An automatic indirect immunofluorescence interpretation method and system, storage medium and program product

Abstract

This application provides a fully automated indirect immunofluorescence interpretation method, system, storage medium, and program product, relating to the field of indirect immunofluorescence detection technology. The method includes: upon obtaining the identification information of the target sample, placing multiple sealed antigen slides in batches into the slide slots of a carrier platform; driving the carrier platform via a transmission mechanism to sequentially move the multiple antigen slides to the imaging area; using a positioning clamping device to perform positioning and fixing operations on the antigen slides entering the imaging area; sequentially performing multi-field scanning on the antigen slides entering the imaging area to obtain multi-field fluorescence images; stitching these fluorescence images to generate a panoramic fluorescence image; extracting features from the panoramic image to determine the fluorescence nucleotype category and target fluorescence intensity value; determining the antibody titer based on the category, intensity value, and dilution information; and associating and mapping the titer, category, and panoramic image based on the identification information to generate a fluorescence interpretation result.

Description

Technical Field

[0001] This application relates to the field of indirect immunofluorescence detection technology, and in particular to a fully automated indirect immunofluorescence interpretation method, system, storage medium and program product. Background Technology

[0002] In the field of clinical laboratory testing, antinuclear antibody (ANA) detection is an important screening method for the diagnosis of autoimmune diseases. Among them, indirect immunofluorescence assay (IIF) is widely used in clinical departments such as rheumatology and dermatology due to its high sensitivity and ability to directly observe fluorescent karyotype characteristics. It is of great significance for improving the accuracy of early diagnosis of diseases and improving patient prognosis.

[0003] In related technologies, a semi-automatic detection and interpretation method combining manual microscopic examination with local imaging is commonly used. Specifically, the laboratory personnel first manually place each sealed antigen slide onto the microscope stage and adjust the focus through the eyepiece to find a suitable observation area. Then, imaging is performed within one or more selected local fields of view to acquire fluorescence images of that area. Next, the cellular fluorescence karyotype and fluorescence intensity in the local field of view are identified and evaluated through manual observation or simple image analysis software. Finally, the laboratory personnel synthesize the local observation results, combine them with their personal experience, and manually record interpretation data such as fluorescence karyotype type and antibody titer, and issue a report. While this method, primarily based on manual operation and local observation, can complete basic detection and interpretation tasks, the entire process relies on human intervention and can only acquire fluorescence image information from a portion of the slide.

[0004] However, the aforementioned semi-automatic detection and interpretation method has several drawbacks. First, relying solely on fluorescence images from a single or a few local fields of view for analysis fails to capture the overall cell distribution and fluorescence expression of the slide. This may lead to the omission of positive cells or characteristic karyotypes in unobserved areas due to field selection bias, thus affecting the accuracy of the interpretation results. Second, manual microscopic examination and interpretation depend on the personal experience of the examiners, and the interpretation standards of different personnel vary, making it difficult to guarantee the consistency of interpretation results. Consequently, the accuracy rate of indirect immunofluorescence interpretation in related technologies is relatively low. Summary of the Invention

[0005] This application provides a fully automated indirect immunofluorescence interpretation method, system, storage medium, and program product to improve the accuracy of indirect immunofluorescence interpretation.

[0006] Firstly, this application provides a fully automated indirect immunofluorescence interpretation method, applied to the aforementioned fully automated indirect immunofluorescence interpretation system. The method includes: upon obtaining the identification information of the target sample, placing multiple sealed antigen slides in batches into slide slots on a carrier platform; driving the carrier platform via a transmission mechanism to sequentially move the multiple antigen slides to the imaging area; and using a positioning clamping device to perform positioning and fixing operations on the antigen slides entering the imaging area; sequentially performing multi-field scanning on the antigen slides entering the imaging area to obtain multi-field fluorescence images; and stitching the multi-field fluorescence images to generate a fluorescence panoramic image; extracting features from the fluorescence panoramic image to determine the fluorescence karyotype category and target fluorescence intensity value; and determining the antibody titer based on the fluorescence karyotype category, the fluorescence intensity value, and the dilution information of the target sample; and associating and mapping the antibody titer, the fluorescence karyotype category, and the fluorescence panoramic image based on the identification information to generate a fluorescence interpretation result.

[0007] By employing the above technical solution, after obtaining the identification information of the target sample, multiple antigen slides, after being sealed, are placed in batches into the slide slots of the carrier platform, achieving batch loading and identification of samples. The transmission mechanism drives the carrier platform to move multiple antigen slides sequentially to the imaging area. This automated transmission method avoids the inefficiency of manual slide-by-slide operation. Simultaneously, the positioning and clamping device performs positioning and fixing operations on the antigen slides entering the imaging area, ensuring the positional stability of the slides during the imaging process. By sequentially scanning the antigen slides entering the imaging area with multiple fields of view, multi-field fluorescence images are obtained, overcoming the limitation of the limited field of view of single-field imaging. The multi-field fluorescence images are stitched together to generate a fluorescence panoramic image, achieving complete imaging of the detection area. Feature extraction is performed on the fluorescence panoramic image to determine the fluorescence karyotype type and target fluorescence intensity value. Based on the fluorescence karyotype type, fluorescence intensity value, and dilution information of the target sample, the antibody titer is determined, achieving quantitative immunofluorescence interpretation. Finally, based on the identification information, the antibody titer, fluorescence karyotype type, and fluorescence panoramic image are correlated and mapped to generate traceable fluorescence interpretation results. This solves the technical problem of low accuracy in indirect immunofluorescence interpretation in related technologies, and achieves the technical effect of improving the accuracy of indirect immunofluorescence interpretation.

[0008] Secondly, embodiments of this application provide a fully automated indirect immunofluorescence interpretation system, which includes: one or more processors and a memory; the memory is coupled to one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the fully automated indirect immunofluorescence interpretation system to perform the method as described in the first aspect and any possible implementation thereof.

[0009] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a fully automated indirect immunofluorescence interpretation system, cause the fully automated indirect immunofluorescence interpretation system to perform the method described in the first aspect and any possible implementation thereof.

[0010] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a fully automated indirect immunofluorescence interpretation system, cause the fully automated indirect immunofluorescence interpretation system to perform the method described in the first aspect and any possible implementation thereof. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating a fully automated indirect immunofluorescence interpretation method in an embodiment of this application; Figure 2 This is an interactive flowchart of a fully automated indirect immunofluorescence interpretation embodiment in this application; Figure 3 This is a schematic diagram of the physical device structure of a fully automated indirect immunofluorescence interpretation system in the embodiments of this application. Detailed Implementation

[0012] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0013] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0014] This application provides a fully automated method for interpreting indirect immunofluorescence, see reference. Figure 1 , Figure 1 This is a flowchart illustrating a fully automated indirect immunofluorescence interpretation method in this application, including the following steps: Step S101: After obtaining the identification information of the target sample, multiple antigen slides that have been sealed are placed in batches in the slide slots of the carrier platform. The carrier platform is driven by the transmission mechanism to move the multiple antigen slides to the imaging area in sequence. The positioning and clamping device is used to perform positioning and fixing operations on the antigen slides that have entered the imaging area. Step S102: Perform multi-field scanning on the antigen slides entering the imaging area sequentially to obtain multi-field fluorescence images, and stitch the multi-field fluorescence images to generate a fluorescence panoramic image. Step S103: Feature extraction is performed on the fluorescence panoramic image to determine the fluorescence nucleotype category and target fluorescence intensity value, and the antibody titer is determined based on the fluorescence nucleotype category, the fluorescence intensity value, and the dilution information of the target sample. Step S104: Based on the identification information, the antibody titer, the fluorescent nucleotype category, and the fluorescent panoramic image are associated and mapped to generate a fluorescence interpretation result.

[0015] The target sample refers to the biological sample that needs to be detected by indirect immunofluorescence, including but not limited to serum samples, plasma samples, and cerebrospinal fluid samples; the identification information refers to the code or mark used to uniquely identify and distinguish different samples, including but not limited to barcodes, QR codes, and sample numbers; the antigen slide refers to the slide carrying a specific antigen, which serves as the detection carrier in indirect immunofluorescence detection, including but not limited to Hep-2 cell slides, liver tissue slides, and kidney tissue slides; the carrier platform refers to the mechanical device used to place and transport the antigen slides, including but not limited to a rotating slide stage, a linear conveyor belt, and a multi-layer slide holder; the slide slot refers to the fixed position or groove on the carrier platform specifically used to place the slide, including... However, this does not include, but is not limited to, circular slots, rectangular slots, trapezoidal slots, etc.; the transmission mechanism refers to the mechanical system that drives the movement of the carrier platform, including but not limited to stepper motor drive systems, servo motor drive systems, linear motor drive systems, etc.; the imaging area is used to represent a specific spatial location for microscope imaging and image acquisition, including but not limited to the microscope stage area, the field of view area of ​​the CCD (Charged Coupled Device) camera, the fluorescence excitation area, etc.; the positioning and clamping device refers to a mechanical device used to accurately position and fix the slide, including but not limited to pneumatic clamps, vacuum adsorption devices, mechanical snap-fit ​​devices, etc.; multi-field scanning is used to represent the division of a single slide into multiple regions at different positions. Scanning imaging includes, but is not limited to, grid scanning, spiral scanning, and random point scanning; multi-field fluorescence images refer to a collection of multiple local fluorescence images obtained through multi-field scanning, including but not limited to 2×2 array images, 3×3 array images, and 5×5 array images; stitching processing is used to describe the process of combining multiple local images into a complete image through algorithmic processing, including but not limited to feature point matching stitching, phase correlation stitching, and template matching stitching; fluorescence panoramic images refer to complete large-field fluorescence images generated after stitching processing, including but not limited to panoramic images of the entire slide, panoramic images of specific detection areas, and panoramic images synthesized from multiple regions; feature extraction is used to describe the extraction of key information for analysis from images. The process of extracting and defining characteristic parameters includes, but is not limited to, morphological feature extraction, texture feature extraction, and fluorescence intensity feature extraction; fluorescence karyotype refers to the type of karyotype classified according to the nuclear fluorescence staining pattern, including but not limited to homogeneous, speckled, and nucleolar types; target fluorescence intensity value is used to represent the representative fluorescence intensity value obtained through measurement and calculation, including but not limited to average fluorescence intensity, median fluorescence intensity, and peak fluorescence intensity; dilution information refers to the dilution factor data of the sample before detection, including but not limited to 1:80 dilution, 1:160 dilution, and 1:320 dilution; antibody titer is used to represent the concentration or activity level of a specific antibody in the sample, including but not limited to 1:80, 1:160, and 1:320.Fluorescence interpretation results refer to the final report data containing all test information, including but not limited to positive result reports, negative result reports, and weakly positive result reports.

[0016] In the above embodiment, taking the detection of ANA antinuclear antibodies as an example, the sample barcode ANA20241205001 is first scanned using an identification information reading device to confirm that the target sample is a serum sample diluted 1:80. Twenty sealed Hep-2 cell antigen slides are placed sequentially into the slide slots of the cuboid-shaped support platform. Each rectangular slot is 1.2 mm deep and 26 mm wide, with a limiting flange height of 1.5 mm to ensure accurate insertion and fixation of the slides in the preset positions. The stepper motor drive system is activated, driving the support platform along the longitudinal guide rail via a lead screw transmission mechanism. The first antigen slide moves at a speed of 5 mm / s to the imaging area below the fluorescence microscope. At this time, the limiting block and the clamping spring of the pneumatic clamping device act simultaneously. The lateral limiting block presses against the edge of the slide, and the upper clamping spring applies a pressure of 0.2 N to fix the slide on the stage, ensuring stable slide position during imaging. The fluorescence microscope employed a 40x objective lens in conjunction with a CCD camera with a resolution of 2048×2048 pixels, corresponding to a physical resolution of 0.24 μm / pixel. A 12 mm × 8 mm Hep-2 cell detection area was scanned using a grid-based multi-field scanning method. Based on a preset field-of-view overlap of 20% and a single field-of-view size of 1.2 mm × 1.2 mm, 96 scanning fields of view (12 columns × 8 rows) were calculated. The stage's precision movement mechanism moved sequentially to each field of view position along a serpentine path starting from the upper left corner. After reaching each position, the focusing mechanism performed 11 coarse localization scans within a ±50 μm range at 10 μm intervals. The Laplacian operator was used to calculate the image sharpness value of each layer, determining the layer with the highest sharpness as the coarse localization focal plane. Subsequently, 10 fine steps were performed within a 5 μm range on either side of this layer with a 1 μm step size to finally determine the precise focal plane layer. A 2048 × 2048 pixel single-field fluorescence image was then acquired at this position.

