Fluorescence in situ hybridization image breakage and fusion collaborative determination method and device and medium

Abstract

The application provides a fluorescence in situ hybridization image breakage and fusion collaborative determination method and device and medium, in order to solve the problems of FISH image breakage and fusion detection fragmentation, insufficient robustness and high misjudgment rate, after separating the cell image channel, the application calculates a binary threshold value based on an adaptive extreme value ratio, combines a connected domain median area filter to extract independent fluorescent points and an equivalent diameter, generates a cell nucleus mask based on fusion morphology and appearance characteristics, performs two-way collaborative determination, globally determines that heterochromatic fluorescent points with a distance less than a fusion threshold value are fused and physically removed, and then connects the remaining points in the same cell nucleus by using a global optimal pairing algorithm, and determines that the breakage threshold value based on the equivalent diameter is greater than the distance dynamically set, and finally outputs a comprehensive classification state. The application is used for precise determination of tumor FISH detection.

Description

Technical Field

[0001] This invention relates to the fields of medical image processing and artificial intelligence, and in particular to a method, apparatus and medium for collaborative discrimination of fragmentation and fusion in fluorescence in situ hybridization images based on adaptive fluorescence dot detection and intelligent spatial decision-making. Background Technology

[0002] Fluorescence in situ hybridization (FISH) is one of the core methods for detecting tumor gene abnormalities, and it is widely used in the detection and analysis of HER2 gene, ALK, ROS1, and other genes in breast cancer. This technology involves applying specific fluorescent probes into the cell nucleus, causing the target gene sequence to exhibit red fluorescence and the reference gene sequence to exhibit green fluorescence. The spatial relationship between the red and green fluorescence signals is then analyzed to determine gene breakage or fusion states, providing a basis for patient diagnosis, classification, and treatment selection.

[0003] In existing FISH image analysis techniques, fluorescence signal detection often employs fixed threshold segmentation or general-purpose deep learning models. Fixed threshold methods struggle to adapt to uneven brightness caused by variations in exposure conditions across different slices, easily resulting in missed detections in low-contrast areas and false positives in high-background areas. While general-purpose deep learning models perform well on standard samples, their robustness remains insufficient in critical scenarios involving signal point adhesion, diffusion, or severe background fluorescence interference. Furthermore, at the signal state determination level, current technologies typically treat gene breakage detection and gene fusion detection as two independent tasks, employing different algorithms and software tools. This not only results in cumbersome procedures but also increases diagnostic uncertainty as the results of the two independent analyses may contradict each other due to algorithmic differences. Regarding breakage determination, existing methods generally use simple nearest-neighbor pairing between red and green signal points and comparison with a fixed distance threshold. This approach struggles to adapt to differences in cell nuclear morphology and variations in signal point orientation, leading to a high false positive rate when the distance is close to the determination threshold.

[0004] Therefore, there is an urgent need for a method, device, and medium for collaborative discrimination of fragmentation and fusion in fluorescence in situ hybridization images based on adaptive fluorescence dot detection and intelligent spatial decision-making. Summary of the Invention

[0005] This invention provides a method, apparatus, and medium for collaborative discrimination of breakage and fusion in fluorescence in situ hybridization images. It addresses the problems of fragmented breakage and fusion detection processes and unstable underlying detection in existing technologies, which lead to low efficiency, high false positive rate in critical states, and conflicting conclusions.

[0006] The core technology of this invention is to construct a dual-path collaborative architecture that includes extreme value ratio adaptive extraction, global fusion signal physical removal, and unbalanced global optimal pairing, so as to achieve precise signal extraction and decoupled discrimination of break-fusion state at both global and nuclear mask scales.

[0007] In a first aspect, the present invention provides a method for the collaborative discrimination of fragmentation and fusion in fluorescence in situ hybridization images, the method comprising the following steps:

[0008] Acquire images of the fluorescent in situ hybridization (FISH) cells to be analyzed, and perform channel separation on the FISH cell images to extract fluorescence images from different channels; The binarization threshold of each channel is calculated based on the adaptive extreme value ratio of image brightness to extract the initial mask. The shape filtering interval is set by combining the median area of ​​the connected domains in the initial mask, and the center coordinates of the independent fluorescent points in each channel are extracted and the equivalent diameter is estimated. Morphological prior features and appearance features are extracted from fluorescence in situ hybridization cell images and feature fusion is performed to generate a cell nuclear mask; Dual-path collaborative determination based on the spatial distance of the center coordinates of independent fluorescent dots: The spatial distance between heterochromatic fluorescent points is calculated globally. Heterochromatic fluorescent point pairs with a spatial distance less than the fusion determination threshold are identified as fusion signals and physically removed. For the remaining heterochromatic fluorescent spots after removal, a global optimal pairing algorithm is used to pair and connect them within the same cell nuclear mask area. If the pairing distance is greater than or equal to the breakage judgment threshold dynamically set based on the equivalent diameter, it is judged as a breakage signal. Based on the determination results of the fracture signal and fusion signal, the comprehensive classification status corresponding to each cell nuclear mask is determined and output.

[0009] Further, the steps of calculating the binarization threshold of each channel based on the adaptive extreme value ratio of image brightness to extract the preliminary mask, and setting the shape filtering interval based on the median area of ​​the connected components within the preliminary mask, include: The maximum pixel value in a single channel image is obtained, and the maximum pixel value is multiplied by an adaptive threshold scaling factor to obtain a binarization threshold. The channel image is then binarized and denoised based on the binarization threshold to obtain a preliminary mask. Extract the effective connected components from the initial mask and calculate the median area of ​​the effective connected components; Multiply the median area by the lower and upper area ratio factors to define the morphological filtering range to remove noise and abnormal clumps.

