Image processing method, device and storage medium for fluorescent slide

Abstract

The application discloses a fluorescent slide image processing method and device and a storage medium, and belongs to the technical field of image processing. The method comprises the following steps: acquiring an image set of a fluorescent slide, taking a scanning image in the image set as a training sample, training a preset model to obtain a special model of the fluorescent slide, identifying light intensity distribution information, background information and noise information corresponding to the scanning image in the image set through the special model, performing image calibration on the scanning image according to the light intensity distribution information, the background information and the noise information, obtaining a single field image, performing image splicing on the single field image, and generating a panoramic image of the fluorescent slide. The application trains a special model for the fluorescent slide, and performs image calibration on the scanning image of the fluorescent slide according to the special model, so that the image display effect of the fluorescent slide is improved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to an image processing method, device and storage medium for a fluorescent glass slide. Background Technology

[0002] During fluorescent slide scanning imaging, the endogenous substances within the slide tissue itself emit autofluorescence upon excitation light, causing spectral aliasing with the target fluorescence signal and resulting in a reduced image signal-to-noise ratio. Related techniques typically address this autofluorescence interference by calibrating the equipment with a fluorescence calibration plate before scanning to generate a uniform compensation file. All subsequent slide scans are then corrected using this compensation file.

[0003] However, due to differences in material and fluorescence properties between the fluorescence calibration plate and the fluorescent slides to be scanned, the light intensity distribution and background compensation parameters generated by the calibration plate cannot accurately reflect the actual situation of each specific slide. This results in inconsistent processing of autofluorescence interference across different slides, and the images still exhibit uneven brightness and background interference.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this application is to provide an image processing method, device, and storage medium for fluorescent slides, aiming to solve the technical problem that the processing results of fluorescent slides exhibit instability across different slides due to autofluorescence interference.

[0006] To achieve the above objectives, this application provides an image processing method for fluorescent glass slides, the method comprising the following steps: A set of images of a fluorescent slide is acquired, and the scanned images in the set are used as training samples to train a preset model to obtain a dedicated model for the fluorescent slide. The dedicated model is used to identify the light intensity distribution information, background information, and noise information corresponding to the scanned images in the image set. Based on the light intensity distribution information, the background information, and the noise information, the scanned image is calibrated to obtain a single field-of-view image; The single-view images are stitched together to generate a panoramic image of the fluorescent slide.

[0007] In one embodiment, before the step of acquiring an image set of a fluorescent slide and using scanned images from the image set as training samples to train a preset model to obtain a dedicated model for the fluorescent slide, the method further includes: The fluorescent slide was pre-scanned to obtain the distribution of autofluorescence intensity of the fluorescent slide under different excitation wavelengths; Based on the autofluorescence intensity distribution, the autofluorescence excitation peak band of the fluorescent glass slide is determined; Select the working excitation wavelength based on the described spontaneous fluorescence excitation peak band; The fluorescent slide is scanned and excited based on the operating excitation wavelength.

[0008] In one embodiment, before the step of acquiring an image set of a fluorescent slide and using scanned images from the image set as training samples to train a preset model to obtain a dedicated model for the fluorescent slide, the method further includes: The fluorescence emitted by the fluorescent slide after it is excited is filtered by a narrow-band filter; The filtered fluorescence is collected by a detector to obtain the scanned image, and the scanned image is stored in the image set; The scanned images in the image set are input into the preset model in batches, and the light intensity distribution prediction value, background mask prediction value and noise distribution prediction value output by the preset model are obtained. The error values ​​of the predicted light intensity distribution, the predicted background mask, and the predicted noise distribution are calculated using a loss function. The weight parameters of each layer of the preset model are updated using the backpropagation algorithm to obtain the special model.

