detailed counts, parameters and visualization results.

[0105] Reference Figure 5 This application also provides a system for identifying dotted fluorescence signals within the cell nucleus. This system includes a data acquisition unit 501, a data processing unit 502, a data determination unit 503, a data multiplication unit 504, a signal recognition unit 505, a data fusion unit 506, and an instance segmentation unit 507, wherein: The data acquisition unit 501 is used to acquire preprocessed tissue section scan images containing dotted fluorescence signal channels and nuclear marker channels; The data processing unit 502 is used to input the preprocessed tissue slice scan image into the cell nuclear segmentation network model, and process the cell nuclear marker channels through the cell nuclear segmentation network model to obtain a binary segmentation mask image. The data determination unit 503 is used to determine the target cell nucleus mask based on the binary segmentation mask image; The data multiplication unit 504 is used to multiply the dotted fluorescence signal channel with the target cell nuclear mask to obtain the multiplication feature result map; The signal recognition unit 505 is used to input the multiplication feature result map into the dotted fluorescence signal recognition model to obtain a signal point heatmap and a preliminary segmentation mask. The data fusion unit 506 is used to fuse the signal point heatmap, the preliminary segmentation mask and the dotted fluorescence signal channel to obtain a fused feature map; The instance segmentation unit 507 is used to input the fused feature map into the signal point instance segmentation model to obtain multiple single signal point instances, and to determine the target ACD score of the tissue slice scan image based on the multiple single signal point instances.

[0106] It should be noted that since the identification system for intranuclear dotted fluorescence signals in this embodiment is based on the same inventive concept as the above-described method for identifying intranuclear dotted fluorescence signals, the corresponding content in the method embodiment is also applicable to this system embodiment, and will not be described in detail here.

[0107] Reference Figure 6 This application also provides an electronic device, which includes: At least one memory; At least one processor; At least one program; The program is stored in memory, and the processor executes at least one program to implement the method for recognizing intranuclear punctate fluorescence signals described above in this disclosure.

[0108] This electronic device can be any smart terminal, including mobile phones, tablets, personal digital assistants (PDAs), and in-vehicle computers.

[0109] The electronic devices according to embodiments of this application will now be described in detail.

[0110] The processor 1600 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this disclosure. The memory 1700 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1700 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1700 and is called and executed by the processor 1600 to execute the method for recognizing dotted fluorescence signals in the cell nucleus according to the embodiments of this disclosure.

[0111] The input / output interface 1800 is used to implement information input and output. The communication interface 1900 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 2000 transmits information between various components of the device (e.g., processor 1600, memory 1700, input / output interface 1800, and communication interface 1900); The processor 1600, memory 1700, input / output interface 1800 and communication interface 1900 are connected to each other within the device via bus 2000.

[0112] This disclosure also provides a storage medium, which is a computer-readable storage medium storing computer-executable instructions for causing a computer to perform the above-described method for identifying dotted fluorescent signals within the cell nucleus.

[0113] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0114] The embodiments described in this disclosure are for the purpose of more clearly illustrating the technical solutions of this disclosure and do not constitute a limitation on the technical solutions provided by this disclosure. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by this disclosure are also applicable to similar technical problems.

[0115] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this disclosure, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0116] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0117] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0118] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0119] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0120] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0121] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0122] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0123] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. The embodiments of this application have been described in detail above with reference to the accompanying drawings, but this application is not limited to the above embodiments. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of this application.

[0124] The embodiments of this application have been described in detail above with reference to the accompanying drawings. However, this application is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of this application.

