A pancreatic cell image segmentation method, device and terminal equipment

Abstract

The application provides a pancreas cell image segmentation method and device and a terminal device, and is suitable for the technical field of image processing. The method comprises the following steps: generating pancreas cell contour information according to to-be-processed pancreas cell image information and a pancreas cell contour extraction model; generating pancreas cell texture classification information according to the to-be-processed pancreas cell image information and a pancreas cell texture classification model; and generating pancreas cell image segmentation information according to the to-be-processed pancreas cell image information, the pancreas cell contour information, the pancreas cell texture classification information, historical pancreas cell image segmentation information and a pancreas cell image segmentation model. The application is used for improving the accuracy of image segmentation of multiple subtypes of pancreas cells, and can be used for providing accurate image segmentation data for pancreas pathological quantitative statistics, early pancreas lesion screening and cell pathological typing research and judgment.

Description

Technical Field

[0001] This application belongs to the field of image processing technology, and in particular relates to pancreatic cell image segmentation methods, apparatus and terminal equipment. Background Technology

[0002] Microscopic imaging of pancreatic pathological sections has become a core technical means for screening pancreatic lesions, evaluating pancreatic islet function, and early assessment of pancreatic cancer. Automatic segmentation of pancreatic cell images, as a key preliminary step in quantitative pathological analysis, has gradually shifted from manual selection and segmentation to intelligent algorithm-based automatic processing in recent years.

[0003] In existing technologies, the entire pancreatic microscopic image to be segmented is usually directly input, and cell edge feature extraction and image pixel segmentation are completed simultaneously by relying on a pre-trained single deep learning model.

[0004] However, existing technologies are prone to problems such as missegmentation of the outline edges of pancreatic cells of different subtypes and missegmentation of cytoplasmic texture. Furthermore, the generalization ability of the model is limited, and the overall segmentation accuracy fluctuates greatly when dealing with non-standard pancreatic slice images with varying staining depths and severe interstitial adhesions. Summary of the Invention

[0005] In view of this, embodiments of this application provide a pancreatic cell image segmentation method, apparatus, and terminal device, aiming to solve the problems of insufficient segmentation accuracy and poor cell contour integrity of multi-subtype pancreatic cells in the prior art.

[0006] The first aspect of this application provides a pancreatic cell image segmentation method, including:

[0007] Acquire image information of multiple pancreatic cells to be processed and segmentation information of multiple historical pancreatic cell images;

[0008] Based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell contour extraction model, multiple pancreatic cell contour information is generated;

[0009] Based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell texture classification model, multiple pancreatic cell texture classification information is generated;

[0010] Based on the multiple pancreatic cell image information to be processed, multiple pancreatic cell contour information, multiple pancreatic cell texture classification information, multiple historical pancreatic cell image segmentation information, and a preset pancreatic cell image segmentation model, multiple pancreatic cell image segmentation information are generated.

[0011] A second aspect of this application provides a pancreatic cell image segmentation apparatus, comprising:

[0012] The information acquisition module is used to acquire multiple pancreatic cell image information to be processed and multiple historical pancreatic cell image segmentation information;

[0013] The pancreatic cell contour information generation module is used to generate multiple pancreatic cell contour information based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell contour extraction model.

[0014] The pancreatic cell texture classification information generation module is used to generate multiple pancreatic cell texture classification information based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell texture classification model.

[0015] The pancreatic cell image segmentation information generation module is used to generate multiple pancreatic cell image segmentation information based on the multiple pancreatic cell image information to be processed, multiple pancreatic cell contour information, multiple pancreatic cell texture classification information, multiple historical pancreatic cell image segmentation information, and a preset pancreatic cell image segmentation model.

[0016] A third aspect of this application provides a terminal device, the terminal device including a memory and a processor, the memory storing a computer program executable on the processor, the processor executing the computer program to implement the steps of the pancreatic cell image segmentation method described in the first aspect above.

[0017] A fourth aspect of this application provides a computer-readable storage medium, comprising: storing a computer program, which, when executed by a processor, implements the steps of the pancreatic cell image segmentation method as described in the first aspect above.

[0018] The beneficial effects of this application embodiment compared with the prior art are: this application is used to improve the accuracy of image segmentation for multi-subtype pancreatic cells, and can be used to provide accurate image segmentation data for pancreatic pathology quantitative statistics, early pancreatic lesion screening and cell pathology classification. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram illustrating the implementation process of the pancreatic cell image segmentation method provided in Embodiment 1 of this application;

[0021] Figure 2This is a schematic diagram illustrating the implementation process of the pancreatic cell image segmentation method provided in Embodiment 2 of this application;

[0022] Figure 3 This is a schematic diagram illustrating the implementation process of the pancreatic cell image segmentation method provided in Embodiment 3 of this application;

[0023] Figure 4 This is a schematic diagram illustrating the implementation process of the pancreatic cell image segmentation method provided in Embodiment 4 of this application;

[0024] Figure 5 This is a schematic diagram illustrating the implementation process of the pancreatic cell image segmentation method provided in Embodiment 5 of this application;

[0025] Figure 6 This is a schematic diagram illustrating the implementation process of the pancreatic cell image segmentation method provided in Embodiment Six of this application;

[0026] Figure 7 This is a schematic diagram illustrating the implementation process of the pancreatic cell image segmentation method provided in Embodiment 7 of this application;

[0027] Figure 8 This is a schematic diagram of the structure of the pancreatic cell image segmentation device provided in the embodiments of this application;

[0028] Figure 9 This is a schematic diagram of the terminal device provided in the embodiments of this application. Detailed Implementation

[0029] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0030] To illustrate the technical solution described in this application, specific embodiments are provided below.

[0031] Figure 1 A flowchart illustrating the implementation of the pancreatic cell image segmentation method provided in Embodiment 1 of this application is shown, and is described in detail below:

[0032] Step S101: Obtain multiple pancreatic cell image information to be processed and multiple historical pancreatic cell image segmentation information.

[0033] In this embodiment, the pancreatic cell image information to be processed can include pancreatic microscopic imaging data under different staining concentrations and different slice fields of view. It serves as the image basis for determining the boundaries and intracellular texture of various pancreatic cells. For example, acinar cells and islet cells will exhibit different ranges of characteristics in terms of cell body grayscale and pore morphology under differential staining conditions. The pancreatic cell image information to be processed can be acquired in real time by field-by-field using a fully automated biological microscopic imaging device. Multiple pancreatic cell image information to be processed can refer to pancreatic cell image information corresponding to multiple different slice fields of view. Historical pancreatic cell image segmentation information can include the labeled segmentation results data of four types of pancreatic cells, namely acinar, islet, duct, and interstitial cells, corresponding to past samples. It can help correct the current image segmentation judgment criteria. For example, historical segmentation contour data of the same type of cells can be used to correct the current segmentation edge deviation. Historical pancreatic cell image segmentation information can be archived and stored for a long time through a pathological database. Multiple historical pancreatic cell image segmentation information can refer to pancreatic cell image segmentation information corresponding to multiple batches of historical pathological samples.

[0034] Step S102: Generate multiple pancreatic cell contour information based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell contour extraction model.

[0035] In this embodiment, the preset pancreatic cell contour extraction model can be pre-defined manually, and an improved U-Net model can be used. This model may include an encoder module, a bottleneck layer module, and a decoder module. The encoder module can then perform downsampling feature extraction on multiple pancreatic cell image information to obtain multi-scale spatial and semantic features. The bottleneck layer module then performs multi-scale feature compression and channel fusion processing. The decoder module then performs upsampling operations to restore the feature map size layer by layer, achieving the splicing and fusion of high-level semantic features and low-level spatial features. Finally, the decoder's terminal convolutional layer performs edge detection and region classification on the fused features, outputting the boundary points, contour confidence, and cell region category probability of each cell region. This accurately extracts the full-dimensional contour data corresponding to acinar cells, islet cells, ductal cells, and stromal cells, generating multiple pancreatic cell contour information.