[0009] In one embodiment, before the step of identifying the light intensity distribution information, background information, and noise information corresponding to the scanned images in the image set using the dedicated model, the method further includes: The radius of the rolling ball is determined based on the maximum diameter of the target signal in the scanned image and the rolling ball algorithm. Using the grayscale value of each pixel in the scanned image as the height, and based on the radius of the rolling ball, the sphere is simulated to roll on the grayscale surface of the scanned image. The pixel areas that the sphere cannot reach are identified as uniformly diffused background areas, and these uniformly diffused background areas are removed from the scanned image.

[0010] In one embodiment, before the step of identifying the light intensity distribution information, background information, and noise information corresponding to the scanned images in the image set using the dedicated model, the method further includes: Obtain the point spread function corresponding to the scanned image; Using the point spread function as input, the scanned image is subjected to iterative deconvolution operations through a blind deconvolution algorithm.

[0011] In one embodiment, the step of identifying the light intensity distribution information, background information, and noise information corresponding to the scanned images in the image set using the dedicated model includes: The scanned image is input into the dedicated model, which classifies each pixel of the scanned image and outputs a target signal mask. The identification information corresponding to each pixel in the target signal mask is used to identify whether the pixel belongs to the target signal region or the background region. The pixels in the target signal mask that are identified as belonging to the background region are used as the background information; The illumination intensity value corresponding to each pixel position of the scanned image output by the dedicated model is used as the illumination intensity distribution information, and the noise feature map of the scanned image is used as the noise information.

[0012] In one embodiment, the step of performing image calibration on the scanned image based on the light intensity distribution information, the background information, and the noise information to obtain a single-field-of-view image includes: Based on the background information, the values ​​of pixels belonging to the background region in the scanned image are replaced with preset background values ​​to obtain an image after background removal; The light intensity distribution information is used to generate a light intensity adjustment layer. The light intensity adjustment layer is then fused pixel by pixel with the background-removed image to correct the brightness unevenness caused by light spot in the background-removed image, thus obtaining a light intensity-corrected image. Based on the noise information, noise stripping is performed on the light intensity corrected image to obtain the single-view image.

[0013] In one embodiment, the step of identifying the light intensity distribution information, background information, and noise information corresponding to the scanned images in the image set using the dedicated model includes: The fluorescence intensity values ​​of each pixel position on the fluorescent glass slide are acquired in multiple wavelength channels using a multispectral imaging system to obtain the complete emission spectrum curve of each pixel. Obtain the reference emission spectrum and reference autofluorescence spectrum of the pre-built target fluorescent dye; For each pixel, a linear blending model is established; The target signal intensity value and autofluorescence intensity value of the pixel are obtained by solving the linear mixing model for each pixel using the least squares method. The scanned image is reconstructed based on the target signal intensity value of the pixel, and the reconstructed scanned image is input into the dedicated model.

[0014] In addition, to achieve the above objectives, this application also provides an image processing apparatus for a fluorescent slide, the apparatus comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the image processing method for a fluorescent slide as described above.

[0015] In addition, to achieve the above objectives, this application also provides a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the image processing method for the fluorescent slide as described above.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: This application acquires an image set of the current fluorescent slide and uses its scanned images as training samples to train a model specific to the fluorescent slide. This model can directly identify the unique light intensity distribution, background information, and noise information of the slide from the actual images of the slide. Then, based on the specific information in the dedicated model obtained in real time, the scanned images are calibrated so that the correction process for background interference and uneven illumination is fully adapted to the imaging characteristics of the current slide. This eliminates the brightness unevenness and background interference caused by differences in material, staining degree, and background fluorescence between different slides, resulting in a fluorescent slide image that truly reflects the target signal and has uniform brightness. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic flowchart of the first embodiment of the image processing method for fluorescent glass slides in this application; Figure 2 This is a schematic flowchart of the second embodiment of the image processing method for fluorescent glass slides in this application; Figure 3 This is a flowchart illustrating the third embodiment of the image processing method for fluorescent glass slides in this application; Figure 4 This is a flowchart illustrating the fourth embodiment of the image processing method for fluorescent glass slides in this application; Figure 5This is a flowchart illustrating the fifth embodiment of the image processing method for fluorescent glass slides in this application; Figure 6 This is a schematic diagram of the structure of the image processing device for the fluorescent glass slide in the hardware operating environment involved in the embodiments of this application.

