A fluorescent image intelligent recognition method, system, device and medium

Abstract

The application provides a fluorescent image intelligent recognition method, system, device and medium, and relates to the technical field of image recognition. The method comprises the following steps: acquiring a plurality of original fluorescent images corresponding to a plurality of dilutions of a sample; extracting texture morphological features, fluorescent bright spot spatial distribution features and average fluorescent intensity of a cell nucleus region in a target original fluorescent image, the target original fluorescent image being any one of the plurality of original fluorescent images; fusing the texture morphological features and the fluorescent bright spot spatial distribution features to obtain a mixed feature vector; judging the mixed feature vector through a preset karyotype classification model to obtain a karyotype category of the target original fluorescent image; and calculating an antibody titer value according to the plurality of dilutions, the karyotype category corresponding to each dilution and the average fluorescent intensity corresponding to each dilution. The technical solution provided by the application improves the efficiency, objectivity and accuracy of detection.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method, system, device and medium for intelligent recognition of fluorescent images. Background Technology

[0002] In the detection of specific autoantibodies, indirect immunofluorescence assay is widely considered the standard method. How to objectively and accurately interpret the karyotypes of the large number of fluorescence images generated during the detection process and quickly determine the final antibody titer has become a key requirement for improving detection accuracy and diagnostic efficiency.

[0003] In existing technologies, antibody titer determination mainly relies on manual interpretation by laboratory physicians under a fluorescence microscope. This process requires physicians to examine samples of different dilutions one by one, subjectively identify the fluorescent karyotypes of cells (such as homogeneous, granular, etc.), and find the highest dilution showing a positive reaction. However, in practical applications, existing technologies are not only inefficient and have limited throughput, but the interpretation results are also easily affected by the physician's subjective experience and fatigue, making it difficult to guarantee consistency. Furthermore, some preliminary automated methods are mostly based on comparing the overall average fluorescence intensity of the image with a preset fixed threshold, which poses a serious risk of false positives or false negatives for samples with complex and diverse fluorescent karyotypes or intensities close to the critical value. Summary of the Invention

[0004] In view of this, this application provides a method, system, device and medium for intelligent recognition of fluorescence images to solve the above problems.

[0005] Firstly, a method for intelligent recognition of fluorescence images is provided, applied to a computer processing device, the method comprising:

[0006] Multiple original fluorescence images corresponding to different dilutions of the sample are obtained, with one original fluorescence image corresponding to each dilution.

[0007] Extract the texture and morphological features of the cell nucleus region, the spatial distribution features of fluorescent bright spots, and the average fluorescence intensity from the original fluorescence image of the target. The original fluorescence image of the target is any one of multiple original fluorescence images.

[0008] The texture morphology features and the spatial distribution features of fluorescent bright spots are fused to obtain a hybrid feature vector;

[0009] The kernel type of the original fluorescence image of the target is obtained by judging the mixed feature vector through a preset kernel type classification model.

[0010] The antibody titer was calculated based on multiple dilutions, the corresponding nucleotypes for each dilution, and the average fluorescence intensity for each dilution.

[0011] The above technical solution acquires multiple fluorescence images corresponding to a series of sample dilutions. The derivation process is not based directly on single brightness information, but rather on deep feature analysis of each image. This involves extracting morphological features describing nucleal texture details, spatial distribution features characterizing the density and positional relationship of fluorescent bright spots, and the overall average fluorescence intensity. Through feature fusion and a pre-defined classification model, each image is objectively assigned a karyotype category. By comprehensively utilizing the sample dilution information, the objectively determined karyotype category, and the average fluorescence intensity of the image, the antibody titer value is calculated. By simulating the comprehensive consideration of morphology, distribution, and brightness during expert interpretation, this approach achieves objective and accurate karyotype interpretation of fluorescence images and automated calculation of antibody titers, improving the efficiency, objectivity, and accuracy of the detection.

[0012] Optionally, extract the texture and morphological features of the cell nucleus region, the spatial distribution features of fluorescent bright spots, and the average fluorescence intensity from the original fluorescence image of the target, specifically including:

[0013] The cell nucleus region is segmented from the original fluorescence image of the target using an image segmentation algorithm;

[0014] Calculate the area, perimeter, circularity, and gray-level co-occurrence matrix of the cell nucleus region, extract contrast, second angular moment, and correlation, and combine the area, perimeter, circularity, contrast, second angular moment, and correlation to form texture morphology features.

[0015] The image within the cell nucleus region was subjected to top-hat transformation and binarization to identify all fluorescent bright spots and determine the coordinate position of each fluorescent bright spot;

[0016] Based on the number and coordinates of all fluorescent bright spots, the average nearest neighbor distance and distribution density are calculated, and the number, average nearest neighbor distance and distribution density together constitute the spatial distribution characteristics of fluorescent bright spots.

[0017] The average fluorescence intensity is obtained by calculating the average gray value of all pixels within the cell nucleus region.

[0018] The above technical solution specifies the detailed calculation methods for three key image features. It transforms the abstract visual information in the original image into a set of quantifiable and concrete numerical features, including texture morphology features describing the physical size and texture of cell nuclei, spatial distribution features describing the number and positional relationships of fluorescent bright spots, and an overall brightness value. This provides accurate and repeatable input data for subsequent feature fusion and classification.

[0019] Optionally, the cell nucleus region is segmented from the original fluorescence image of the target using an image segmentation algorithm, specifically including:

[0020] Gaussian filtering is applied to the original fluorescence image to obtain a smoothed image;

[0021] Calculate the gradient of the smoothed image and generate the gradient energy field;

[0022] Set the initial contour line in the gradient energy field;

[0023] The total energy value of the initial contour line is calculated based on the preset internal energy function and external energy function. The position of the initial contour line is iteratively adjusted until the total energy value reaches the minimum and the edge of the cell nucleus is outlined.

[0024] The area enclosed by the adjusted initial outline is defined as the cell nucleus region.

[0025] The above technical solution, by applying an active contour model based on energy function iterative optimization, achieves accurate segmentation and extraction of the cell nucleus region from the original fluorescence image. It accurately defines the target region for feature analysis, eliminating interference from background and other non-target regions, thereby ensuring the accuracy and reliability of subsequent feature extraction.

[0026] Optionally, the image within the cell nucleus region is subjected to a top-cap transform and binarized to identify all fluorescent bright spots and determine the coordinates of each bright spot. Specifically, this includes:

[0027] A top-cap morphological transformation is performed on the image within the cell nucleus region using a preset structuring element to obtain an enhanced image. The size of the structuring element is smaller than the fluorescent bright spot to be identified.

[0028] The optimal segmentation threshold for the enhanced image is calculated using the big law method, and the enhanced image is binarized using the optimal segmentation threshold to obtain a binary image containing fluorescent bright pixels.