[0010] Furthermore, the adaptive threshold scaling factor ranges from 0.3 to 0.5; the lower limit scaling factor is 0.1, and the upper limit scaling factor is 3.0.

[0011] Further, the steps of extracting morphological prior features and appearance features from fluorescence in situ hybridization cell images and performing feature fusion to generate a nuclear mask include: Prior cell morphology features and multi-scale appearance features were extracted in parallel from fluorescence in situ hybridization cell images; Using cell morphology prior features as query vectors and multi-scale appearance features as key and value vectors, cross-attention calculation is performed to generate optimized fusion features. Based on optimized fusion features, a decoder network generates a nuclear mask containing the boundaries of each cell and removes abnormal masks that do not meet the preset morphological feature lower limit.

[0012] Furthermore, the steps of performing pairing connections using a globally optimal pairing algorithm within the same nuclear mask area include: For heterochromatic fluorescent spots of different channels with unequal numbers remaining in the same cell nuclear mask, a corresponding number of virtual nodes are added to align the total number of spots. The actual Euclidean distance between real heterochromatic fluorescent points and the maximum penalty cost involving virtual nodes are combined and spliced ​​together to construct an extended non-equilibrium cost matrix; The Hungarian algorithm is used to solve the unbalanced cost matrix to obtain the globally optimal pairing connection that avoids local pairing interference.

[0013] Furthermore, the fracture determination threshold is the product of the average equivalent diameter of different channels and the preset fracture distance ratio coefficient; the fracture distance ratio coefficient ranges from 2.0 to 3.0.

[0014] Furthermore, the steps for determining and outputting the overall classification state corresponding to each nuclear mask include: If both breakage and fusion signals are detected within the nuclear mask, it is classified as a complex state; if only fusion signals are detected, it is classified as a fusion state; if only breakage signals are detected, it is classified as a breakage state; if neither breakage nor fusion signals are detected, and heterochromatic fluorescent spots are also present, it is classified as a normal co-occurrence state. The output of each comprehensive classification state is visualized and rendered, the boundary outlines of the cell nuclei in different states are drawn with corresponding colors, and the judgment results are converted into a standardized data format for output.

[0015] Secondly, the present invention provides a fluorescence in situ hybridization image fragmentation and fusion co-discrimination device, comprising: The image acquisition and preprocessing module is used to acquire images of fluorescence in situ hybridization cells to be analyzed, and to perform channel separation on the fluorescence in situ hybridization cell images to extract fluorescence images of different channels; The adaptive feature detection module is used to calculate the binarization threshold of each channel based on the adaptive extreme value ratio of image brightness to extract the preliminary mask, and to set the morphological filtering interval by combining the median area of ​​the connected domains in the preliminary mask, extract the center coordinates of independent fluorescent points in each channel and estimate the equivalent diameter. The morphology-enhanced cell segmentation module is used to extract morphological prior features and appearance features from fluorescence in situ hybridization cell images and perform feature fusion to generate a cell nuclear mask. The intelligent dual-path collaborative decision-making module is used to calculate the spatial distance between heterochromatic fluorescent dots in the global scope, extract heterochromatic fluorescent dot pairs whose spatial distance is less than the fusion judgment threshold, judge them as fusion signals and physically remove them; for the remaining heterochromatic fluorescent dots after removal, they are paired and connected using a global optimal pairing algorithm within the same cell nuclear mask area. If the pairing distance is greater than or equal to the breakage judgment threshold dynamically set based on the equivalent diameter, it is judged as a breakage signal. The results generation and output module is used to determine the comprehensive classification state corresponding to each cell nuclear mask based on the judgment results of the fracture signal and fusion signal, and to perform visualization rendering and standardized output.

[0016] Thirdly, the present invention provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to execute the above-described fluorescence in situ hybridization image fragmentation and fusion co-discrimination method.

[0017] Fourthly, the present invention provides a readable storage medium storing a computer program, the computer program including program code for controlling a process to execute the process, the process including the above-described fluorescence in situ hybridization image fragmentation and fusion co-discrimination method.

[0018] The main contributions and innovations of this invention are as follows: 1. This invention employs an adaptive binarization threshold based on the ratio of extreme image brightness values, avoiding missed or false detections caused by fixed thresholds under different exposure conditions. Simultaneously, it dynamically sets the area filtering interval based on the median area of ​​connected components, adapting to the physical size of signal points under different magnification and staining conditions, effectively eliminating noise and abnormally adhered regions. Experiments show that this invention achieves an F1 score of 96.5% for fluorescence point detection on challenging samples with low contrast and strong backgrounds, significantly outperforming fixed threshold methods and general deep learning methods.

[0019] 2. This invention simultaneously executes the fusion determination path and the breakage determination path within a unified framework: the fusion determination path first identifies and eliminates extremely close pairs of dissimilar fluorescent points, fundamentally eliminating the logical risk of the fused signal being misjudged as a "short-distance breakage"; the breakage determination path uses a globally optimal pairing algorithm to pair the remaining signal points, avoiding misjudgments caused by local shortest-distance pairing in multi-signal-point scenarios. The dual-path collaborative architecture reduces the contradiction rate of conclusions to zero, ensuring that the two types of determination results are logically mutually exclusive and consistent.