[0020] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0022] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0023] The main solution of this application embodiment is as follows: acquire an image set of a fluorescent slide, and use the scanned images in the image set as training samples to train a preset model to obtain a dedicated model of the fluorescent slide. Through the dedicated model, identify the light intensity distribution information, background information and noise information corresponding to the scanned images in the image set. Based on the light intensity distribution information, background information and noise information, perform image calibration on the scanned images to obtain single-field images. Perform image stitching on the single-field images to generate a panoramic image of the fluorescent slide.

[0024] In existing fluorescent slide scanning imaging techniques, the intrinsic substances within the slide tissue itself emit autofluorescence upon excitation light, causing spectral aliasing with the target fluorescence signal and resulting in a reduced image signal-to-noise ratio. Related techniques typically address this autofluorescence interference by using a fluorescence calibration plate to calibrate the equipment's intensity before scanning, generating a uniform compensation file that is then applied to all subsequent slide scans. However, due to differences in material and fluorescence characteristics between the calibration plate and the slides being scanned, the intensity distribution and background compensation parameters generated by the calibration plate cannot accurately reflect the actual conditions of each specific slide. This leads to inconsistent results in the processing of autofluorescence interference across different slides, resulting in persistent issues of uneven brightness and background interference in the images.

[0025] This application acquires the image set of the current fluorescent slide itself and uses its scanned images as training samples to train a model specific to the fluorescent slide. The model can directly identify the unique light intensity distribution information, background information, and noise information of the slide from the actual image of the slide. Then, based on these real-time acquired specific information, the scanned image is calibrated so that the correction process for background interference and uneven illumination is fully adapted to the imaging characteristics of the current slide itself. This eliminates the brightness unevenness and background interference caused by differences in material, staining degree, and background fluorescence between different slides, resulting in a fluorescent slide image that truly reflects the target signal and has uniform brightness.

[0026] To better understand the above technical solutions, exemplary embodiments of this application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of this application and to fully convey the scope of this application to those skilled in the art.

[0027] It should be noted that the executing entity in this embodiment can be an image processing system for a fluorescent slide, or a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or image processing device for a fluorescent slide capable of the above functions. This embodiment does not specifically limit the specific implementation. The following uses an image processing system for a fluorescent slide as an example to describe this embodiment and the following embodiments.

[0028] Based on this, embodiments of this application provide an image processing method for fluorescent glass slides, referring to... Figure 1 , Figure 1 This is a schematic flowchart of the first embodiment of the image processing method for fluorescent glass slides in this application.

[0029] In this embodiment, the image processing method for the fluorescent slide includes steps S10 to S40: Step S10: Obtain an image set of the fluorescent slide, and use the scanned images in the image set as training samples to train the preset model to obtain a dedicated model for the fluorescent slide; In this embodiment, the image processing system controls the scanning device to perform a full-field scan of the currently placed fluorescent slide, storing all acquired single-field images into an image set. The images in this image set are then used as training samples to train a preset model, resulting in a dedicated model specific to that fluorescent slide. The image set of the fluorescent slide refers to a collection of digital image files named according to the scanning position, generated by scanning the same fluorescent slide field-by-field with the scanning device. A scanned image refers to the original fluorescence microscopic image corresponding to a single field of view in the image set, carrying both the target fluorescence signal and the autofluorescence interference signal. Training samples refer to image data extracted from the image set and used for model training. The preset model refers to a pre-initialized convolutional neural network architecture capable of learning and outputting light intensity distribution patterns, background morphological features, and noise distribution patterns from input images, but with weight parameters not yet adjusted for the current fluorescent slide. The dedicated model refers to a model instance where, based on the preset model and trained using the scanned images of the current fluorescent slide itself, the weight parameters have converged to accurately characterize the imaging features of that specific slide.