[0029] Connectivity analysis is performed on the binary image to identify each independent connected component as an independent fluorescent bright spot;

[0030] Calculate the geometric centroid of each fluorescent spot, and determine the coordinates of the geometric centroid to determine the coordinate position of each fluorescent spot.

[0031] The above technical solution, through a series of image processing steps including top-hat morphological transformation, large-law threshold segmentation, and connected component analysis, achieves the automatic identification and localization of all tiny fluorescent bright spots within the cell nucleus region. It accurately identifies each individual fluorescent bright spot and outputs its quantity and precise coordinates, providing fundamental data for calculating the spatial distribution characteristics of fluorescent bright spots.

[0032] Optionally, texture morphology features and the spatial distribution features of fluorescent bright spots can be fused to obtain a hybrid feature vector, specifically including:

[0033] By concatenating the texture morphology features with the spatial distribution features of fluorescent bright spots, a high-dimensional feature vector is obtained.

[0034] Applying principal component analysis to high-dimensional eigenvectors yields multiple principal components and their respective contribution rates.

[0035] From multiple principal components, target principal components are selected sequentially in descending order of their contribution rates until the cumulative contribution rate of all target principal components reaches or exceeds a preset threshold for the first time, and all target principal components are combined to form a hybrid feature vector.

[0036] The above technical solution achieves effective feature fusion by concatenating features from multiple different sources and applying principal component analysis algorithm for dimensionality reduction. It transforms a high-dimensional feature vector with high dimensionality and potential information redundancy into a hybrid feature vector with lower dimensionality while retaining the essence of the main information. This not only reduces the computational burden of the subsequent classification model but also helps to improve the classification performance of the model.

[0037] Optionally, the mixed feature vector is judged by a preset kernel type classification model to obtain the kernel type category of the original fluorescence image of the target, specifically including:

[0038] The mixed feature vectors are input into multiple support vector machine classifiers in the kernel type classification model, where each support vector machine classifier corresponds to a preset kernel type category;

[0039] The decision function value of the mixture feature vector is calculated using each support vector machine classifier. The decision function value is used to characterize the probability of belonging to the kernel type class corresponding to the support vector machine classifier.

[0040] The kernel type corresponding to the support vector machine classifier with the largest decision function value is determined as the kernel type of the original fluorescence image of the target.

[0041] The above technical solution achieves multi-class classification by inputting feature vectors into a set of support vector machine classifiers corresponding to different kernel types and comparing the decision function values ​​of each classifier. It can automatically assign a best-matching, preset kernel type to the corresponding original fluorescence image based on the input mixed feature vectors, thus completing intelligent classification of image content.

[0042] Optionally, the antibody titer value can be calculated based on multiple dilutions, the corresponding karyotype for each dilution, and the average fluorescence intensity for each dilution. Specifically, this includes:

[0043] Among the multiple karyotype categories identified for multiple original fluorescence images, the target karyotype category is determined by a preset decision rule, and the positive interpretation fluorescence intensity threshold corresponding to the target karyotype category is queried and determined from a preset threshold mapping table based on the target karyotype category.

[0044] Each average fluorescence intensity corresponding to various dilutions is compared with the positive interpretation fluorescence intensity threshold.

[0045] Among all the first average fluorescence intensities that are higher than the positive interpretation fluorescence intensity threshold, the second average fluorescence intensity is selected. The dilution corresponding to the second average fluorescence intensity is the highest, and the dilution corresponding to the second average fluorescence intensity is determined as the antibody titer value.

[0046] The above technical solution defines a set of clear rules for determining the final antibody titer from multiple test results. A unified target karyotype category is determined from potentially inconsistent karyotype classification results using decision rules; based on this category, a precise positive interpretation fluorescence intensity threshold is found; finally, by comparing the intensity at each dilution with this threshold, the highest dilution meeting the positive standard is identified, thereby calculating an objective and consistent antibody titer value.

[0047] Secondly, a fluorescent image intelligent recognition system is provided, comprising an image acquisition module, a feature extraction module, a feature fusion module, a karyotype classification module, and a titer calculation module, wherein:

[0048] The image acquisition module is configured to acquire multiple original fluorescence images corresponding to different dilutions of the sample, with one original fluorescence image corresponding to each dilution.

[0049] The feature extraction module is configured to extract the texture morphology features of the cell nucleus region, the spatial distribution features of fluorescent bright spots, and the average fluorescence intensity in the target original fluorescence image. The target original fluorescence image is any one of multiple original fluorescence images.

[0050] The feature fusion module is configured to fuse texture morphology features and fluorescent bright spot spatial distribution features to obtain a hybrid feature vector;

[0051] The kernel type classification module is configured to judge the mixed feature vectors through a preset kernel type classification model to obtain the kernel type category of the original fluorescence image of the target;

[0052] The titer calculation module is configured to calculate the antibody titer value based on multiple dilutions, the nucleotype category corresponding to each dilution, and the average fluorescence intensity corresponding to each dilution.

[0053] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described above.

[0054] A fourth aspect of this application provides a computer-readable storage medium storing instructions that, when executed, perform the method described in any of the preceding descriptions.

[0055] In summary, implementing one or more technical solutions provided in this application has at least the following technical effects or advantages:

[0056] By acquiring multiple fluorescence images corresponding to a series of sample dilutions, the derivation process is not directly based on a single brightness information. Instead, it performs deep feature analysis on each image, extracting morphological features that describe the details of cell nuclear texture, spatial distribution features that characterize the density and positional relationship of fluorescent bright spots, and the overall average fluorescence intensity. Through feature fusion and a pre-defined classification model, each image is objectively assigned a karyotype category. By comprehensively utilizing the sample dilution information, the objectively determined karyotype category, and the average fluorescence intensity of the image, the antibody titer value is calculated. By simulating the comprehensive consideration of morphology, distribution, and brightness during expert interpretation, objective and accurate karyotype interpretation of fluorescence images and automated calculation of antibody titers are achieved, improving the efficiency, objectivity, and accuracy of detection. Attached Figure Description

[0057] Figure 1 This is an exemplary system architecture diagram of a fluorescent image intelligent recognition method or a fluorescent image intelligent recognition system applying the present application;

[0058] Figure 2 This is a flowchart illustrating a fluorescent image intelligent recognition method according to an embodiment of this application;

[0059] Figure 3 This is a schematic diagram of a fluorescent image intelligent recognition system according to an embodiment of this application;

[0060] Figure 4 This is a schematic diagram of the structure of an electronic device disclosed in the application embodiment.

