Nuclear cytoplasmic bridge image recognition method and system based on artificial intelligence

Abstract

The invention provides an artificial intelligence-based cytoplasmic bridge image recognition method and system. The method comprises the following steps of: setting a recognition constraint condition of a cytoplasmic bridge image to be recognized based on a cross-nuclear continuity feature of a cytoplasmic bridge; identifying isolated noise point distribution in the cytoplasmic bridge image under the identification constraint condition, and constructing a pixel focusing window when local pixel feature identification is performed on the cytoplasmic bridge image according to the isolated noise point distribution; dividing a binuclear cell region and a cytoplasmic bridge adhesion region in the cytoplasmic bridge image based on the pixel focusing window, and further determining a bimodal index corresponding to the binuclear cell region and pixel gradual change loss corresponding to the cytoplasmic bridge adhesion region; according to the bimodal index and the pixel gradual change loss, determining the fine-grained boundary characteristics of the binuclear cell and the karyoplasm bridge in the connection state; and segmenting a target cytoplasmic bridge contour from the cytoplasmic bridge image according to the fine-grained boundary features. According to the technical scheme provided by the invention, accurate identification of the adhesion boundary of the nucleoplasm bridge and the binuclear cell can be realized under the influence of the randomly distributed isolated noisy points.

Description

Technical Field

[0001] The present application relates to the field of intelligent image recognition technology, and more specifically, to a nucleocytoplasmic bridge image recognition method and system based on artificial intelligence. Background Art

[0002] Intelligent image recognition is an important branch of the field of artificial intelligence and has developed rapidly in recent years. With the improvement of computer hardware performance and breakthroughs in deep learning algorithms, image recognition has gradually moved from the laboratory to practical applications. In image recognition, computers automatically learn images and extract image features to achieve detection, classification and recognition of target objects. Image recognition is widely used in security monitoring, autonomous driving, medical imaging diagnosis, industrial testing and other fields, providing strong technical support for the construction of an intelligent society.

[0003] Existing intelligent recognition solutions for nucleoplasmic bridge images typically use deep learning techniques to automatically learn local features of nucleoplasmic bridge images. For example, U-Net deep learning technology, part of the convolutional neural network (CNN) architecture, automatically learns local features of nucleoplasmic bridge images, such as the shape of the cell nucleus, the distribution of plastids, and the connection morphology of the nucleoplasmic bridge. This allows for the recognition of characteristic information in nucleoplasmic bridge images and improves recognition accuracy. However, due to the traditional U-Net's inadequate global modeling of gradient directional features and its sensitivity to noise, the accuracy of adhesion boundary recognition can be reduced. Furthermore, due to the influence of randomly distributed isolated noise points in nucleoplasmic bridge images, existing methods are unable to deeply analyze the characteristic distribution of the adhesion boundary between the nucleoplasmic bridge and binucleated cells in the pixel gradient direction. This makes it difficult to accurately identify the morphological features of the nucleoplasmic bridge and binucleated cell adhesion boundary, resulting in boundary offset when identifying the target nucleoplasmic bridge in the nucleoplasmic bridge image. Therefore, how to accurately identify the adhesion boundary between the nucleoplasmic bridge and binucleated cells under the influence of randomly distributed isolated noise points has become a challenge facing the industry. Summary of the Invention

[0004] The present application provides an artificial intelligence-based nucleocytoplasmic bridge image recognition method and system, which can achieve accurate recognition of the nucleocytoplasmic bridge and the adhesion boundary of binucleated cells under the influence of randomly distributed isolated noise points.

[0005] In a first aspect, the present application provides a method for recognizing nucleoplasmic bridge images based on artificial intelligence, comprising the following steps: Acquiring a nucleoplasmic bridge image to be identified, and setting identification constraint conditions for the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image based on the cross-nuclear continuity characteristics of the nucleoplasmic bridge; Under the identification constraint conditions, identifying the distribution of isolated noise points in the nucleoplasmic bridge image that are different from the nucleoplasmic bridge connection structure, and constructing a pixel focusing window for local pixel feature recognition of the nucleoplasmic bridge image based on the isolated noise distribution; dividing a binucleated cell region and a nucleoplasmic bridge adhesion region in the nucleoplasmic bridge image based on the pixel focus window, and then determining a bimodal index of binucleated cells in the binucleated cell region in a pixel gradient direction and a pixel gradient loss of an adhesion boundary in the nucleoplasmic bridge adhesion region; Fuzzy suppressing the edge structure of the binucleated cell and the nucleoplasmic bridge in the nucleoplasmic bridge image according to the bimodal index and the pixel gradient loss, thereby obtaining fine-grained boundary features of the binucleated cell and the nucleoplasmic bridge in a connected state; The target nucleoplasmic bridge contour is segmented from the nucleoplasmic bridge image according to the fine-grained boundary features.

[0006] In some embodiments, setting the identification constraint conditions of the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image based on the cross-nuclear continuity feature of the nucleoplasmic bridge specifically includes: identifying all connected domains in the nucleoplasmic bridge image; Extract the skeleton structure of each connected domain, and select the connected domains with a skeleton structure length greater than a preset threshold as candidate connecting nuclei of the nucleoplasmic bridge; The length of the connection between each two candidate connecting nuclei was taken as the cross-nuclear continuity characteristic of the nucleoplasmic bridge; The identification constraint conditions of the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image are determined according to all cross-nuclear continuity features and the center radius of all candidate connection nuclei.

[0007] In some embodiments, under the identification constraint condition, identifying the distribution of isolated noise points in the nucleoplasmic bridge image that are different from the nucleoplasmic bridge connection structure specifically includes: identifying discrete pixel blocks that do not satisfy the identification constraint conditions from the nucleoplasmic bridge image as isolated noise points; The distribution of isolated noise points in the nucleoplasmic bridge image that are different from the nucleoplasmic bridge connection structure is determined from all the isolated noise points.

[0008] In some embodiments, constructing a pixel focus window for local pixel feature recognition of the nucleoplasmic bridge image based on the isolated noise point distribution specifically includes: generating a noise point association window corresponding to each isolated noise point based on the isolated noise point distribution; Determine the gradient entropy of the pixels within the window associated with each noise point; Determining the noise focus feature of the nucleoplasmic bridge image according to the gradient entropy of pixels within each noise association window; A pixel focusing window is constructed based on the noise focus feature when performing local pixel feature recognition on the nucleoplasmic bridge image.