[0030] Specifically, the scanning device employs a line-by-line scanning method. Driven by a stepper motor, the stage moves equidistantly to the right from the upper left corner of the slide, stopping at each preset field-of-view position. An excitation light source illuminates the field of view, and a detector collects fluorescence signals and generates a digital image. Upon receiving the image, the image processing system immediately names it according to a combination of row and column coordinates and writes it to the storage path of the image set. After storage, the image processing system sends a command to the scanning device to move to the next position without waiting for further image processing. After the full-field scan is completed, the image set contains the original images of all fields of view for the slide. The image processing system reads all images from the image set and uses them as training samples, inputting them batch by batch into a preset convolutional neural network model. The model performs multi-layer convolution and pooling operations on the input image to extract features at different scales, including low-frequency features that characterize the overall illuminance trend of the light spot and high-frequency features that characterize the local morphology of autofluorescence in the tissue. The predicted values ​​of light intensity distribution, background mask, and noise distribution output by the model are calculated against the preset loss function. The weight parameters of each layer of the model are updated through the backpropagation algorithm. After multiple rounds of iterative training until the loss function converges, a dedicated model specific to the current fluorescent slide is obtained.

[0031] As an optional implementation, the image processing system performs data augmentation operations on the images in the image set before training. Augmentation operations include randomly rotating the images by 0, 90, 180, or 270 degrees, changing the spatial orientation of the light intensity distribution map while maintaining its relative relationship with the image content; horizontally or vertically mirroring the images with a 50% probability; and multiplying the overall contrast of each image by a random coefficient uniformly sampled between 0.8 and 1.2 while preserving the dynamic range of the image signal. These augmentation operations are performed within the image set, without introducing data from outside the fluorescent slide, resulting in a specialized model that is more adaptable to subtle changes in field of view and brightness. The number of augmented images is expanded to several times the number of original images, supporting model training with a larger sample size and making the loss function decrease more smoothly.

[0032] As an alternative implementation, when the fluorescent slide contains two or more tissue types with significant differences in autofluorescence intensity and spectral characteristics, the image processing system does not distinguish between tissue types during the training phase. Instead, it mixes all images from the field of view and inputs them into a preset model for training. The preset model employs a U-shaped convolutional neural network architecture. The encoding path compresses the input image into a multi-scale feature map through continuous convolution and downsampling operations. The bottleneck layer integrates global information into a latent spatial representation. The decoding path gradually restores the spatial resolution through upsampling and skip connections, ultimately outputting three branches corresponding to the light intensity distribution map, background mask, and noise feature map, respectively. During training, the shallow convolutional layers of the encoding path extract low-level features such as image edges and textures, while the deep convolutional layers combine these low-level features to form high-level semantic features such as light intensity trends and background region connectivity. After mixed training of images of different tissue types, the features encoded by the model weights are a comprehensive expression of the features of each tissue type. During inference, the model can automatically adapt to the feature patterns of the tissue type to which the current input image belongs without pre-labeling the tissue category.

[0033] Optionally, the system filters the fluorescence emitted by the excited fluorescent slide using a narrowband filter, collects the filtered fluorescence with a detector to obtain a scanned image, and stores the scanned image in an image set. The scanned images in the image set are then input batch by batch into a preset model to obtain the predicted values ​​of light intensity distribution, background mask, and noise distribution output by the preset model. The error values ​​of the predicted values ​​of light intensity distribution, background mask, and noise distribution are calculated using a loss function. The weight parameters of each layer of the preset model are updated using a backpropagation algorithm to obtain a dedicated model.