[0036] In this embodiment, the preset pancreatic cell contour extraction model can be a trained improved U-Net model. The specific training process can be as follows: first, pancreatic slice image data is selected from a manually constructed multi-subtype pancreatic cell annotation dataset; feature extraction is performed through multi-level downsampling in the encoder module to obtain feature mapping maps at different scales; cross-scale feature fusion is completed in the bottleneck layer module through stacked convolution and residual connections; feature resolution is gradually restored in the decoder module through deconvolution upsampling, and skip connections are used to reuse the low-level contour features of the encoder; finally, the cell contour mask is output by the end classification convolution. During the training phase, the contour confidence judgment threshold is set to 0.75. This threshold can be preset manually. When the model outputs a contour confidence greater than 0.75, it is judged as a valid cell contour, and then qualified contour data is selected to participate in subsequent statistical operations.

[0037] Step S103: Generate multiple pancreatic cell texture classification information based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell texture classification model.

[0038] In this embodiment, the preset pancreatic cell texture classification model can be manually preset. Specifically, it can include preset sub-models for pancreatic acinar cell texture, pancreatic islet cell texture, pancreatic duct cell texture, and pancreatic stromal cell texture. All four sub-models uniformly adopt the MobileNetV3 deep learning network architecture. For the original acinar cell image data in the pancreatic cell image information to be processed, the preset pancreatic acinar cell texture classification sub-model can be used to obtain cytoplasmic texture anchor point features and calculate the aspect ratio of acinar cytoplasmic pores. Combined with the proportion of statistical feature annotations using a sliding window, the sliding window size can be manually preset, with the window side length uniformly set to 32 pixels. Then, based on the proportion threshold, the typical acinar texture is determined. The critical threshold for the proportion of typical acinar texture annotations can be manually preset, specifically set to 0.62. For the original islet cell image data, the preset islet cell texture classification sub-model can be used... The cell texture classification sub-model obtains the feature points of pancreatic islet nuclei and islet granules, calculates the minimum Euclidean distance between the granules and the center of the nucleus, and statistically analyzes the feature annotation ratio through a sliding window. The islet texture discrimination distance limit threshold can be preset by the user, specifically set to 6.8 micrometers. The critical threshold for the proportion of islet feature texture annotations can also be preset by the user, specifically set to 0.58. For the original image data of pancreatic duct cells, the preset pancreatic duct cell texture classification sub-model can be used to extract the duct wall texture features and calculate the texture density ratio. For the original image data of pancreatic stromal cells, the preset pancreatic stromal cell texture classification sub-model can be used to distinguish between fibrous texture and blank stromal regions. Then, the classification results of each sub-model are integrated to generate specific texture classification data such as acinar cell texture determination information, islet cell texture determination information, pancreatic duct cell texture determination information, and pancreatic stromal cell texture determination information, i.e., multiple pancreatic cell texture classification information.

[0039] Step S104: Generate multiple pancreatic cell image segmentation information based on the multiple pancreatic cell image information to be processed, multiple pancreatic cell contour information, multiple pancreatic cell texture classification information, multiple historical pancreatic cell image segmentation information, and a preset pancreatic cell image segmentation model.

[0040] In this embodiment, the preset pancreatic cell image segmentation model can be manually preset, or it can be a trained random forest fusion model. Specifically, it can be trained based on a grid search optimization iterative algorithm. Then, the continuous pixel grayscale data and discrete decision data from the pancreatic cell image information to be processed, along with the pancreatic cell contour information and pancreatic cell texture classification information, are integrated into multidimensional feature sequence data. This data serves as the input data for the preset pancreatic cell image segmentation model. After analysis and calculation by the preset pancreatic cell image segmentation model, the corresponding pancreatic cell image segmentation information is obtained. It can be understood that a pancreatic cell image segmentation information can be obtained by combining a set of pancreatic cell image information to be processed, a set of pancreatic cell contour information, a set of pancreatic cell texture classification information, and corresponding historical pancreatic cell image segmentation information. Multiple pancreatic cell image segmentation information can be batch-calculated in the same way.

[0041] In this embodiment, during the training of the pancreatic cell image segmentation model, the feature splitting threshold, decision tree depth parameters, and feature weight coefficients of the model can be initialized by combining multiple historical pancreatic cell image segmentation information. The maximum depth of a single decision tree can be preset by human intervention, with a value set to 12. The minimum number of samples for feature splitting can be preset by human intervention, with a value set to 5. Then, the single feature classification error and multi-feature fusion classification error are obtained by verifying the data outside the sample bag. The model parameters are then updated and convergence is determined. The convergence threshold can be preset by human intervention, with a value set to 0.001. After the model converges, the optimal path retrieval algorithm can be used to determine the cell category to which the pixel belongs based on the multidimensional feature sequence, thereby generating pancreatic cell image segmentation information for the corresponding acinar cell region, islet cell region, ductal cell region, and interstitial region. Historical pancreatic cell image segmentation information can include various pancreatic cell segmentation annotation results retained from past pathological samples, as well as image grayscale data, contour annotation data, and texture classification data from past slices. In other words, it can include microscopic imaging grayscale information, various cell contour point information, and various cell texture determination information from past samples. It can be used to reflect the segmentation change pattern of pancreatic cells of the same type at the sample batch level and can provide a reference for the segmentation determination of pancreatic cell images in the current field of view. Historical pancreatic cell image segmentation information can be continuously recorded and cumulatively updated through the pathological database system.

[0042] In this embodiment, the specific training process of the pancreatic cell image segmentation model is as follows:

[0043] First, a feature dimension set can be determined, which includes three types of data: image grayscale dimension, contour boundary dimension, and intracellular texture dimension. Historical pancreatic cell image segmentation information can be obtained by acquiring and determining grayscale data (continuous values) of a single field of view based on microscopic imaging equipment; or by determining cell partition data (discrete values) of a single field of view based on contour extraction results and texture classification results, including acinar partitions, islet partitions, duct partitions, and interstitial partitions. Then, historical sample feature vectors can be determined based on historical pancreatic cell image segmentation information. A grid search optimization iterative algorithm can be used to train the pancreatic cell image segmentation model. Before starting training, the parameters of the pancreatic cell image segmentation model need to be initialized. This can be done by first initializing the feature weight allocation coefficients of a single decision tree, followed by initializing the feature splitting threshold matrix within the model. The single-dimensional feature thresholds within the matrix are selected from a pre-set parameter range of 0.01 to 0.99. The sample observation distribution probabilities corresponding to each type of feature are also initialized. For each cell partition category, the continuous grayscale variable follows a Gaussian distribution, and the discrete partition variable follows a multinomial distribution. In the forward error calculation, the single-feature classification error is calculated, and in the backward parameter back-substitution, the multi-feature fusion classification error is calculated. The model parameters are iteratively updated using these two types of error values. The overall logarithmic loss of the model is calculated. The convergence of the grid search optimization iterative algorithm can be determined by checking if it has reached a convergence threshold of 0.001. When the change in loss value is less than the preset convergence threshold, the pancreatic cell image segmentation model has stabilized its fit to historical pancreatic cell image segmentation information, and further iterations will result in minimal parameter changes. After training the pancreatic cell image segmentation model using the grid search optimization iterative algorithm, an optimal path retrieval algorithm can be used to determine the cell partition to which a pixel belongs based on the multi-dimensional feature sequence. Understandably, the pancreatic cell image segmentation model can be fine-tuned using the grid search optimization iterative algorithm based on updated historical pancreatic cell image segmentation information, updated feature splitting thresholds, decision tree depth parameters, and various feature weight coefficients.

[0044] The pancreatic cell image segmentation method provided in this application improves the accuracy of segmenting multi-subtype pancreatic cell images and can be used to provide accurate image segmentation data for pancreatic pathology quantitative statistics, early pancreatic lesion screening, and cell pathology typing.