[0061] Explanation of reference numerals in the attached figures: 100, System architecture; 101, First terminal device; 102, Second terminal device; 103, Third terminal device; 104, Network; 105, Server; 301, Image acquisition module; 302, Feature extraction module; 303, Feature fusion module; 304, Kernel type classification module; 305, Titer calculation module; 401, Processor; 402, Communication bus; 403, User interface; 404, Network interface; 405, Memory. Detailed Implementation

[0062] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0063] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0064] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0065] Figure 1 A schematic diagram of an exemplary system architecture is shown, illustrating an embodiment of a fluorescent image intelligent recognition method or a fluorescent image intelligent recognition system to which this application can be applied.

[0066] like Figure 1 As shown, the system architecture 100 may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0067] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as model training applications, video recognition applications, web browser applications, social platform software, etc.

[0068] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices with displays, including but not limited to smartphones, tablets, e-book readers, MP3 (Moving Picture Experts Group Audio Layer III) players, MP4 (Moving Picture Experts Group Audio Layer IV) players, laptops, and desktop computers, etc. When terminal devices 101, 102, and 103 are software, they can be installed in the aforementioned electronic devices. They can be implemented as multiple software programs or software modules (e.g., multiple software programs or software modules used to provide distributed services) or as a single software program or software module. No specific limitations are imposed here.

[0069] When terminals 101, 102, and 103 are hardware devices, video capture devices can also be installed on them. These video capture devices can be various devices capable of capturing video, such as cameras, sensors, etc. Users can use the video capture devices on terminals 101, 102, and 103 to capture video.

[0070] Server 105 can be a server that provides various services, such as a backend server for processing data displayed on terminal devices 101, 102, and 103. The backend server can analyze and process the received data and can feed back the processing results (such as recognition results) to the terminal devices.

[0071] It should be noted that a server can be either hardware or software. When the server is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When the server is software, it can be implemented as multiple software programs or software modules (e.g., multiple software programs or software modules used to provide distributed services), or as a single software program or software module. No specific limitations are made here.

[0072] It should be understood that Figure 1The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included. In particular, if the target data does not need to be obtained remotely, the above system architecture may exclude the network and include only terminal devices or servers.

[0073] Figure 2 This is a flowchart illustrating a fluorescence image intelligent recognition method according to an embodiment of this application. This method can be implemented using a computer program, a microcontroller, or run on a fluorescence image intelligent recognition system. The computer program can be integrated into the application or run as a standalone utility application. The specific steps of the fluorescence image intelligent recognition method are described in detail below.

[0074] S201: Obtain multiple original fluorescence images corresponding to different dilutions of the sample, with one original fluorescence image corresponding to each dilution.

[0075] In the embodiments of this application, multiple dilutions of a sample refer to preparing a series of test liquids of different concentrations from the original test sample (e.g., biological serum) through a gradient dilution method, which are used to represent different concentration levels of the antibody. For example, the original sample is diluted sequentially to a series of concentration gradients such as 1:40, 1:80, 1:160, and 1:320.

[0076] Specifically, this step is typically performed within an automated experimental analysis workflow. The automated equipment aspirates samples from the original test sample and performs a series of gradient dilutions using diluents to obtain samples of various dilutions. Each dilution is then individually added to a slide containing an experimental substrate (e.g., HEp-2 cells) for reaction, incubation, and washing. After fluorescent labeling, the slide is fed into the imaging system of an automated fluorescence microscope. The imaging system automatically focuses and photographs the region corresponding to each dilution sample; the resulting unprocessed digital image is the raw fluorescence image. During acquisition, the system binds and records the numerical information of each acquired raw fluorescence image with its corresponding dilution, ultimately forming a dataset containing multiple raw fluorescence images and their respective dilution information for subsequent processing.

[0077] S202: Extract the texture and morphological features of the cell nucleus region, the spatial distribution features of fluorescent bright spots, and the average fluorescence intensity from the target original fluorescence image. The target original fluorescence image is any one of multiple original fluorescence images.

[0078] For example, for a target raw fluorescence image where the cell nucleus region has been segmented, the system performs the following feature extraction process: The system performs geometric parameter and texture analysis on the segmented cell nucleus region. For example, the area is obtained by calculating the total number of pixels in the region (e.g., 12500 pixels), the perimeter is obtained by calculating the length of the contour line (e.g., 400 pixels), and the roundness (e.g., 0.92) is calculated based on these two values. The system constructs a gray-level co-occurrence matrix with specified parameters (e.g., offset of 1 pixel, direction of 0 degrees) and extracts contrast (e.g., 0.15), angular second moment (e.g., 0.88), and correlation (e.g., 0.95) from it. These values ​​are combined into a six-dimensional texture morphology feature vector: [12500, 400, 0.92, 0.15, 0.88, 0.95]. The system uses a preset disk-shaped structural element with a radius of 3 pixels to perform a top-cap morphological transformation on the cell nucleus region image to highlight fluorescent bright spots. The system automatically determines and processes the binarization threshold using a large-scale method, then performs connected component analysis on the generated binary image to identify, for example, 35 independent fluorescent bright spots and record their coordinate positions. Based on the number and coordinates of these 35 bright spots, the system calculates the average nearest neighbor distance (e.g., 4.5 pixels) and distribution density (e.g., 35 / 12500 = 0.0028). These values ​​are combined into a three-dimensional feature vector of fluorescent bright spot spatial distribution: [35, 4.5, 0.0028]. The system accumulates the gray values ​​of all 12,500 pixels within the cell nucleus region and divides them by the area of ​​12,500 to obtain the average fluorescence intensity value of that region (e.g., 188.5). Through the above steps, the system successfully extracts three specific and quantified feature sets from the original fluorescent image of the target, providing a data foundation for subsequent feature fusion and classification steps.

[0079] In one possible implementation, the textural morphology features, spatial distribution features of fluorescent bright spots, and average fluorescence intensity of the cell nucleus region in the original target fluorescence image are extracted. Specifically, this includes: segmenting the cell nucleus region from the original target fluorescence image using an image segmentation algorithm; calculating the area, perimeter, circularity, and gray-level co-occurrence matrix of the cell nucleus region, extracting contrast, second moment of angle, and correlation, and combining the area, perimeter, circularity, contrast, second moment of angle, and correlation to constitute the textural morphology features; performing a top-hat transform and binarization on the image within the cell nucleus region to identify all fluorescent bright spots and determine the coordinate position of each fluorescent bright spot; calculating the average nearest neighbor distance and distribution density based on the number and coordinate position of all fluorescent bright spots, and combining the number, average nearest neighbor distance, and distribution density to constitute the spatial distribution features of fluorescent bright spots; and calculating the average gray value of all pixels within the cell nucleus region to obtain the average fluorescence intensity.

[0080] In the embodiments of this application, texture morphology features refer to a set of numerical measures used to quantify the physical shape, size, and internal pixel grayscale distribution of the cell nucleus region. It integrates the geometric shape of the target (such as size and roundness) with surface texture (such as roughness, graininess, and directionality). For example, the texture morphology feature values ​​of a smooth and rounded cell nucleus and a cell nucleus with irregular contours and coarse granular interior will be significantly different.