[0020] 3. This invention dynamically sets the fracture detection threshold based on the equivalent diameter of fluorescent dots, allowing the detection scale to adaptively adjust with the physical scale of the image itself. This overcomes the problem that a fixed threshold cannot cover differences in cell morphology and chromosome spatial conformation. Experiments show that this invention achieves an accuracy of 94.5% in identifying fractures and fusions of cells in critical states, with a sensitivity of 95.2%, a specificity of 93.8%, and a clinical diagnostic accuracy (Kappa coefficient) of 0.91, significantly outperforming traditional independent workflow methods and existing deep learning methods.

[0021] 4. This invention is based on a traditional image processing and globally optimal pairing algorithm architecture, avoiding complex multi-model collaborative reasoning. The average processing time per image is 1.2 seconds, which is much faster than existing deep learning methods. Simultaneously, by converting the judgment results and cell mask data into a standardized output format, it achieves seamless integration with mainstream pathology analysis platforms, supporting pathologists in visual verification and interactive correction, forming a complete clinical workflow of "automatic analysis—human-computer interaction—result archiving".

[0022] Details of one or more embodiments of the present invention are set forth in the following drawings and description, so that other features, objects and advantages of the invention will be more readily understood. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is the flowchart of the fluorescence in situ hybridization image fragmentation and fusion co-discrimination method according to an embodiment of the present invention; Figure 2 This is a comparison chart of the effects of adaptive threshold detection and fixed threshold detection; Figure 3 This is a diagram comparing the globally optimal pairing and the locally greedy pairing of the Hungarian algorithm; Figure 4 This is the final visual rendering; Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0024] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0025] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0026] This invention provides a collaborative discrimination method for fragmentation and fusion in fluorescence in situ hybridization (FISH) images based on adaptive fluorescence dot detection and intelligent spatial decision-making. The overall process of this method is as follows: Figure 1 As shown, the process includes five sequentially executed steps: image acquisition and standardized preprocessing, adaptive fluorescent dot robust detection, cell nucleus segmentation, spatial distance-based dual-path collaborative discrimination, and result visualization and standardized output. The specific implementation methods for each step are described in detail below.

[0027] Example 1 This invention aims to propose a method for the collaborative discrimination of fragmentation and fusion in fluorescence in situ hybridization images. Specifically, referring to... Figure 1 The method includes: (I) Cell image acquisition and standardized preprocessing (i.e.) Figure 1 S1, graphic preprocessing) Acquire images of the cells to be analyzed in fluorescence in situ hybridization. These images are typically acquired using a fluorescence microscope under specific exposure conditions and contain information from both the red and green fluorescence channels. Channel separation is performed on the images, and the red fluorescence channel images are extracted separately. With green fluorescence channel image Each channel image is a single-channel grayscale matrix, with pixel values ​​reflecting the intensity of the corresponding fluorescence signal. Optionally, the separated channel images can be preprocessed using standardization, including but not limited to grayscale normalization, contrast enhancement, or noise filtering, to improve the stability of subsequent detection.

[0028] (II) Robust Detection of Dual-Channel Fluorescent Dots Based on Adaptive Threshold and Morphology Optimization (i.e., Figure 1 S2, adaptive dual-channel fluorescence dot detection) For each channel image The following adaptive detection process will be executed sequentially: S2-1, Adaptive Global Threshold Calculation: Calculate the maximum pixel value of this channel's image. Based on adaptive threshold scaling factor Calculate the binarization threshold The calculation formula is:

[0029] Based on binarization threshold Convert the channel image into a preliminary binary mask. The rules are as follows: Pixel values ​​greater than... The pixel value is set to 255 (foreground), and the rest are set to 0 (background).

[0030] This adaptive thresholding mechanism is calculated based on the image's own brightness extremes, and can automatically adapt to overall brightness changes caused by differences in exposure and staining depth in different slices. Extensive experimental verification has shown that when... When the value ranges from 0.3 to 0.5, it can effectively balance the missed detection in low-brightness areas and the false detection in high-background areas.

[0031] In this embodiment, preferably At this point, the detection accuracy of fluorescent spots in unevenly exposed FISH sections can reach over 98%. When the value is too large (e.g., greater than 0.5), low-contrast signals are easily missed; when When the value is too small (e.g., less than 0.3), background noise is easily misdetected. When the value is 0.3, it can effectively preserve the weak signal in low-contrast areas, making it suitable for cell images with dark backgrounds; when... When the value is 0.5, it can effectively suppress noise in strong background areas and is suitable for cell images with strong exposure.

[0032] S2-2, Morphological optimization denoising: Denoising of the initial binary mask Perform opening operations, using The use of square structural elements eliminates fine noise and artifacts, resulting in an optimized mask. The opening operation is a standard morphological operation that involves erosion followed by dilation. Its function is to filter out noise points and fine line artifacts that are too small while maintaining the overall shape of the target area.

[0033] S2-3, Connectivity Analysis and Median Area Filtering: Optimizing Masks Extract all connected components and calculate the set of areas of all connected components with an area greater than 0. Find the median area With median area As a benchmark, set a lower limit for area filtering. and upper limit The calculation formula is:

[0034]

[0035] Retained area and Connected regions between them are used as candidate fluorescent spot regions to obtain the final fluorescent spot mask. ; Remove areas smaller than The noise and area are greater than Large, adhered clumps.