[0017] In the above embodiment, after scanning 96 field-of-view units, a multi-field fluorescence image dataset consisting of 96 single-field fluorescence images is obtained. The image stitching program extracts the top-left corner image with spatial location number (1,1) as the first image to be stitched, and extracts the adjacent image at position (1,2) as the second image to be stitched. Feature points are extracted from the two images respectively. 247 feature points are detected in the first image to be stitched, and 231 feature points are detected in the second image to be stitched. 78 initial feature point pairs are obtained through the nearest neighbor matching algorithm. After reliability analysis and screening, 52 reliable feature point pairs are retained. The spatial transformation matrix is ​​calculated based on these point pairs, and affine transformation processing is performed on the second image to be stitched to register it to the coordinate system of the first image to be stitched. There is a 240×2048 pixel overlapping area between the two images. A weighted average fusion algorithm is used to process the overlapping area, with the weight coefficient linearly transitioning from 0.1 at the overlapping boundary to 0.9. The merged overlapping region is stitched together with the non-overlapping regions of the two images to obtain a 2808×2048 pixel intermediate stitched image. The remaining 94 images are processed in the same manner to generate a 14400×9600 pixel fluorescence panoramic image. Otsu's thresholding method is used for background separation on the panoramic image, setting the pixel value of the cell nucleus region to 255 and the pixel value of the background region to 0, resulting in a binary segmentation image. An 8-connected component algorithm is used to extract 1847 cell nucleus connected components. After removing noisy connected components with an area smaller than 100 pixels, 1623 effective cell nucleus connected components are retained.

[0018] In the above embodiments, comprehensive feature extraction is performed on each nuclear connected region: eight morphological features, including area, perimeter, roundness, and aspect ratio, are extracted from a morphological perspective; six intensity distribution features, including average intensity, variance, skewness, and kurtosis, are extracted through fluorescence intensity histogram analysis; four spatial distribution features, including Moran's index and Geary coefficient, are extracted using spatial autocorrelation analysis; and eight texture features, including contrast, correlation, energy, and entropy, are calculated using the gray-level co-occurrence matrix. These 26 feature values ​​are normalized and then concatenated to form a 26-dimensional comprehensive feature vector for each nuclear connected region. The 1623 comprehensive feature vectors are input into a preset random forest karyotype classification model, which contains 500 decision trees and outputs the probability distributions of four karyotypes: homogeneous, speckled, nucleolar, and cytoplasmic. With a probability threshold of 0.6 and a confidence threshold of 0.85, 1245 homogeneous nuclei, 312 speckled nuclei, 58 nucleolar nuclei, and 8 cytoplasmic nuclei were identified, confirming homogeneity as the dominant nucleotype. Fluorescence images were extracted from the connected regions of the 1245 homogeneous nuclei, and the average fluorescence intensity of each nucleus was measured, yielding an intensity distribution range of 85-156 grayscale values. The target fluorescence intensity value was calculated to be 118 using the median. Combining the 1:80 dilution information and the fluorescence intensity threshold standard (positive result ≥100), the antibody titer was determined to be 1:80 positive. Finally, based on the identification information ANA20241205001, the antibody titer of 1:80 positive, the homogeneous fluorescence nucleotype, and the 14400×9600 pixel fluorescence panoramic image were correlated and mapped to generate a standardized fluorescence interpretation result report containing patient information, detection parameters, quantitative results, qualitative conclusions, and image evidence. The report was then automatically transmitted to the hospital laboratory information system (LIS) via a data interface for result archiving.

[0019] Through the above steps, after obtaining the identification information of the target sample, multiple antigen slides with complete mounting are placed in batches into the slide slots of the carrier platform, realizing batch loading and identification of samples. The transmission mechanism drives the carrier platform to move multiple antigen slides sequentially to the imaging area. This automated transmission method avoids the inefficiency of manual slide-by-slide operation. At the same time, the positioning and clamping device performs positioning and fixing operations on the antigen slides entering the imaging area, ensuring the positional stability of the slides during the imaging process. By sequentially scanning the antigen slides entering the imaging area with multiple fields of view, multi-field fluorescence images are obtained, overcoming the limitation of the limited field of view of single-field imaging. The multi-field fluorescence images are stitched together to generate a fluorescence panoramic image, realizing complete imaging of the detection area. Feature extraction is performed on the fluorescence panoramic image to determine the fluorescence nucleotype category and target fluorescence intensity value. Based on the fluorescence nucleotype category, fluorescence intensity value, and dilution information of the target sample, the antibody titer is determined, realizing quantitative immunofluorescence interpretation. Finally, the antibody titer, fluorescence nucleotype category, and fluorescence panoramic image are correlated and mapped according to the identification information to generate traceable fluorescence interpretation results. This solves the technical problem of low accuracy in indirect immunofluorescence interpretation in related technologies, and achieves the technical effect of improving the accuracy of indirect immunofluorescence interpretation.

[0020] The entity performing the above steps can be a system, such as a fully automated indirect immunofluorescence interpretation system, or a device, or a controller or processor in the device or system, or a separate controller or processor, or other processing devices or processing units with similar processing functions, but is not limited to these.

[0021] In an optional embodiment, the antigen slide entering the imaging region is sequentially scanned using multiple fields of view to obtain a multi-field fluorescence image. Specifically, this includes sequentially performing the following operations on the antigen slide entering the imaging region: pre-scanning the antigen slide detection sub-region of the target antigen slide using a fluorescence microscope in low-magnification mode to obtain the boundary coordinate information of the antigen slide detection sub-region; the target antigen slide being the antigen slide currently entering the imaging region; obtaining a preset field-of-view overlap rate and the single-field-of-view size of the fluorescence microscope; and determining the number of scan columns required in the length direction and the number of scan rows required in the width direction of the antigen slide detection sub-region based on the boundary coordinate information, the preset field-of-view overlap rate, and the single-field-of-view size. The preset field-of-view overlap rate is the area of ​​the overlapping region between two adjacent scan field-of-view units and the area of ​​a single scan field-of-view unit. The ratio of the area of ​​the antigen slide is used to divide the antigen slide detection sub-region into multiple scanning field-of-view units according to the number of scanning columns and rows. Each scanning field-of-view unit is assigned a spatial position number and a scanning sequence number. Each scanning field-of-view unit is activated sequentially according to the scanning sequence number. When the target scanning field-of-view unit is activated, the stage of the fluorescence microscope is moved to the target spatial position corresponding to the spatial position number of the target scanning field-of-view unit. The target scanning field-of-view unit is the currently activated scanning field-of-view unit. Adaptive focusing is performed on the target scanning field-of-view unit to obtain a single-field fluorescence image of the target scanning field-of-view unit. After obtaining the single-field fluorescence image of each scanning field-of-view unit, the single-field fluorescence images of each scanning field-of-view unit are arranged and combined according to the spatial position number to obtain a multi-field fluorescence image.

[0022] Here, fluorescence microscope refers to optical microscope equipment specifically used for fluorescence imaging, including but not limited to inverted fluorescence microscopes, upright fluorescence microscopes, and confocal fluorescence microscopes; low magnification mode refers to the working mode of the microscope using a lower magnification for observation, including but not limited to 4x objective mode, 10x objective mode, and 20x objective mode; antigen slide detection sub-region refers to the effective area on the antigen slide that is actually detected, including but not limited to cell adhesion area, tissue section area, and antigen coating area; pre-scan refers to a low-resolution rapid scan performed before the formal high-resolution scan, including but not limited to boundary recognition scan, focus pre-positioning scan, and region division scan; boundary coordinate information is used to represent the spatial location data of the detection area boundary, including but not limited to rectangular boundary coordinates, irregular polygon coordinates, and circular area coordinates; preset field of view overlap rate refers to the preset proportion of the overlapping area between adjacent scan fields to the area of ​​a single field of view, including but not limited to 10% overlap rate, 15% overlap rate, and 20% overlap rate; single field of view size is used to represent the actual physical size covered by the microscope in a single imaging, including but not limited to the physical size corresponding to 512×512 pixels, the physical size corresponding to 1024×1024 pixels, and 2048 pixels. The physical dimensions corresponding to ×2048 pixels, etc.; the number of scan columns refers to the number of scan fields required along the length of the detection area, including but not limited to 5-column scans, 8-column scans, 12-column scans, etc.; the number of scan rows indicates the number of scan fields required along the width of the detection area, including but not limited to 4-row scans, 6-row scans, 10-row scans, etc.; a scan field of view unit refers to each independent scan block after dividing the detection area, including but not limited to square field of view units, rectangular field of view units, overlapping field of view units, etc.; the spatial location number is used to indicate the spatial location of each scan field of view unit, including but not limited to row and column coordinates. Numbering, sequential numbering, alphanumeric combination numbering, etc.; Scanning sequence numbering refers to the numbering that determines the order in which scans are performed, including but not limited to left-to-right line-by-line scanning numbering, serpentine path scanning numbering, spiral scanning numbering, etc.; Stage refers to the platform device in the microscope that carries the sample and can be moved precisely, including but not limited to XY-axis motorized stage, three-axis precision stage, rotating stage, etc.; Single field-of-view fluorescence image refers to the fluorescence imaging result corresponding to a single scanning field of view, including but not limited to 512×512 resolution image, 1024×1024 resolution image, 2048×2048 resolution image, etc.

[0023] In the above embodiment, taking an antigen slide containing Hep-2 cells as an example, after the slide is moved to the imaging area by the carrier platform transmission mechanism, the 10x objective lens of the inverted fluorescence microscope is first activated to enter low-magnification mode. The XY axis precision movement mechanism of the stage moves the slide to the preset pre-scan starting position, which is 2mm away from the upper left edge of the slide. The LED blue light source configured in the fluorescence microscope emits 470nm excitation light, which is irradiated on the surface of the Hep-2 cell detection sub-region through the objective lens. The FITC fluorescence signal emitted by the excited cells enters the CCD camera after passing through a 525nm bandpass filter. During the pre-scan, the stage moves along the X-axis at a constant speed of 10mm / s, and the CCD camera continuously acquires low-resolution preview images of 512×512 pixels at a collection frequency of 2 frames / second. When the stage moves to the right edge of the slide, the image edge detection algorithm identifies that the fluorescence signal intensity suddenly drops from a gray value of 15 in the background to below a gray value of 5, and automatically records the X-axis coordinate at this time as 23.5mm. The stage was then moved 0.5 mm along the Y-axis and a reverse X-axis scan was performed. The upper boundary coordinates (Y=8.2 mm), lower boundary coordinates (Y=16.8 mm), and left boundary coordinates (X=2.3 mm) of the detection sub-region were identified sequentially using the same method. Finally, the rectangular boundary coordinates of the Hep-2 cell detection sub-region were determined to be the upper left corner (2.3 mm, 8.2 mm) and the lower right corner (23.5 mm, 16.8 mm), with an actual detection area size of 21.2 mm × 8.6 mm. The preset field-of-view overlap rate was read from the parameter configuration file as 15%, and the single-field-of-view size of the 10x objective lens with the CCD camera was obtained as 1.31 mm × 1.31 mm. Based on the boundary coordinate information, the effective coverage requirement of the detection sub-region is calculated as follows: In the length direction, considering a 15% overlap rate, the effective coverage width between adjacent fields of view is 1.31×(1-0.15)=1.1135mm, so the required number of scanning columns is ⌈21.2÷1.1135⌉=20 columns; in the width direction, the required number of scanning rows is ⌈8.6÷1.1135⌉=8 rows, forming a grid array of 20×8=160 scanning field of view units in total.

[0024] In the above embodiments, each scanning field of view is assigned a spatial position number in row and column coordinates, starting from the top left corner and numbered sequentially as (1,1), (1,2), ..., (1,20), (2,1), (2,2), ..., (8,20). Simultaneously, a scanning sequence number is assigned according to a serpentine scanning strategy: the first row is numbered 001-020 from left to right, the second row from right to left as 021-040, the third row from left to right as 041-060, and so on until the eighth row, numbered 141-160. This serpentine path design minimizes the movement distance of the stage between adjacent fields of view, improving scanning efficiency. During the scanning execution phase, each scanning field of view is activated sequentially according to its scanning sequence number. When the target scanning field of view unit numbered 001 is activated, the target spatial coordinates corresponding to its spatial position number (1,1) are calculated as X=2.3+0.655=2.955mm and Y=8.2+0.655=8.855mm, which is the center position of the field of view. The XY-axis motorized stage moves to the target position at a speed of 20mm / s, with the positioning accuracy controlled within ±2μm. After reaching the target position, the microscope automatically switches to 40x high magnification mode, and the single field of view size is adjusted accordingly to 0.328mm×0.328mm. The adaptive focusing program is started, and the Z-axis focusing mechanism first performs 11 levels of coarse positioning scanning within ±25μm of the current position at 5μm intervals. At each focal plane level, the CCD camera acquires an initial fluorescence image of 1024×1024 pixels, and the Laplacian variance sharpness value is calculated for each image. After measurement, the sharpness values ​​of the 11 levels were: 148, 203, 267, 298, 356, 389, 351, 312, 278, 231, and 189. Among them, the sharpness value of level 6, 389, was the highest, and it was determined to be the coarse positioning focal plane level.