[0081] Specifically, after successfully segmenting the cell nucleus region from the original fluorescence image using an image segmentation algorithm, the system performs a series of calculations to extract three core features: texture morphology features, spatial distribution features of fluorescent bright spots, and average fluorescence intensity. Texture morphology features are extracted by calculating the geometric parameters of the cell nucleus region, where area represents the size of the nucleus, perimeter represents the length of its outline, and roundness measures how closely its shape approximates a standard circle. To describe the internal texture details, the system calculates the gray-level co-occurrence matrix of the region, a matrix reflecting the spatial distribution of pixel gray levels. Based on this matrix, it further extracts contrast (representing image sharpness and texture depth), angular second moment (representing the uniformity of image gray-level distribution and texture coarseness), and correlation (representing the linear dependence of local gray levels in the image). These six metrics—area, perimeter, roundness, contrast, angular second moment, and correlation—are combined to form a comprehensive texture morphology feature vector. Extracting the spatial distribution features of fluorescent bright spots: The system performs a top-hat transform on the image within the segmented cell nucleus region. This is a morphological operation that significantly enhances tiny bright spots in the image. Binarization is then performed to separate the enhanced fluorescent bright spots from the background. Through connected component analysis of the binary image, each individual fluorescent bright spot is identified and its coordinate position is determined. Based on the number and coordinate positions of all identified fluorescent bright spots, the system further calculates their average nearest neighbor distance (representing the average distance between bright spots) and distribution density (representing the density of bright spots within the cell nucleus region). The number of bright spots, average nearest neighbor distance, and distribution density together constitute the spatial distribution features of the fluorescent bright spots. Extracting the average fluorescence intensity: The system traverses every pixel within the cell nucleus region, reads its grayscale value, and calculates the arithmetic mean of all these pixel grayscale values ​​to obtain an average fluorescence intensity value that represents the overall fluorescence brightness of the cell nucleus.

[0082] In one possible implementation, the cell nucleus region is segmented from the original fluorescence image using an image segmentation algorithm. Specifically, this includes: performing Gaussian filtering on the original fluorescence image to obtain a smooth image; calculating the gradient of the smooth image to generate a gradient energy field; setting an initial contour line in the gradient energy field; calculating the total energy value of the initial contour line according to preset internal and external energy functions; iteratively adjusting the position of the initial contour line until the total energy value reaches a minimum and delineates the edge of the cell nucleus; and determining the region enclosed by the adjusted initial contour line as the cell nucleus region.

[0083] In this embodiment, the internal energy function and the external energy function refer to two constraints that jointly determine the shape and position of the contour line in the active contour model. The internal energy function is used to control the properties of the contour line itself, such as keeping it smooth and continuous, and preventing sharp inflections or breaks. The external energy function, on the other hand, originates from the features of the image itself and is used to attract the contour line to the edge of interest in the image. For example, in areas with large gradient values ​​(i.e., edges), the value of the external energy function is usually designed to be minimized to guide the contour line to move towards that location.

[0084] Specifically, this image segmentation process aims to accurately identify and extract cell nucleus regions from the background. The system performs Gaussian filtering on the input raw fluorescence image to effectively suppress inherent random noise, resulting in a smooth image. The system calculates the gradient of this smooth image, generating a gradient energy field; in this energy field, regions with drastic pixel grayscale changes (i.e., potential edges of cell nuclei) exhibit higher energy values. The system pre-defines an initial contour line in this energy field, the initial position of which can be determined through a simple preprocessing step, for example, by setting it at the geometric center of the entire image or at the centroid of the largest connected region obtained through preliminary thresholding. In the iterative adjustment phase, the system calculates the total energy value of the current initial contour line based on a pre-defined internal energy function (constraining the smoothness of the contour line) and an external energy function (guiding the contour line towards high-gradient edges in the energy field). The algorithm iteratively fine-tunes the positions of various points on the contour line, with each adjustment aiming to reduce the total energy value until it reaches a minimum. At this point, the position of the contour line stabilizes and no longer changes significantly, thus closely outlining the edge of the cell nucleus. When the iteration process terminates, the system ultimately determines the closed region enclosed by this adjusted and optimized initial contour line as the cell nucleus region for subsequent feature extraction. For example, in a 5x5 pixel region representing the edge of the cell nucleus, the actual gradient value of the cell nucleus edge is as high as 90, while there is an isolated noise point with a gradient value of 50 nearby. The system calculates the total energy value of the contour line based on a preset internal energy function (penalizing sharp corners) and an external energy function (attracted by high gradient values). Suppose an initial contour line consisting of three points P1, P2, and P3 is being adjusted, where P1 and P3 are already on the edge. The algorithm needs to decide between two possible positions for the middle point P2: point A, located on the high gradient edge between P1 and P3, moving to this point will form a smooth straight line; point B, the noise point with a gradient value of 50, moving to this point will form a sharp corner. When calculating the total energy value, if P2 moves to point A, its external energy is extremely low (approximately 0.01), and its internal energy is 0 (smooth morphology), resulting in a total energy value of approximately 0.01. However, if P2 moves to point B, although its external energy is also low (approximately 0.02), the internal energy will incur a significant penalty value (e.g., 5.0) due to the corner formation, causing the total energy value to spike to 5.02. Therefore, the constraint of the internal energy function on the morphology guides the algorithm to select point A, which has a much lower total energy value, thus preventing distortion caused by noise points affecting the contour line. This iterative fine-tuning process is repeated for all points on the entire contour line, with each adjustment aiming to reduce the total energy value until it reaches a local minimum, thereby achieving precise segmentation of the cell nucleus region.

[0085] In one possible implementation, a top-hat transformation and binarization are performed on the image within the cell nucleus region to identify all fluorescent bright spots and determine the coordinate position of each bright spot. Specifically, this includes: performing a top-hat morphological transformation on the image within the cell nucleus region using a preset structuring element to obtain an enhanced image, wherein the size of the structuring element is smaller than the fluorescent bright spots to be identified; calculating the optimal segmentation threshold of the enhanced image using the big rule method, and using the optimal segmentation threshold to binarize the enhanced image to obtain a binary image containing fluorescent bright spot pixels; performing connected component analysis on the binary image to identify each independent connected component as an independent fluorescent bright spot; calculating the geometric centroid of each fluorescent bright spot, and determining the coordinates of the geometric centroid as the coordinate position of each fluorescent bright spot.

[0086] In the embodiments of this application, top-hat morphological transformation refers to a morphological operation in digital image processing used to extract bright features smaller than a specific structural element in an image. By subtracting the image version obtained through opening operations, from which these bright features have been filtered out, from the original image, it can very effectively correct uneven lighting backgrounds and significantly enhance local feature points that are brighter than their surroundings, such as the fluorescent bright spots that need to be identified in this application.