[0009] In some embodiments, generating a noise point association window corresponding to each isolated noise point based on the isolated noise point distribution specifically includes: Selecting an isolated noise point in the isolated noise point distribution as a selected isolated noise point; Extract the minimum bounding rectangle of the selected isolated noise point; Using the length of the minimum circumscribed rectangle as the scale of the noise point association window to construct the noise point association window for the selected isolated noise point; Continue to determine the noise point association windows corresponding to the remaining isolated noise points in the isolated noise point distribution.

[0010] In some embodiments, determining the bimodal index of binucleated cells in the binucleated cell region in the pixel gradient direction and the pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion region specifically comprises: Performing radial gradient scanning on the binucleated cell region to obtain a direction histogram of pixel gradients; determining a bimodal index of binucleated cells in the binucleated cell region in the pixel gradient direction according to the main peak amplitude and the secondary peak amplitude in the direction histogram; extracting boundary pixel points of the adhesion boundary in the nucleoplasmic bridge adhesion region; Determining a local associated sub-region of each boundary pixel point, and determining a pixel change feature corresponding to each boundary pixel point according to the local associated sub-region of each boundary pixel point; The pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion region is determined according to the pixel change characteristics corresponding to each boundary pixel point.

[0011] In some embodiments, images of the nucleoplasmic bridges to be identified are acquired by optical microscopy.

[0012] In a second aspect, the present application provides an artificial intelligence-based nucleocytoplasmic bridge image recognition system, comprising: an acquisition module, configured to acquire a nucleoplasmic bridge image to be identified, and set identification constraint conditions for the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image based on the cross-nuclear continuity feature of the nucleoplasmic bridge; a processing module, configured to identify, under the identification constraint conditions, a distribution of isolated noise points in the nucleoplasmic bridge image that are distinct from the nucleoplasmic bridge connection structure, and construct a pixel focusing window for performing local pixel feature recognition on the nucleoplasmic bridge image based on the isolated noise distribution; The processing module is further configured to divide the binucleated cell region and the nucleoplasmic bridge adhesion region in the nucleoplasmic bridge image based on the pixel focusing window, and then determine the bimodal index of the binucleated cells in the binucleated cell region in the pixel gradient direction and the pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion region; The processing module is further configured to perform fuzzy suppression on the edge structure of the binucleated cell and the nucleoplasmic bridge in the nucleoplasmic bridge image based on the bimodal index and the pixel gradient loss, so as to obtain fine-grained boundary features of the binucleated cell and the nucleoplasmic bridge in a connected state; An execution module is used to segment the target nucleoplasmic bridge contour from the nucleoplasmic bridge image according to the fine-grained boundary features.

[0013] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores code, and the processor is configured to obtain the code and execute the above-mentioned artificial intelligence-based nucleocytoplasmic bridge image recognition method.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, which implements the above-mentioned artificial intelligence-based nucleocytoplasmic bridge image recognition method when executed by a processor.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: In the artificial intelligence-based nucleoplasmic bridge image recognition method and system provided in the present application, first, a nucleoplasmic bridge image to be identified is acquired, and recognition constraints of the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image are set based on the cross-nuclear continuity characteristics of the nucleoplasmic bridge; secondly, under the recognition constraints, the distribution of isolated noise points that are different from the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image is identified, and a pixel focusing window for local pixel feature recognition of the nucleoplasmic bridge image is constructed based on the isolated noise point distribution; further, based on the pixel focusing window, a binuclear cell area and a nucleoplasmic bridge adhesion area in the nucleoplasmic bridge image are divided, and then the bimodal index of the binuclear cells in the binuclear cell area in the pixel gradient direction and the pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion area are determined; then, the edge structure of the binuclear cells and the nucleoplasmic bridge in the nucleoplasmic bridge image is fuzzy suppressed based on the bimodal index and the pixel gradient loss, so as to obtain fine-grained boundary features of the binuclear cells and the nucleoplasmic bridge in the connection state; finally, the target nucleoplasmic bridge contour is segmented from the nucleoplasmic bridge image based on the fine-grained boundary features.

[0016] It can be seen that the present application can realize the accurate recognition of the nuclear-cytoplasmic bridge and the adhesion boundary of binucleated cells under the influence of randomly distributed isolated noise points; first, based on the cross-nuclear continuity characteristics of the nuclear-cytoplasmic bridge, the recognition constraint conditions of the nuclear-cytoplasmic bridge connection structure in the nuclear-cytoplasmic bridge image to be identified are set, thereby providing a judgment standard for accurately distinguishing the real nuclear-cytoplasmic bridge connection structure from non-specific noise, thereby avoiding the influence of randomly distributed isolated noise points in the nuclear-cytoplasmic bridge image; secondly, under the recognition constraint conditions, the distribution of isolated noise points in the nuclear-cytoplasmic bridge image that is different from the nuclear-cytoplasmic bridge connection structure is identified, and a pixel focusing window for local pixel feature recognition of the nuclear-cytoplasmic bridge image is constructed based on the isolated noise distribution, so as to effectively analyze the local area associated with the nuclear-cytoplasmic bridge connection structure, enhance the perception of fine-grained features such as edge transition, connection state and directional change, and thus deeply analyze the characteristic distribution of the adhesion boundary between the nuclear-cytoplasmic bridge and binucleated cells in the pixel gradient direction; further, based on the pixel focusing window, the binucleated cells in the nuclear-cytoplasmic bridge image are divided. The nuclear cell area and the nuclear-cytoplasmic bridge adhesion area are determined, and then the bimodal index of the binuclear cells in the binuclear cell area in the pixel gradient direction and the pixel gradient loss of the adhesion boundary in the nuclear-cytoplasmic bridge adhesion area are determined. By determining the bimodal index and pixel gradient loss, the connection structure characteristics of the binuclear cells and the nuclear-cytoplasmic bridge can be effectively identified, thereby improving the accuracy and reliability of nuclear-cytoplasmic bridge discrimination in image recognition; then, the adhesion boundary of the binuclear cells and the nuclear-cytoplasmic bridge in the nuclear-cytoplasmic bridge image is morphologically analyzed based on the bimodal index and pixel gradient loss, and the fine-grained boundary features of the binuclear cells and the nuclear-cytoplasmic bridge in the connection state are obtained, which can effectively identify and distinguish the edge features of the binuclear cells and the nuclear-cytoplasmic bridge, thereby avoiding the boundary offset of the target nuclear-cytoplasmic bridge in the nuclear-cytoplasmic bridge image; finally, the target nuclear-cytoplasmic bridge contour is segmented from the nuclear-cytoplasmic bridge image according to the fine-grained boundary features; in summary, the technical solution provided by the present application can realize the accurate recognition of the nuclear-cytoplasmic bridge and the adhesion boundary of the binuclear cells under the influence of randomly distributed isolated noise points. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of an application scenario architecture of the artificial intelligence-based nucleocytoplasmic bridge image recognition method according to some embodiments of the present application; Figure 2 is an exemplary flow chart of an artificial intelligence-based nucleoplasmic bridge image recognition method according to some embodiments of the present application; Figure 3 is an exemplary flow chart of determining identification constraints according to some embodiments of the present application; Figure 4 is a schematic structural diagram of a nucleocytoplasmic bridge image recognition system based on artificial intelligence according to some embodiments of the present application; Figure 5 It is a structural schematic diagram of a computer device for implementing an artificial intelligence-based nucleocytoplasmic bridge image recognition method according to some embodiments of the present application. DETAILED DESCRIPTION