[0036] The filtering interval is dynamically adjusted based on the median area of ​​the connected components, which can adapt to the physical size of signal points under different amplification factors and coloring conditions. In this embodiment, it is preferred that... , .when When the value is less than 0.05, fine noise cannot be effectively filtered out; when... When the value is greater than 4.0, abnormally large signals of clustered adhesion are easily retained by mistake, which leads to the subsequent diameter estimation deviation and thus affects the accuracy of fracture determination.

[0037] S2-4. Fluorescent dot center extraction and diameter estimation: Based on the final fluorescent dot mask. Calculate the centroids of each connected region and the center coordinates of all detected fluorescence spots. Output the set of center coordinates of red fluorescence spots in the red channel. The green channel outputs the set of coordinates of the centers of green fluorescent dots. At the same time, based on the median area Approximate the fluorescent dots as circles, and use the formula for the area of ​​a circle to inversely deduce the equivalent diameter of the fluorescent dots. The calculation formula is:

[0038] equivalent diameter This provides an adaptive benchmark scale for subsequent fracture distance determination.

[0039] See Figure 2 , Figure 2 This demonstrates a comparison of the performance of adaptive threshold detection and fixed threshold detection. Among them, Figure 2 Part a is a real fish image; Figure 2Part b shows the results of detection using a fixed low threshold, where a large number of weak signals and background noise are misdetected as fluorescent dots. Figure 2 Part c shows the results of detection using a fixed high threshold, indicating that some weak signals were missed. Figure 2 The d part is the adaptive threshold adopted in this invention ( The test results accurately detected all real fluorescent spots while effectively eliminating noise interference.

[0040] (III) Dual-path fusion cell nucleus segmentation based on morphological constraint attention optimization (i.e.) Figure 1 S3, morphological constraints, attention, and cell nucleus segmentation The original FISH image is input into a pre-built and trained dual-path fusion cell nucleus segmentation model to obtain a pixel-level mask for each cell nucleus. It also calculates the morphological characteristics of each cell, including area, perimeter, and roundness.

[0041] The dual-path fusion cell nucleus segmentation model employs a morphologically constrained attention mechanism and comprises the following four core components: S3-1, Morphological Prior Encoder: Employs Convolutional layers extract prior morphological features such as cell boundaries, area, and roundness from the original FISH image. This encoder focuses on capturing the geometric structure information of cells, providing morphological constraint cues for subsequent attention fusion.

[0042] S3-2, Appearance Feature Encoder: Employs a pre-trained lightweight visual encoder (e.g., MobileNetV3 network) to extract multi-scale appearance features from the raw FISH image. This encoder utilizes its general feature extraction capabilities, acquired through pre-training on large-scale datasets, to capture appearance information such as cell texture, brightness, and color.

[0043] S3-3, Constrained Attention Fusion Layer: Incorporating prior morphological features As a query, appearance features As keys and values, cross-attention calculations are performed to generate optimized fusion features. This mechanism enables the model to dynamically focus on corresponding morphological correction cues in the image based on morphological priors, such as real cell membrane boundaries and nuclear textures, thereby improving segmentation accuracy.

[0044] S3-4, Refined Mask Generation: Optimizing Blending Features Input the decoder network to generate the final high-precision set of cell nuclear masks. The morphological characteristics of each cell are calculated based on a mask. An effective cell mask must meet the following conditions: its area must be within a preset threshold range and its roundness must be greater than the lower limit; cells with abnormal or incomplete morphology are removed. The formula for calculating roundness is:

[0045] in, For cell area, This represents the cell perimeter. The closer the roundness is to 1, the closer the cell nucleus is to a round shape.

[0046] The dual-path fusion segmentation scheme is a serial enhancement process of "basic segmentation and morphological constraint optimization". By combining morphological knowledge with deep learning attention mechanisms, it specifically addresses the problems of blurred cell boundaries and morphological distortion in FISH images, ensuring segmentation accuracy while avoiding complex multi-model collaboration and improving processing efficiency.

[0047] (iv) Dual-path collaborative intelligent discrimination based on spatial distance (i.e.) Figure 1 S4, intelligent system discrimination) Receive S2 output , , And the cell mask output by S3 The algorithm performs dual-path collaborative discrimination. At the algorithm execution level, it first extracts and removes extremely close fusion pairs to prevent them from being mistakenly assigned to distant points by the subsequent global pairing algorithm, thereby ensuring the accuracy of the two independent judgments.

[0048] S4-1, Fusion Decision Path: The global coordinate set output by S2 and Using the input as input, calculate the spatial Euclidean distance between all red and green fluorescent dots. For red fluorescent spots With green fluorescent dots The distance calculation formula is as follows:

[0049] A greedy nearest neighbor merging strategy is adopted, which merges neighbors whose distance is less than the merging distance threshold. The red and green fluorescent dots are identified as fusion signals. When the red fluorescent dots... With green fluorescent dots Spatial distance between The midpoint between the coordinates of two fluorescent spots is taken as the fusion point (i.e., the point that appears as a yellow signal on the image) when the following conditions are met: (1) ; (2) and None of them were paired before this determination. The formula for calculating the coordinates of the fusion point is:

[0050] After merging, the paired red and green fluorescent dots are removed from their respective coordinate sets. and Removed from the middle.

[0051] Fusion distance threshold It can be an absolute distance in pixels, or it can be relative to the equivalent diameter. The proportional distance. Preferably, when a proportional distance relative to the equivalent diameter is used, The value ranges from 0.3 to 0.8 times the equivalent diameter. When using absolute distance in pixels, The value ranges from 2 to 8 pixels, and the specific value is determined based on the probe design and imaging conditions.