Claims

1. A method for identifying punctate fluorescence signals within the cell nucleus, characterized in that, The method includes: Acquire preprocessed tissue section scan images containing dotted fluorescence signal channels and nuclear marker channels; The preprocessed tissue slice scan image is input into a nuclear segmentation network model. The nuclear segmentation network model processes the nuclear marker channels to obtain a binary segmentation mask image, including: A semantic segmentation network model is constructed using an encoder-decoder structure. The Dice loss and classification loss are weighted and summed to construct a composite loss function; Based on the composite loss function, the segmentation network model is trained using a training dataset to obtain a cell nucleus segmentation network model. The training dataset contains labels that indicate whether each cell nucleus is valid or invalid. The preprocessed tissue slice scan image is input into the cell nuclear segmentation network model, and the cell nuclear marker channel is processed by the cell nuclear segmentation network model to obtain a binary segmentation mask image; Based on the binary segmentation mask image, the target cell nucleus mask is determined; The dotted fluorescence signal channel is multiplied with the target cell nuclear mask to obtain a multiplication feature result map; The multiplication feature result image is input into the point-like fluorescence signal recognition model to obtain a signal point heatmap and a preliminary segmentation mask, including: Two parallel decoding heads are connected after the shared feature encoder to construct a dotted fluorescence signal recognition model; After the multiplication feature result map is input into the shared feature encoder for data processing, the output of the shared feature encoder is decoded by two parallel decoding heads to obtain a signal point heatmap and a preliminary segmentation mask. The signal point heatmap, the preliminary segmentation mask, and the dotted fluorescence signal channel are fused to obtain a fused feature map; The fused feature map is input into the signal point instance segmentation model to obtain multiple single signal point instances. Based on the multiple single signal point instances, the target ACD score of the tissue slice scan image is determined, including: Construct a signal point instance segmentation model that includes a self-attention mechanism, a spatial pyramid pooling module, and a label-controlled watershed algorithm; The fused feature map is input into the spatial pyramid pooling module to obtain the output result of the spatial pyramid pooling module; The output of the spatial pyramid pooling module is input into the self-attention mechanism to obtain the signal instance offset vector field and the signal semantic enhancement map. Based on the signal instance offset vector field and the signal semantic enhancement map, the attribution metric matrix is ​​determined; Based on the attribution metric matrix, the label-controlled watershed algorithm is used to perform instance segmentation on the fused feature map to obtain a single signal point instance segmentation mask; Multiple single-signal-point instances are obtained by segmenting the mask based on the single-signal-point instance.

2. The method for recognizing punctate fluorescence signals within the cell nucleus according to claim 1, characterized in that, Determining the target cell nucleus mask based on the binary segmentation mask image includes: Connectivity analysis is performed on the binary segmentation mask to retain the nucleus regions that are classified as valid. Based on the effective cell nucleus region, a target cell nucleus mask is generated.

3. The method for recognizing punctate fluorescence signals within the cell nucleus according to claim 1, characterized in that, The step of fusing the signal point heatmap, the preliminary segmentation mask, and the dotted fluorescence signal channel to obtain a fused feature map includes: Upsample the signal point heatmap and the preliminary segmentation mask to obtain a first upsampled result corresponding to the signal point heatmap and a second upsampled result corresponding to the preliminary segmentation mask; The first upsampling result, the second upsampling result, and the dotted fluorescence signal channel are spliced ​​together to obtain the spliced ​​feature. The concatenated features are then fused and dimensionality reduced using a lightweight convolutional module to obtain a fused feature map.

4. The method for recognizing punctate fluorescence signals within the cell nucleus according to claim 1, characterized in that, The determination of the target ACD score of the tissue slice scan image based on the multiple single signal point instances includes: Based on the multiple single signal point instances, the target point density and the target cluster ratio are calculated. The target point density and the target cluster ratio are input into the target ACD scoring model to obtain the target ACD score of the tissue slice scan image.

5. A system for recognizing dotted fluorescent signals within the cell nucleus, characterized in that, The method for identifying punctate fluorescence signals within the cell nucleus according to any one of claims 1 to 4, wherein the system comprises: The data acquisition unit is used to acquire preprocessed tissue section scan images containing dotted fluorescence signal channels and nuclear marker channels; The data processing unit is used to input the preprocessed tissue slice scan image into the cell nuclear segmentation network model, and process the cell nuclear marker channels through the cell nuclear segmentation network model to obtain a binary segmentation mask image. The data determination unit is used to determine the target cell nucleus mask based on the binary segmentation mask image; The data multiplication unit is used to multiply the dotted fluorescence signal channel with the target cell nuclear mask to obtain a multiplication feature result map; The signal recognition unit is used to input the multiplication feature result map into the dotted fluorescence signal recognition model to obtain a signal point heatmap and a preliminary segmentation mask. The data fusion unit is used to fuse the signal point heatmap, the preliminary segmentation mask, and the dotted fluorescence signal channel to obtain a fused feature map; The instance segmentation unit is used to input the fused feature map into the signal point instance segmentation model to obtain multiple single signal point instances, and to determine the target ACD score of the tissue slice scan image based on the multiple single signal point instances.

6. An electronic device, characterized in that, It includes at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, which, when executed by the at least one control processor, enable the at least one control processor to perform the method for identifying punctate fluorescent signals in the cell nucleus as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for causing a computer to perform the method for identifying dotted fluorescent signals in the cell nucleus as described in any one of claims 1 to 4.

Images