[0034] For example, after scanning a liver tissue fluorescence slide, the image set contains 500 scanned images. The image processing system rotates and flips these images one by one to enhance them, resulting in 2,000 enhanced images. These 2,000 enhanced images are then used to train a preset model. The model's loss function decreases rapidly in the first ten iterations and stabilizes after the twentieth iteration. The trained model has learned the unique spot distribution shape, tissue autofluorescence texture, and sensor fixed pattern noise specific to the liver slide.

[0035] For example, a fluorescent slide contains two different types of tissue regions: kidney tissue and colon tissue. Images of these two types of regions in the image set differ in light intensity distribution and background morphology. During training, the image processing system mixes all images and inputs them into a preset model. The model automatically learns these differences through convolutional layers and encodes them uniformly in the weight parameters. After training, a specialized model that can simultaneously adapt to the processing needs of both types of tissue regions is obtained.

[0036] Step S20: Using the dedicated model, identify the light intensity distribution information, background information, and noise information corresponding to the scanned images in the image set; In this embodiment, the trained dedicated model is switched to inference mode, processing each scanned image in the image set one by one. During a single forward propagation, the model simultaneously completes three recognition tasks: classifying image pixels into target signal regions or background regions to extract background information; performing illuminance value regression on all pixels in the image to obtain light intensity distribution information; and removing random fluctuation components unrelated to both signal and background from the image to constitute noise information. Light intensity distribution information refers to a data array describing the relative illuminance differences at each pixel position in a single scanned image caused by the uneven distribution of the excitation light spot shape and intensity. The value at each position in this data array represents the coefficient that the pixel should be compensated for or attenuated under the assumption of ideal uniform illumination. Background information refers to the segmentation results identifying which pixels in the scanned image belong to the autofluorescent background region and which belong to the target fluorescence signal region. This result is represented as a mask data of the same size as the original image. Noise information refers to data characterizing the spatial distribution and statistical intensity of random interference signals in the scanned image, composed of sensor readout noise, dark current noise, and ambient stray light noise. It is presented in the form of a noise feature map of the same size as the original image.

[0037] Specifically, when a scanned image is input into a dedicated model, the model's encoding path performs multi-layer convolutional feature extraction. The first convolutional layer uses 64 3x3 convolutional kernels with a stride of 1, generating 64 feature channels to capture basic textures such as edges and corners of the image. The subsequent downsampling layer uses 2x2 max pooling to halve the feature map size, and the second convolutional layer uses 128 convolutional kernels to further combine basic textures into local structural features. After four downsampling passes, the feature map size is reduced to one-sixteenth of the original input, and the number of channels increases to 512. At this point, the deep features have encoded high-level semantic information such as the illumination trend and background region connectivity of the entire image. The model's decoding path then begins upsampling to restore spatial resolution. For the background information recognition branch, the decoder ends with a 1x1 convolutional layer and a sigmoid activation function, outputting a single-channel probability map. The image processing system binarizes this probability map with a threshold of 0.5 to obtain a target signal mask. The set of pixels in the mask that are identified as background regions is the background information of the image. For the branch identifying light intensity distribution information, the decoder is connected to another 1x1 convolutional layer and a linear activation function, outputting a single-channel illuminance prediction map as light intensity distribution information. For the branch identifying noise information, the decoder is connected to a third 1x1 convolutional layer, outputting a noise feature map incorporating a residual learning mechanism as noise information.

[0038] As an optional implementation, to improve the accuracy of background segmentation, the image processing system automatically determines the background category based on the pixel value distribution of the image before inputting the scanned image into the dedicated model. If the overall image histogram shows a bimodal distribution and the peak area of ​​the low grayscale region is large, the background is determined to be a dark background. In this case, the probability map threshold output by the model is set to an adaptive threshold calculated individually for each image by the Otsu algorithm, to more accurately adapt to the grayscale distribution characteristics of the target signal against a dark background. If the overall image histogram is biased towards high grayscale and the background region has high grayscale, the background is determined to be a bright background, and the threshold remains unchanged at 0.5 to ensure stable separation of the target signal from the background against a bright background.