[0052] It is important to note that this fusion determination path is executed globally and is not constrained by individual cell nuclear masks. This design aligns with the physical characteristics of FISH imaging: the physical formation of fusion signals (yellow signal points) is not contingent on cell boundaries, and fusion signals located at cell edges should not be cut off by cell masks.

[0053] S4-2, Fracture Detection Path: For each cell Filtering located in its mask Red fluorescent dot subset within With green fluorescent dot subset When both red and green fluorescent spots are present in the cell, the Hungarian algorithm is used to perform global optimal pairing of the remaining red and green fluorescent spots after the fusion point is removed by S4-1.

[0054] Specifically, firstly, the cost matrix is ​​constructed. Its elements Indicates the first The red fluorescent dot and the first Euclidean distance between green fluorescent dots:

[0055] When the number of red fluorescent dots With the number of green fluorescent dots When there is asymmetry, the cost matrix is ​​expanded into a square matrix by adding virtual nodes. For example, add A virtual red fluorescent dot will be used to represent the cost matrix. from Expand to square array The extended rules are as follows:

[0056] In the formula, This represents the actual Euclidean distance between the real red fluorescent spot and the real green fluorescent spot. The maximum penalty value is set to this maximum value when a virtual red fluorescent dot is paired with a real green fluorescent dot. This can be set to the image diagonal length; the cost is set to 0 when pairing virtual nodes to avoid virtual pairings consuming real cost. This extended strategy ensures that the Hungarian algorithm performs globally optimal pairing only between real red and real green fluorescent dots. When, add A virtual green fluorescent dot expands the cost matrix to The cost between virtual green fluorescent dots and real red fluorescent dots in the square matrix is ​​also set to... The cost between virtual nodes is set to 0. This extended strategy ensures that the Hungarian algorithm performs globally optimal pairing only between real red and real green fluorescent dots. The Hungarian algorithm solves for the optimal pairing function. This minimizes the total cost of all pairings.

[0057] Subsequently, the dynamic threshold for fracture determination is calculated. And make a judgment, the formula is:

[0058]

[0059] in, This is the fracture distance proportionality coefficient. Estimate the average equivalent diameter for the red and green channels. For cells The Boolean value of the fracture state. In order to be with the first Index of the best-paired green fluorescent dots for each red fluorescent dot.

[0060] The Hungarian algorithm ensures globally optimal pairing in multi-signal point scenarios, preventing misjudgments caused by local shortest-distance pairing. A dynamic threshold is set based on a multiple of the estimated diameter to adapt to different cell and signal scales. A break is identified as long as the distance between any two optimal pairing points exceeds the dynamic threshold. (Break distance proportionality coefficient) The value range is from 2.0 to 3.0. In this embodiment, it is preferred that... This value is derived based on the physical properties of the FISH probe and statistical analysis of chromosome spatial conformation. At this time, normal co-occurring signals are easily misinterpreted as breaks, increasing false positives; when At times, some genuine fracture signals were missed due to their proximity, increasing the false negative rate. Experiments show that... The accuracy of detecting cell breakage in the critical state is improved by about 15% compared to the fixed threshold.

[0061] See Figure 3 The figure shows the incorrect connections caused by simple nearest neighbor pairing (local shortest distance pairing) in a scenario with multiple red fluorescent dots and multiple green fluorescent dots, as well as the globally optimal correct connection obtained by using the Hungarian algorithm, verifying the effectiveness of the globally optimal pairing.

[0062] S4-3, Collaborative Decision Making and Cell Classification: By combining the outputs of the breakage and fusion determination paths, each cell is co-classified. Define Boolean variables. Represents cells The fracture state (if there is a distance greater than or equal to) The optimal pairing of red and green fluorescent dots is then... ,otherwise ), Boolean variable Represents cells The fusion state (if there is a distance less than) The red fluorescent dots and green fluorescent dots are then ,otherwise ).

[0063] The decision-making logic for collaborative classification is as follows: 1. If and If the result is negative, the cell is classified as a complex state. This state can occur in complex gene rearrangements (such as staining abnormalities involving multiple sites), representing heterogeneity where signal fusion and separation coexist within the same cell.

[0064] 2. If and If so, the cell is classified as a fused state.

[0065] 3. If and If so, the cell is classified as broken.

[0066] 4. If and If both red and green fluorescent spots are present within the cell, then the cell is classified as a normal co-occurrence state.

[0067] This decision-making logic achieves compatibility and decoupling for different clinical testing needs within a single framework. Fusion and breakage are two independent discrimination criteria designed for different probes, respectively characterizing the "extreme proximity" and "extreme separation" states of signal points in physical space. By adding "complex states," the system fully covers all possible clinical scenarios, including intracellular heterogeneity, ensuring the physical exclusivity and logical consistency of the output conclusions.

[0068] (v) Results visualization, statistics, and standardized output (i.e.) Figure 1 S5, standardized output) S5-1, Visualization and Rendering: Based on the co-classification results, draw boundary outlines of different colors for cell nuclei of different classifications. Specifically, red boundaries mark fused cells, green boundaries mark broken cells, white boundaries mark cells whose roundness does not meet the screening criteria, yellow boundaries mark cells whose area does not meet the screening criteria, and light purple boundaries mark cells that do not simultaneously contain both red and green fluorescent dots. Simultaneously, the positions of red fluorescent dots, green fluorescent dots, and yellow fused dots are marked on the image.

[0069] See Figure 4 , Figure 4 The final visual rendering shows the color coding of the boundary contours of cells in each category and the effect of fluorescent dot marking.