[0045] Figure 2 The flowchart illustrating the implementation of the pancreatic cell image segmentation method provided in Embodiment 2 of this application is shown. The difference between this method and Embodiment 1 described above is that:

[0046] The preset pancreatic cell texture classification model includes a preset pancreatic acinar cell texture classification sub-model, a preset pancreatic islet cell texture classification sub-model, a preset pancreatic duct cell texture classification sub-model, and a preset pancreatic interstitial cell texture classification sub-model.

[0047] The pancreatic cell texture classification information includes pancreatic acinar cell texture classification information, pancreatic islet cell texture classification information, pancreatic duct cell texture classification information, and pancreatic interstitial cell texture classification information.

[0048] Step S103 specifically includes:

[0049] Step S201: Generate multiple pancreatic acinar cell texture classification information based on the multiple pancreatic cell image information to be processed and the preset pancreatic acinar cell texture classification sub-model.

[0050] In this embodiment, the preset pancreatic acinar cell texture classification sub-model can be pre-defined by the user. It adopts the MobileNetV3 deep learning network model. The pancreatic cell image information to be processed is microscopic section image data of the pancreatic acinar region, which includes the texture details and grayscale features of acinar cytoplasm and acinar cavity. This is the core basis for determining the typical texture features of acinar cells. Then, the pancreatic cell image information to be processed is input into the preset pancreatic acinar cell texture classification sub-model. The model accurately extracts the texture anchor points and cavity structure features of acinar cytoplasm, then quantifies and calculates the morphological parameters and texture distribution features of acinar cavities, and then combines the preset texture judgment criteria to complete feature identification, thereby generating multiple pancreatic acinar cell texture classification information to accurately distinguish typical acinar texture regions from atypical acinar texture regions.

[0051] Step S202: Generate multiple pancreatic islet cell texture classification information based on the multiple pancreatic cell image information to be processed and the preset pancreatic islet cell texture classification sub-model.

[0052] In this embodiment, the preset pancreatic islet cell texture classification sub-model can be manually preset. It adopts the MobileNetV3 deep learning network model. The multiple pancreatic cell image information to be processed is microscopic section image data of the islet cell region, including the distribution texture and gray-level difference features of the islet cell nucleus and endocrine granules. This is the key basis for identifying the islet cell-specific texture. Then, the multiple pancreatic cell image information to be processed is input into the preset islet cell texture classification sub-model. The model locates the core structure of the islet and extracts the granule distribution texture features. Then, it calculates the distribution correlation feature parameters between granules and cell nuclei. Finally, it completes feature classification according to the preset islet texture judgment rules, thereby generating multiple islet cell texture classification information and accurately identifying the characteristic texture region of islet cells.

[0053] Step S203: Generate multiple pancreatic duct cell texture classification information based on the multiple pancreatic cell image information to be processed and the preset pancreatic duct cell texture classification sub-model.

[0054] In this embodiment, the preset pancreatic duct cell texture classification sub-model can be pre-defined manually. It adopts the MobileNetV3 deep learning network model. The multiple pancreatic cell image information to be processed is microscopic section image data of the pancreatic duct region, which includes differential features of dense texture of duct wall and blank texture of lumen. It is an important basis for distinguishing duct cells from other pancreatic cell textures. Then, the multiple pancreatic cell image information to be processed is input into the preset pancreatic duct cell texture classification sub-model. The model extracts core features such as duct wall texture density and texture direction, then quantifies and statistically analyzes the texture distribution ratio of the duct region, and then combines the preset duct texture judgment threshold to complete the classification judgment, thereby generating multiple pancreatic duct cell texture classification information and accurately dividing the pancreatic duct cell texture region.

[0055] Step S204: Generate multiple pancreatic stromal cell texture classification information based on the multiple pancreatic cell image information to be processed and the preset pancreatic stromal cell texture classification sub-model.

[0056] In this embodiment, the preset pancreatic stromal cell texture classification sub-model can be pre-defined manually. It adopts the MobileNetV3 deep learning network model. The multiple pancreatic cell image information to be processed is microscopic section image data of the pancreatic stromal region, which includes the feature differences of the disordered texture of stromal fibers and the transparent texture of blank stromal tissue. This is the main basis for identifying the texture of the pancreatic stromal region. Then, the multiple pancreatic cell image information to be processed is input into the preset pancreatic stromal cell texture classification sub-model. The model distinguishes the feature differences between fibrous texture and blank stromal texture, and then counts the distribution range and density of stromal texture. Finally, the classification is completed according to the preset stromal texture judgment criteria, thereby generating multiple pancreatic stromal cell texture classification information and accurately segmenting the pancreatic stromal texture region.

[0057] The pancreatic cell image segmentation method provided in this application enables refined and differentiated identification of multi-subtype pancreatic cell textures, effectively avoids the problem of confusion caused by texture features of a single model, significantly improves the accuracy of pancreatic cell texture classification, and provides comprehensive texture feature data support for subsequent high-precision pancreatic cell image segmentation.

[0058] Figure 3 The flowchart illustrating the pancreatic cell image segmentation method provided in Embodiment 3 of this application is shown. The difference between this method and Embodiment 2 is that step S201 specifically includes:

[0059] Step S301: Based on the multiple pancreatic cell image information to be processed and the preset pancreatic acinar cell texture classification sub-model, generate multiple acinar cytoplasmic feature point information.

[0060] In this embodiment, the multiple pancreatic cell image information to be processed is microscopic imaging image data of the pancreatic acinar region, which includes complete acinar cytoplasm and cavity structure image features. The preset pancreatic acinar cell texture classification sub-model can be preset by humans. The MobileNetV3 deep learning network model is used. The multiple pancreatic cell image information to be processed is then input into the sub-model. The model performs feature traversal detection on the acinar cytoplasm region, and then accurately locates key feature points such as cytoplasmic texture protrusions, depressions, and cavity boundaries. Then, all effective feature points are summarized and integrated to generate multiple acinar cytoplasmic feature point information.

[0061] Step S302: Based on the multiple acinar cytoplasmic feature point information, calculate the length-to-width ratio information of multiple acinar cavities.

[0062] In this embodiment, the multiple acinar cytoplasmic feature point information includes the coordinate data of all boundary points corresponding to the acinar cavity, which can completely delineate the outline of the acinar cavity. Then, the longitudinal extreme points and transverse extreme points of the acinar cavity are screened, and the maximum longitudinal length and maximum transverse width of the cavity are calculated respectively. Then, the quantification process is completed by calculating the ratio of length to width, thereby obtaining the length-to-width ratio information of multiple acinar cavities, and accurately quantifying the morphological features of a single acinar cavity.

[0063] Step S303: Determine whether the aspect ratio of the acinar cavity is greater than a preset threshold value; if yes, proceed to step S304; if no, proceed to step S305.

[0064] In this embodiment, the preset acinar cavity aspect ratio threshold can be manually set, specifically 1.8. This threshold is a critical standard obtained based on the statistical analysis of a large number of normal pancreatic acinar cell texture samples. It can effectively distinguish between typical acinar cavity texture and abnormal atypical acinar cavity texture. Then, the calculated acinar cavity aspect ratio information is compared with the preset threshold one by one. Based on the comparison result, it is determined whether the current acinar cavity conforms to typical texture features. Then, based on the determination result, the corresponding processing step is jumped to ensure the accuracy of texture determination.

[0065] Step S304: Generate pancreatic acinar cell texture marker information.

[0066] In this embodiment, when the aspect ratio of the acinar cavity is greater than the preset threshold, it indicates that the acinar cavity corresponding to the current feature point has a regular shape and conforms to the typical texture features of pancreatic acinar cells. Then, the system positively marks the acinar cytoplasmic feature points and binds the corresponding cavity shape parameters and point coordinate data to generate pancreatic acinar cell texture marker point information containing complete feature parameters.