[0025] In the above embodiment, fine focusing is then performed. The Z-axis focusing mechanism moves 12 times within a ±3μm range around this level, with a step size of 0.5μm. At each step position, a 1024×1024 pixel fluorescence image is acquired, and the sharpness value is calculated: 375, 382, ​​389, 401, 415, 423, 418, 408, 395, 383, 371, 358. The sharpness value of 423 at the 6th step position is the highest, and it is determined to be the precise focal plane level. The focusing mechanism stays at this position, and the CCD camera formally acquires a 1024×1024 pixel single-field fluorescence image of the target scanning field of view unit. The actual physical resolution of the image is 0.32μm / pixel. After imaging all 160 scanning field of view units in sequence, 160 single-field fluorescence images are obtained. The image management program arranges these images in an 8-row × 20-column array format according to their spatial location numbers: the first row contains 20 images with location numbers (1,1) to (1,20), the second row contains 20 images with location numbers (2,1) to (2,20), and so on up to the eighth row. Each image retains its original spatial location information and imaging parameters in the data structure, forming a complete multi-field fluorescence image dataset.

[0026] In an optional embodiment, an adaptive focusing operation is performed on the target scanning field of view unit to obtain a single-field fluorescence image of the target scanning field of view unit. Specifically, this includes: after the stage moves to the target spatial position, controlling the focusing mechanism of the fluorescence microscope to move along the optical axis within a preset focusing range to multiple preset focal plane levels, and acquiring initial fluorescence images at each of the multiple preset focal plane levels to obtain multiple initial fluorescence images. The optical axis is the direction of the objective optical axis of the fluorescence microscope, and the multiple preset focal plane levels are multiple focal plane positions distributed at preset level intervals along the optical axis within a preset focusing range; performing a first sharpness analysis on the multiple initial fluorescence images to determine multiple initial sharpness values ​​corresponding one-to-one with the multiple preset focal plane levels; arranging the multiple initial sharpness values ​​in descending order, so that the preset focal plane level corresponding to the first initial sharpness value after arrangement is determined as the coarseness of the target scanning field of view unit. The focal plane is positioned at a specific level. Using the coarse focal plane as the center, the focusing mechanism is controlled to move in steps within a preset offset range on both sides of the coarse focal plane along the optical axis. After each step, a step fluorescence image is acquired, resulting in multiple step fluorescence images. The single step size of each step is less than the preset level interval. A second sharpness analysis is performed on the multiple step fluorescence images to determine multiple step sharpness values ​​that correspond one-to-one with each image. These step sharpness values ​​are arranged in descending order, and the step movement position corresponding to the first step sharpness value is determined as the precise focal plane level of the target scanning field of view. The focusing mechanism is controlled to remain at the precise focal plane level, and the fluorescence microscope is switched from low-magnification mode to high-magnification mode. In high-magnification mode, the CCD camera acquires images of the target scanning field of view at the precise focal plane level, resulting in a single-field fluorescence image.

[0027] The focusing mechanism refers to the mechanical device that controls the focal position of the microscope, including but not limited to stepper motor focusing mechanisms, piezoelectric ceramic focusing mechanisms, and voice coil motor focusing mechanisms. The optical axis direction refers to the direction of the optical axis of the microscope objective lens, usually perpendicular to the Z-axis of the stage, including but not limited to the vertically upward direction, vertically downward direction, and tilt angle direction. The preset focus range indicates the spatial range of the focusing search, including but not limited to ±50 micrometers, ±100 micrometers, and ±200 micrometers. The preset focal plane hierarchy refers to multiple focal positions distributed at fixed intervals within the focus range, including but not limited to focal planes spaced 5 micrometers apart. The focal planes are defined as follows: focal planes, focal planes every 10 micrometers, focal planes every 20 micrometers, etc.; initial fluorescence images are used to represent the original images acquired at each preset focal plane level, including but not limited to coarse positioning scan images, fast acquisition images, low-quality preview images, etc.; first sharpness analysis refers to the algorithmic processing that quantifies the image sharpness, including but not limited to gradient magnitude analysis, Laplacian operator analysis, variance analysis, etc.; initial sharpness values ​​are numerical indicators used to represent the image sharpness, including but not limited to gradient variance values, edge strength values, contrast scores, etc.; coarse positioning focal plane levels refer to the approximate optimal focus determined through preliminary focusing. Position, including but not limited to the initial optimal Z-axis position, coarse focus position, and near-optimal level; stepping movement is used to represent the fine displacement operation of the focusing mechanism, including but not limited to 1-micron stepping, 0.5-micron stepping, and 0.1-micron stepping; stepping fluorescence image refers to the image acquired during the fine stepping process, including but not limited to fine focus image, high-precision scan image, and micro-stepping image; second sharpness analysis is used to represent a more accurate sharpness assessment of the stepping image, including but not limited to fine gradient analysis, high-precision edge detection, and sub-pixel sharpness assessment; stepping sharpness value refers to the fineness of each position during the stepping movement. Precise sharpness values ​​include, but are not limited to, fine sharpness scores, high-precision sharpness indicators, and micron-level sharpness values; precise focal plane level indicates the optimal focal position determined through fine adjustment, including but not limited to optimal Z-axis coordinates, precise focus position, and sub-micron-level optimal level; high magnification mode refers to the microscope switching to a high magnification working mode, including but not limited to 40x objective mode, 63x objective mode, and 100x objective mode; CCD camera refers to charge-coupled device cameras, which are digital image acquisition devices, including but not limited to cooled CCD cameras, high-sensitivity CCD cameras, and color CCD cameras.

[0028] In the above embodiment, taking the target scanning field of view unit with spatial location number (3,8) as an example, when the XY-axis motorized stage is precisely moved to the target spatial position corresponding to this field of view unit (X=9.655mm, Y=10.055mm), the adaptive focusing operation program is initiated. At this time, the fluorescence microscope remains in the low magnification mode of the 10x objective lens for rapid focusing search during the coarse positioning stage. The focusing mechanism adopts a precision lead screw structure driven by a stepper motor, which can achieve a minimum displacement resolution of 0.1μm. First, the system configuration parameters are read to determine the preset focusing range as ±30μm above and below the current Z-axis position, with a total search range of 60μm. The preset focal plane levels are distributed within this range at fixed intervals of 6μm, forming 11 discrete focal plane positions: Z-30μm, Z-24μm, Z-18μm, Z-12μm, Z-6μm, Z0μm, Z+6μm, Z+12μm, Z+18μm, Z+24μm, and Z+30μm, where Z0μm is the initial Z-axis coordinate when the stage reaches the target position. The coarse positioning stage begins, with the focusing mechanism moving sequentially to each preset focal plane level at a constant speed of 5μm / s. At each level, it pauses for 200ms for mechanical vibration damping, and then the CCD camera acquires an initial 512×512 pixel fluorescence image at the current focal plane, with an exposure time set to 50ms and a gain value set to 2.5x. The images acquired at the first level, Z-30μm, were generally blurry, with indistinct cell nucleus edges. Image sharpness gradually improved as the focal plane moved upwards. At the Z+6μm level, the fluorescence signal of the Hep-2 cell nuclei reached its strongest, with a clear boundary between the cytoplasm and nucleus. A first-stage sharpness analysis was performed on the 11 initial fluorescence images. Laplacian convolution kernels [-1,-1,-1;-1,8,-1;-1,-1,-1] were used to enhance the edges of each image, and the pixel variance of the processed image was calculated as the sharpness evaluation index. The calculation results showed that the initial sharpness values ​​corresponding to each level were: 158, 203, 267, 324, 389, 445, 472, 436, 398, 341, and 285. After arranging these values ​​in descending order, the Z+6μm level, corresponding to the highest value of 472, was determined as the coarse focal plane level.

[0029] In the above embodiment, the fine focusing stage is centered on the Z+6μm level, performing a high-precision search within a preset offset range of ±4μm on both sides of this level. The stepping movement parameter of the focusing mechanism is adjusted to a single movement step size of 0.5μm, which is much smaller than the preset level spacing of 6μm in the coarse positioning stage, ensuring the resolution capability of fine focusing. Seventeen stepping positions are set in the range of Z+2μm to Z+10μm: Z+2μm, Z+2.5μm, Z+3μm, ..., Z+9.5μm, Z+10μm. The focusing mechanism moves to each stepping position sequentially at a slow speed of 2μm / s, pausing at each position for 300ms for precise positioning. Subsequently, the CCD camera acquires a 1024×1024 pixel stepped fluorescence image, with the exposure time extended to 80ms and the gain value increased by 3.0 times to obtain higher image quality and signal-to-noise ratio. At the Z+6.5μm position, the acquired image clearly shows the fine internal structure of the cell nucleus, with uniform fluorescence intensity distribution and sharp edges. A second sharpness analysis was performed on 17 step-through fluorescence images. A more refined Sobel edge detection operator was used to calculate the image gradient in both the X and Y directions, and the average gradient amplitude was then used as the step-through sharpness value. The sharpness value sequence corresponding to each step position is: 426, 438, 451, 467, 485, 503, 521, 547, 563, 571, 568, 552, 534, 515, 496, 478, 461. Arranging the step-through sharpness values ​​from largest to smallest, the highest value of 571, corresponding to the Z+8.5μm step-through position, was determined as the precise focal plane level of the target scanning field of view unit. After determining the precise focal plane level, the focusing mechanism precisely stopped at the Z+8.5μm position, with positional stability controlled within ±0.05μm. The fluorescence microscope was then switched from a 10x low-magnification mode to a 40x high-magnification mode. The objective turret rotated 90 degrees under the drive of a stepper motor to complete the objective switch, which took 1.2 seconds. After the switch, the optical path configuration was automatically adjusted, reducing the excitation power from 30mW to 15mW to avoid fluorescence bleaching. The filter combination was switched from a broadband filter for low-magnification to a narrowband filter for high-magnification, with a center wavelength of 520nm and a full width at half maximum (FWHM) of 15nm.

[0030] In the above embodiment, in 40x high-magnification mode, the single field-of-view coverage is adjusted to 0.328mm × 0.328mm, with a corresponding physical resolution of 0.16μm / pixel, meeting the system's technical requirement of ≤0.25μm / pixel. The cooled CCD camera operates at -15°C to reduce dark current noise and is set to a full-resolution acquisition mode of 2048 × 2048 pixels. At the precise focal plane level Z+8.5μm, the CCD camera performs the final formal image acquisition, with an exposure time set to 120ms and a gain adjusted to 2.8 times to ensure an image signal-to-noise ratio greater than 35dB. The acquired single-field fluorescence image shows that the internal structure of Hep-2 cell nuclei is clearly layered, the cytoplasmic background signal is uniform, and the fluorescence intensity dynamic range fully utilizes the 0-4095 grayscale levels of the CCD camera's 12-bit A / D conversion. The image files are saved in 16-bit lossless TIFF format, including complete acquisition parameter metadata: objective magnification 40×, numerical aperture 0.75, exposure time 120ms, gain 2.8×, focal plane position Z+8.5μm, acquisition timestamp, etc.

[0031] In an optional embodiment, the multi-view fluorescence images are stitched together to generate a fluorescence panoramic image. Specifically, this includes: extracting a single-view fluorescence image with the first spatial location number from the multi-view fluorescence images as the first image to be stitched; extracting a single-view fluorescence image spatially adjacent to the first image to be stitched as the second image to be stitched; using the first and second images to be stitched as loop variables, performing the following image stitching operation until a fluorescence panoramic image is generated: extracting first feature points from the first image to be stitched to obtain a first feature point set; and extracting second feature points from the second image to be stitched to obtain a second feature point set; matching the first and second feature point sets to obtain multiple feature point pairs; performing reliability analysis on the multiple feature point pairs to obtain a reliable feature point pair set; determining the spatial transformation matrix between the first and second images to be stitched based on the reliable feature point pair set; and using the spatial transformation matrix to stitch the second image to be stitched... The images to be stitched undergo coordinate transformation to transform the second image to be stitched into the coordinate system of the first image to be stitched, resulting in a transformed image to be stitched. The overlapping area between the first image to be stitched and the transformed image to be stitched is determined, and image fusion is performed on the overlapping area to obtain a fused overlapping area. The non-overlapping areas of the first image to be stitched, the fused overlapping area, and the non-overlapping areas of the transformed image to be stitched are stitched together to obtain an intermediate stitched image. It is determined whether there are any incompletely stitched single-field fluorescence images in the multi-field fluorescence images. If it is determined that there are incompletely stitched single-field fluorescence images in the multi-field fluorescence images, the intermediate stitched image is updated to the first image to be stitched, and the single-field fluorescence images adjacent to the intermediate stitched image and not yet stitched, extracted from the multi-field fluorescence images according to their spatial location numbers, are updated to the second image to be stitched. If it is determined that there are no incompletely stitched single-field fluorescence images in the multi-field fluorescence images, the intermediate stitched image is determined as a fluorescence panoramic image.