[0087] Specifically, this process aims to automatically identify and precisely locate all individual fluorescent bright spots from a segmented image of the cell nucleus region. The system uses a preset structuring element (a small-sized shape template, its size set smaller than the fluorescent bright spots to be identified) to perform a top-hat morphological transformation on the image within the cell nucleus region. This transformation effectively suppresses uneven background brightness within the cell nucleus while highlighting individual fluorescent bright spots, resulting in an enhanced image. In this enhanced image, large-scale background grayscale variations in the original image are suppressed, making the grayscale values ​​of the background region more uniform, while improving the contrast between the fluorescent bright spots and their local background. To completely separate these bright spots from the background, the system uses Otsu's method to analyze the enhanced image. This method automatically calculates an optimal segmentation threshold to distinguish bright spot pixels from background pixels in a way that maximizes the inter-class variance. After binarizing the enhanced image using this threshold, a binary image containing only black and white pixels is obtained, where the white pixel areas represent all fluorescent bright spot pixels. The system performs connected component analysis on the binary image, grouping interconnected white pixels together. Each individual pixel cluster (i.e., connected component) is identified as an independent fluorescent bright spot. To assign a unique coordinate to each identified fluorescent bright spot, the system calculates the geometric centroid of each fluorescent bright spot (i.e., each connected component) and determines the coordinates of that centroid as the coordinate position of the fluorescent bright spot.

[0088] S203: The texture morphology features and the spatial distribution features of fluorescent bright spots are fused to obtain a hybrid feature vector.

[0089] For example, to effectively integrate the previously extracted multi-source features, the texture morphology feature vector and the fluorescent bright spot spatial distribution feature vector are concatenated end-to-end to integrate them into a high-dimensional feature vector containing all the information. To eliminate any redundant information and reduce the complexity of this vector, the system then applies PCA (Principal Component Analysis), a statistical method for data dimensionality reduction, to transform the high-dimensional data into a set of principal components arranged in descending order of contribution rate (i.e., the degree to which they explain the total variance of the original data). The system sequentially selects the principal components with the highest contribution rates until the cumulative contribution rate of the selected target principal components reaches or exceeds a preset threshold. These selected core principal components then collectively constitute a hybrid feature vector. This hybrid feature vector has a lower dimension than the high-dimensional feature vector and concentrates the main variation information in the original features limited by the cumulative contribution rate threshold (e.g., 95%) for subsequent classification processing.

[0090] In one possible implementation, the texture morphology features and the spatial distribution features of fluorescent bright spots are fused to obtain a hybrid feature vector. Specifically, this includes: concatenating the texture morphology features and the spatial distribution features of fluorescent bright spots to obtain a high-dimensional feature vector; applying principal component analysis to the high-dimensional feature vector to obtain multiple principal components and their corresponding contribution rates; selecting target principal components from the multiple principal components in descending order of their contribution rates until the cumulative contribution rate of all target principal components reaches or exceeds a preset threshold for the first time, and then combining all target principal components to form a hybrid feature vector.

[0091] In this embodiment, PCA refers to a commonly used statistical method for data dimensionality reduction. Its core idea is to transform a set of potentially correlated high-dimensional variables into a set of linearly uncorrelated low-dimensional variables through orthogonal transformation. This new set of low-dimensional variables is called principal components. The algorithm aims to retain as much variation information as possible from the original data using the fewest number of principal components. The first principal component explains the largest amount of variation, and subsequent principal components explain the largest portion of the remaining variation in turn.

[0092] Specifically, this feature fusion and dimensionality reduction process aims to integrate previously extracted multi-source features into a compact and information-rich single feature vector. The system concatenates the extracted texture morphology feature vector (e.g., a six-dimensional vector) with the fluorescent bright spot spatial distribution feature vector (e.g., a three-dimensional vector), directly joining the two vectors to form a high-dimensional feature vector containing all the original feature information but with a higher dimension (e.g., nine dimensions). To eliminate potentially redundant information and reduce the complexity of this high-dimensional feature vector, the system applies principal component analysis (PCA) to it. This algorithm calculates multiple (e.g., nine) new, mutually orthogonal principal components, along with their corresponding contribution rates, which represent the extent to which each principal component explains the total variance of the original data. The system selects features according to preset rules: it selects target principal components in descending order of their contribution rates and calculates the cumulative contribution rate of all selected target principal components in real time. The selection process terminates when the cumulative value first reaches or exceeds a preset threshold (for example, preferably, the threshold can be set between 90% and 99%, and 95% is used as an example in this embodiment). All the finally selected target principal components (for example, the cumulative contribution rate of only the first four principal components may exceed 95%) will be combined to form a new, significantly reduced-dimensional hybrid feature vector for subsequent classification processing.

[0093] S204: The mixed feature vector is judged by a preset kernel type classification model to obtain the kernel type category of the original fluorescence image of the target.

[0094] For example, a highest confidence selection strategy is employed, the core of which is a pre-defined model integrating multiple Support Vector Machine (SVM) classifiers. Each SVM classifier, serving as a supervised learning model, is specifically trained to identify a particular pre-defined kernel type. During the decision-making process, the input mixed feature vector is simultaneously fed into all SVM classifiers. Each classifier independently computes a decision function value, which characterizes the confidence or probability that the input sample belongs to its corresponding class. The system compares all output decision function values ​​and determines the kernel type corresponding to the classifier that produces the highest value as the final classification result for the image.

[0095] Regarding the training process of the preset karyotype classification model, exemplarily, the construction of the preset karyotype classification model (i.e., the preset process) can be completed through the following steps: Prepare a large-scale, professionally labeled training dataset. This dataset contains a sufficient number of original fluorescence images, and each image has been manually interpreted and labeled with its exact karyotype category (e.g., homogeneous, granular, nucleolar, etc.) by one or more experienced laboratory physicians. For each image in the training dataset, the system will automatically extract its features strictly according to the methods described in steps S202 and S203 above, and generate the final mixed feature vector. This forms a training sample library where feature vectors correspond one-to-one with the true karyotype category labels. The system uses a highest confidence selection strategy to train multiple SVM classifiers. Taking the training of the homogeneous classifier as an example, the system will use all sample feature vectors labeled as homogeneous as positive samples, and all sample feature vectors labeled as heterogeneous (e.g., granular, nucleolar, etc.) as negative samples, and train the SVM classifier based on this, enabling it to distinguish between homogeneous and heterogeneous types. This training process is repeated for each preset kernel type, resulting in multiple specialized SVM classifiers. The collection of these fully trained classifiers constitutes the final preset kernel type classification model used for classification.