[0067] Step S305: Generate unlabeled point information of pancreatic acinar cell texture.

[0068] In this embodiment, when the aspect ratio of the acinar cavity is not greater than the preset aspect ratio threshold, it indicates that the acinar cavity corresponding to the current feature point has a deformed shape and abnormal texture, and does not belong to the standard acinar cell texture features. Then the system reverse-marks the acinar cytoplasmic feature points, and records the abnormal morphological parameters and point information, thereby generating unmarked point information of pancreatic acinar cell texture.

[0069] Step S306: Perform statistical processing on the multiple pancreatic acinar cell texture marker points and multiple pancreatic acinar cell texture unmarked points to obtain the number of pancreatic acinar cell texture marker points and the number of pancreatic acinar cell texture unmarked points.

[0070] In this embodiment, the entire acinar image can be traversed through a fixed sliding window with a side length of 32 pixels, which can be preset by the user. Then, all pancreatic acinar cell texture marker points and pancreatic acinar cell texture unmarked points are batch filtered within the sliding window range. The total number of the two types of marker information is then counted separately, and the total number of effective typical texture points and abnormal atypical texture points within the window is accurately recorded. Then, the data is organized into standardized statistical data to obtain the number of pancreatic acinar cell texture marker points and the number of pancreatic acinar cell texture unmarked points.

[0071] Step S307: Based on the information on the number of texture markers in pancreatic acinar cells and the information on the number of unmarked texture points in pancreatic acinar cells, calculate the percentage of texture markers in pancreatic acinar cells.

[0072] In this embodiment, the total number of pancreatic acinar cell texture markers and unmarked points within the sliding window is first summarized. Then, a division operation is performed with the number of pancreatic acinar cell texture markers as the numerator and the total number of feature points within the window as the denominator. The result is then standardized and converted to four decimal places to obtain the pancreatic acinar cell texture marker ratio information that can intuitively reflect the ratio of typical textures in the region.

[0073] Step S308: Determine whether the proportion of pancreatic acinar cell texture markers is greater than a preset threshold for the proportion of pancreatic acinar cell texture markers; if yes, proceed to step S309; ​​if no, do not generate multiple pancreatic acinar cell texture classification information.

[0074] In this embodiment, the preset threshold for the proportion of pancreatic acinar cell texture markers can be preset manually, specifically set to 0.62. This threshold can effectively distinguish between acinar regions with local texture abnormalities and those with overall texture abnormalities. Then, the real-time calculated proportion of pancreatic acinar cell texture markers is compared with the preset threshold to determine whether the current image region is a standard acinar cell texture region. Finally, based on the comparison result, it is determined whether to generate final classification information, thus avoiding judgment errors caused by local noise.

[0075] Step S309: Generate multiple pancreatic acinar cell texture classification information.

[0076] In this embodiment, when the proportion of pancreatic acinar cell texture markers is greater than the preset threshold for the proportion of pancreatic acinar cell texture markers, it indicates that the proportion of typical acinar texture in the current image area meets the standard and conforms to the pancreatic acinar cell texture recognition standard. Then, the system integrates the texture proportion data, cavity morphology parameters, and feature point distribution data of the current window to complete the determination of regional texture attributes, thereby generating multiple pancreatic acinar cell texture classification information containing regional texture type, confidence level, and feature parameters.

[0077] The pancreatic cell image segmentation method provided in this application embodiment achieves refined quantitative recognition of pancreatic acinar cell texture, which solves the problems of subjective judgment and insufficient accuracy in traditional recognition methods, and effectively improves the stability and accuracy of acinar cell texture classification.

[0078] Figure 4 The flowchart illustrating the pancreatic cell image segmentation method provided in Embodiment 4 of this application is shown. The difference between this method and Embodiment 2 is that step S202 specifically includes:

[0079] Step S401: Based on the multiple pancreatic cell image information to be processed and the preset pancreatic islet cell texture classification sub-model, generate multiple pancreatic islet cell nuclear feature site information and multiple pancreatic islet endocrine granule feature site information.

[0080] In this embodiment, the multiple pancreatic cell image information to be processed is microscopic section image data of the islet cell region, including the grayscale differences and spatial distribution characteristics of the islet cell nucleus and endocrine granules. The preset islet cell texture classification sub-model can be preset by humans. The MobileNetV3 deep learning network model is used. The multiple pancreatic cell image information to be processed is then input into the sub-model. The model accurately identifies the contour region of the islet cell nucleus and the tiny dot-like region of the endocrine granules. Then, the edge feature points and center points of the two types of core structures are extracted. Then, all effective feature points are filtered and denoised to generate multiple islet cell nucleus feature site information and multiple islet endocrine granule feature site information.

[0081] Step S402: Based on the multiple pancreatic islet cell nuclear characteristic site information, calculate the multiple pancreatic islet cell nuclear center site information.

[0082] In this embodiment, the information on multiple islet cell nuclei feature sites includes the coordinate data of all boundary points that outline the complete contour of the islet cell nucleus, which can accurately characterize the morphology and position of the cell nucleus. Then, the coordinates of all boundary feature points corresponding to a single islet cell nucleus are collected, and the average values ​​of the abscissa and ordinate of all points are calculated to obtain the center coordinates of a single cell nucleus. Then, all islet cell nuclei feature point data are processed in batches, and all cell nucleus center coordinate data are summarized to obtain information on multiple islet cell nuclei center sites.

[0083] Step S403: Based on the information of the central sites of the multiple pancreatic islet cell nuclei and the characteristic sites of the multiple pancreatic islet endocrine granules, the distance information between the multiple pancreatic islet cell nuclei and the secretory granules is calculated.

[0084] In this embodiment, the information on the central location of the pancreatic islet cell nucleus accurately records the central coordinate position of each cell nucleus, and the information on the characteristic location of the pancreatic islet endocrine granules includes the distribution coordinate position of all granules. Then, for each characteristic location of the pancreatic islet endocrine granule, the Euclidean straight-line distance between it and the corresponding central location of the pancreatic islet cell nucleus is calculated. Subsequently, the matching distance value between each group of granules and cell nuclei is recorded, and then all distance calculation results are summarized to obtain multiple information on the distance between pancreatic islet cell nuclei and secretory granules.

[0085] Step S404: Based on the multiple pancreatic islet endocrine granule feature site information, filter the minimum value of multiple islet cell nuclei and secretory granule distance information to obtain multiple islet cell nuclei and secretory granule distance information to be determined.

[0086] In this embodiment, the information on the distance between multiple pancreatic islet cell nuclei and secretory granules covers the distance data between all endocrine granules and cell nuclei within a single field of view. The minimum distance can most intuitively reflect the effective attachment and distribution characteristics of the granules. Then, multiple sets of distance values ​​corresponding to a single granule are filtered one by one, and the minimum distance is extracted as the core judgment parameter of the granule. Then, the minimum distance data of all granules are processed in batches to obtain multiple information on the distance between pancreatic islet cell nuclei and secretory granules to be judged.

[0087] Step S405: Determine whether the distance information between the pancreatic islet cell nucleus and the secretory granule to be determined is less than the preset threshold for the distance between the pancreatic islet cell nucleus and the secretory granule; if yes, proceed to step S406; if no, proceed to step S407.

[0088] In this embodiment, the preset threshold for the distance between the islet cell nucleus and secretory granules can be preset manually, specifically set to 6.8 micrometers. This threshold is a critical value obtained based on the statistical analysis of the physiological characteristics of normal islet cell granule distribution. It can effectively distinguish between effective endocrine granule texture and invalid noise points. Then, the distance information between the islet cell nucleus and secretory granules to be judged is compared with the preset threshold one by one to determine whether the current granule belongs to the effective texture features of islet cells. Then, based on the judgment result, the corresponding marking step is jumped to ensure the accuracy of granule texture recognition.