[0018] In order to better understand the technical solution of the present application, the technical solution of the present application will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0019] refer to Figure 1 , this figure is a schematic diagram of an application scenario architecture of the artificial intelligence-based nucleoplasmic bridge image recognition method shown in some embodiments of the present application. The application scenario architecture includes an acquisition terminal, a communication network, a server end and a data storage system. The acquisition terminal and the server end are directly or indirectly connected through the communication network. The acquisition terminal obtains the nucleoplasmic bridge image to be identified and uploads it to the server terminal. The server end sets the recognition constraint conditions of the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image based on the cross-nuclear continuity characteristics of the nucleoplasmic bridge; under the recognition constraint conditions, the isolated noise distribution in the nucleoplasmic bridge image that is different from the nucleoplasmic bridge connection structure is identified, and a localization of the nucleoplasmic bridge image is constructed based on the isolated noise distribution. a pixel focusing window for partial pixel feature recognition; based on the pixel focusing window, a binuclear cell area and a nucleoplasmic bridge adhesion area in the nucleoplasmic bridge image are divided, and then a bimodal index of binuclear cells in the binuclear cell area in the pixel gradient direction and a pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion area are determined; according to the bimodal index and the pixel gradient loss, the edge structure of the binuclear cells and the nucleoplasmic bridge in the nucleoplasmic bridge image is fuzzy suppressed to obtain fine-grained boundary features of the binuclear cells and the nucleoplasmic bridge in a connected state; according to the fine-grained boundary features, a target nucleoplasmic bridge outline is segmented from the nucleoplasmic bridge image, and finally, the server side stores the target nucleoplasmic bridge outline in a data storage system.

[0020] refer to Figure 2 , which is an exemplary flow chart of a nucleoplasmic bridge image recognition method based on artificial intelligence according to some embodiments of the present application. The nucleoplasmic bridge image recognition method based on artificial intelligence mainly includes the following steps: In step 101, a nucleoplasmic bridge image to be identified is acquired, and identification constraints for the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image are set based on the cross-nuclear continuity feature of the nucleoplasmic bridge.

[0021] In specific implementation, the image of the nucleoplasmic bridge to be identified can be obtained through an optical microscope. The nucleoplasmic bridge image represents an image obtained through microscopic imaging technology that shows the connection structure between the cell nucleus and the cytoplasm. In cell biology, the nucleoplasmic bridge is a special connection channel between the cell nucleus and the cytoplasm. It plays an important role in the cell's material transport, signal transduction, and gene expression regulation. By obtaining the nucleoplasmic bridge image, the nucleoplasmic bridge in the image can be effectively identified.

[0022] In some embodiments, reference Figure 3 As shown in FIG. 1 , this figure is an exemplary flow chart for determining identification constraints according to some embodiments of the present application. In this embodiment, the identification constraints for the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image are set based on the cross-nuclear continuity feature of the nucleoplasmic bridge, which can be achieved by the following steps: First, in step 1011, all connected domains in the nucleoplasmic bridge image are identified; Next, in step 1012, the skeleton structure of each connected domain is extracted, and the connected domains whose skeleton structure length is greater than a preset threshold are selected as candidate connecting cores of the nucleoplasmic bridge; Then, in step 1013 , the connection length between each two candidate connected nuclei is used as the cross-nuclear continuity feature of the nucleoplasmic bridge; Finally, in step 1014, the identification constraint conditions of the nucleocytoplasmic bridge connection structure in the nucleocytoplasmic bridge image are determined based on all cross-nuclear continuity features and the center radius of all candidate connection nuclei.

[0023] It should be noted that the identification constraints in this application represent the judgment criteria set for accurately distinguishing between real nucleocytoplasmic bridge connection structures and non-specific noise when analyzing nucleocytoplasmic bridge images. The identification constraints combine the cross-nuclear connection lengths between candidate connected nuclei and their respective minimum center radii. By setting a composite restriction that the connection length is not less than a certain threshold and the connecting nucleus has sufficient center thickness, it ensures that only when there is a connection path with strong spatial extensibility and high morphological continuity between the two cell nuclei will it be identified as a nucleocytoplasmic bridge structure, thereby improving the specificity and reliability of the identification.