[0032] Here, the first digit number represents the first identifier in the spatial location number, including but not limited to number 001, location A1, coordinates (0,0), etc.; the first image to be stitched refers to the starting image used as the stitching reference, including but not limited to the top left image, the reference image, the origin image, etc.; the second image to be stitched represents the image adjacent to the first image to be stitched and prepared for stitching, including but not limited to the right adjacent image, the bottom adjacent image, the diagonal adjacent image, etc.; image stitching operation refers to the process of combining two or more images into a complete image, including but not limited to feature point stitching, pixel-level stitching, geometric transformation stitching, etc.; the first feature point extraction is used to represent the extraction of key features from the first image to be stitched. The process of feature point extraction; the first feature point set refers to the set of all feature points extracted from the first image to be stitched, including but not limited to corner point sets, edge point sets, texture feature point sets, etc.; the second feature point extraction is used to represent the process of extracting key feature points from the second image to be stitched; the second feature point set refers to the set of all feature points extracted from the second image to be stitched; feature point matching is used to represent the process of finding the correspondence between two feature point sets, including but not limited to nearest neighbor matching, bidirectional matching, ratio test matching, etc.; feature point pairs refer to the combination of feature points in two feature point sets that are determined to have a correspondence, including but not limited to matching point pairs, corresponding point pairs, and same-name point pairs, etc.; reliability analysis is used to represent the evaluation of the matching quality of feature point pairs. The evaluation process; a reliable feature point pair set refers to a high-quality feature point pair set retained after reliability screening, including but not limited to inlier sets, stable matching point sets, and high-confidence point pair sets; a spatial transformation matrix is ​​a mathematical matrix used to represent the geometric transformation relationship between two images, including but not limited to affine transformation matrices, perspective transformation matrices, and rigid transformation matrices; coordinate transformation processing refers to the process of geometrically correcting images using transformation matrices, including but not limited to affine transformation processing, perspective correction processing, and rigid body transformation processing; the image to be stitched is used to represent the second image to be stitched after coordinate transformation, including but not limited to geometrically corrected images, registered images, and transformed images; overlapping regions refer to the regions where two images overlap during stitching. In overlapping common areas, including but not limited to edge overlap areas, corner overlap areas, and center overlap areas; image fusion is used to represent the processing of reasonably combining pixel information in overlapping areas, including but not limited to weighted average fusion, multi-band fusion, and Poisson fusion; fused overlapping area refers to the overlapping area result after fusion processing, including but not limited to seamless stitching area, smooth transition area, and optimized fusion area; non-overlapping area is used to represent unique areas in an image that do not overlap with other images, including but not limited to unique boundary areas, unique corner areas, and unique center areas; stitching combination refers to the process of merging the fused overlapping area with the non-overlapping areas of each image, including but not limited to region merging, pixel recombination, and image integration;Intermediate stitched images are used to represent intermediate results during the stitching process, including but not limited to partially stitched images, temporary composite images, and process result images.

[0033] In the above embodiment, taking a multi-field fluorescence image composed of 160 single-field fluorescence images (8 rows × 20 columns) as an example, the image stitching program first extracts the first single-field fluorescence image with spatial location number (1,1) from the data storage array as the first image to be stitched. This image has a size of 2048×2048 pixels, corresponding to the upper left corner of the detection area, and is a 12-bit grayscale image with pixel values ​​ranging from 0 to 4095. Based on the adjacency relationship of the serpentine scanning path, the single-field fluorescence image with spatial location number (1,2) is extracted as the second image to be stitched. This image is adjacent to the first image to be stitched in the X-axis direction and has a 15% preset overlap area. The first loop of the stitching operation begins to execute, and SIFT feature point extraction is performed on the first image to be stitched. First, a Gaussian pyramid is constructed, using an initial scale parameter of σ=1.6, and the scale factor is increased by multiples of √2 at each of the four octave levels. Each octave contains five scale layers, and extreme points are detected as keypoint candidates using the difference of Gaussians operator. After locating 312 initial keypoints in scale space, further precise localization and low-contrast point removal were performed using a contrast threshold of 0.03 and an edge threshold of 10, ultimately retaining 267 stable keypoints. A 128-dimensional SIFT descriptor vector was calculated for each keypoint, forming the first feature point set, containing complete descriptive information for all 267 feature points. Simultaneously, the same SIFT feature point extraction process was performed on the second image to be stitched. Since this image has similar cell distribution and fluorescence characteristics to the first image, 295 initial keypoints were detected. After precise localization and filtering, 251 valid feature points were retained, forming the second feature point set. Each feature point also contains position coordinates (x, y), scale parameter σ, principal direction angle θ, and a 128-dimensional descriptor vector.

[0034] In the above embodiment, the feature point matching stage employs a nearest neighbor distance ratio test. For each feature point in the first feature point set, the two candidate matching points with the closest Euclidean distance in the second feature point set are searched. The ratio of the nearest distance to the second nearest distance is calculated, and a match is considered valid when the ratio is less than a threshold of 0.7. This method identifies 89 initial feature point pairs in a 267×251 matching combination. Each feature point pair contains the coordinate information of two feature points: a point (x1, y1) in the first image and the corresponding point (x2, y2) in the second image. Reliability analysis uses the RANSAC random consensus algorithm to screen the 89 feature point pairs. The algorithm sets an affine transformation model, requiring at least 3 pairs of matching points to determine the transformation parameters. 1000 random sampling iterations are performed, each time randomly selecting 3 pairs of feature points to calculate the affine transformation matrix, and then the reprojection error of the remaining feature point pairs under this transformation is statistically analyzed. A 2-pixel inlier distance threshold is set, and the number of inliers supporting this transformation model is counted. After iterative optimization, the transformation model with the highest support was determined to contain 62 interior points, forming a set of reliable feature point pairs. The optimal 2×3 affine transformation matrix was then calculated based on these 62 reliable feature point pairs. [a,b,tx] The transformation matrix [c,d,ty] represents the rotation and scaling parameters: a=0.9987, b=-0.0156, c=0.0152, d=0.9989, and the translation parameters: tx=1743.2, ty=12.8. This transformation matrix reflects the small rotation (approximately 0.9 degrees) and the main horizontal translation (approximately 1743 pixels) between the two images. In the coordinate transformation stage, bilinear interpolation is used to perform an affine transformation on each pixel of the second image to be stitched. For each pixel position (x',y') in the output image, the corresponding position (x,y) in the original image is calculated through an inverse transformation, and then the new pixel value is determined using the weighted average of the four neighboring pixels. After transformation, the resulting image to be stitched is expanded to 3791×2060 pixels to accommodate the transformed complete image content.

[0035] In the above embodiment, the overlapping region is determined by comparing pixel coordinates. The effective range of the first image to be stitched is (0,0) to (2048,2048), and the effective range of the image to be stitched and transformed is (1743,13) to (3791,2061). The intersection of the two is (1743,13) to (2048,2048), forming an overlapping region of 305×2035 pixels. This region contains the common cellular structures and fluorescence signals of the two images, requiring precise fusion processing. Image fusion employs a multi-band fusion algorithm. First, the two images within the overlapping region are decomposed into four frequency levels using Laplacian pyramid decomposition. The low-frequency level reflects the overall brightness distribution of the image, while the high-frequency level contains detailed texture information. At each frequency level, a fusion weight is calculated based on the distance from the pixel to the boundary of the overlapping region, with a weight of 0.5 at the boundary, gradually transitioning to 1.0 or 0.0 towards the inwards. The fusion weighting function uses the sigmoid function to achieve a smooth transition: w(x) = 1 / (1 + exp(-k*(x-x0))), where k = 0.1 controls the transition steepness, and x0 is the center line position of the overlapping region. After fusion, a fused overlapping region of 305×2035 pixels is obtained through Laplacian pyramid reconstruction, eliminating stitching gaps and brightness differences. In the stitching combination stage, three regions are integrated: the non-overlapping region (0,0) to (1743,2048) of the first image to be stitched, the fused overlapping region (1743,13) to (2048,2048), and the non-overlapping region (2048,13) to (3791,2061) of the image to be stitched and transformed. Through pixel duplication and region merging operations, an intermediate stitched image of size 3791×2061 pixels is generated, successfully completing the first stitching operation. Checking the multi-field fluorescence image array, it was found that there are still 158 single-field fluorescence images that have not been stitched. According to the serpentine scanning sequence, the next image to be stitched is the single-field fluorescence image with spatial location number (1,3). The current intermediate image to be stitched is updated to the new first image to be stitched, and the image at position (1,3) is set as the new second image to be stitched, thus starting the second round of stitching.

[0036] In the above embodiment, in the second round of stitching, since the first image to be stitched has been expanded to a large size of 3791×2061 pixels, SIFT feature point extraction is performed across the entire image. To improve computational efficiency, a block processing strategy is adopted, dividing the large image into overlapping sub-blocks of 1024×1024 pixels for feature extraction, and then merging the feature point results of all sub-blocks. The new first image to be stitched detected 459 valid feature points, and the second image to be stitched still detected 248 feature points. The matching and reliability analysis process is the same as in the first round, but due to the increased image size, the matching search range is also expanded accordingly. The final number of reliable feature point pairs is 58, and the calculated affine transformation matrix shows that it is mainly a horizontal translation, with a translation amount of approximately 1740 pixels, which is basically consistent with the expected overlap rate. After a similar coordinate transformation, overlapping area fusion, and stitching combination process, a new intermediate stitched image is generated, with its size further expanded to 5534×2061 pixels. The process continues, with each stitch fusing the current intermediate stitched image with the next single-view image, gradually constructing a complete panoramic image. After completing the horizontal stitching of the first row of 20 images, the second row begins processing. Based on the serpentine scanning path, the 21st image, spatially numbered (2,20), is located directly below the last image in the first row. At this point, the stitching operation changes from horizontal to vertical, with the affine transformation matrix primarily representing a vertical translation of approximately 1740 pixels. The entire stitching process repeats 160 times, each cycle performing complete feature extraction, matching, reliability analysis, transformation calculation, coordinate transformation, overlap fusion, and region combination operations. Finally, when no incompletely stitched single-view fluorescence images are found, the current intermediate stitched image is determined as the fluorescence panoramic image. The final generated fluorescence panoramic image has a size of 34880×13728 pixels and a physical coverage area of ​​21.2mm×8.6mm, achieving complete coverage of the entire Hep-2 cell detection area.

[0037] In an optional embodiment, feature extraction is performed on the fluorescence panoramic image to determine the fluorescence karyotype category and target fluorescence intensity value. Specifically, this includes: performing background separation processing on the fluorescence panoramic image to obtain a binary segmentation image; extracting connected components from the binary segmentation image to obtain multiple nuclear connected components; performing comprehensive feature extraction on each of the multiple nuclear connected components to obtain a comprehensive feature vector for each nuclear connected component; inputting the comprehensive feature vector into a preset karyotype classification model to obtain the karyotype category of each nuclear connected component and determining the number of nuclear connected components corresponding to each karyotype category; and based on the cell nucleus... The number of connected regions determines the dominant karyotype category, which is then used as the fluorescent karyotype category. Connected nuclear regions with the same karyotype as the dominant karyotype category are selected from multiple nuclear connected regions to obtain the target karyotype nuclear connected region set. Nuclear fluorescence images of each nuclear connected region in the target karyotype nuclear connected region set are extracted from the fluorescence panoramic image. Fluorescence intensity is measured for each nuclear fluorescence image to obtain the mononuclear fluorescence intensity value. Central tendency analysis is performed on the mononuclear fluorescence intensity values ​​of all nuclear fluorescence images in the target karyotype nuclear connected region set to obtain the target fluorescence intensity value.

[0038] Background separation processing refers to image processing that separates the target object from the background region in an image, including but not limited to threshold segmentation, region growing segmentation, and clustering segmentation; binary segmentation image refers to the image containing only black and white pixel values ​​after background separation, including but not limited to images with black background and black target pixels, images with black target pixels and white background, and masked binary images; connected component extraction is used to represent the process of identifying and extracting interconnected pixel regions from a binary image, including but not limited to 8-connected component extraction, 4-connected component extraction, and morphological connected component extraction; cell nucleus connected component refers to the connected pixel region corresponding to a single cell nucleus in a binary image, including but not limited to complete cell nucleus regions, partial cell nucleus regions, and overlapping cell nucleus regions; comprehensive feature extraction is used to represent the comprehensive process of extracting multiple types of feature information from the target object, including but not limited to morphological feature extraction, texture feature extraction, and spectral feature extraction; comprehensive feature vector refers to a numerical vector containing multiple feature information; pre-set karyotype classification model is used to represent a pre-trained machine learning model for cell karyotype classification; karyotype Category refers to the classification type of cell nuclei, including but not limited to homogeneous, speckled, and nucleolar types; dominant karyotype category indicates the most numerous or predominant karyotype type among all detected karyotypes, including but not limited to the karyotype with the largest proportion, statistically dominant karyotype, and dominant karyotype; target karyotype nuclear connected region set refers to the set of all nuclear connected regions that are the same as the dominant karyotype category, including but not limited to the set of similar karyotype regions, the dominant type set, and the target category set; nuclear fluorescence image refers to the fluorescence image corresponding to a single cell nucleus extracted from the panoramic image, including but not limited to single-nucleus fluorescence slices, nuclear ROI images, and local fluorescence images; fluorescence intensity measurement refers to the process of quantifying the brightness of fluorescence images, including but not limited to average brightness measurement, peak brightness measurement, and integrated brightness measurement; single-nucleus fluorescence intensity value refers to the fluorescence intensity value of a single cell nucleus, including but not limited to average fluorescence intensity, maximum fluorescence intensity, and integrated fluorescence intensity; central tendency analysis refers to the method of statistically analyzing the central position of a set of data, including but not limited to the calculation of the mean, median, and mode.