[0089] Step S406: Generate pancreatic islet cell texture marker information.

[0090] In this embodiment, when the distance between the pancreatic islet cell nucleus and the secretory granules is less than the preset threshold for the distance between the pancreatic islet cell nucleus and the secretory granules, it indicates that the current endocrine granules are distributed close to the cell nucleus, which conforms to the typical texture distribution characteristics of pancreatic islet cells. Then, the system performs positive texture marking on the feature points of the granules, and then binds the corresponding distance parameters and point coordinate data to generate pancreatic islet cell texture marking point information.

[0091] Step S407: Generate unmarked point information of pancreatic islet cell texture.

[0092] In this embodiment, when the distance between the pancreatic islet cell nucleus and the secretory granules is not less than the preset threshold for the distance between the pancreatic islet cell nucleus and the secretory granules, it indicates that the current granule point is too far from the cell nucleus, belongs to discrete noise or non-islet texture structure, and does not meet the requirements of pancreatic islet cell texture features. Then the system reverse-marks the feature point, and then records the abnormal distance parameters and point information, thereby generating unmarked point information of pancreatic islet cell texture.

[0093] Step S408: Perform quantity statistics processing based on the pancreatic islet cell texture marker point information and pancreatic islet cell texture unmarked point information to generate pancreatic islet cell texture marker point quantity information and pancreatic islet cell texture unmarked point quantity information.

[0094] In this embodiment, a fixed sliding window with a side length of 32 pixels is used to traverse the entire pancreatic islet image. The size of this window can be preset by the user. Then, within the sliding window, the number of all pancreatic islet cell texture marker points and pancreatic islet cell texture unmarked points is counted in batches. The total number of valid pancreatic islet texture points and invalid non-islet texture points within the window is accurately recorded. Then, the statistical results are organized into standardized numerical data to generate information on the number of pancreatic islet cell texture marker points and the number of pancreatic islet cell texture unmarked points.

[0095] Step S409: Based on the information on the number of pancreatic islet cell texture markers and the information on the number of unmarked points in the pancreatic islet cell texture, calculate the percentage of pancreatic islet cell texture markers.

[0096] In this embodiment, the total number of pancreatic islet cell texture markers and unmarked points within the sliding window is first counted. Then, a division operation is performed with the number of pancreatic islet cell texture markers as the numerator and the total number of feature points within the window as the denominator. The result is then standardized and converted, retaining four decimal places, thereby obtaining the proportion of pancreatic islet cell texture markers that can reflect the purity of the regional islet texture.

[0097] Step S410: Determine whether the proportion of the number of pancreatic islet cell texture markers is greater than a preset threshold for the proportion of the number of pancreatic islet cell texture markers; if yes, proceed to step S411; if no, do not generate multiple pancreatic islet cell texture classification information.

[0098] In this embodiment, the preset threshold for the proportion of pancreatic islet cell texture markers can be preset manually, specifically set to 0.58. This threshold can accurately distinguish between real islet texture regions and mixed noise regions. Then, the real-time calculated proportion of pancreatic islet cell texture markers is compared with the preset threshold to determine whether the current image region is a standard islet cell texture region. Based on the comparison result, it is determined whether to generate classification information, effectively avoiding misjudgment problems caused by scattered noise points.

[0099] Step S411: Generate multiple pancreatic islet cell texture classification information.

[0100] In this embodiment, when the proportion of pancreatic islet cell texture markers is greater than the preset threshold for the proportion of pancreatic islet cell texture markers, it indicates that the secretory particles in the current image area are evenly distributed and have typical texture features, which meets the pancreatic islet cell texture recognition standard. Then, the system integrates the texture proportion data, particle spacing parameters, and feature distribution information of the current area to accurately determine the pancreatic islet cell texture attributes, thereby generating multiple pancreatic islet cell texture classification information containing texture type, confidence level, and feature parameters.

[0101] The pancreatic cell image segmentation method provided in this application enables accurate identification of the fine texture of pancreatic islet cells, effectively solving the technical problem of fragmented pancreatic islet cell texture being easily confused with interstitial noise, and significantly improving the overall accuracy of pancreatic islet cell texture classification and image segmentation.

[0102] Figure 5 The flowchart illustrating the implementation of the pancreatic cell image segmentation method provided in Embodiment 5 of this application is shown. Its difference from Embodiment 1 described above lies in:

[0103] The preset pancreatic cell contour extraction model includes a preset pancreatic cell membrane upper edge extraction sub-model, a preset pancreatic cell membrane lower edge extraction sub-model, a preset pancreatic cell membrane left boundary extraction sub-model, a preset pancreatic cell membrane right boundary extraction sub-model, a preset pancreatic cell nuclear membrane circumferential contour extraction sub-model, and a preset pancreatic cell closed boundary extraction sub-model.

[0104] Step S102 specifically includes:

[0105] Step S501: Based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell membrane upper edge extraction sub-model, preset pancreatic cell membrane lower edge extraction sub-model, preset pancreatic cell membrane left boundary extraction sub-model, preset pancreatic cell membrane right boundary extraction sub-model, preset pancreatic cell nuclear membrane circumferential contour extraction sub-model, and preset pancreatic cell closed boundary extraction sub-model, multiple pancreatic cell membrane upper edge information, multiple pancreatic cell membrane lower edge information, multiple pancreatic cell membrane left boundary information, multiple pancreatic cell membrane right boundary information, multiple pancreatic cell nuclear membrane circumferential contour information, and multiple pancreatic cell closed boundary information are generated.

[0106] In this embodiment, the pancreatic cell image information to be processed can be pancreatic microscopic section image data with different staining depths, different section clarity, and different field-of-view densities. It fully includes structural features such as the edge of the pancreatic cell membrane, the outline of the nuclear membrane, and the overall cell boundary, which is the basis for finely segmenting the outline of the cell in each direction. The preset pancreatic cell membrane upper edge extraction sub-model, preset pancreatic cell membrane lower edge extraction sub-model, preset pancreatic cell membrane left boundary extraction sub-model, preset pancreatic cell membrane right boundary extraction sub-model, preset pancreatic cell nuclear membrane circumferential outline extraction sub-model, and preset pancreatic cell closed boundary extraction sub-model can all be preset by humans and all belong to the improved U-Net deep learning network model. Each sub-model is designed for the membrane structure, nuclear membrane, and other structural features of the pancreatic cell in different directions. Specialized training was performed on the membrane contour and closed boundary features. Then, multiple pancreatic cell image information to be processed was input into the corresponding sub-models. The encoder module of each sub-model completed multi-scale contour feature downsampling and extraction. Then, the bottleneck layer module completed edge feature fusion and noise reduction. Finally, the decoder module upsampled and restored the contour feature resolution, accurately capturing the feature points and contour trajectories of the upper edge, lower edge, left boundary, right boundary, circumferential nuclear membrane, and overall closed boundary of the cell membrane. This generated multiple pancreatic cell membrane upper edge information, multiple pancreatic cell membrane lower edge information, multiple pancreatic cell membrane left boundary information, multiple pancreatic cell membrane right boundary information, multiple pancreatic cell membrane circumferential contour information, and multiple pancreatic cell closed boundary information.

[0107] Step S502: Generate multiple pancreatic cell contour information based on the multiple pancreatic cell membrane upper edge information, multiple pancreatic cell membrane lower edge information, multiple pancreatic cell membrane left boundary information, multiple pancreatic cell membrane right boundary information, multiple pancreatic cell nuclear membrane circumferential contour information, and multiple pancreatic cell closed boundary information.