[0096] In one possible implementation, the kernel type of the target original fluorescence image is obtained by judging the mixed feature vector through a preset kernel type classification model. Specifically, this includes: inputting the mixed feature vector into multiple SVM classifiers included in the kernel type classification model, wherein each SVM classifier corresponds to a preset kernel type; calculating the decision function value of the mixed feature vector using each SVM classifier, the decision function value being used to characterize the probability of belonging to the kernel type corresponding to the SVM classifier; and determining the kernel type corresponding to the SVM classifier with the largest decision function value as the kernel type of the target original fluorescence image.

[0097] In this embodiment, the SVM classifier refers to a supervised learning model whose basic function is to determine which of two preset categories a given input data (i.e., feature vector) should belong to. It achieves classification by finding an optimal hyperplane in the feature space that can separate the two classes of sample points with the maximum margin. For example, an SVM classifier can be trained specifically to determine whether the features of a fluorescence image are homogeneous or heterogeneous.

[0098] Specifically, this classification process aims to assign the most accurate karyotype category to the mixed feature vector generated in the previous step. Since fluorescence images typically have multiple karyotype categories (e.g., homogeneous, granular, nucleolar, etc.), the pre-defined karyotype classification model employs a highest-confidence selection strategy. This model integrates multiple SVM classifiers, each specifically trained to identify only one pre-defined karyotype category. Upon receiving the input mixed feature vector, the system feeds it into all the SVM classifiers included in the model. Each classifier independently analyzes the vector and calculates a decision function value. In this embodiment, the decision function value characterizes the confidence or probability that the input vector belongs to the specific karyotype category corresponding to that classifier (the higher the value, the higher the probability of belonging to that category). After all classifiers have completed their calculations, the system compares all output decision function values ​​and determines the karyotype category of the original fluorescence image based on the SVM classifier with the highest decision function value.

[0099] S205: The antibody titer value is calculated based on multiple dilutions, the nuclear type corresponding to each dilution, and the average fluorescence intensity corresponding to each dilution.

[0100] For example, using preset decision rules, potentially inconsistent karyotype results from multiple images are identified as a unified target karyotype category. Based on this target karyotype category, the system retrieves a corresponding, specific positive fluorescence intensity threshold from a preset mapping table and compares the average fluorescence intensity at each dilution with this threshold to filter out all positive results. The highest dilution among all positive results is then determined as the objective antibody titer value, representing the highest dilution factor capable of detecting a positive reaction.

[0101] In one possible implementation, the antibody titer value is calculated based on multiple dilutions, the karyotype category corresponding to each dilution, and the average fluorescence intensity corresponding to each dilution. Specifically, this includes: determining a target karyotype category from multiple karyotype categories identified for multiple original fluorescence images using a preset decision rule; querying and determining the positive fluorescence intensity threshold corresponding to the target karyotype category from a preset threshold mapping table based on the target karyotype category; comparing each average fluorescence intensity corresponding to the multiple dilutions with the positive fluorescence intensity threshold; selecting a second average fluorescence intensity from all first average fluorescence intensities higher than the positive fluorescence intensity threshold, wherein the dilution corresponding to the second average fluorescence intensity is the highest, and determining the dilution corresponding to the second average fluorescence intensity as the antibody titer value.

[0102] In the embodiments of this application, antibody titer value refers to an indicator used in immunological detection to semi-quantitatively represent the concentration of antibodies in a sample. It is determined by finding the highest dilution factor that can detect a positive reaction after performing a series of serial dilutions on the sample. This dilution factor is the titer value, for example, 1:160, which indicates that the antibody can still be effectively detected after being diluted 160 times.

[0103] Specifically, this calculation process aims to derive a final, objective antibody titer value based on the series of data generated from all previous steps. The system summarizes the results of multiple karyotype categories corresponding to all processed original fluorescence images. Since the classification results of different images may have slight differences (e.g., most are classified as homogeneous, and a few as particulate), the system determines a unique target karyotype category that can represent the state of the entire sample through a preset decision rule (e.g., a majority vote). Based on the determined target karyotype category, the system queries a preset threshold mapping table to obtain a corresponding positive fluorescence intensity threshold. For example, after determining that the target karyotype is homogeneous, the corresponding positive fluorescence intensity threshold is found to be 150. The system compares each average fluorescence intensity corresponding to various dilutions with this queried positive fluorescence intensity threshold (150). Of all the comparison results, the system will filter out those with a first average fluorescence intensity higher than the threshold (i.e., all positive results), and select a second average fluorescence intensity from these. This second average fluorescence intensity is characterized by its corresponding highest dilution (for example, if the intensities of 1:40, 1:80, and 1:160 are all higher than 150, then 1:160 is the highest dilution). The system will then determine the dilution corresponding to this second average fluorescence intensity (i.e., 1:160) as the final output antibody titer value.

[0104] In another embodiment, the preset decision rule is a confidence-weighted rule. Specifically, in step S204, the decision function value output by each support vector machine classifier can be regarded as the confidence level of the classification result. Assume that for images diluted to 1:80, 1:160, and 1:320, the classification results are {homogeneous: 0.95}, {homogeneous: 0.88}, and {granular: 0.7}, respectively. The system can be configured to only adopt the kernel type (homogeneous in this case) as the target kernel type when its average confidence level exceeds a certain preset value (e.g., 0.9). This approach can further eliminate the interference of accidental classification results with low confidence.

[0105] For example, the construction of the preset threshold mapping table can be accomplished through statistical analysis of a large number of clinical samples with known results: A large dataset of positive samples containing multiple known karyotype categories is collected. For each karyotype category (e.g., homogeneous), the system processes fluorescence images of all positive samples under that category at different dilutions and calculates the average fluorescence intensity of each image according to the method described in step S202. For the specific category of homogeneous, the system statistically analyzes the distribution of the average fluorescence intensity values ​​of all positive samples. Statistical methods (e.g., Receiver Operating Characteristic (ROC) curves) are used to determine an optimal positive interpretation fluorescence intensity threshold. This threshold achieves the best balance between sensitivity and specificity when distinguishing between positive and negative samples. For example, analysis shows that an intensity value of 150 is the optimal distinction threshold for homogeneous samples. This process is repeated for all preset karyotype categories, and an optimal threshold is calculated for each category. By storing all kernel types and their corresponding optimal thresholds (e.g., {homogeneous: 150, granular: 120, ...}) in key-value pairs, a preset threshold mapping table is formed for system calls.

[0106] Figure 3 This is a schematic diagram of a fluorescent image intelligent recognition system according to an embodiment of this application. This system can be implemented through software, hardware, or a combination of both, forming all or part of the overall system. For example... Figure 3 As shown, the system includes: an image acquisition module 301, a feature extraction module 302, a feature fusion module 303, a kernel type classification module 304, and a titer calculation module 305, wherein:

[0107] Image acquisition module 301 is configured to acquire multiple original fluorescence images corresponding to various dilutions of the sample, with one original fluorescence image corresponding to each dilution.