[0039] In the above embodiment, taking a fluorescence panoramic image with a size of 34880×13728 pixels as an example, the image is first subjected to background separation processing. Given that the fluorescence image has a distinct bimodal distribution characteristic, namely high fluorescence intensity in the cell nucleus region and low fluorescence intensity in the background region, an improved Otsu adaptive threshold segmentation algorithm is adopted. Gray-level histogram statistics of the panoramic image reveal that the image gray-level values ​​are distributed within a 12-bit dynamic range of 0-4095, with background pixels mainly concentrated in the 0-150 gray-level value range and cell nucleus pixels mainly distributed in the 200-3800 gray-level value range. The optimal segmentation threshold is automatically determined by maximizing the inter-class variance. The inter-class variance σ²(t) = ω0(t) × ω1(t) × [μ0(t) - μ1(t)]² corresponding to each possible threshold t (from 1 to 4094) is calculated, where ω0 and ω1 are the weight ratios of the two classes, and μ0 and μ1 are the average gray-level values ​​of the two classes, respectively. After calculation, the optimal threshold corresponding to the maximum inter-class variance is T = 185. The panoramic image is binarized: pixels with a grayscale value ≥ 185 are set to 255 (white, representing the cell nucleus), and pixels with a grayscale value < 185 are set to 0 (black, representing the background), generating a 34880×13728 pixel binary segmentation image. The connected component extraction stage uses an 8-connected-component algorithm to process the binary segmentation image. The algorithm scans pixel by pixel starting from the top left corner. When a foreground pixel with a value of 255 is encountered, the connectivity of its 8 neighboring pixels is checked. For each newly discovered connected region, a unique label number is assigned, and the pixel coordinates of that region are recorded. To improve processing efficiency, a two-pass scanning algorithm is used: the first pass establishes initial labels and an equivalence relation table; the second pass merges identical connected regions based on the equivalence relations. After processing, 2847 initial connected components are detected. Considering the influence of noise points and cell debris, an area threshold is set for filtering: connected components with an area less than 80 pixels are judged as noise and removed; connected components with an area greater than 8000 pixels are judged as cell clusters and morphological segmentation is performed. After screening and segmentation, 2156 valid cell nucleus connected domains were retained. Each connected domain contains basic geometric parameters such as centroid coordinates, bounding box coordinates, area, and perimeter.

[0040] In the above embodiments, the comprehensive feature extraction stage performs multi-dimensional feature analysis on the connected components of each cell nucleus. First, based on the bounding box coordinates of the connected components, the corresponding rectangular regions are extracted from the fluorescence panoramic image as the regions of interest (ROI) for each individual cell nucleus. To ensure the integrity of feature extraction, the bounding box is extended outward by 10 pixels as a buffer. Morphological feature extraction includes eight dimensions: area A = total number of connected component pixels, perimeter P = total length of boundary pixels, circularity C = 4πA / P², aspect ratio R = major axis / minor axis, compactness M = P² / A, convexity H = A / area of ​​convex hull, eccentricity E = √(1 - (minor axis / major axis)²), and firmness S = A / area of ​​bounding box. Taking the connected nuclear region numbered 1247 as an example, its morphological feature values ​​are: A=1856, P=162.4, C=0.884, R=1.247, M=14.2, H=0.923, E=0.345, S=0.756. Texture feature extraction is based on the Gray-Level Co-occurrence Matrix (GLCM) algorithm, calculating co-occurrence matrices at distances of 1 pixel in the four directions of 0°, 45°, 90°, and 135°. For each direction of GLCM, four Haralick texture features are calculated: Contrast = Σᵢⱼ(ij)²P(i,j), Correlation = Σᵢⱼ(i-μᵢ)(j-μⱼ)P(i,j) / σᵢσⱼ, Energy = ΣᵢⱼP(i,j)², and Entropy = -ΣᵢⱼP(i,j)log2P(i,j). The average value of the four directions is taken as the texture feature of the cell nucleus, resulting in a 4-dimensional texture feature vector. Fluorescence intensity features are extracted and analyzed to determine the distribution characteristics of fluorescence signals within the cell nucleus. Six intensity statistical features are calculated: Mean = Σpᵢ / n, Standard Deviation Std = √(Σ(pᵢ-Mean)² / (n-1)), Skewness Skew = E[(x-μ)³] / σ³, and Kurt = E[(x-μ)]. 4 ] / σ 4 -3. Minimum value (Min) and maximum value (Max). Where pᵢ is the pixel grayscale value, and n is the total number of pixels. Spatial distribution feature extraction evaluates the spatial organization pattern of fluorescence signals within the cell nucleus. Four spatial features are calculated: centroid offset = distance from centroid to geometric center / equivalent radius; non-uniformity = maximum radial intensity / minimum radial intensity; aggregation = degree of aggregation of high-intensity pixels; symmetry = correlation coefficient of intensity distribution in the upper and lower halves.

[0041] In the above embodiment, morphological (8-dimensional), texture (4-dimensional), intensity (6-dimensional), and spatial distribution (4-dimensional) features are standardized and then concatenated to form a 22-dimensional comprehensive feature vector for each cell nucleus's connected domain. Standardization uses the Z-score method: x'=(x-μ) / σ, where μ is the mean of the feature across all cell nuclei, and σ is the standard deviation. During the karyotype classification stage, 2156 22-dimensional comprehensive feature vectors are input into a pre-defined random forest karyotype classification model. This model contains 500 decision trees, and each tree randomly selects √22≈5 features for node splitting during training, with a maximum tree depth of 15. The model outputs the probability distribution of four karyotypes: homogeneous, speckled, nucleolar, and cytoplasmic. The classification results show: homogeneous 1689 (78.3%), speckled 312 (14.5%), nucleolar 128 (5.9%), and cytoplasmic 27 (1.3%). Since homogeneous nuclei were the most numerous, accounting for 78.3% of the total, they were determined as the dominant karyotype category and used as the final fluorescent karyotype category. The target karyotype nucleus connected component set was screened, extracting 1689 homogeneous nucleus connected components from 2156. Corresponding nucleus fluorescence images were extracted from the fluorescence panoramic image based on the bounding box coordinates of each connected component. During extraction, the original pixel values ​​and spatial resolution were maintained; the size of each nucleus fluorescence image varied according to the size of the connected component, with an average size of approximately 64×64 pixels. In the fluorescence intensity measurement stage, each of the 1689 nucleus fluorescence images underwent quantitative analysis. For each nucleus fluorescence image, intensity was calculated only for pixels within the connected component, excluding background pixels within the bounding box. Intensity measurement used the average fluorescence intensity method: Intensity = Σ(pixel_value × mask) / Σ(mask), where mask is the binary mask of the connected component. After measurement, the single-nucleus fluorescence intensity values ​​of the 1689 homogeneous nuclei were distributed in the grayscale range of 156-892. The intensity distribution exhibits an approximately normal distribution, with peak values ​​concentrated in the 280-320 grayscale range. Central tendency analysis employed the median method to calculate the target fluorescence intensity value to mitigate the impact of outliers. The 1689 mononuclear fluorescence intensity values ​​were arranged in ascending order; the median position was (1689+1) / 2=845, corresponding to a fluorescence intensity value of 298 grayscale. The mean (302 grayscale) and mode (286 grayscale) were also calculated as references, ultimately determining the target fluorescence intensity value to be 298.

[0042] In an optional embodiment, comprehensive feature extraction is performed on each of the multiple nuclear connected domains to obtain a comprehensive feature vector for each nuclear connected domain. Specifically, this includes: sequentially traversing each nuclear connected domain included in the multiple nuclear connected domains, and performing the following operations on the currently traversed nuclear connected domain: obtaining the position coordinates of the target nuclear connected domain, and extracting the target nuclear fluorescence image corresponding to the target nuclear connected domain from the fluorescence panoramic image based on the position coordinates, wherein the target nuclear connected domain is the currently traversed nuclear connected domain; extracting morphological features from the target nuclear fluorescence image to obtain a morphological feature vector; performing fluorescence intensity distribution analysis on the target nuclear fluorescence image to obtain a fluorescence intensity feature vector; performing fluorescence spatial distribution analysis on the target nuclear fluorescence image to obtain a fluorescence spatial feature vector; extracting texture features from the target nuclear fluorescence image to obtain a texture feature vector; and concatenating and combining the morphological feature vector, fluorescence intensity feature vector, fluorescence spatial feature vector, and texture feature vector to obtain the target comprehensive feature vector of the target nuclear connected domain.

[0043] Among them, position coordinates represent the spatial location information of the cell nucleus connected region in the image, including but not limited to centroid coordinates, bounding box coordinates, and contour coordinates; target cell nucleus fluorescence image refers to the fluorescence image corresponding to a single cell nucleus cropped from the panoramic image based on position coordinates, including but not limited to square ROI images, rectangular ROI images, and irregular contour images; morphological feature extraction is used to represent the process of extracting geometric shape-related features from the image, including but not limited to area and perimeter extraction, roundness extraction, and aspect ratio extraction; morphological feature vector refers to a numerical vector containing morphological feature information, including but not limited to geometric parameter vectors, shape description vectors, and contour feature vectors; fluorescence intensity distribution analysis is used to represent the process of statistically analyzing the spatial distribution of fluorescence intensity in the image, including but not limited to intensity histogram analysis, intensity gradient analysis, and intensity clustering analysis; fluorescence intensity feature vector refers to a numerical vector describing the characteristics of fluorescence intensity distribution; fluorescence spatial distribution analysis uses... The term "fluorescence" refers to the process of analyzing the spatial distribution pattern of fluorescence signals, including but not limited to spatial autocorrelation analysis, clustering pattern analysis, and distribution uniformity analysis. Fluorescence spatial feature vectors are numerical vectors describing the spatial distribution characteristics of fluorescence, including but not limited to spatial statistical vectors, distribution pattern vectors, and spatial relationship vectors. Texture feature extraction represents the process of extracting texture-related information from an image, including but not limited to gray-level co-occurrence matrix feature extraction, local binary pattern extraction, and wavelet texture feature extraction. Texture feature vectors are numerical vectors containing image texture information, including but not limited to texture statistical vectors, texture description vectors, and texture parameter vectors. Concatenation and combination represent the operation of merging multiple vectors into a single comprehensive vector, including but not limited to vector concatenation, feature fusion, and dimension merging. Target comprehensive feature vectors are complete vectors containing all feature information of the target cell nucleus, including but not limited to multidimensional feature vectors, comprehensive description vectors, and complete feature sets.

[0044] In the above embodiment, taking 2156 connected nuclear domains as an example, each connected nuclear domain is traversed sequentially according to its label number. When the target nuclear domain numbered 1247 is reached, its position coordinates are first obtained from the connected domain attribute table. The centroid coordinates calculated by the connected domain analysis algorithm are (15842, 7364), and the bounding box coordinates are the upper left corner (15798, 7336) and the lower right corner (15886, 7392), forming a rectangular area of ​​88×56 pixels. Based on the bounding box coordinates, the corresponding area is extracted from the fluorescence panoramic image with a size of 34880×13728 pixels. To ensure the integrity of feature extraction and the complete preservation of edge information, the original bounding box is extended by 12 pixels in each of the four directions as a safety buffer. Finally, the coordinates of the extracted area are adjusted to the upper left corner (15786, 7324) and the lower right corner (15898, 7404), resulting in a target nuclear fluorescence image of 112×80 pixels. The image retains the original fluorescence panoramic image's 12-bit grayscale depth and 0.16 μm / pixel physical resolution. In the morphological feature extraction stage, geometric shape analysis was performed on the target cell nucleus fluorescence image. The extracted fluorescence image was binarized using the same threshold T=185 as the panoramic image to obtain the precise contour of the cell nucleus region. Subsequently, eight core morphological feature parameters were calculated: Area calculation was achieved by counting the total number of pixels with a value of 255 in the binary image; the area of ​​the connected region of this cell nucleus was A=1856 pixels, corresponding to a physical area of ​​0.0475 mm². Perimeter calculation used the Freeman chain code algorithm to trace the boundaries of the connected region, counting the total length of boundary pixels to obtain a perimeter P=162.4 pixels. The roundness calculation formula is C=4πA / P², reflecting the degree to which the cell nucleus shape is close to round; the calculation result C=0.884, indicating that the cell nucleus shape is relatively regular.