[0108] In this embodiment, multiple pancreatic cell membrane upper edge information, multiple pancreatic cell membrane lower edge information, multiple pancreatic cell membrane left boundary information, and multiple pancreatic cell membrane right boundary information completely cover the contour features of the pancreatic cell membrane structure. Multiple pancreatic cell nuclear membrane circumferential contour information can accurately characterize the complete contour morphology of the cell nucleus. Multiple pancreatic cell closed boundary information can lock the overall outer boundary range of a single cell. Then, the cell membrane contour information, cell nuclear membrane contour information, and cell closed boundary information from all directions are feature aligned and fused. Then, the broken points, noise points, and redundant edges in each contour information are removed. Then, the scattered multi-directional contour features are spliced ​​and completed to repair the missing and broken areas of the cell contour, thereby generating multiple pancreatic cell contour information with complete boundaries, accurate positions, and distinct layers.

[0109] The pancreatic cell image segmentation method provided in this application effectively solves the problems of edge loss, local misalignment, and blurred nuclear membrane contours that are prone to occur in traditional single-model contour extraction. It comprehensively improves the integrity and accuracy of pancreatic cell contour extraction, and provides solid contour feature support for subsequent high-precision pancreatic cell image segmentation.

[0110] Figure 6 The flowchart illustrating the implementation of the pancreatic cell image segmentation method provided in Embodiment Six of this application is shown. The difference between this method and Embodiment One is that, after step S104, the method further includes:

[0111] Step S601: Obtain multiple historical pancreatic cell image information.

[0112] In this embodiment, multiple historical pancreatic cell image information can be multiple batches of standard pancreatic pathological slide image data that have been archived and stored in the system database for a long time. These images include original pancreatic cell images of different subtypes, different staining states, and different field densities. They can reflect the morphology, size, distribution density, contour features, and texture patterns of normal pancreatic cells. Multiple historical pancreatic cell image information can be directly retrieved in batches from the pathological database. Multiple historical pancreatic cell image information can refer to original pancreatic cell image data corresponding to multiple batches of different pathological samples and different acquisition devices.

[0113] Step S602: Obtain pancreatic cell quantity information based on the segmentation information of the multiple pancreatic cell images.

[0114] In this embodiment, multiple pancreatic cell image segmentation information has completed the precise segmentation and annotation of acinar cells, islet cells, ductal cells, and interstitial regions, which can clearly identify the independent segmentation regions and individual ranges of various types of pancreatic cells in a single field of view. Then, the independent cell segmentation regions in all pancreatic cell image segmentation information are traversed, and then the effective complete cell regions are counted one by one. Incomplete cell regions, overlapping noise regions, and invalid background regions are removed. Then, the statistical values ​​are normalized to obtain pancreatic cell quantity information that accurately represents the total number of various types of pancreatic cells in a single field of view.

[0115] Step S603: Generate pancreatic cell number correction information based on the pancreatic cell number information, multiple historical pancreatic cell image information, multiple historical pancreatic cell image segmentation information, and preset pancreatic cell number correction weights.

[0116] In this embodiment, the preset pancreatic cell number correction weight can be manually preset, including the weight coefficient corresponding to the pancreatic cell number information obtained from real-time segmentation and the weight coefficient corresponding to the statistical quantity of historical samples. The real-time quantity weight coefficient is set to 0.6, the historical quantity weight coefficient is set to 0.4, and the sum of the two coefficients is 1. The weight value is set based on a comprehensive consideration of the accuracy of real-time data and the universality of historical samples. Then, the historical standard pancreatic cell number mean is obtained by statistically analyzing multiple historical pancreatic cell image information and multiple historical pancreatic cell image segmentation information. The real-time pancreatic cell number information and the historical standard number mean are then weighted and fused using the corresponding weight coefficients. Finally, the extreme value of the fused quantity value is checked to remove abnormal deviation values, thereby generating pancreatic cell number correction information that conforms to the distribution pattern of pathological samples and has smaller errors.

[0117] The pancreatic cell image segmentation method provided in this application effectively avoids the problems of single-field segmentation counting bias and large random errors, and greatly improves the accuracy and stability of pancreatic cell number statistics, providing accurate and reliable quantitative statistics for pancreatic pathological quantitative analysis and lesion degree assessment.

[0118] Figure 7 The flowchart illustrating the implementation of the pancreatic cell image segmentation method provided in Embodiment Seven of this application is shown. The difference between this method and Embodiment Six is ​​that, after step S603, the method further includes:

[0119] Step S701: Generate optimized pancreatic cell image segmentation information based on the multiple pancreatic cell image segmentation information, pancreatic cell number correction information, and multiple preset pancreatic cell image segmentation optimization rules.

[0120] In this embodiment, the pancreatic cell image segmentation information consists of initial pancreatic cell image data after segmentation, including complete segmentation masks and pixel classification results for acinar cell regions, islet cell regions, ductal cell regions, and interstitial regions. It contains minor issues such as cell adhesion segmentation deviations, missing small cells, edge contour spurs, and residual interstitial noise, serving as the foundational data for subsequent image optimization. The pancreatic cell quantity correction information is accurate cell quantity statistics after weighted correction based on historical samples. It provides a standard quantity benchmark for various types of pancreatic cells in a single field of view, which can be used to constrain the overall accuracy range of image segmentation optimization, avoiding over-correction or under-correction during the optimization process. Multiple preset pancreatic cell image segmentation optimization rules can be manually preset, covering a full range of optimization strategies such as pancreatic cell adhesion separation, small cell supplementation, contour spur repair, background noise filtering, abnormal region removal, segmentation scale calibration, and multi-class cell balance optimization. These rules can specifically correct various defects in the initial segmented image.

[0121] In this embodiment, multiple preset pancreatic cell image segmentation optimization rules can be preset manually. Specifically, they include six core rules: cell adhesion separation optimization rules, micro-cell supplementation segmentation optimization rules, contour edge repair optimization rules, interstitial noise filtering optimization rules, abnormal and deformed cell removal optimization rules, and segmentation scale calibration optimization rules. All optimization rules are trained and calibrated based on a large number of labeled pancreatic pathological image samples to adapt to pancreatic microscopic image scenarios with different staining concentrations, different slice clarity, and different cell densities.

[0122] In this embodiment, the preset cell adhesion splitting optimization rules can be preset manually, and the adhesion judgment pixel threshold can be preset manually, specifically 8 pixels. When the pixel width of the adjacent cell adhesion area is greater than 8 pixels, it is judged as severe cell adhesion. Then, the boundary splitting of the adhesion cells is completed by contour curvature analysis and texture difference features, accurately distinguishing the adhesion acinar cells, pancreatic islet cells and duct cells, and solving the segmentation confusion problem caused by multi-cell stacking.

[0123] In this embodiment, the preset micro-cell supplementation segmentation optimization rules can be preset manually, and the micro-cell pixel area threshold can be preset manually, specifically set to 120 pixels. For micro-pancreatic islet cells and small duct cells with a pixel area greater than 120 pixels that were missed in the initial segmentation, the supplementation segmentation process is completed through secondary matching of texture features and contour feature retrieval, thus completing the micro-cell targets missed in the initial segmentation and improving the segmentation integrity of small-sized pancreatic cells.

[0124] In this embodiment, the preset contour edge repair optimization rules can be preset by humans. They can address defects such as burrs, gaps, jagged edges, and local missing parts in the initial segmentation contour. Then, through contour smoothing algorithms and edge feature completion strategies, they can refine the segmentation contours of various pancreatic cells, regulate the cell boundary morphology, and improve the smoothness and integrity of the cell contours.

[0125] In this embodiment, the preset interstitial noise filtering optimization rules can be preset manually. They can accurately identify scattered noise points, false contours, and invalid texture interference areas in the pancreatic interstitial region, thereby filtering out interstitial background noise in batches, preserving pure interstitial segmentation regions, avoiding cell segmentation misjudgment caused by noise interference, and improving the overall image segmentation cleanliness.