[0024] In the specific implementation, first, the independent closed areas in the nucleoplasmic bridge image are identified as connected domains through the existing connected domain analysis method (the connected domain analysis method based on 8 neighborhoods can be used); secondly, the skeleton structure is extracted for each connected domain (the pixel center line of the connected domain can be extracted as the skeleton structure by using the Zhang-Sun thinning algorithm), and the connected domains whose skeleton structure length exceeds a preset threshold (the preset threshold can be set according to the typical spacing between cell nuclei, such as 20 pixels) are screened as candidate connected nuclei, wherein the Zhang-Sun thinning algorithm is an algorithm that gradually removes the edge points of foreground pixels without destroying the connectivity and topological structure of the image, and ultimately retains a slender and connected center line; then, the connection length between each two candidate connected nuclei is used as the cross-nuclear continuity feature of the nucleoplasmic bridge, and the connection length can be calculated according to the Euclidean distance. No further details will be given here; finally, the identification constraint conditions of the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image are determined based on all the cross-nuclear continuity features and the central radius of all candidate connection nuclei, namely: the minimum cross-nuclear continuity feature and the maximum cross-nuclear continuity feature are extracted from all the cross-nuclear continuity features, and combined into a cross-nuclear continuity feature interval, the minimum central radius and the longitudinal radius are extracted from the central radius of all candidate connection nuclei, and combined into a central radius interval, and the cross-nuclear continuity feature interval and the central radius interval are combined into the identification constraint conditions of the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image, wherein the central radius is the distance from the center point of the candidate connection nucleus to the edge of the candidate connection nucleus with the center of gravity as the center point. No further details will be given here. The central radius represents the distance extending outward from the center point to the edge of the candidate connection nucleus.

[0025] It should be noted that in this embodiment, the candidate connecting nucleus represents the cell nucleus connected by the selected nucleocytoplasmic bridge; the transnuclear continuity feature in this embodiment represents the structural feature of the continuity of the connection path between the two cell nuclei. The transnuclear continuity feature reflects the degree of spatial connectivity of the connection path and is the key criterion for distinguishing real nucleocytoplasmic bridges from isolated structures or pseudo-connections.

[0026] In step 102, under the identification constraint condition, the distribution of isolated noise points in the nucleoplasmic bridge image that are different from the nucleoplasmic bridge connection structure is identified, and a pixel focusing window for local pixel feature recognition of the nucleoplasmic bridge image is constructed based on the isolated noise point distribution.

[0027] It should be noted that the isolated noise point distribution in this application represents a combination of discrete pixel blocks that are spatially non-contacting and closest to the cell nucleus or nucleoplasmic bridge connection structure. The isolated noise point distribution reflects the local aggregation or dispersion characteristics of non-structural noise in the nucleoplasmic bridge image, which is usually caused by random staining particles, image acquisition errors or background interference. By extracting the isolated noise point distribution, the pixel processing weight can be dynamically adjusted in the subsequent image recognition process, effectively reducing the interference of pseudo-structures on nucleoplasmic bridge recognition and improving the accuracy and robustness of image analysis.

[0028] identifying discrete pixel blocks that do not satisfy the identification constraint conditions from the nucleoplasmic bridge image as isolated noise points; The distribution of isolated noise points in the nucleoplasmic bridge image that are different from the nucleoplasmic bridge connection structure is determined from all the isolated noise points.

[0029] In a specific implementation, first, discrete pixel blocks that do not satisfy the identification constraints are identified from the nucleoplasmic bridge image as isolated noise points, that is, the image processing tool OpenCV can be used to identify discrete pixel blocks that do not satisfy the identification constraints from the nucleoplasmic bridge image as isolated noise points. Details will not be given here. For example, discrete pixel blocks whose cross-nuclear continuity features are not within the cross-nuclear continuity feature interval within the identification constraints or whose central radius of the connected area is not within the central radius interval are regarded as isolated noise points. Then, the distribution of isolated noise points in the nucleoplasmic bridge image that are different from the nucleoplasmic bridge connection structure is determined by all the isolated noise points, that is, all the isolated noise points can be marked in the original nucleoplasmic bridge image to obtain the distribution of isolated noise points in the nucleoplasmic bridge image that are different from the nucleoplasmic bridge connection structure.

[0030] It should be noted that, in this embodiment, the nucleoplasmic bridge connection structure represents a combination of a cell nucleus structure and a connection structure between cell nuclei; in this embodiment, isolated noise points represent independent pixel blocks that have no direct contact with the cell nucleus or the nucleoplasmic bridge connection structure in space. These independent pixel blocks are usually randomly distributed in the form of points or blocks, and are usually caused by uneven staining, image acquisition noise, background stray light or microscope imaging errors. They are non-structural interference information in the image. Isolated noise points do not have the continuity, directionality or connectivity characteristics that a nucleoplasmic bridge should have. Therefore, they need to be identified and eliminated through connectivity analysis and spatial distance judgment in image recognition to avoid misjudgment or missegmentation of target structure extraction.

[0031] In some embodiments, constructing a pixel focus window for local pixel feature recognition of the nucleoplasmic bridge image based on the isolated noise point distribution can be achieved by the following steps, namely: generating a noise point association window corresponding to each isolated noise point based on the isolated noise point distribution; Determine the gradient entropy of the pixels within the window associated with each noise point; Determining the noise focus feature of the nucleoplasmic bridge image according to the gradient entropy of pixels within each noise association window; A pixel focusing window is constructed based on the noise focus feature when performing local pixel feature recognition on the nucleoplasmic bridge image.

[0032] In some embodiments, generating a noise point association window corresponding to each isolated noise point based on the isolated noise point distribution may be implemented by the following steps, namely: Selecting an isolated noise point in the isolated noise point distribution as a selected isolated noise point; Extract the minimum bounding rectangle of the selected isolated noise point; Using the length of the minimum circumscribed rectangle as the scale of the noise point association window to construct the noise point association window for the selected isolated noise point; Continue to determine the noise point association windows corresponding to the remaining isolated noise points in the isolated noise point distribution.

[0033] In the specific implementation, first, a noise association window corresponding to each isolated noise point is generated based on the isolated noise point distribution; secondly, the gradient entropy of the pixels in each noise association window is determined, that is: for each noise association window, the gradient amplitude of each pixel in the noise association window is obtained by the Sobel operator, and the amplitude ratio of each gradient amplitude to the total amplitude of the window is counted, and the amplitude ratio of each gradient amplitude to the total amplitude of the window is input as an input variable into the Shannon entropy function, and the Shannon entropy function outputs the gradient entropy of the pixels in the noise association window, thereby obtaining the gradient entropy of the pixels in each noise association window; then, due to the distribution of isolated noise points in the nucleoplasmic bridge image, the gradient entropy of the pixels in the nucleoplasmic bridge image is obtained. The distribution is uneven. In order to effectively measure the focusing property of isolated noise points in the nucleoplasmic bridge image, the mean gradient entropy of pixels in all noise association windows can be used as the noise focusing feature of the nucleoplasmic bridge image. Finally, based on the noise focusing feature, a pixel focusing window is constructed for local pixel feature recognition of the nucleoplasmic bridge image, that is, the gradient entropy corresponding to each noise association window is obtained, and the noise association windows with gradient entropy greater than the noise focusing feature are extracted. The scale mean of all the extracted noise association windows is used as the scale of the pixel focusing window, and then the pixel focusing window for local pixel feature recognition of the nucleoplasmic bridge image is constructed.