[0108] The feature extraction module 302 is configured to extract the texture morphology features of the cell nucleus region, the spatial distribution features of fluorescent bright spots, and the average fluorescence intensity in the target original fluorescence image. The target original fluorescence image is any one of multiple original fluorescence images.

[0109] The feature fusion module 303 is configured to fuse texture morphology features and fluorescent bright spot spatial distribution features to obtain a hybrid feature vector;

[0110] The kernel type classification module 304 is configured to judge the mixed feature vector through a preset kernel type classification model to obtain the kernel type category of the original fluorescence image of the target;

[0111] The titer calculation module 305 is configured to calculate the antibody titer value based on multiple dilutions, the nucleotype category corresponding to each dilution, and the average fluorescence intensity corresponding to each dilution.

[0112] Based on the above embodiments, as an optional embodiment, the feature extraction module 302 is further configured to segment the cell nucleus region from the target original fluorescence image using an image segmentation algorithm; calculate the area, perimeter, circularity, and gray-level co-occurrence matrix of the cell nucleus region, extract contrast, angular second moment, and correlation, and combine the area, perimeter, circularity, contrast, angular second moment, and correlation to constitute texture morphology features; perform top-hat transformation and binarization on the image within the cell nucleus region, identify all fluorescent bright spots, and determine the coordinate position of each fluorescent bright spot; calculate the average nearest neighbor distance and distribution density based on the number and coordinate position of all fluorescent bright spots, and combine the number, average nearest neighbor distance, and distribution density to constitute the spatial distribution features of fluorescent bright spots; calculate the average gray value of all pixels within the cell nucleus region to obtain the average fluorescence intensity.

[0113] Based on the above embodiments, as an optional embodiment, the feature extraction module 302 is further used to perform Gaussian filtering on the original fluorescence image to obtain a smooth image; calculate the gradient of the smooth image to generate a gradient energy field; set an initial contour line in the gradient energy field; calculate the total energy value of the initial contour line according to the preset internal energy function and external energy function, iteratively adjust the position of the initial contour line until the total energy value reaches the minimum and outlines the edge of the cell nucleus; and determine the area enclosed by the adjusted initial contour line as the cell nucleus region.

[0114] Based on the above embodiments, as an optional embodiment, the feature extraction module 302 is further configured to perform a top-hat morphological transformation on the image within the cell nucleus region using a preset structuring element to obtain an enhanced image, wherein the size of the structuring element is smaller than the fluorescent bright spot to be identified; calculate the optimal segmentation threshold of the enhanced image using the big rule method, and perform binarization processing on the enhanced image using the optimal segmentation threshold to obtain a binary image containing fluorescent bright spot pixels; perform connected component analysis on the binary image to identify each independent connected component as an independent fluorescent bright spot; calculate the geometric centroid of each fluorescent bright spot, and determine the coordinates of the geometric centroid as the coordinate position of each fluorescent bright spot.

[0115] Based on the above embodiments, as an optional embodiment, the feature fusion module 303 is further used to concatenate the texture morphology features and the spatial distribution features of fluorescent bright spots to obtain a high-dimensional feature vector; apply principal component analysis algorithm to the high-dimensional feature vector to obtain multiple principal components and their corresponding contribution rates; select target principal components from the multiple principal components in descending order of their contribution rates until the cumulative contribution rate of all target principal components reaches or exceeds a preset threshold for the first time, and then combine all target principal components to form a hybrid feature vector.

[0116] Based on the above embodiments, as an optional embodiment, the kernel classification module 304 is further configured to input the mixed feature vectors into multiple support vector machine classifiers included in the kernel classification model, wherein each support vector machine classifier corresponds to a preset kernel type; calculate the decision function value of the mixed feature vector using each support vector machine classifier, the decision function value being used to characterize the probability of belonging to the kernel type corresponding to the support vector machine classifier; and determine the kernel type corresponding to the support vector machine classifier with the largest decision function value as the kernel type of the target original fluorescence image.

[0117] Based on the above embodiments, as an optional embodiment, the titer calculation module 305 is further configured to determine the target karyotype category among multiple karyotype categories determined for multiple original fluorescence images by means of a preset decision rule, and query and determine the positive interpretation fluorescence intensity threshold corresponding to the target karyotype category from a preset threshold mapping table according to the target karyotype category; compare each average fluorescence intensity corresponding to multiple dilutions with the positive interpretation fluorescence intensity threshold; select the second average fluorescence intensity from all first average fluorescence intensities higher than the positive interpretation fluorescence intensity threshold, the dilution corresponding to the second average fluorescence intensity is the highest, and determine the dilution corresponding to the second average fluorescence intensity as the antibody titer value.

[0118] It should be noted that the system provided in the above embodiments is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0119] This embodiment also discloses an electronic device, as shown in the reference. Figure 4 The electronic device may include: at least one processor 401, at least one communication bus 402, user interface 403, network interface 404, and at least one memory 405.

[0120] The communication bus 402 is used to enable communication between these components.

[0121] The user interface 403 may include a display screen and a camera. Optionally, the user interface 403 may also include a standard wired interface and a wireless interface.

[0122] The network interface 404 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0123] The processor 401 may include one or more processing cores. The processor 401 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 405, and by calling data stored in memory 405. Optionally, the processor 401 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 401 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor 401.

[0124] The memory 405 may include random access memory (RAM) or read-only memory. Optionally, the memory 405 may include a non-transitory computer-readable storage medium. The memory 405 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 405 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 405 may also be at least one storage device located remotely from the aforementioned processor 401. Figure 4 As shown, the memory 405, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a fluorescent image intelligent recognition method.

[0125] exist Figure 4 In the electronic device shown, the user interface 403 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 401 can be used to call an application program stored in the memory 405 for a fluorescent image intelligent recognition method. When executed by one or more processors 401, the electronic device executes one or more methods as described in the above embodiments.

[0126] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0127] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0128] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some service interfaces; indirect couplings or communication connections between apparatuses or units may be electrical or other forms.

[0129] 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.

[0130] 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.

[0131] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory 405 and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory 405 includes various media capable of storing program code, such as a USB flash drive, external hard drive, magnetic disk, or optical disk.

[0132] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the disclosure in this specification. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are considered exemplary only, and the scope of this application is defined by the claims.