[0126] In this embodiment, the preset abnormal and deformed cell elimination optimization rules can be preset by humans. They can combine the standard quantitative benchmark of pancreatic cell number correction information with the normal cell morphological characteristics, thereby eliminating invalid pseudo-cell regions in the field of view that are morphologically deformed, abnormally sized, or lack complete cytoplasmic and nuclear structures, ensuring that the segmentation results are all effective pancreatic cell target regions.

[0127] In this embodiment, the preset segmentation scale calibration optimization rule can be preset by humans. It can unify the segmentation scale standard of the same type of pancreatic cells in a single field of view based on the standard physiological scale parameters of pancreatic cells, correct abnormal areas of local over-segmentation and under-segmentation, and achieve balanced optimization of segmentation accuracy for multiple cells and multiple regions.

[0128] In this embodiment, the standard cell count and cell scale benchmark of the current image field of view can be locked by the pancreatic cell count correction information. Then, multiple preset pancreatic cell image segmentation optimization rules are called in sequence to perform adhesion separation, micro-cell completion, contour repair, noise filtering and abnormal region removal processing on a region-by-region and pixel-by-pixel basis on multiple pancreatic cell image segmentation information. Then, the segmentation scale and region boundary of various pancreatic cells are uniformly calibrated to complete multi-dimensional fine optimization correction, thereby generating optimized pancreatic cell image segmentation information with higher accuracy, fewer defects and stronger standardization.

[0129] The pancreatic cell image segmentation method provided in this application effectively solves the technical defects of traditional pancreatic cell image segmentation, such as missegmentation due to cell adhesion, missed segmentation of tiny cells, irregular contour edges, background noise interference, and residual abnormal areas. It significantly improves the completeness, accuracy, and standardization of multi-subtype pancreatic cell image segmentation, and can provide more accurate, stable, and reliable image segmentation data support for pancreatic pathology quantitative statistics, early lesion screening, and cell pathology typing, significantly improving the overall effect and clinical application value of intelligent pancreatic pathology analysis.

[0130] Corresponding to the method in the above embodiments, Figure 8 A structural block diagram of the pancreatic cell image segmentation device provided in an embodiment of this application is shown. For ease of explanation, only the parts related to the embodiment of this application are shown. Figure 8 The example pancreatic cell image segmentation device can be the execution subject of the pancreatic cell image segmentation method provided in the aforementioned embodiment 1.

[0131] Reference Figure 8 The pancreatic cell image segmentation device includes:

[0132] The information acquisition module 810 is used to acquire multiple pancreatic cell image information to be processed and multiple historical pancreatic cell image segmentation information;

[0133] The pancreatic cell contour information generation module 820 is used to generate multiple pancreatic cell contour information based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell contour extraction model.

[0134] The pancreatic cell texture classification information generation module 830 is used to generate multiple pancreatic cell texture classification information based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell texture classification model.

[0135] The pancreatic cell image segmentation information generation module 840 is used to generate multiple pancreatic cell image segmentation information based on the multiple pancreatic cell image information to be processed, multiple pancreatic cell contour information, multiple pancreatic cell texture classification information, multiple historical pancreatic cell image segmentation information, and a preset pancreatic cell image segmentation model.

[0136] The process by which each module in the pancreatic cell image segmentation device provided in this application implements its respective function can be specifically referred to the foregoing. Figure 1 The description of Embodiment 1 shown will not be repeated here.

[0137] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0138] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0139] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0140] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0141] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance. It should also be understood that although the terms "first," "second," etc., are used in the text to describe various elements in some embodiments of this application, these elements should not be limited by these terms. These terms are merely used to distinguish one element from another. For example, a first table may be named a second table, and similarly, a second table may be named a first table, without departing from the scope of the various described embodiments. Both the first table and the second table are tables, but they are not the same table.

[0142] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0143] The pancreatic cell image segmentation method provided in this application embodiment can be applied to terminal devices such as mobile phones, tablets, wearable devices, in-vehicle devices, augmented reality / virtual reality devices, laptops, super mobile personal computers, netbooks, and personal digital assistants. This application embodiment does not impose any restrictions on the specific type of terminal device.

[0144] For example, the terminal device may be a station in a WLAN, a cellular phone, a cordless phone, a session initiation protocol phone, a wireless local loop station, a personal digital processing device, a handheld device with wireless communication capabilities, a computing device or other processing device connected to a wireless modem, an in-vehicle device, a vehicle networking terminal, a computer, a laptop computer, a handheld communication device, a handheld computing device, a satellite wireless device, a wireless modem card, a set-top box, a user premises equipment, and / or other devices for communication over a wireless system, as well as next-generation communication systems, such as mobile terminals in 5G networks or mobile terminals in future evolved public terrestrial mobile networks, etc.

[0145] As an example and not a limitation, when the terminal device is a wearable device, the term "wearable device" can also refer to any device that utilizes wearable technology to intelligently design and develop everyday wearables, such as glasses, gloves, watches, clothing, and shoes. Wearable devices are portable devices worn directly on the body or integrated into a user's clothing or accessories. Wearable devices are not merely hardware devices; they achieve powerful functions through software support, data interaction, and cloud interaction. Broadly defined, wearable smart devices include those with comprehensive functions, large sizes, and the ability to perform complete or partial functions without relying on a smartphone, such as smartwatches or smart glasses, as well as those focused on a specific application function that require interaction with other devices such as smartphones, such as various smart bracelets and smart jewelry for vital sign monitoring.

[0146] Figure 9 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. For example... Figure 9 As shown, the terminal device 9 of this embodiment includes: at least one processor 90 ( Figure 9 (Only one is shown in the image) A memory 91 stores a computer program 92 that can run on the processor 90. When the processor 90 executes the computer program 92, it implements the steps in the various pancreatic cell image segmentation method embodiments described above, for example... Figure 1 Steps S101 to S104 are shown. Alternatively, when the processor 90 executes the computer program 92, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 8 The functions of modules 810 to 840 are shown.

[0147] The terminal device 9 can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor 90 and a memory 91. Those skilled in the art will understand that... Figure 9This is merely an example of terminal device 9 and does not constitute a limitation on terminal device 9. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal device may also include input transmission devices, network access devices, buses, etc.

[0148] The processor 90 may be a central processing unit, or it may be other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0149] In some embodiments, the memory 91 may be an internal storage unit of the terminal device 9, such as a hard disk or memory of the terminal device 9. The memory 91 may also be an external storage device of the terminal device 9, such as a plug-in hard disk, smart memory card, secure digital card, flash memory card, etc., equipped on the terminal device 9. Furthermore, the memory 91 may include both internal and external storage units of the terminal device 9. The memory 91 is used to store operating systems, applications, bootloaders, data, and other programs, such as the program code of computer programs. The memory 91 can also be used to temporarily store data that has been sent or will be sent.

[0150] 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.

[0151] This application also provides a terminal device, which includes at least one memory, at least one processor, and a computer program stored in the at least one memory and executable on the at least one processor. When the processor executes the computer program, it causes the terminal device to implement the steps in any of the above method embodiments.

[0152] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0153] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0154] If the integrated module / 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, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0155] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0156] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0157] 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.

[0158] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for segmenting pancreatic cell images, characterized in that, include: Acquire image information of multiple pancreatic cells to be processed and segmentation information of multiple historical pancreatic cell images; Based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell contour extraction model, multiple pancreatic cell contour information is generated; Based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell texture classification model, multiple pancreatic cell texture classification information is generated; Based on the multiple pancreatic cell image information to be processed, multiple pancreatic cell contour information, multiple pancreatic cell texture classification information, multiple historical pancreatic cell image segmentation information, and a preset pancreatic cell image segmentation model, multiple pancreatic cell image segmentation information are generated.