[0034] It should be noted that, in this embodiment, the noise association window represents the noise association range delineated with the isolated noise point as the center, and the noise association window is used to capture the pixel gradient characteristics and texture changes of the isolated noise point and its neighborhood; in this embodiment, the gradient entropy represents an indicator for measuring the complexity of the image texture changes within the noise association window; in this embodiment, the noise focus feature represents a characteristic indicator that describes the degree of overall texture disturbance in the nucleoplasmic bridge image, and the noise focus feature reflects the interference intensity of the isolated noise point on the image pixel gradient pattern in the nucleoplasmic bridge image; in this application, the pixel focus window represents a local pixel analysis window delineated on the nucleoplasmic bridge image, and the pixel focus window focuses on identifying areas in the nucleoplasmic bridge image with drastic texture changes or clear structural edges. The pixel focus window can effectively analyze local areas associated with the nucleoplasmic bridge connection structure, enhance the perception of fine-grained features such as edge transitions, connection states and directional changes, thereby improving the recognition accuracy and efficiency of nucleoplasmic bridges.

[0035] In step 103, the binucleated cell region and the nucleoplasmic bridge adhesion region in the nucleoplasmic bridge image are divided based on the pixel focusing window, and then the bimodal index of the binucleated cells in the binucleated cell region in the pixel gradient direction and the pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion region are determined.

[0036] It should be noted that the binuclear cell area in the present application refers to a structural area composed of two morphologically closed, clearly-bounded and close-to-each-other cell nuclei in the nucleoplasmic bridge image. The binuclear cell area is usually the basic structural unit for the existence of the nucleoplasmic bridge, which is manifested as two nuclear structures having strong grayscale contrast and edge closure in the microscopic image, and are connected to each other by a slender, low-grayscale or gradient-continuous connecting band (i.e., the nucleoplasmic bridge). The key to identifying the binuclear cell area is to extract two nuclear bodies with biological morphological consistency and exclude isolated nuclei or multi-nuclear aggregate structures; the nucleoplasmic bridge adhesion area in the present application refers to a strip-shaped pixel area initially identified in the nucleoplasmic bridge image, located between two adjacent cell nuclei and showing a continuous connection state, which is usually manifested as a transition structure with relatively low grayscale value but consistency in the gradient direction.

[0037] In some embodiments, dividing the binucleated cell region and the nucleoplasmic bridge adhesion region in the nucleoplasmic bridge image based on the pixel focus window can be achieved by the following steps, namely: Sliding processing is performed on the nucleoplasmic bridge image by the pixel focusing window, and a multi-scale operator is used to detect the closed features of the cell nucleus edge to locate the binucleate cell area; The pixel gradient in the nucleoplasmic bridge image is tracked by the pixel focus window, and the nucleoplasmic bridge adhesion area is segmented by dynamic thresholding.

[0038] In a specific implementation, first, according to a preset sliding step size, the pixel focusing window is used to slide the nucleoplasmic bridge image, and within each pixel focusing window, an existing multi-scale operator is used to detect all connected domains, and the center distance between any two connected domains is used as a closed feature, and two connected domains whose closed feature is less than a preset threshold (which can be positioned as 1.5 times the average nuclear diameter) are combined into a binuclear cell region, wherein the sliding step size can be set according to actual needs, for example, it can be set to 3 pixels, which is not limited here; then, according to the preset sliding step size, the pixel focusing window is used to slide the nucleoplasmic bridge image, and within each pixel focusing window, a Sobel operator is used to track the pixel gradient along the gradient direction starting from the binuclear cell boundary, and a connected path area where the gradient amplitude on the path continuously changes and satisfies the length constraint is used as the nucleoplasmic bridge adhesion area, and the length constraint can be set based on machine learning of the lengths of a large number of nucleoplasmic bridges connecting the two cell nuclei or based on expert knowledge, which is not limited here.

[0039] It should be noted that, in this embodiment, the length constraint represents a parameter for identifying the effective connection length between two cell nuclei.

[0040] In some embodiments, determining the bimodal index of binucleated cells in the binucleated cell region in the pixel gradient direction and the pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion region can be achieved by the following steps, namely: Performing radial gradient scanning on the binucleated cell region to obtain a direction histogram of pixel gradients; determining a bimodal index of binucleated cells in the binucleated cell region in the pixel gradient direction according to the main peak amplitude and the secondary peak amplitude in the direction histogram; extracting boundary pixel points of the adhesion boundary in the nucleoplasmic bridge adhesion region; Determining a local associated sub-region of each boundary pixel point, and determining a pixel change feature corresponding to each boundary pixel point according to the local associated sub-region of each boundary pixel point; The pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion region is determined according to the pixel change characteristics corresponding to each boundary pixel point.

[0041] It should be noted that the bimodal index in this application represents an indicator for quantifying the symmetry of binuclear cells. The bimodal index reflects the symmetry and directional separation of the two cell nuclei, and is a key basis for identifying nucleocytoplasmic bridges in a binuclear structural state; the pixel gradient loss in this application represents an indicator for measuring the spatial difference in the pixel gradient of the adhesion boundary in the nucleocytoplasmic bridge adhesion area. The pixel gradient loss is an important indicator for judging whether the adhesion boundary in the nucleocytoplasmic bridge adhesion area is a real nucleocytoplasmic bridge. A real nucleocytoplasmic bridge usually has a smooth, slender and directional boundary morphology, rather than random curvature or messy connection. By calculating the pixel gradient loss, the pseudo-bridge structure in the nucleocytoplasmic bridge image can be effectively identified, thereby improving the accuracy and reliability of nucleocytoplasmic bridge discrimination in image recognition.