[0039] As an alternative implementation, for images acquired by a multispectral imaging system, the number of input channels for the dedicated model is expanded to be the same as the number of spectral channels, with each channel corresponding to fluorescence intensity within a narrow band. The encoding path extracts cross-spectral spatial features uniformly from the multi-channel input, while the three branches of the decoding path output multi-channel results equal to the number of input channels. The light intensity distribution information recognition branch outputs not a single-channel illuminance map, but a multi-channel illuminance tensor containing independent illuminance coefficients for each channel, which can correct for chromatic brightness unevenness caused by differences in the optical path of different wavelength channels in the optical system. The background information recognition branch performs pixel classification for each spectral channel and takes the consistent region between channels as the final background mask. The noise feature map output by the noise information recognition branch also has a multi-channel structure, with each channel independently representing the noise distribution within its corresponding spectral band.

[0040] For example, when processing a renal tubular fluorescence-stained scan image, the dedicated model identifies that the illumination intensity in the central region of the image is higher than that in the edge region. The generated light intensity distribution map shows a gradual change pattern of bright in the center and dark around the edges. Simultaneously, it identifies the renal tubular epithelial cell region as the target signal region and the lumen and interstitial region as the background region. The generated mask accurately outlines the renal tubular cell contours. The noise feature map mainly shows a high-frequency granular distribution in the dark areas of the image, consistent with the sensor readout noise pattern. For example, when processing a brain tissue fluorescence scan image, the dedicated model identifies a distinct band-shaped bright area on the left side of the image due to off-axis illumination. The light intensity distribution map generated by the model accurately reflects this band-shaped bright feature. Simultaneously, it segments the neuronal cell body region as the target signal region and the glial cell autofluorescence region as the background region. The noise feature map shows a higher amplitude in the dark areas on the right side of the image.

[0041] In this embodiment, a dedicated model simultaneously outputs three independent types of information—light intensity distribution, background information, and noise information—during a single forward inference process. This avoids the redundant computation overhead and consistency issues between the outputs of different modules when calling different algorithm modules for serial processing. Moreover, all three types of information are specific to the actual imaging situation of the current slide and the current field of view.

[0042] Step S30: Based on the light intensity distribution information, the background information, and the noise information, perform image calibration on the scanned image to obtain a single field-of-view image; In this embodiment, the image processing system utilizes three types of information—light intensity distribution information, background information, and noise information—to sequentially perform three operations on the original scanned image: background removal, light intensity correction, and noise stripping. The execution order of these three operations is fixed: background removal first, light intensity correction second, and noise stripping third. The output of the previous step serves as the input for the next step, ensuring that each operation targets an image that has already undergone pre-processing optimization. Image calibration refers to performing background removal, light intensity correction, and noise stripping sequentially on the original scanned image, eliminating spontaneous fluorescence interference, restoring uniform illumination brightness, suppressing random noise, and preserving the true intensity and morphology of the target fluorescence signal in the processed image. A single-field-of-view image refers to a single field-of-view image obtained after the original scanned image has undergone a complete image calibration process. This image contains only the target fluorescence signal and has normalized brightness and noise; it serves as the input unit for subsequent stitching operations.

[0043] By applying three types of interference information to image calibration in sequence, each correction step can be performed on a cleaner and more uniform image basis, resulting in a single-view image with a pure background, uniform illumination, and controllable noise, thus fully preserving the true morphology and quantitative relationship of the target fluorescence signal.