[0045] In the above embodiment, the aspect ratio is determined by principal component analysis to determine the directions of the principal and secondary axes of the cell nucleus. The covariance matrix of the connected component pixel coordinates is calculated, and the eigenvalues ​​λ1 and λ2 (λ1 ≥ λ2) are solved. The major axis length is 2√λ1 = 54.2 pixels, the minor axis length is 2√λ2 = 43.5 pixels, and the aspect ratio R = 1.247. The eccentricity is calculated as E = √(1 - (λ2 / λ1)) = 0.345, indicating that the cell nucleus shape is slightly elliptical. The compactness M = P² / (4πA) = 1.42, measuring the tightness of the shape. Convexity calculation requires first generating the convex hull of the connected components. Using the Graham scan algorithm, the convex hull area is obtained as 2012 pixels, and the convexity H = A / convex hull area = 0.923. The solidity S = A / bounding box area = 1856 / 4928 = 0.377. The morphological feature vector is represented as [1856, 162.4, 0.884, 1.247, 1.42, 0.923, 0.345, 0.377]. Fluorescence intensity distribution analysis calculates the statistical characteristics of pixel intensity within the target cell nuclear fluorescence image. Only effective pixels within connected components are analyzed, excluding background pixels in the buffer zone. A binary mask is applied to filter out 1856 effective pixels, whose fluorescence intensity values ​​are distributed within the grayscale range of 198-456. Six intensity distribution features are calculated: the average intensity Mean = ∑pᵢ / n = 298.7, where pᵢ is the pixel grayscale value and n is the total number of pixels. The standard deviation Std = √(∑(pᵢ-Mean)² / (n-1)) = 34.2, reflecting the dispersion of the intensity distribution. The skewness Skew = E[(x-μ)³] / σ³ = 0.187, indicating a slight shift in the intensity distribution towards higher values. Kurt = E[(x-μ)] 4 ] / σ 4 -3 = -0.523 indicates a more gradual distribution than a normal distribution. Maximum intensity Max = 456, minimum intensity Min = 198. The fluorescence intensity feature vector is represented as [298.7, 34.2, 0.187, -0.523, 456, 198]. Fluorescence spatial distribution analysis assesses the spatial organization pattern of the fluorescence signal within the cell nucleus. The offset between the centroid coordinates of the connected region (relative to the local coordinates (56, 40) of the extracted image) and the geometric center coordinates (56, 40) is calculated. The centroid offset is calculated as offset distance / equivalent radius, where the equivalent radius r = √(A / π) = 24.3 pixels, the offset distance is 0.8 pixels, and the centroid offset = 0.8 / 24.3 = 0.033.

[0046] In the above embodiments, the non-uniformity feature was calculated using radial intensity analysis. With the centroid as the center, the average intensity along 24 radial lines from the center to the boundary was calculated at 0.5 pixel intervals, resulting in a radial intensity distribution sequence. The maximum radial intensity was 328.5, the minimum radial intensity was 267.2, and the non-uniformity was 328.5 / 267.2 = 1.23. Aggregation was assessed by calculating the spatial clustering of high-intensity pixels (grayscale value > 350). Using nearest neighbor distance analysis, the average nearest neighbor distance between 392 high-intensity pixels was calculated to be 2.34 pixels. Compared to the expected distance of 2.89 pixels for random distribution, the aggregation was 2.89 / 2.34 = 1.24, indicating a slight clustering pattern among the high-intensity pixels. Symmetry was determined by dividing the cell nucleus into upper and lower parts along the principal axis and calculating the correlation coefficient between the intensity distributions of the two parts. The average intensity of the upper part was 302.1, and the average intensity of the lower part was 295.3, with a correlation coefficient of 0.847, indicating good symmetry in the nuclear fluorescence distribution. The fluorescence spatial feature vector is represented as [0.033, 1.23, 1.24, 0.847]. Texture feature extraction employs the Gray-Level Co-occurrence Matrix (GLCM) method to analyze the texture characteristics of the image. Co-occurrence matrices with a distance of 1 pixel are calculated in four directions: 0°, 45°, 90°, and 135°, with a matrix size of 256×256 to accommodate 8-bit quantization of gray levels. The target cell nuclear fluorescence image is first subjected to gray-level normalization, linearly mapping the 12-bit image to the 0-255 range. For each direction of GLCM, four Haralick texture features are calculated: contrast Con=∑ᵢ∑ⱼ(ij)²P(i,j), which measures the degree of local variation in the image; correlation Cor=∑ᵢ∑ⱼ(i-μᵢ)(j-μⱼ)P(i,j) / (σᵢσⱼ), which measures the linear correlation of pixel pairs; energy Ene=∑ᵢ∑ⱼP(i,j)², which measures the uniformity of the texture; and entropy Ent=-∑ᵢ∑ⱼP(i,j)log2P(i,j), which measures the complexity of the texture.

[0047] In the above embodiment, the feature calculation results for the four directions are: 0° direction [Con=12.34, Cor=0.832, Ene=0.0345, Ent=6.78], 45° direction [Con=13.67, Cor=0.819, Ene=0.0327, Ent=6.89], 90° direction [Con=11.89, Cor=0.845, Ene=0.0362, Ent=6.65], and 135° direction [Con=12.98, Cor=0.828, Ene=0.0338, Ent=6.82]. The average value of the four directions is taken as the final texture feature: contrast=12.72, correlation=0.831, energy=0.0343, entropy=6.79. The texture feature vector is represented as [12.72, 0.831, 0.0343, 6.79]. In the assembly stage, four types of feature vectors are concatenated. First, each feature vector is standardized to eliminate dimensional differences between features: x'=(x-μ) / σ, where μ and σ are the mean and standard deviation of the feature across all 2156 cell nuclei, respectively. The standardized feature vectors are: morphological feature vector (8-dimensional), fluorescence intensity feature vector (6-dimensional), fluorescence spatial feature vector (4-dimensional), and texture feature vector (4-dimensional). Through concatenation, the components are arranged sequentially in the order of morphology → intensity → space → texture, forming a 22-dimensional comprehensive feature vector for the connected domain of the target cell nucleus, numbered 1247: [Standardized morphological features (8 components), standardized intensity features (6 components), standardized spatial features (4 components), standardized texture features (4 components)]. This comprehensive feature vector fully describes the nucleus's geometry, fluorescence intensity distribution, spatial organization pattern, and texture characteristics, providing a comprehensive feature information foundation for subsequent karyotype classification. The same feature extraction process is then applied to the remaining 2155 cell nucleus connected domains, ultimately generating a feature matrix of 2156 22-dimensional comprehensive feature vectors, which are then input into a preset karyotype classification model for intelligent recognition.

[0048] In an optional embodiment, the comprehensive feature vector is input into a preset karyotype classification model to obtain the karyotype category of each nuclear connected region and to determine the number of nuclear connected regions corresponding to each karyotype category. Specifically, this includes: inputting the target comprehensive feature vector into the preset karyotype classification model to obtain the karyotype category probability distribution output by the preset karyotype classification model after classifying and recognizing the target comprehensive feature vector; determining the category probability values ​​greater than a preset probability threshold from the karyotype category probability distribution; using the karyotype category corresponding to the category probability as the candidate karyotype category of the target nuclear connected region, and using the category probability value as the classification confidence score; comparing the classification confidence score with a preset confidence threshold to obtain the confidence score. The comparison results are as follows: If the classification confidence is greater than a preset confidence threshold, the candidate karyotype category is determined as the target karyotype category of the target cell nuclear connected region; if the classification confidence is less than or equal to the preset confidence threshold, the target cell nuclear connected region is marked as a low-confidence cell nucleus, and the candidate karyotype category is used as a temporary karyotype category of the target cell nuclear connected region. The candidate karyotype category and the probability distribution of the karyotype category are stored in the database to be reviewed; if the karyotype category of each cell nuclear connected region is determined, multiple cell nuclear connected regions are grouped according to each karyotype category to determine the number of cell nuclear connected regions corresponding to each karyotype category.

[0049] Here, the preset kernel type classification model refers to the machine learning model for kernel type recognition trained on a large amount of data; the kernel type category probability distribution refers to the probability value distribution corresponding to various kernel type categories output by the model; the preset probability threshold is used to represent the threshold parameter used to filter high-probability categories, including but not limited to 0.5 threshold, 0.7 threshold, 0.8 threshold, etc.; the category probability value refers to the probability value corresponding to a specific kernel type category, including but not limited to homogeneous probability value, speckled probability value, nucleolar probability value, etc.; the candidate kernel type category is used to represent the potential kernel type classification results with a probability higher than the threshold, including but not limited to high-probability candidate categories, preliminary classification results, categories to be confirmed, etc.; the classification confidence refers to the evaluation index of the model's reliability of the classification results, including but not limited to classification probability value, confidence score, credibility index, etc.; the preset confidence threshold is used to represent the confidence standard for judging whether the classification result is reliable, including but not limited to the 0.8 confidence threshold. Values, 0.9 confidence threshold, 0.95 confidence threshold, etc.; confidence comparison refers to the process of comparing the classification confidence with a preset threshold; confidence comparison results are used to represent the conclusion of the confidence comparison, including but not limited to high confidence results, low confidence results, boundary confidence results, etc.; target karyotype category refers to the final determined cell nucleus type classification result, including but not limited to confirmed karyotype, final classification, target type, etc.; low confidence cell nuclei are used to represent cell nucleus objects with insufficient classification confidence, including but not limited to difficult samples, uncertain samples, samples to be reviewed, etc.; temporary karyotype category refers to the temporary classification result under the condition of insufficient confidence, including but not limited to provisional category, preliminary category, candidate category, etc.; database to be reviewed is used to represent the database storing samples that need to be manually reviewed, including but not limited to difficult sample database, quality control database, manual review database, etc.; grouping refers to the process of classifying cell nuclei according to karyotype category.

[0050] In the above embodiment, taking the 22-dimensional comprehensive feature vector of the connected domain of the target cell nucleus (numbered 1247) as an example, this feature vector is input into a preset karyotype classification model. This classification model uses a random forest algorithm architecture, containing 500 decision trees, with a maximum depth of 15 layers per tree and a minimum number of samples per leaf node set to 5. During the training phase, the model uses a dataset containing 56,000 labeled samples, covering four karyotype categories: homogeneous, speckled, nucleolar, and cytoplasmic, with 23,800, 16,500, 10,200, and 5,500 samples for each category, respectively. After the feature vector enters the model, it undergoes data preprocessing through the input layer. The dimensionality of the feature vector is checked, confirming that it contains 8-dimensional morphological features, 6-dimensional fluorescence intensity features, 4-dimensional fluorescence spatial features, and 4-dimensional texture features, totaling 22 components, all of which are valid values. Subsequently, feature scaling is performed, using the mean and standard deviation parameters saved during training to standardize the input features: x_scaled=(x-mean_train) / std_train. 500 decision trees are used in parallel to process the standardized feature vectors. During training, each tree randomly selects approximately 5 features (√2² ≈ 5) for node splitting to avoid overfitting. Taking the 127th decision tree as an example, its root node first splits based on the contrast value in the texture features, with a threshold of 0.345 (after standardization). Since the standardized contrast value of the target cell nucleus is 0.267, which is less than the threshold, the branch enters the left subtree. The left subtree node splits based on the roundness in the morphological features, with a threshold of 0.756. The standardized roundness value of the target cell nucleus is 0.891, which is greater than the threshold, so the branch enters the right subtree. After 15 layers of recursive splitting, the 127th tree finally reaches a leaf node, which contains 158 cell nuclei from the training samples: 142 homogeneous, 12 speckled, 3 nucleolar, and 1 cytoplasmic. Therefore, the class probabilities output by this decision tree are: homogeneous type 0.899, speckled type 0.076, nucleolar type 0.019, and cytoplasmic type 0.006.