[0042] In the specific implementation, first, a 360° radial gradient scan is performed with the midpoint of the line connecting the centroids of the binuclear cell region as the origin (wherein the angular resolution is 1°), and the average gradient amplitude in each direction is counted to generate a directional histogram of the pixel gradient, wherein the directional histogram is a histogram with the direction as the horizontal axis and the average gradient amplitude as the vertical axis; secondly, the amplitude ratio of the secondary peak amplitude to the main peak amplitude is used as the bimodal index of the binuclear cells in the binuclear cell region in the pixel gradient direction; then, the boundary pixel points of the adhesion boundary in the nucleoplasmic bridge adhesion region are extracted by the image processing tool OpenCV, and the boundary pixel points represent the pixel points on the adhesion boundary in the nucleoplasmic bridge adhesion region; further, the eight boundary pixel points can be The neighborhood area is used as the local correlation sub-area to obtain the local correlation sub-area of ​​each boundary pixel point. For each boundary pixel point, the occurrence frequency of each pixel point in the local correlation sub-area corresponding to the boundary pixel point is calculated, and the occurrence frequency of each pixel point is input into the Shannon entropy function as an input variable, and the output result of the Shannon entropy function is used as the pixel change feature corresponding to the boundary pixel point, thereby obtaining the pixel change feature corresponding to each boundary pixel point. Finally, since the higher the degree of pixel chaos in the image, the more obvious the pixel change is, therefore, in order to measure the spatial difference change of the adhesion boundary in the nucleoplasmic bridge adhesion area, the variance of all pixel change features can be used as the pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion area.

[0043] It should be noted that, in this embodiment, the directional histogram represents a graphical representation method for statistically analyzing the directional distribution of pixel gradients in an image. The directional histogram can reflect the distribution characteristics of edges or structures in the image in various directions. In the analysis of binucleated cell regions, the directional histogram can be used to identify whether there are peaks in two main directions, thereby assisting in determining whether the region has typical binucleated structural characteristics. In this embodiment, the local correlation sub-region represents a local region composed of pixel points adjacent to boundary pixels. In this embodiment, the pixel change feature represents an indicator for measuring the complexity of the pixel value distribution in the image region.

[0044] In step 104, the edge structure of the binucleated cell and the nucleoplasmic bridge in the nucleoplasmic bridge image is fuzzy suppressed based on the bimodal index and the pixel gradient loss to obtain fine-grained boundary features of the binucleated cell and the nucleoplasmic bridge in a connected state.

[0045] In this application, the fine-grained boundary features represent the fine-grained features that describe the boundary between binucleated cells and nucleoplasmic bridges in the nucleoplasmic bridge image. The fine-grained boundary features reflect the smoothness and transition of the edge pixels in the nucleoplasmic bridge image in terms of spatial position. The real nucleoplasmic bridge connection usually presents a smooth and continuous gradient transition, while the non-real or pseudo-connection structure often has a sudden or chaotic gradient direction. By extracting the fine-grained boundary features, the edge features of binucleated cells and nucleoplasmic bridges can be effectively identified and distinguished, thereby avoiding the boundary offset of the target nucleoplasmic bridge in the nucleoplasmic bridge image.

[0046] In some embodiments, blurring and suppressing the edge structure of the binucleated cell and the nucleoplasmic bridge in the nucleoplasmic bridge image based on the bimodal index and the pixel gradient loss to obtain fine-grained boundary features of the binucleated cell and the nucleoplasmic bridge in a connected state can be achieved by the following steps, namely: determining the morphological change parameters of the connection between the binucleated cells and the nucleoplasmic bridge in the nucleoplasmic bridge image according to the bimodal index and the pixel gradient loss; The morphological change parameters are used to set a fine-grained convolution kernel for morphological analysis of the adhesion boundary between binucleated cells and the nucleoplasmic bridge in the nucleoplasmic bridge image; Performing morphological recognition on the nucleoplasmic bridge image using the fine-grained convolution kernel to obtain a morphological feature map; The fine-grained boundary features of the binucleate cell and the nucleoplasmic bridge in the connected state are determined based on the morphological feature map.

[0047] In the specific implementation, first, since the binuclear cells and the nucleoplasmic bridge in the nucleoplasmic bridge image are interconnected and the two have local correlation, in order to effectively identify the comprehensive structural relationship between the binuclear cells and the nucleoplasmic bridge, the bimodal index and the pixel gradient loss can be normalized to between 0 and 1 through minimum-maximum normalization, so that the bimodal index and the pixel gradient loss are dimensionally unified, and then the weighted sum of the normalized bimodal index and the pixel gradient loss is used as the morphological change parameter of the connection between the binuclear cells and the nucleoplasmic bridge in the nucleoplasmic bridge image, wherein the weight values ​​of the bimodal index and the pixel gradient loss can be set according to actual needs or according to neural network learning. For example, the weight values ​​of the bimodal index and the pixel gradient loss can be set to 0.3 and 0.7 respectively, which are not limited here; secondly, the morphological change parameter is compared with the change parameter threshold. For comparison, when the morphological change parameter is greater than the change parameter threshold, the fine-grained convolution kernel for morphological analysis of the adhesion boundary between the binucleated cells and the nucleoplasmic bridge in the nucleoplasmic bridge image is set to 5×5; when the morphological change parameter is less than or equal to the change parameter threshold, the fine-grained convolution kernel for morphological analysis of the adhesion boundary between the binucleated cells and the nucleoplasmic bridge in the nucleoplasmic bridge image is set to 3×3, wherein the change parameter threshold can be set according to actual needs or according to expert knowledge, and is not limited here; then, the fine-grained convolution kernel performs a convolution operation on each pixel area of ​​the nucleoplasmic bridge image, and the image obtained after the convolution operation is used as a morphological feature map; finally, the Sobel edge detection algorithm in the prior art can be used to perform edge recognition on the morphological feature map, and the recognized edge is used as the fine-grained boundary feature of the binucleated cells and the nucleoplasmic bridge in the connected state.