[0070] S5-2, Effectiveness Filtering and Statistics: Effectiveness filtering is performed based on the morphological characteristics of cells, and the roundness of each cell is calculated. Excluding roundness Less than the preset threshold Cell nuclei with abnormal area or those touching the edge of the image. Circularity threshold. The value ranges from 0.5 to 0.8, with the specific value determined based on cell type and imaging conditions. It is used to remove cell nuclei with severely morphologically distorted or incomplete structures. For the selected effective cell nuclei, the breakage ratio (the proportion of cells determined to be in a broken state out of the total number of effective cells) and the fusion ratio (the proportion of cells determined to be in a fused state out of the total number of effective cells) are statistically analyzed. These ratios are key quantitative bases for clinical diagnostic classification.

[0071] S5-3, Standardized Output: The judgment results and spatial masks are converted into standard GeoJSON format, and base64 encoded cell mask data is embedded within them, achieving pixel-level lossless integration with mainstream pathology platforms such as QuPath. Pathologists can directly perform visual verification, interactive correction, and archiving of results within their own platform, forming a clinical closed loop of "automatic analysis - human-computer interaction - result archiving".

[0072] Example 2 This embodiment uses an adaptive threshold scaling factor. Area filtration coefficient and The impact of different values ​​of on detection performance is explained.

[0073] The applicant constructed a test set of 200 clinical FISH images with different exposure conditions and magnifications, covering a total of 2337 cell nuclei, of which 30% were challenging samples with low contrast and strong backgrounds. Pixel-level annotation by two senior pathologists was used as the gold standard. The performance of fluorescence spot detection under different parameter settings was evaluated on the challenging sample subset.

[0074] Using the optimal parameter combination ( , , When the F1 score reached 96.5%, the false negative rate was 1.8% and the false positive rate was 1.7%.

[0075] when When the value deviates from the optimal value, the detection performance decreases significantly: When the F1 score dropped to 88.1%, the false detection rate rose sharply to 9.9%, indicating that the excessively low threshold coefficient caused a large number of background noise points to be falsely detected as fluorescent points. When the F1 score dropped to 89.5%, the false negative rate rose sharply to 8.2%, indicating that the excessively high threshold coefficient caused low-contrast signals to be missed.

[0076] when When deviating from the preferred value: When the F1 score dropped to 92.4%, the false positive rate rose to 6.1%, indicating that an excessively small area lower limit cannot effectively filter out fine noise.

[0077] when When deviating from the preferred value: When the F1 score dropped to 90.8%, the false positive rate rose to 7.1%, indicating that the excessively large area limit caused abnormally large signals of cluster adhesion to be mistakenly retained.

[0078] The above experimental results show that The value ranges from 0.3 to 0.5. The value can be 0.1. The value of 3.0 is an optimal range that has been experimentally verified, and it can achieve the best detection balance under different exposure conditions and magnification.

[0079] Example 3 This embodiment uses the fracture distance proportionality coefficient. The impact of different values ​​of on the discrimination performance is explained.

[0080] Experiments were conducted on a full test set containing a large number of critical state cells (cells whose red-green signal distance is close to the decision threshold, accounting for 20% of the test set).

[0081] use At that time, the accuracy of identifying cell breakage in the critical state was improved by about 15% compared with the method using a fixed threshold, and the overall accuracy reached 94.5%.

[0082] when When the fracture detection threshold is too low, the proportion of normal co-occurring signals misjudged as fractures increases significantly, and the false positive rate rises markedly; when At that time, the fracture detection threshold was too high, and some true fracture signals were missed due to their proximity, resulting in a significantly increased false negative rate. The above results demonstrate that... The rationality of the value range of 2.0 to 3.0 and the necessity of the preferred value of 2.5.

[0083] Example 4 This embodiment illustrates the overall effectiveness of the dual-path collaborative discrimination architecture for fracture and fusion.

[0084] On the complete test set, the complete method of the present invention was compared with the traditional independent process method (fixed threshold segmentation plus nearest neighbor pairing plus rigid distance determination) and the deep learning-based method. The results are shown in Tables 1 to 3.

[0085] Table 1 Comparison of fluorescence dot detection performance and parameter sensitivity

[0086] As shown in Table 1, the F1 score of the preferred parameter combination of this invention reaches 96.5%, significantly outperforming general deep learning methods and the fixed threshold method. or When the value deviates from the preferred value, the rate of missed detection or false detection increases significantly, which verifies the necessity of the key parameter settings of the present invention and its ability to suppress the underlying detection error.

[0087] Table 2 Comparison of accuracy between fracture / fusion co-discrimination and critical state

[0088] As shown in Table 2, the accuracy of the present invention in identifying cells in critical state reaches 94.5%, which is about 15% higher than that of traditional methods. At the same time, the dual-path collaborative architecture reduces the rate of contradiction in conclusions to 0%, completely eliminating the logical risk of the fused signal being misjudged as "short-distance break", and realizing the decoupling and unification of the two types of judgments.

[0089] Table 3 Comparison of computational efficiency and clinical evaluation indicators

[0090] As shown in Table 3, the architecture of this invention based on traditional image processing and the Hungarian algorithm has the highest computational efficiency, requiring only 1.2 seconds per image. At the same time, its clinical diagnostic accuracy (Kappa coefficient 0.91) is significantly better than that of traditional methods and general deep learning methods, proving that this invention can provide clinical judgment results that are highly consistent with those of pathologists while ensuring efficient computation.