[0051] In the above embodiment, after the 500 decision trees completed their inference, the outputs of all trees were integrated and voted on. Statistical results showed that 476 trees predicted a homogeneous karyotype, 19 trees predicted a speckled karyotype, 4 trees predicted a nucleolar karyotype, and 1 tree predicted a cytoplasmic karyotype. The final karyotype probability distribution was calculated using a soft voting mechanism: homogeneous P_homogeneous = 0.932, speckled P_speckled = 0.045, nucleolar P_nucleolar = 0.018, and cytoplasmic P_cytoplasmic = 0.005. The total probability was 1.000, meeting the normalization requirement. Class probability values ​​greater than a preset probability threshold of 0.6 were selected from the karyotype probability distribution. After comparison, the homogeneous probability of 0.932 was greater than the threshold and was determined as a valid candidate; the speckled probability of 0.045, the nucleolar probability of 0.018, and the cytoplasmic probability of 0.005 were all less than the threshold and were excluded. Therefore, homogeneous karyotype was identified as the candidate karyotype category for the target nuclear connected region, with a category probability value of 0.932 used as the classification confidence level. In the confidence comparison phase, the classification confidence level of 0.932 was compared with a preset confidence threshold of 0.85. The comparison showed that 0.932 > 0.85, indicating a high confidence level. Based on this high confidence level, the candidate karyotype category of homogeneous karyotype was determined as the target karyotype category for the target nuclear connected region. This classification result has sufficient reliability and does not require further review. The classification information for connected region number 1247 was updated in the nuclear attribute database: the karyotype category was marked as homogeneous, the classification confidence level was recorded as 0.932, the classification timestamp was recorded as the current system time, and the classification status was marked as confirmed. This classification result was directly applied to subsequent statistical analysis and titer calculation. In contrast, when processing the nuclear connected region numbered 1489, the random forest model output the following karyotype probability distributions: homogeneous 0.523, speckled 0.387, nucleolar 0.067, and cytoplasmic 0.023. The homogeneous probability of 0.523 is less than the preset probability threshold of 0.6, but it is still identified as the candidate karyotype category due to its highest probability. Comparing the classification confidence score of 0.523 with the preset confidence threshold of 0.85, the result is 0.523 < 0.85, indicating a low confidence level.

[0052] In the above embodiment, based on the low-confidence result, the nuclear connected component numbered 1489 was marked as a low-confidence nucleus, and the homogeneous candidate karyotype category was set as a temporary karyotype category. The relevant information of this nucleus was stored in the database to be reviewed: nucleus number 1489, temporary karyotype category homogeneous, complete karyotype category probability distribution [0.523, 0.387, 0.067, 0.023], centroid coordinates (18456, 9287), and nuclear fluorescence image data extracted from the fluorescence panoramic image. This record can be provided to professional technicians for manual review in subsequent quality control processes. After classifying all 2156 nuclear connected components one by one, complete classification statistics were obtained. A total of 1987 nuclei were classified with high confidence, and 169 nuclei were classified with low confidence, achieving a confidence pass rate of 92.2%. Among the 1987 high-confidence nuclei, they were grouped and statistically analyzed according to the target karyotype: homogeneous 1689 (85.0%), speckled 201 (10.1%), nucleolar 78 (3.9%), and cytoplasmic 19 (1.0%). For the 169 low-confidence nuclei, they were initially grouped according to provisional karyotype: homogeneous 112, speckled 39, nucleolar 14, and cytoplasmic 4. These provisional classification results are for preliminary statistical reference only; the final karyotype classification needs to be manually verified. Based on the combined high-confidence and low-confidence preliminary classification results, the number of nuclear connected domains corresponding to each karyotype category was determined as follows: homogeneous type 1801 (including 1689 confirmed + 112 pending verification), speckled type 240 (including 201 confirmed + 39 pending verification), nucleolar type 92 (including 78 confirmed + 14 pending verification), and cytoplasmic type 23 (including 19 confirmed + 4 pending verification). Based on the statistical results, the homogeneous type, accounting for 83.5%, became the dominant karyotype category.

[0053] It should also be noted that the examples of all the specific values ​​mentioned above are merely exemplary embodiments, and the specific values ​​mentioned above are not limited to the examples.

[0054] It should be noted that the embodiments described above are only some embodiments of this application, and not all embodiments. The present application will be described in detail below with reference to specific embodiments.

[0055] This application provides a fully automated indirect immunofluorescence interpretation system. The system comprises a sample loading module, a fully automated fluorescence imaging module, an image analysis and karyotype recognition module, and a result output and reporting module. The sample loading mechanism automatically loads the sealed antigen slides onto the scanning platform via a mechanical structure, supporting batch loading of multiple slides and ensuring their stable position during imaging, thus overcoming the inefficiency of traditional manual slide placement. Secondly, the fluorescence imaging system consists of a high-resolution fluorescence microscope and a CCD camera. Once the slide enters the imaging area, the system automatically performs multi-field scanning and autofocuses in different areas, ensuring all fluorescence modes are covered and the imaging resolution remains stable at ≤0.25 μm / pixel, thereby avoiding imaging deviations caused by human error. Furthermore, the image analysis and karyotype recognition system uses accompanying control software to stitch, extract features, and classify the acquired images. It can automatically identify common karyotypes such as homogeneous, speckled, nucleolar, and cytoplasmic karyotypes, and calculate antibody titers by combining sample dilution and fluorescence intensity, thus eliminating subjective differences caused by reliance on human experience. Finally, the results output and reporting system automatically transmits and archives the interpretation results through a data interface with the hospital's LIS platform, while generating standardized reports, achieving unified management and traceability of the results. The various systems are sequentially connected in function and structure, forming a complete automated system from sample loading, image acquisition, intelligent analysis to result output.

[0056] 1. Sample loading module: The sample loading module consists of a carrying platform, a transmission mechanism, positioning and clamping devices, and connecting seats. These components are fixedly installed and positioned to form an overall frame. The carrying platform, as the main structure, is rectangular in shape, with multiple rows of regularly arranged slide slots machined on its surface. The width and depth of the slots are designed according to the dimensions of standard glass slides, and the slots are separated by partitions. Limiting flanges are provided on the inner walls of the slots, with the flange height slightly exceeding the thickness of the glass slide, structurally defining the insertion direction of the slides. A transmission mechanism is installed below the carrying platform, consisting of a slide rail and a drive assembly. The drive assembly preferably uses a stepper motor combined with a lead screw or synchronous belt structure, driving the slide rail to move the glass slides longitudinally. Positioning and clamping devices are installed at both ends of the carrying platform. These devices include limiting blocks and clamping springs. The limiting blocks are positioned laterally on the platform to laterally restrict the edges of the glass slides, while the clamping springs are positioned above the inserted glass slides to vertically clamp them. The carrying platform and the transmission mechanism are structurally connected. The transmission mechanism and the positioning and clamping device maintain coordinated movement, ensuring that the slide remains in a fixed position within the slot during longitudinal transmission. The front end of the carrying platform connects to the scanning platform of the imaging module via a connecting seat. The connecting seat uses a threaded or snap-fit ​​connection to ensure stable installation of the loading module and the imaging module. The carrying platform has a multi-layered stacked structure with internal guide rails arranged longitudinally and fitting against the edge of the slide to limit lateral displacement during transmission. Through the arrangement and installation of these components, the sample loading module forms an integrated structure with the carrying platform as the foundation, the transmission mechanism as the drive, the positioning and clamping device as the constraint, and the connecting seat as the docking point. All components are interconnected through fixed installation and mechanical cooperation.

[0057] 2. Fully automated fluorescence imaging module: The fully automated fluorescence imaging module consists of a fluorescence microscope and a CCD camera, which are fixedly connected to the system's scanning platform. The fluorescence microscope is mounted on the main frame of the interpretation system, with its lower part docking with the scanning platform. The scanning platform and the slide transported by the sample loading module are kept at the same horizontal plane, ensuring accurate entry of the slide into the imaging area. The microscope's imaging channel is coaxially connected to the CCD camera, which is fixed at the microscope's imaging port and connected to the control computer via a data interface. Structurally, this module maintains an integrated layout with the scanning platform, ensuring that the slide, after being transported to the designated position by the loading module, is within the microscope's focal plane. The microscope and CCD camera are fixedly positioned, with the CCD camera directly receiving the fluorescence signal acquired by the microscope and transmitting it to the control system. The scanning platform, fixedly supporting the slide below the microscope, maintains a precise positional relationship with the imaging device, thus ensuring stable coverage of the imaging area. The combination of the fluorescence microscope and CCD camera forms a complete imaging path. The module, through sequential connection with the loading module, scanning platform, and image analysis module, constitutes a continuous imaging link within the system.

[0058] 3. Image analysis and kernel pattern recognition model: The image analysis and karyotype recognition module, as the core processing component of the fully automated indirect immunofluorescence interpretation system, mainly consists of a control computer and its supporting software system, and is directly connected to the CCD camera of the imaging module via a data interface. The control computer is fixedly installed on the side or bottom of the system rack, and its host unit is connected to the CCD camera via a high-speed data cable. The cable is routed through the internal wiring channels of the equipment to avoid interference with the transmission mechanism. Fluorescence images acquired by the CCD camera at the microscope imaging end are transmitted to the control computer in real time as digital signals. The computer's processing and storage units are structurally integrated within the same chassis, enabling immediate image reception and storage upon data input. Inside the control computer, pre-installed image analysis software and hardware processor work together. The software interface is output through a monitor, which is fixedly connected to the host unit via a video cable and installed at the front of the rack for easy operator monitoring. After receiving images, the software interacts with the database, which is stored on the same host unit's hard drive, enabling archiving and management of sample images from different batches and with different numbers. The module's software functions include image stitching, segmentation, and feature extraction. These processing functions operate with the support of the host memory and processor, and are synchronized with external interfaces via an internal bus. In the karyotype recognition stage, the module utilizes the computer's processing unit to classify and discriminate the feature data of the fluorescence image. The recognition results are directly bound to the image file. This binding relationship is stored in the software as a fixed file number and interfaces with the hospital information system through an internal interface. The entire image analysis module is structurally sequentially connected to the imaging module. The CCD camera output is connected to the computer input, and the computer output is connected to the result interpretation module. The front-end image input and the back-end data output form a continuous path in terms of wiring and interfaces. Overall, the image analysis module consists of a computer host, monitor, data interface, and internal storage components. These components are physically connected by cables and logically operate collaboratively through a bus and software. It sequentially connects with the imaging module and information output module via interfaces, forming a complete channel from image acquisition to image processing and result transmission.

[0059] 4. Result Interpretation Module: The results interpretation module consists of a control computer, a display terminal, a data interface, and connecting components for interfacing with the hospital information system. It is installed at the system's output end and directly connected to the image analysis module. The control computer, as the core component of this module, shares the same host with the image analysis module. The host is equipped with data processing and storage units to receive and aggregate analysis data from the front end. The display terminal is fixedly connected to the host via a video cable and is located at the front of the rack to maintain synchronized output with the host. For connections to external systems, the control computer has a standardized data interface on its back or side. This interface extends to the outside of the rack via a network cable and connects to the LIS platform of the hospital information system. The cable is routed along the internal wiring channels of the rack, isolating it from the electrical wiring of other modules. The data interface and storage components are directly connected via the host's internal bus. The storage components, located inside the host, correspond to the report generation program, forming a data transmission and storage path. Positionally, the results interpretation module is located after the image analysis module in the overall structure. Its input end is connected to the output end of the image analysis module, and its output end interfaces with the LIS platform of the hospital information system via a data interface, forming a continuous data transmission path. The display terminal, control computer, and data interface are all fixedly installed on the rack, and the cables are distributed in an orderly manner through the wiring channels. The components form a complete structural system in terms of physical connection and signal path.

[0060] This application also provides a fully automated indirect immunofluorescence interpretation workflow, see below. Figure 2 , Figure 2 This is an interactive flowchart of a fully automated indirect immunofluorescence interpretation embodiment of this application, including the following steps: I. Preparation Stage: Step 1, Register Sample Information: The user / operator inputs the sample information into the photo control system; Step 2, Place the sample slide: The user / operator places the slide into the sample loading module; II. Sample Scheduling and Location: Step 3, Moving the sample loading module: The photo control system instructs the sample loading module to move the carrier to its position (and waits for confirmation of its position). Step 4: Use a low-power microscope to scan and locate the antigen slide: The image capture control system requires the fully automatic fluorescence imaging module to use a low-power microscope to scan and initially locate the entire antigen slide. Step 5, Transmit Preview Image: The fully automated fluorescence imaging module transmits the scanned preview image to the image analysis and model recognition module; Step 6, Identify tissue / cell regions: The image analysis and model recognition module processes the preview image to identify tissue or cell regions in the image; Step 7, Return Coordinate List: The image analysis and model recognition module returns a list of precise coordinates of the identified tissue / cell regions to the image capture control system; III. Image acquisition loop for each interpretation region / field of view: Step 8, Move to precise coordinates: The photo control system instructs the fully automatic fluorescence imaging module to move the stage to the precisely specified coordinate position (and wait for confirmation of arrival). Step 9, switch to high-power objective: The image control system instructs the fully automatic fluorescence imaging module to switch the microscope to the high-power objective; Step 10, Set Fluorescence Notification: The photo control system sends a command to the fully automatic fluorescence imaging module to set the fluorescence. Step 11, Focusing: The camera control system requires the fully automatic fluorescence imaging module to perform autofocus. Step 12, Capture high-resolution fluorescence images: The camera control system instructs the fully automatic fluorescence imaging module to capture high-resolution fluorescence images of the current field of view; Step 13, Transmit raw image data: The fully automatic fluorescence imaging module transmits the captured raw image data to the image analysis and model recognition module; IV. Image Analysis and Interpretation: Step 14, Image Preprocessing (Noise Reduction, Stitching): The image analysis and model recognition module preprocesses the raw image data (and potentially accumulated multi-view images) transmitted each time, including noise reduction and stitching. Step 15, Feature Extraction and Pattern Recognition: The image analysis and model recognition module extracts features from the preprocessed image and performs pattern recognition; Step 16, Intelligent interpretation based on fluorescence pattern: The image analysis and model recognition module uses the built-in analysis model to intelligently interpret the fluorescence pattern based on the extracted features and patterns. Step 17, Generate titer / result: The image analysis and model recognition module ultimately derives the titer value or judgment result; V. Results Integration and Output: Step 18, Upload Interpretation Results and Images: The image analysis and model recognition module uploads the final interpretation results and relevant image data from the analysis process. This uploaded data will trigger or flow to the subsequent storage and report generation stages; Step 19, Store the results in the database: The interpretation results and image data are stored in the database (LIS / local); Step 20, Generate Final Report: The photo control system (or the part of the system connected to the user interface) generates a final report based on the results stored in the database; Step 21, Display and print the interpretation report: Users / operators can view the interpretation report on the screen and print it through the photo control system.