[0048] It should be noted that, in this embodiment, the morphological change parameter represents an indicator for measuring the degree of change in the spatial morphology of the structural boundary in the nucleoplasmic bridge image; in this embodiment, the fine-grained convolution kernel represents a convolution kernel for extracting tiny and fine structural features in the image, and the fine-grained convolution kernel can effectively amplify the subtle differences in edge transitions and connection adhesion structures in the nucleoplasmic bridge image, and assist the subsequent network in identifying the key morphological changes between the real connection and the pseudo connection of the nucleoplasmic bridge; in this embodiment, the morphological feature map represents a two-dimensional feature image generated after morphological analysis of the nucleoplasmic bridge image.

[0049] It should also be noted that, in the present application, fuzzy suppression refers to the process of suppressing fuzzy pixels of the edge structure in the nucleoplasmic bridge image, that is, the process of pixel enhancement of the edge structure in the nucleoplasmic bridge image, wherein the edge structure of the binuclear cells and the nucleoplasmic bridge in the nucleoplasmic bridge image is fuzzy suppressed based on the bimodal index and the pixel gradient loss, that is: the morphological change parameters of the connection between the binuclear cells and the nucleoplasmic bridge in the nucleoplasmic bridge image are determined based on the bimodal index and the pixel gradient loss; the fine-grained convolution kernel for morphological analysis of the adhesion boundary of the binuclear cells and the nucleoplasmic bridge in the nucleoplasmic bridge image is set by the morphological change parameters; the nucleoplasmic bridge image is morphologically recognized by the fine-grained convolution kernel to obtain a morphological feature map; the fine-grained boundary features of the binuclear cells and the nucleoplasmic bridge in the connection state are determined based on the morphological feature map, that is, the fine-grained boundary features are used as the result of fuzzy suppression, thereby completing the fuzzy suppression of the edge structure of the binuclear cells and the nucleoplasmic bridge in the nucleoplasmic bridge image.

[0050] In step 105, the target nucleoplasmic bridge contour is segmented from the nucleoplasmic bridge image according to the fine-grained boundary features.

[0051] In some embodiments, segmenting the target nucleoplasmic bridge contour from the nucleoplasmic bridge image according to the fine-grained boundary features can be achieved by using the following steps, namely: Marking the segmentation contour of the target nucleoplasmic bridge from the nucleoplasmic bridge image based on the fine-grained boundary features; A target nucleoplasmic bridge is segmented from the nucleoplasmic bridge image according to the segmentation contour.

[0052] In the specific implementation, first, the fine-grained boundary features are matched with the nucleoplasmic bridge edge features in the preset nucleoplasmic bridge edge feature database through the Hu invariant moment in the existing technology, and the edge with the greatest matching degree with the nucleoplasmic bridge edge feature in the fine-grained boundary features is used as the segmentation contour of the target nucleoplasmic bridge; then, the image processing tool OpenCV is used to segment the nucleoplasmic bridge image according to the segmentation contour to obtain the target nucleoplasmic bridge contour.

[0053] In addition, in another aspect of the present application, in some embodiments, the present application provides a nucleoplasmic bridge image recognition system based on artificial intelligence, referring to Figure 4 , which is a schematic diagram of the structure of a nucleoplasmic bridge image recognition system based on artificial intelligence according to some embodiments of the present application. The nucleoplasmic bridge image recognition system based on artificial intelligence includes: an acquisition module 201, a processing module 202, and an execution module 203, which are described as follows: Acquisition module 201, in this application, acquisition module 201 is mainly used to acquire a nucleoplasmic bridge image to be identified, and set identification constraints of the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image based on the cross-nuclear continuity characteristics of the nucleoplasmic bridge; Processing module 202, in the present application, is mainly used to identify the distribution of isolated noise points in the nucleoplasmic bridge image that are different from the nucleoplasmic bridge connection structure under the identification constraint conditions, and to construct a pixel focusing window for local pixel feature recognition of the nucleoplasmic bridge image based on the isolated noise distribution; The processing module 202 is further configured to segment the binucleated cell region and the nucleoplasmic bridge adhesion region in the nucleoplasmic bridge image based on the pixel focusing window, and further determine the bimodal index of the binucleated cells in the binucleated cell region in the pixel gradient direction and the pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion region; In addition, the processing module 202 is further configured to perform fuzzy suppression on the edge structure of the binucleated cell and the nucleoplasmic bridge in the nucleoplasmic bridge image based on the bimodal index and the pixel gradient loss, so as to obtain fine-grained boundary features of the binucleated cell and the nucleoplasmic bridge in a connected state; The execution module 203 in this application is mainly used to segment the target nucleoplasmic bridge contour from the nucleoplasmic bridge image according to the fine-grained boundary features.

[0054] In addition, the present application also provides a computer device, which includes a memory and a processor, wherein the memory stores code, and the processor is configured to obtain the code and execute the above-mentioned artificial intelligence-based nucleocytoplasmic bridge image recognition method.

[0055] In some embodiments, reference Figure 5 , which is a schematic diagram of the structure of a computer device for implementing the method for recognizing nucleoplasmic bridge images based on artificial intelligence according to some embodiments of the present application. The method for recognizing nucleoplasmic bridge images based on artificial intelligence in the above embodiments can be performed by Figure 5 The computer device shown in FIG3 is implemented as shown in FIG3 , which includes at least one processor 301 , a communication bus 302 , a memory 303 and at least one communication interface 304 .

[0056] The processor 301 can be a general-purpose central processing unit (CPU), or an application-specific integrated circuit (ASIC) or one or more processors for controlling the execution of the artificial intelligence-based nucleocytoplasmic bridge image recognition method in the present application.

[0057] The communication bus 302 may be used to transmit information between the aforementioned components.

[0058] Memory 303 may be, but is not limited to, a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), a magnetic disk or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer. Memory 303 may be independent and connected to processor 301 via communication bus 302. Memory 303 may also be integrated with processor 301.

[0059] Memory 303 is used to store program code for executing the solution of the present application, and is controlled by processor 301 for execution. Processor 301 is used to execute the program code stored in memory 303. The program code may include one or more software modules. The determination of the artificial intelligence-based nucleoplasmic bridge image recognition method in the above embodiment can be implemented by processor 301 and one or more software modules in the program code in memory 303.

[0060] The communication interface 304 uses any device such as a transceiver to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area network (WLAN), etc.