Claims

1. A method for intelligent recognition of fluorescence images, characterized in that, Applied to a computer processing device, the method includes: Multiple original fluorescence images corresponding to different dilutions of the sample are obtained, with one original fluorescence image corresponding to each dilution. Extract the texture and morphological features of the cell nucleus region, the spatial distribution features of fluorescent bright spots, and the average fluorescence intensity from the target original fluorescence image, wherein the target original fluorescence image is any one of the multiple original fluorescence images; The texture morphology features and the spatial distribution features of the fluorescent bright spots are fused to obtain a hybrid feature vector; The mixed feature vector is judged by a preset kernel type classification model to obtain the kernel type category of the original fluorescence image of the target; The antibody titer value is calculated based on the various dilutions, the corresponding karyotype for each dilution, and the average fluorescence intensity for each dilution; wherein, The antibody titer value is calculated based on multiple dilutions, the nuclear type corresponding to each dilution, and the average fluorescence intensity corresponding to each dilution, specifically including: Among the multiple karyotype categories determined for multiple original fluorescence images, a target karyotype category is determined by a preset decision rule, and a positive fluorescence intensity threshold corresponding to the target karyotype category is queried and determined from a preset threshold mapping table according to the target karyotype category; Each of the average fluorescence intensities corresponding to the various dilutions is compared with the positive read fluorescence intensity threshold. Among all the first average fluorescence intensities that are higher than the positive interpretation fluorescence intensity threshold, the second average fluorescence intensity is selected, and the dilution corresponding to the second average fluorescence intensity is the highest. The dilution corresponding to the second average fluorescence intensity is then determined as the antibody titer value.

2. The method according to claim 1, characterized in that, The extraction of texture morphology features, spatial distribution features of fluorescent bright spots, and average fluorescence intensity from the original fluorescence image of the target specifically includes: The cell nucleus region is segmented from the original fluorescence image of the target using an image segmentation algorithm; Calculate the area, perimeter, circularity, and gray-level co-occurrence matrix of the cell nucleus region, extract the contrast, second moment of angle, and correlation, and combine the area, perimeter, circularity, contrast, second moment of angle, and correlation to form the texture morphology feature; The image within the cell nucleus region is subjected to top-hat transformation and binarization to identify all fluorescent bright spots and determine the coordinate position of each fluorescent bright spot; Based on the number of all fluorescent bright spots and their coordinate positions, the average nearest neighbor distance and distribution density are calculated, and the number, the average nearest neighbor distance, and the distribution density together constitute the spatial distribution characteristics of the fluorescent bright spots. The average fluorescence intensity is obtained by calculating the average gray value of all pixels within the cell nucleus region.

3. The method according to claim 2, characterized in that, The step of segmenting the cell nucleus region from the original fluorescence image of the target using an image segmentation algorithm specifically includes: The original fluorescence image is subjected to Gaussian filtering to obtain a smoothed image; Calculate the gradient of the smoothed image and generate a gradient energy field; An initial contour line is set in the gradient energy field; The total energy value of the initial contour line is calculated based on the preset internal energy function and external energy function. The position of the initial contour line is iteratively adjusted until the total energy value reaches the minimum and the edge of the cell nucleus is outlined. The region enclosed by the adjusted initial contour line is defined as the cell nucleus region.

4. The method according to claim 2, characterized in that, The step of performing a top-hat transform and binarization on the image within the cell nucleus region to identify all fluorescent bright spots and determine the coordinate position of each fluorescent bright spot specifically includes: A top-cap morphological transformation is performed on the image within the cell nucleus region using a preset structuring element to obtain an enhanced image, wherein the size of the structuring element is smaller than the fluorescent bright spot to be identified. The optimal segmentation threshold of the enhanced image is calculated using the big law method, and the enhanced image is binarized using the optimal segmentation threshold to obtain a binary image containing fluorescent bright pixels. Connectivity analysis is performed on the binary image to identify each independent connected component as an independent fluorescent bright spot; Calculate the geometric centroid of each fluorescent spot, and determine the coordinates of the geometric centroids to determine the coordinate position of each fluorescent spot.

5. The method according to claim 1, characterized in that, The process of fusing the texture morphology features and the spatial distribution features of the fluorescent bright spots to obtain a hybrid feature vector specifically includes: The texture morphology features and the fluorescent bright spot spatial distribution features are concatenated end to end to obtain a high-dimensional feature vector; Applying principal component analysis to the high-dimensional feature vector yields multiple principal components and their corresponding contribution rates. From the plurality of principal components, target principal components are selected sequentially in descending order of their contribution rates until the cumulative contribution rate of all target principal components reaches or exceeds a preset threshold for the first time, and all target principal components are combined to form the hybrid feature vector.

6. The method according to claim 1, characterized in that, The step of determining the kernel type of the target original fluorescence image by judging the mixed feature vector through a preset kernel type classification model specifically includes: The mixed feature vectors are respectively input into multiple support vector machine classifiers included in the kernel type classification model, wherein each support vector machine classifier corresponds to a preset kernel type category; The decision function value of the mixture feature vector is calculated using each of the support vector machine classifiers. The decision function value is used to characterize the probability of belonging to the kernel type corresponding to the support vector machine classifier. The kernel type corresponding to the support vector machine classifier with the largest decision function value is determined as the kernel type of the original fluorescence image of the target.

7. A fluorescent image intelligent recognition system, characterized in that, The system includes an image acquisition module, a feature extraction module, a feature fusion module, a kernel type classification module, and a titer calculation module, among which: The image acquisition module is configured to acquire multiple original fluorescence images corresponding to various dilutions of the sample, with one original fluorescence image corresponding to each dilution. The feature extraction module is configured to extract the texture morphology features of the cell nucleus region, the spatial distribution features of fluorescent bright spots, and the average fluorescence intensity in the target original fluorescence image, wherein the target original fluorescence image is any one of the multiple original fluorescence images; The feature fusion module is configured to fuse the texture morphology features and the fluorescent bright spot spatial distribution features to obtain a hybrid feature vector; The kernel type classification module is configured to judge the mixed feature vector through a preset kernel type classification model to obtain the kernel type category of the target original fluorescence image; The titer calculation module is configured to calculate the antibody titer value based on multiple dilutions, the karyotype corresponding to each dilution, and the average fluorescence intensity corresponding to each dilution; wherein... The antibody titer value is calculated based on multiple dilutions, the nuclear type corresponding to each dilution, and the average fluorescence intensity corresponding to each dilution, specifically including: Among the multiple karyotype categories determined for multiple original fluorescence images, a target karyotype category is determined by a preset decision rule, and a positive fluorescence intensity threshold corresponding to the target karyotype category is queried and determined from a preset threshold mapping table according to the target karyotype category; Each of the average fluorescence intensities corresponding to the various dilutions is compared with the positive read fluorescence intensity threshold. Among all the first average fluorescence intensities that are higher than the positive interpretation fluorescence intensity threshold, the second average fluorescence intensity is selected, and the dilution corresponding to the second average fluorescence intensity is the highest. The dilution corresponding to the second average fluorescence intensity is then determined as the antibody titer value.

8. An electronic device, characterized in that, The device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1-6.

Images