[0061] The fully automated indirect immunofluorescence interpretation system in the embodiments of this invention is described below from the perspective of hardware processing. (See attached document.) Figure 3 , Figure 3 This is a schematic diagram of the physical device structure of a fully automated indirect immunofluorescence interpretation system in the embodiments of this application.

[0062] It should be noted that, Figure 3 The structure of the fully automated indirect immunofluorescence interpretation system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0063] like Figure 3 As shown, the fully automated indirect immunofluorescence interpretation system includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 302 or a program loaded from storage section 308 into random access memory (RAM) 303, such as executing the methods described in the above embodiments. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0064] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including a hard disk, etc.; and communication section 309 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0065] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs the various functions defined in the present invention.

[0066] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

Claims

1. A fully automated indirect immunofluorescence interpretation method, characterized in that, include: Once the identification information of the target sample is obtained, multiple antigen slides that have been sealed are placed in batches into the slide slots of the carrier platform. The carrier platform is driven by the transmission mechanism to move the multiple antigen slides sequentially to the imaging area, and the positioning and clamping device is used to perform positioning and fixing operations on the antigen slides that have entered the imaging area. The antigen slides entering the imaging area are sequentially scanned using multiple fields of view to obtain multi-field fluorescence images. These multi-field fluorescence images are then stitched together to generate a panoramic fluorescence image. Feature extraction is performed on the fluorescence panoramic image to determine the fluorescence nucleus type and target fluorescence intensity value, specifically including: The fluorescence panoramic image is subjected to background separation processing to obtain a binary segmentation image; Connected component extraction is performed on the binary segmented image to obtain multiple cell nucleus connected components; A comprehensive feature vector is obtained by extracting comprehensive features from each of the multiple nuclear connected domains, specifically including: Iterate through each of the plurality of connected nuclear regions in turn, and perform the following operation on the currently traversed connected nuclear region: Obtain the position coordinates of the target cell nucleus connected region, and extract the target cell nucleus fluorescence image corresponding to the target cell nucleus connected region from the fluorescence panoramic image based on the position coordinates, wherein the target cell nucleus connected region is the currently traversed cell nucleus connected region; Morphological features are extracted from the target cell nuclear fluorescence image to obtain a morphological feature vector; Fluorescence intensity distribution analysis was performed on the target cell nuclear fluorescence image to obtain a fluorescence intensity feature vector; Fluorescence spatial distribution analysis is performed on the fluorescence image of the target cell nucleus to obtain a fluorescence spatial feature vector. The fluorescence spatial feature vector includes centroid offset, non-uniformity, aggregation, and symmetry. The fluorescence spatial distribution analysis is used to evaluate the spatial organization pattern of the fluorescence signal included in the fluorescence image of the target cell nucleus. The spatial organization pattern is the centroid offset, non-uniformity, aggregation, and symmetry of the fluorescence signal in the connected domain of the target cell nucleus. Texture features are extracted from the target cell nuclear fluorescence image to obtain a texture feature vector; The morphological feature vector, the fluorescence intensity feature vector, the fluorescence spatial feature vector, and the texture feature vector are concatenated and combined to obtain the target comprehensive feature vector of the target cell nuclear connected domain; The comprehensive feature vector is input into a preset karyotype classification model to obtain the karyotype category of each nuclear connected region and to determine the number of nuclear connected regions corresponding to each karyotype category. The dominant karyotype category is determined based on the number of connected nuclear regions, and the dominant karyotype category is used as the fluorescent karyotype category. From the plurality of nuclear connected domains, nuclear connected domains with the same karyotype as the dominant karyotype category are selected to obtain a set of target karyotype nuclear connected domains; Extract the nuclear fluorescence image of each nuclear connected region in the target karyotype cell nuclear connected region set from the fluorescence panoramic image; Fluorescence intensity was measured for each of the cell nuclei fluorescence images to obtain the single-nucleus fluorescence intensity value for each cell nuclei fluorescence image; The target fluorescence intensity value is obtained by performing central tendency analysis on the single-nuclear fluorescence intensity values ​​of all cell nuclear fluorescence images in the target karyotype cell nuclear connected region set; The antibody titer is determined based on the fluorescent nucleotype, the fluorescence intensity value, and the dilution information of the target sample. The antibody titer, the fluorescent nucleotype category, and the fluorescent panoramic image are associated and mapped according to the identification information to generate a fluorescence interpretation result.

2. The method according to claim 1, characterized in that, The step of sequentially performing multi-field scanning on the antigen slide entering the imaging region to obtain a multi-field fluorescence image specifically includes: The following operations are performed sequentially on the antigen slide that enters the imaging area: The antigen slide detection sub-region of the target antigen slide is pre-scanned using a fluorescence microscope in low-magnification mode to obtain the boundary coordinate information of the antigen slide detection sub-region. The target antigen slide is the antigen slide currently entering the imaging area. The preset field-of-view overlap rate and the single field-of-view size of the fluorescence microscope are obtained. Based on the boundary coordinate information, the preset field-of-view overlap rate and the single field-of-view size, the number of scanning columns required in the length direction and the number of scanning rows required in the width direction of the antigen slide detection sub-region are determined. The preset field-of-view overlap rate is the ratio of the area of ​​the overlapping region between two adjacent scanning field-of-view units to the area of ​​a single scanning field-of-view unit. The antigen slide detection sub-region is divided into multiple scanning field units according to the number of scanning columns and the number of scanning rows, and each of the multiple scanning field units is assigned a spatial position number and a scanning sequence number. Each scanning field of view unit is activated sequentially according to the scanning sequence number. When the target scanning field of view unit is activated, the stage of the fluorescence microscope is controlled to move to the target spatial position corresponding to the spatial position number of the target scanning field of view unit. The target scanning field of view unit is the currently activated scanning field of view unit. An adaptive focusing operation is performed on the target scanning field of view unit to obtain a single-field fluorescence image of the target scanning field of view unit; Having acquired a single-field fluorescence image for each scanning field of view unit, the single-field fluorescence images of each scanning field of view unit are arranged and combined according to their spatial location numbers to obtain the multi-field fluorescence image.

3. The method according to claim 2, characterized in that, The step of performing adaptive focusing on the target scanning field of view (RFP) to obtain a single-field fluorescence image of the RFP specifically includes: After the stage is moved to the target spatial position, the focusing mechanism of the fluorescence microscope is controlled to move along the optical axis within a preset focusing range to multiple preset focal plane levels, and initial fluorescence images are acquired at each of the multiple preset focal plane levels to obtain multiple initial fluorescence images. The optical axis is the direction of the objective optical axis of the fluorescence microscope, and the multiple preset focal plane levels are multiple focal plane positions distributed at preset level intervals along the optical axis within the preset focusing range. A first sharpness analysis is performed on the plurality of initial fluorescence images to determine a plurality of initial sharpness values ​​that correspond one-to-one with the plurality of preset focal plane levels; The initial sharpness values ​​are arranged in descending order, so that the preset focal plane level corresponding to the first initial sharpness value after arrangement is determined as the coarse positioning focal plane level of the target scanning field of view unit. Centered on the coarse positioning focal plane layer, the focusing mechanism is controlled to move in steps within a preset offset range on both the positive and negative sides of the coarse positioning focal plane layer along the optical axis direction, and a step fluorescence image is acquired after each step movement to obtain multiple step fluorescence images. The single step size of the step movement is less than the preset layer spacing. A second sharpness analysis is performed on the plurality of stepped fluorescence images to determine a plurality of stepped sharpness values ​​that correspond one-to-one with the plurality of stepped fluorescence images; The multiple step sharpness values ​​are arranged in descending order so that the step movement position corresponding to the first step sharpness value after arrangement is determined as the precise focal plane level of the target scanning field of view unit. The focusing mechanism is controlled to remain at the precise focal plane level, and the fluorescence microscope is controlled to switch from the low magnification mode to the high magnification mode. In the high magnification mode, the CCD camera is used to acquire images of the target scanning field of view unit at the precise focal plane level to obtain a single field of view fluorescence image.

4. The method according to claim 2, characterized in that, The step of stitching together the multi-view fluorescence images to generate a fluorescence panoramic image specifically includes: From the multi-field fluorescence image, extract the single-field fluorescence image with the first spatial position number as the first image to be stitched, and extract the single-field fluorescence image that is adjacent to the first image to be stitched in terms of spatial position number as the second image to be stitched. Using the first image to be stitched and the second image to be stitched as loop variables, perform the following image stitching operation until the fluorescent panoramic image is generated: First feature points are extracted from the first image to be stitched to obtain a first feature point set; and second feature points are extracted from the second image to be stitched to obtain a second feature point set. The first set of feature points is matched with the second set of feature points to obtain multiple pairs of feature points. Reliability analysis is performed on the multiple feature point pairs to obtain a set of reliable feature point pairs; The spatial transformation matrix between the first image to be stitched and the second image to be stitched is determined based on the set of reliable feature points; The second image to be stitched is transformed into the coordinate system of the first image to be stitched using the spatial transformation matrix, thereby obtaining the transformed image to be stitched. The overlapping region between the first image to be stitched and the image to be transformed is determined, and the overlapping region is fused to obtain the fused overlapping region. The non-overlapping regions of the first image to be stitched, the fused overlapping regions, and the non-overlapping regions of the image to be stitched and transformed are stitched together to obtain an intermediate stitched image; Determine whether there are any incompletely stitched single-field fluorescence images in the multi-field fluorescence images; If it is determined that there is an incomplete single-field fluorescence image in the multi-field fluorescence image, the intermediate stitching image is updated to the first image to be stitched, and the single-field fluorescence image that is adjacent to the intermediate stitching image and is incompletely stitched, extracted from the multi-field fluorescence image according to the spatial location number, is updated to the second image to be stitched. If it is determined that there are no incompletely stitched single-field fluorescence images in the multi-field fluorescence images, the intermediate stitched image is determined as the fluorescence panoramic image.

5. The method according to claim 1, characterized in that, The step of inputting the comprehensive feature vector into a preset karyotype classification model to obtain the karyotype category of each nuclear connected region and determining the number of nuclear connected regions corresponding to each karyotype category specifically includes: The target comprehensive feature vector is input into a preset kernel type classification model to obtain the kernel type probability distribution output by the preset kernel type classification model after classifying and recognizing the target comprehensive feature vector; Determine the category probability value that is greater than a preset probability threshold from the kernel type category probability distribution; The karyotype category corresponding to the category probability is used as the candidate karyotype category of the target cell nuclear connected domain, and the category probability value is used as the classification confidence. The classification confidence level is compared with a preset confidence threshold to obtain the confidence comparison result; If the classification confidence is greater than the preset confidence threshold based on the confidence comparison result, the candidate karyotype category is determined as the target karyotype category of the target cell nuclear connected domain; If, based on the confidence comparison result, the classification confidence is less than or equal to the preset confidence threshold, the target cell nucleus connected region is marked as a low-confidence cell nucleus, and the candidate karyotype category is used as a temporary karyotype category of the target cell nucleus connected region. The candidate karyotype category and the probability distribution of the karyotype category are then stored in the database to be reviewed. Once the karyotype of each nuclear connected region is determined, the plurality of nuclear connected regions are grouped according to each karyotype, and the number of nuclear connected regions corresponding one-to-one with each karyotype is determined.

6. A fully automated indirect immunofluorescence interpretation system, characterized in that, The fully automated indirect immunofluorescence interpretation system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the fully automated indirect immunofluorescence interpretation system to perform the method as described in any one of claims 1-5.

7. A computer-readable storage medium comprising instructions, characterized in that, When the instructions are executed on a fully automated indirect immunofluorescence interpretation system, the fully automated indirect immunofluorescence interpretation system performs the method as described in any one of claims 1-5.

8. A computer program product, characterized in that, When the computer program product is run on a fully automated indirect immunofluorescence interpretation system, the fully automated indirect immunofluorescence interpretation system performs the method as described in any one of claims 1-5.

Images