[0061] In a specific implementation, as an example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. A processor herein may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0062] The aforementioned computer device can be a general-purpose computer device or a dedicated computer device. In a specific implementation, the computer device can be a desktop computer, a portable computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiments of this application do not limit the type of computer device.

[0063] In addition, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned artificial intelligence-based nucleocytoplasmic bridge image recognition method.

[0064] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0065] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A method for recognizing nucleoplasmic bridge images based on artificial intelligence, characterized in that: The steps include: Acquiring a nucleoplasmic bridge image to be identified, and setting identification constraint conditions for the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image based on the cross-nuclear continuity characteristics of the nucleoplasmic bridge; Under the identification constraint conditions, identifying the distribution of isolated noise points in the nucleoplasmic bridge image that are different from the nucleoplasmic bridge connection structure, and constructing a pixel focusing window for local pixel feature recognition of the nucleoplasmic bridge image based on the isolated noise distribution; dividing a binucleated cell region and a nucleoplasmic bridge adhesion region in the nucleoplasmic bridge image based on the pixel focus window, and then determining a bimodal index of binucleated cells in the binucleated cell region in a pixel gradient direction and a pixel gradient loss of an adhesion boundary in the nucleoplasmic bridge adhesion region; Fuzzy suppressing the edge structure of the binucleated cell and the nucleoplasmic bridge in the nucleoplasmic bridge image according to the bimodal index and the pixel gradient loss, thereby obtaining fine-grained boundary features of the binucleated cell and the nucleoplasmic bridge in a connected state; The target nucleoplasmic bridge contour is segmented from the nucleoplasmic bridge image according to the fine-grained boundary features.

2. The method according to claim 1, wherein The identification constraint conditions for the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image are set based on the cross-nuclear continuity characteristics of the nucleoplasmic bridge, specifically including: identifying all connected domains in the nucleoplasmic bridge image; Extract the skeleton structure of each connected domain, and select the connected domains with a skeleton structure length greater than a preset threshold as candidate connecting nuclei of the nucleoplasmic bridge; The length of the connection between each two candidate connecting nuclei was taken as the cross-nuclear continuity characteristic of the nucleoplasmic bridge; The identification constraint conditions of the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image are determined according to all cross-nuclear continuity features and the center radius of all candidate connection nuclei.

3. The method according to claim 1, wherein Under the identification constraint condition, identifying the distribution of isolated noise points in the nucleoplasmic bridge image that are different from the nucleoplasmic bridge connection structure specifically includes: identifying discrete pixel blocks that do not satisfy the identification constraint conditions from the nucleoplasmic bridge image as isolated noise points; The distribution of isolated noise points in the nucleoplasmic bridge image that are different from the nucleoplasmic bridge connection structure is determined from all the isolated noise points.

4. The method according to claim 1, wherein Constructing a pixel focus window for local pixel feature recognition of the nucleoplasmic bridge image according to the isolated noise point distribution specifically includes: generating a noise point association window corresponding to each isolated noise point based on the isolated noise point distribution; Determine the gradient entropy of the pixels within the window associated with each noise point; Determining the noise focus feature of the nucleoplasmic bridge image according to the gradient entropy of pixels within each noise association window; A pixel focusing window is constructed based on the noise focus feature when performing local pixel feature recognition on the nucleoplasmic bridge image.

5. The method according to claim 4, wherein Generating a noise point association window corresponding to each isolated noise point based on the isolated noise point distribution specifically includes: Selecting an isolated noise point in the isolated noise point distribution as a selected isolated noise point; Extract the minimum bounding rectangle of the selected isolated noise point; Using the length of the minimum circumscribed rectangle as the scale of the noise point association window to construct the noise point association window for the selected isolated noise point; Continue to determine the noise point association windows corresponding to the remaining isolated noise points in the isolated noise point distribution.

6. The method according to claim 1, wherein Determining the bimodal index of binucleated cells in the binucleated cell region in the pixel gradient direction and the pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion region specifically includes: Performing radial gradient scanning on the binucleated cell region to obtain a direction histogram of pixel gradients; determining a bimodal index of binucleated cells in the binucleated cell region in the pixel gradient direction according to the main peak amplitude and the secondary peak amplitude in the direction histogram; extracting boundary pixel points of the adhesion boundary in the nucleoplasmic bridge adhesion region; Determining a local associated sub-region of each boundary pixel point, and determining a pixel change feature corresponding to each boundary pixel point according to the local associated sub-region of each boundary pixel point; The pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion region is determined according to the pixel change characteristics corresponding to each boundary pixel point.

7. The method according to claim 1, wherein Images of the nucleoplasmic bridges to be identified are acquired by light microscopy.

8. An artificial intelligence-based nucleocytoplasmic bridge image recognition system, characterized in that: include: an acquisition module, configured to acquire a nucleoplasmic bridge image to be identified, and set identification constraint conditions for the nucleoplasmic bridge connection structure in the nucleoplasmic bridge image based on the cross-nuclear continuity feature of the nucleoplasmic bridge; a processing module, configured to identify, under the identification constraint conditions, a distribution of isolated noise points in the nucleoplasmic bridge image that are distinct from the nucleoplasmic bridge connection structure, and construct a pixel focusing window for performing local pixel feature recognition on the nucleoplasmic bridge image based on the isolated noise distribution; The processing module is further configured to divide the binucleated cell region and the nucleoplasmic bridge adhesion region in the nucleoplasmic bridge image based on the pixel focusing window, and then determine the bimodal index of the binucleated cells in the binucleated cell region in the pixel gradient direction and the pixel gradient loss of the adhesion boundary in the nucleoplasmic bridge adhesion region; The processing module is further configured to perform fuzzy suppression on the edge structure of the binucleated cell and the nucleoplasmic bridge in the nucleoplasmic bridge image based on the bimodal index and the pixel gradient loss, so as to obtain fine-grained boundary features of the binucleated cell and the nucleoplasmic bridge in a connected state; An execution module is used to segment the target nucleoplasmic bridge contour from the nucleoplasmic bridge image according to the fine-grained boundary features.

9. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a code, and the processor is configured to obtain the code and execute the artificial intelligence-based nucleoplasmic bridge image recognition method according to 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, the artificial intelligence-based nucleoplasmic bridge image recognition method according to any one of claims 1 to 7 is implemented.

Images