[0044] Specifically, background removal is performed first. The image processing system reads the target signal mask, iterates through each pixel position of the mask, and when the mask value indicates that the pixel belongs to the background region, the gray value of the corresponding pixel in the original scanned image is replaced with a uniform background compensation value. The background compensation value is obtained by calculating the average gray value of all pixels in the original image that are marked by the mask as outside the target signal region. After the replacement is completed, all pixels identified as autofluorescent background are set to the same gray value, while the pixel values ​​of the target signal region remain unchanged. The light intensity correction operation is performed on the image after background removal. The image processing system reads the light intensity distribution map, iterates through each pixel of the image, and divides the current gray value of the pixel by the corresponding correction coefficient. The correction coefficient for pixels in dark corner areas is less than 1, so the division operation increases their gray value; the correction coefficient for pixels in bright areas at the center of the spot is greater than 1, so the division operation decreases their gray value. The correction coefficient is obtained by smoothing the illumination distribution of the image itself, without including high-frequency details of the tissue structure. The division operation flattens the overall brightness while completely preserving the relative intensity differences within the target signal. The noise stripping operation is performed on the image after intensity correction. The image processing system reads the noise feature map, and areas with amplitudes below a preset threshold are considered clean areas and are not smoothed. For areas with amplitudes above the threshold, a weighted mean filter is applied using a neighborhood window centered on the pixel. The weight of each pixel in the neighborhood is determined by the amplitude difference between the pixel and the center pixel in the noise feature map. The smaller the difference, the greater the weight, indicating that the two pixels are in the same noise environment.

[0045] As an optional implementation, for extreme cases where the fluorescence signal is extremely weak and the background is extremely strong, the background compensation value in the background removal operation is not determined by the image's own background mean, but rather by a preset fixed value. This fixed value is set with reference to the grayscale value of a blank area image acquired under the same conditions as the current slide, ensuring that the weak signal is not submerged after background subtraction. The light intensity correction operation is performed on the image where the background has been replaced with a fixed value, and the correction coefficient is still provided by the light intensity distribution map. The corrected weak signal target area reveals the previously suppressed signal-to-noise ratio because the high background threshold has been removed.

[0046] As an alternative implementation, the noise stripping operation employs a guided filtering approach. The intensity-corrected image itself serves as the guide map, and the noise feature map serves as the noise weight indicator map. The guided filter applies strong smoothing to the high-noise regions indicated by the noise feature map while preserving the edges of the guide map, and weaker smoothing or no smoothing to the low-noise regions indicated by the noise feature map. Within a filtering window centered on each pixel, the guided filter solves a local linear model to ensure that the output image is structurally consistent with the guide map. The noise-stripped image eliminates high-frequency random noise while fully preserving the sharp transitions of the target signal edges.

[0047] For example, when performing image calibration on a scanned image with a large area of ​​autofluorescent fixative, a light intensity distribution that is centrally bright and peripherally dark, and significant readout noise in the dark peripheral areas under high-gain acquisition, the image processing system replaces the autofluorescent areas with the background mean color to make the background clean and uniform. It then uses the light intensity distribution map to perform division correction on the entire image, increasing the overall edge grayscale while decreasing the central grayscale. The system smooths the noise pixels in the dark peripheral areas indicated by the noise feature map, but not the target signal pixels in the central bright areas. The resulting single-view image has a clean background, uniform brightness, and clear target signal. For example, when performing image calibration on a scanned image with a relatively uniform light intensity distribution but with dot-like bright autofluorescent particles in the background and slight Poisson noise, the mask accurately segments the dot-like autofluorescent particles and replaces them with the background mean color. The light intensity correction step makes almost no adjustment to most pixels because the light intensity distribution map value is close to one. The noise stripping step performs low-intensity smoothing on the entire image to target the uniformly distributed slight noise. The resulting single-view image eliminates granular interference and slight noise.

[0048] Step S40: Stitch the single-view image to generate a panoramic image of the fluorescent slide.