2. The pancreatic cell image segmentation method as described in claim 1, characterized in that, The preset pancreatic cell texture classification model includes a preset pancreatic acinar cell texture classification sub-model, a preset pancreatic islet cell texture classification sub-model, a preset pancreatic duct cell texture classification sub-model, and a preset pancreatic interstitial cell texture classification sub-model. The pancreatic cell texture classification information includes pancreatic acinar cell texture classification information, pancreatic islet cell texture classification information, pancreatic duct cell texture classification information, and pancreatic interstitial cell texture classification information. The step of generating multiple pancreatic cell texture classification information based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell texture classification model specifically includes: Based on the multiple pancreatic cell image information to be processed and the preset pancreatic acinar cell texture classification sub-model, multiple pancreatic acinar cell texture classification information is generated. Based on the multiple pancreatic cell image information to be processed and the preset pancreatic islet cell texture classification sub-model, multiple pancreatic islet cell texture classification information are generated; Based on the multiple pancreatic cell image information to be processed and the preset pancreatic duct cell texture classification sub-model, multiple pancreatic duct cell texture classification information is generated; Based on the multiple pancreatic cell image information to be processed and the preset pancreatic stromal cell texture classification sub-model, multiple pancreatic stromal cell texture classification information are generated.

3. The pancreatic cell image segmentation method as described in claim 2, characterized in that, The step of generating multiple pancreatic acinar cell texture classification information based on the multiple pancreatic cell image information to be processed and the preset pancreatic acinar cell texture classification sub-model specifically includes: Based on the multiple pancreatic cell image information to be processed and the preset pancreatic acinar cell texture classification sub-model, multiple acinar cytoplasmic feature point information are generated. Based on the multiple acinar cytoplasmic feature point information, the aspect ratio of multiple acinar cavities is calculated; Determine whether the aspect ratio of the acinar cavity is greater than a preset threshold value. If so, then generate pancreatic acinar cell texture marker information; If not, then generate unmarked point information of pancreatic acinar cell texture; Based on the multiple pancreatic acinar cell texture marker points and multiple pancreatic acinar cell texture unmarked points, a statistical analysis is performed to obtain the number of pancreatic acinar cell texture marker points and the number of pancreatic acinar cell texture unmarked points. Based on the information on the number of labeled points in the pancreatic acinar cell texture and the information on the number of unlabeled points in the pancreatic acinar cell texture, the percentage of labeled points in the pancreatic acinar cell texture is calculated. When the proportion of pancreatic acinar cell texture markers is greater than the preset threshold for the proportion of pancreatic acinar cell texture markers, multiple pancreatic acinar cell texture classification information are generated.

4. The pancreatic cell image segmentation method as described in claim 2, characterized in that, The step of generating multiple pancreatic islet cell texture classification information based on the multiple pancreatic cell image information to be processed and the preset islet cell texture classification sub-model specifically includes: Based on the multiple pancreatic cell image information to be processed and the preset pancreatic islet cell texture classification sub-model, multiple pancreatic islet cell nuclear feature site information and multiple pancreatic islet endocrine granule feature site information are generated. Based on the aforementioned multiple pancreatic islet cell nuclear characteristic site information, multiple pancreatic islet cell nuclear central site information were calculated; Based on the information of the central sites of multiple pancreatic islet cell nuclei and the characteristic sites of multiple pancreatic islet endocrine granules, the distance information between multiple pancreatic islet cell nuclei and secretory granules is calculated. Based on the characteristic site information of multiple pancreatic islet endocrine granules, the minimum value of multiple islet cell nuclei and secretory granule distance information is selected to obtain multiple islet cell nuclei and secretory granule distance information to be determined. Determine whether the distance between the islet cell nucleus and the secretory granule to be determined is less than a preset threshold for the distance between the islet cell nucleus and the secretory granule; If so, then generate pancreatic islet cell texture marker information; If not, then generate unmarked point information of pancreatic islet cell texture; Based on the pancreatic islet cell texture marker point information and pancreatic islet cell texture unmarked point information, a quantity statistical processing is performed to generate pancreatic islet cell texture marker point quantity information and pancreatic islet cell texture unmarked point quantity information. Based on the information on the number of marker points in the islet cell texture and the information on the number of unmarked points in the islet cell texture, the percentage of marker points in the islet cell texture is calculated. When the proportion of the number of pancreatic islet cell texture markers is greater than the preset threshold for the proportion of the number of pancreatic islet cell texture markers, multiple pancreatic islet cell texture classification information are generated.

5. The pancreatic cell image segmentation method as described in claim 1, characterized in that, The preset pancreatic cell contour extraction model includes a preset pancreatic cell membrane upper edge extraction sub-model, a preset pancreatic cell membrane lower edge extraction sub-model, a preset pancreatic cell membrane left boundary extraction sub-model, a preset pancreatic cell membrane right boundary extraction sub-model, a preset pancreatic cell nuclear membrane circumferential contour extraction sub-model, and a preset pancreatic cell closed boundary extraction sub-model. The step of generating multiple pancreatic cell contour information based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell contour extraction model specifically includes: Based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell membrane upper edge extraction sub-model, preset pancreatic cell membrane lower edge extraction sub-model, preset pancreatic cell membrane left boundary extraction sub-model, preset pancreatic cell membrane right boundary extraction sub-model, preset pancreatic cell nuclear membrane circumferential contour extraction sub-model, and preset pancreatic cell closed boundary extraction sub-model, multiple pancreatic cell membrane upper edge information, multiple pancreatic cell membrane lower edge information, multiple pancreatic cell membrane left boundary information, multiple pancreatic cell membrane right boundary information, multiple pancreatic cell nuclear membrane circumferential contour information, and multiple pancreatic cell closed boundary information are generated. Multiple pancreatic cell contour information is generated based on the multiple pancreatic cell membrane upper edge information, multiple pancreatic cell membrane lower edge information, multiple pancreatic cell membrane left boundary information, multiple pancreatic cell membrane right boundary information, multiple pancreatic cell nuclear membrane circumferential contour information, and multiple pancreatic cell closed boundary information.

6. The pancreatic cell image segmentation method as described in claim 1, characterized in that, After the step of generating multiple pancreatic cell image segmentation information based on the multiple pancreatic cell image information to be processed, multiple pancreatic cell contour information, multiple pancreatic cell texture classification information, multiple historical pancreatic cell image segmentation information, and a preset pancreatic cell image segmentation model, the method further includes: Acquire multiple historical pancreatic cell image information; Based on the segmentation information of the multiple pancreatic cell images, the number of pancreatic cells is obtained; Based on the pancreatic cell number information, multiple historical pancreatic cell image information, multiple historical pancreatic cell image segmentation information, and preset pancreatic cell number correction weights, pancreatic cell number correction information is generated.

7. The pancreatic cell image segmentation method as described in claim 6, characterized in that, After the step of generating pancreatic cell number correction information based on the pancreatic cell number information, multiple historical pancreatic cell image information, multiple historical pancreatic cell image segmentation information, and preset pancreatic cell number correction weights, the method further includes: Based on the multiple pancreatic cell image segmentation information, pancreatic cell number correction information, and multiple preset pancreatic cell image segmentation optimization rules, optimized pancreatic cell image segmentation information is generated.

8. A pancreatic cell image segmentation device, characterized in that, include: The information acquisition module is used to acquire multiple pancreatic cell image information to be processed and multiple historical pancreatic cell image segmentation information; The pancreatic cell contour information generation module is used to generate multiple pancreatic cell contour information based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell contour extraction model. The pancreatic cell texture classification information generation module is used to generate multiple pancreatic cell texture classification information based on the multiple pancreatic cell image information to be processed and the preset pancreatic cell texture classification model. The pancreatic cell image segmentation information generation module is used to generate multiple pancreatic cell image segmentation information based on the multiple pancreatic cell image information to be processed, multiple pancreatic cell contour information, multiple pancreatic cell texture classification information, multiple historical pancreatic cell image segmentation information, and a preset pancreatic cell image segmentation model.

9. A terminal device, characterized in that, The terminal device includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images