[0091] Example 5 This embodiment provides a collaborative discrimination device for fragmentation and fusion of fluorescence in situ hybridization images based on adaptive fluorescence dot detection and intelligent spatial decision-making. This device strictly corresponds to the steps of the above method embodiment and includes the following modules: The image acquisition and preprocessing module is used to acquire images of fluorescence in situ hybridization cells to be analyzed, and to perform channel separation on the cell images to extract fluorescence images from different channels.

[0092] The adaptive feature detection module is used to calculate the binarization threshold of each channel based on the adaptive extreme value ratio of image brightness to extract the preliminary mask, and to set the morphological filtering interval by combining the median area of ​​the connected domains in the preliminary mask, extract the center coordinates of independent fluorescent points in each channel and estimate the equivalent diameter.

[0093] The morphology enhancement cell segmentation module is used to extract morphological prior features and appearance features from cell images and perform feature fusion to generate a cell nuclear mask.

[0094] The intelligent dual-path collaborative decision-making module is used to calculate the spatial distance between heterochromatic fluorescent dots globally, extract heterochromatic fluorescent dot pairs whose spatial distance is less than the fusion determination threshold, determine them as fusion signals, and physically remove them; for the remaining heterochromatic fluorescent dots after removal, pair them up using a global optimal pairing algorithm within the same cell nuclear mask area. If the pairing distance is greater than or equal to the breakage determination threshold dynamically set based on the equivalent diameter, it is determined to be a breakage signal.

[0095] The results generation and output module is used to determine the comprehensive classification state corresponding to each cell nuclear mask based on the judgment results of the fracture signal and fusion signal, and to perform visualization rendering and standardized output.

[0096] The specific functional implementation details of each module are consistent with the corresponding method steps in Examples 1 to 4, and will not be repeated here.

[0097] Example 6 This embodiment also provides an electronic device, see reference. Figure 5 It includes a memory 402 and a processor 401, the memory 402 storing a computer program and the processor 401 being configured to run the computer program to perform the steps in any of the above method embodiments.

[0098] Specifically, the processor 401 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of the present invention.

[0099] The memory 402 may include a mass storage device for data or instructions. For example, and not limitingly, the memory 402 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 402 may include removable or non-removable (or fixed) media. Where appropriate, the memory 402 may be internal or external to a data processing device. In a particular embodiment, the memory 402 is non-volatile memory. In a particular embodiment, the memory 402 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random Access Memory (FPMDRAM), Extended Data Out Dynamic Random Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0100] The memory 402 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 401.

[0101] The processor 401 implements any of the methods described above by reading and executing computer program instructions stored in the memory 402.

[0102] Optionally, the electronic device may further include a transmission device 403 and an input / output device 404, wherein the transmission device 403 is connected to the processor 401 and the input / output device 404 is connected to the processor 401.

[0103] The transmission device 403 can be used to receive or send data via a network. Specific examples of the network described above may include wired or wireless networks provided by the communication provider of the electronic device. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 403 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0104] Input / output device 404 is used for inputting or outputting information. It can be a speaker, microphone, monitor, or keyboard.

[0105] Example 7 This embodiment also provides a readable storage medium storing a computer program, the computer program including program code for controlling a process to execute the process, the process including the fluorescence in situ hybridization image fragmentation and fusion co-discrimination method according to Embodiment 1.

[0106] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0107] Generally, various embodiments can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention can be implemented in hardware, while others can be implemented by firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, by way of non-limiting example, these blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0108] Embodiments of the present invention can be implemented by computer software, which may be executable by a data processor of a mobile device, such as a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products) including software routines, applets, and / or macros can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. The computer program product may include one or more computer-executable components configured to perform the embodiments when the program is run. The one or more computer-executable components may be at least one piece of software code or a portion thereof. Additionally, it should be noted in this respect that, as Figure 1 Any box in the logical flow can represent a program step, or interconnected logic circuits, boxes and functions, or a combination of program steps and logic circuits, boxes and functions. Software can be stored on physical media such as memory chips or blocks of storage implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as DVDs and their data variants, CDs, etc. The physical medium is a non-transient medium.

[0109] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0110] The above embodiments are merely illustrative of several implementations of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the appended claims.

Claims

1. A method for collaboratively discriminating breaks and fusions in fluorescence in situ hybridization images, characterized in that, Includes the following steps: Acquire images of the fluorescence in situ hybridization (FISH) cells to be analyzed, and perform channel separation on the FISH cell images to extract fluorescence images from different channels; The binarization threshold of each channel is calculated based on the adaptive extreme value ratio of image brightness to extract the preliminary mask. The shape filtering interval is set by combining the median area of ​​the connected domains in the preliminary mask, and the center coordinates of the independent fluorescent points in each channel are extracted and the equivalent diameter is estimated. The morphological prior features and appearance features of the fluorescence in situ hybridization cell images are extracted and fused to generate a cell nuclear mask; Dual-path collaborative determination is performed based on the spatial distance between the center coordinates of the independent fluorescence points: The spatial distance between heterochromatic fluorescent points is calculated globally. Heterochromatic fluorescent point pairs with a spatial distance less than the fusion determination threshold are identified as fusion signals and physically removed. For the remaining heterochromatic fluorescent dots after removal, a global optimal pairing algorithm is used to pair and connect them within the same cell nuclear mask area. If the pairing distance is greater than or equal to the breakage judgment threshold dynamically set based on the equivalent diameter, it is determined to be a breakage signal. The breakage judgment threshold is the product of the average equivalent diameter of different channels and the preset breakage distance ratio coefficient. The value range of the fracture distance proportionality coefficient is 2.0 to 3.0; Based on the determination results of the fracture signal and the fusion signal, the comprehensive classification status corresponding to each of the cell nuclear masks is determined and output.

2. The method for collaborative discrimination of image fragmentation and fusion in fluorescence in situ hybridization as described in claim 1, characterized in that, The steps include calculating the binarization threshold of each channel based on the adaptive extreme value ratio of image brightness to extract the initial mask, and setting the morphological filtering interval based on the median area of ​​the connected components within the initial mask, including: The maximum pixel value in a single channel image is obtained, and the maximum pixel value is multiplied by an adaptive threshold scaling factor to obtain the binarization threshold. The channel image is then binarized and denoised according to the binarization threshold to obtain the preliminary mask. Extract the effective connected components from the initial mask and calculate the median area of ​​the effective connected components; The median area is multiplied by the lower limit area ratio coefficient and the upper limit area ratio coefficient to define the morphological filtering interval in order to remove noise and abnormal clumps.

3. The method for collaborative discrimination of fragmentation and fusion in fluorescence in situ hybridization images as described in claim 2, characterized in that, The adaptive threshold scaling factor ranges from 0.3 to 0.5; the lower limit scaling factor is 0.1, and the upper limit scaling factor is 3.

0.

4. The method for collaborative discrimination of fragmentation and fusion in fluorescence in situ hybridization images as described in claim 1, characterized in that, The steps of extracting morphological prior features and appearance features from the fluorescence in situ hybridization cell image and performing feature fusion to generate a nuclear mask include: Prior cell morphology features and multi-scale appearance features were extracted in parallel from the fluorescence in situ hybridization cell images. The cell morphology prior features are used as query vectors, and the multi-scale appearance features are used as key vectors and value vectors for cross-attention calculation to generate optimized fusion features. Based on the optimized fusion features, a decoder network generates a nuclear mask containing the boundaries of each cell, and removes abnormal masks that do not meet the preset morphological feature lower limit.

5. The method for collaborative discrimination of fragmentation and fusion in fluorescence in situ hybridization images as described in claim 1, characterized in that, The steps of performing pairing and connection using a globally optimal pairing algorithm within the same nuclear mask area include: For heterochromatic fluorescent dots of different channels with unequal numbers remaining in the same cell nuclear mask, a corresponding number of virtual nodes are added to align the total number of dots. The actual Euclidean distance between real heterochromatic fluorescent points and the maximum penalty cost involving the virtual nodes are combined and spliced ​​to construct an extended non-equilibrium cost matrix; The Hungarian algorithm is used to solve the unbalanced cost matrix to obtain the globally optimal pairing connection that avoids local pairing interference.

6. The method for collaborative discrimination of fragmentation and fusion in fluorescence in situ hybridization images as described in claim 1, characterized in that, The steps for determining and outputting the overall classification state corresponding to each of the aforementioned nuclear masks include: If both the breakage signal and the fusion signal are simultaneously detected within the nuclear mask, it is classified as a complex state; if only the fusion signal is present, it is classified as a fusion state; if only the breakage signal is present, it is classified as a breakage state; if neither the breakage signal nor the fusion signal is detected, and heterochromatic fluorescent spots are present, it is classified as a normal co-occurrence state. The output of each comprehensive classification state is visualized and rendered, the boundary outlines of the cell nuclei in different states are drawn with corresponding colors, and the judgment results are converted into a standardized data format for output.

7. A fluorescence in situ hybridization image fragmentation and fusion co-discrimination device, characterized in that, include: The image acquisition and preprocessing module is used to acquire images of fluorescence in situ hybridization cells to be analyzed, and to perform channel separation on the fluorescence in situ hybridization cell images to extract fluorescence images of different channels; An adaptive feature detection module is used to calculate the binarization threshold of each channel based on the adaptive extreme value ratio of image brightness to extract the preliminary mask, and to set the morphological filtering interval in combination with the median area of ​​the connected domains in the preliminary mask, extract the center coordinates of independent fluorescent points in each channel and estimate the equivalent diameter. The morphology-enhanced cell segmentation module is used to extract the morphological prior features and appearance features of the fluorescence in situ hybridization cell image and perform feature fusion to generate a cell nuclear mask. The intelligent dual-path collaborative decision-making module is used to calculate the spatial distance between heterochromatic fluorescent points in the global scope, extract heterochromatic fluorescent point pairs whose spatial distance is less than the fusion judgment threshold, judge them as fusion signals, and physically remove them; For the remaining heterochromatic fluorescent dots after removal, a global optimal pairing algorithm is used to pair and connect them within the same cell nuclear mask area. If the pairing distance is greater than or equal to a breakage judgment threshold dynamically set based on the equivalent diameter, it is determined to be a breakage signal. The breakage judgment threshold is the product of the average equivalent diameter of different channels and a preset breakage distance ratio coefficient. The breakage distance ratio coefficient ranges from 2.0 to 3.

0. The result generation and output module is used to determine the comprehensive classification state corresponding to each of the cell nuclear masks based on the judgment results of the fracture signal and the fusion signal, and to perform visualization rendering and standardized output.

8. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to execute the fluorescence in situ hybridization image breakage and fusion co-discrimination method according to any one of claims 1 to 6.

9. A readable storage medium, characterized in that, The readable storage medium stores a computer program, the computer program including program code for controlling the process to execute the process, the process including the fluorescence in situ hybridization image fragmentation and fusion co-discrimination method according to any one of claims 1 to 6.

Images