[0049] In this embodiment, the image processing system aligns all single-field images in a two-dimensional grid according to the row and column coordinates encoded in the filename of each single-field image, performs weighted fusion on the overlapping areas between adjacent images, and generates a continuous panoramic image that seamlessly covers the entire imaging area of ​​the slide. Image stitching refers to the digital image processing process of geometrically aligning and fusionling multiple single-field images with spatial relationships according to their positional information on a two-dimensional plane to generate a continuous panoramic image covering the entire scanning area of ​​the slide. A panoramic image is a seamless, large-size image generated after stitching together all single-field images, which can completely present the distribution of the target fluorescence signal on the entire fluorescent slide. This image can be directly used for pathological diagnosis or quantitative analysis.

[0050] Specifically, the image processing system parses the filename of each single-view image, extracts the row and column indices, and constructs a two-dimensional image matrix based on these indices. The single-view image corresponding to the index is placed at the i-th row and j-th column position of the matrix. The scanning device maintains a fixed step size when moving to adjacent views, with the step size set slightly smaller than the side length of the detector's imaging area. Adjacent view images have a pixel overlap region of constant width in both the horizontal and vertical directions. The image processing system traverses each pair of horizontally adjacent images in the grid, determines the pixel range of their overlap region, and calculates the ratio of the distance from any pixel position within the overlap region to the non-overlapping regions of the left and right images as a fusion weight. The weight of the left image is equal to the nearest distance from the pixel to the non-overlapping region of the left image divided by the total distance to the non-overlapping regions of both images. The weight of the right image is one minus the weight of the left image. The fused grayscale value is equal to the grayscale value of the left image multiplied by its weight, plus the grayscale value of the right image multiplied by its weight. At the left boundary of the overlap region, the weight of the left image approaches one, and the image displays the content of the left image. At the right boundary, the weight of the right image approaches one, and the image displays the content of the right image. The two images transition smoothly in the middle region. Overlapping regions of vertically adjacent images are processed using the same row-by-row weighted fusion method. After all adjacent pairs have been fused, the image processing system crops the unimaged areas at the grid edges using the bounding rectangles of the actual positions of all single-view images, and outputs a full-resolution panoramic image.

[0051] As an alternative implementation, when panoramic images need to be stored or transmitted with minimal space usage while maintaining full-resolution information, the image processing system generates a panoramic image file with a multi-resolution pyramid structure. The base image is a full-resolution image, and the upper layers are thumbnails that are downsampled by a factor of two at each level. The pathology diagnostic software dynamically loads the layer corresponding to the resolution in the pyramid based on the current zoom level of the display viewport. When viewing the panoramic image in thumbnail mode, a low-resolution thumbnail is loaded, and when zooming in to local details, the local tiles corresponding to the full-resolution base image are loaded. This significantly reduces image loading latency in remote image viewing and digital pathology consultation scenarios.

[0052] As an alternative implementation, for a single-view image with a unified background grayscale, if the width of the overlapping region set during the original scan is sufficient, the image processing system performs sub-pixel-level image registration based on normalized cross-correlation on the overlapping region before seam fusion. Cross-correlation calculation is performed within local windows of the two images corresponding to the overlapping region. The sub-pixel offset is determined by finding the peak position of the cross-correlation function, and this offset is used to compensate for translation in one of the images before weighted fusion. This registration operation compensates for potential backtracking errors that may exist in the stage stepping mechanism during scanning, further improving the continuity of the anatomical structure at the seam.

[0053] For example, a panoramic image of liver tissue generated by stitching together 500 single-field images, because each single-field image has undergone background homogenization and brightness normalization in the calibration step, the background grayscale values ​​and target signal brightness values ​​in the overlapping areas of adjacent fields are highly consistent, and the transition after seam fusion is invisible. The generated panoramic image presents a seamless continuous pathological image, with liver lobule structures and portal vascular fluorescent markers clearly distinguishable and without stitching artifacts. For example, a panoramic image of kidney tissue generated by stitching together 300 single-field images, because the inter-row light intensity between each field of view has been corrected to the same level in the calibration step during scanning, there are no alternating bright and dark stitching stripes between rows, and the overall brightness of the image is continuous and uniform, meeting the requirements of illumination uniformity and signal fidelity for quantitative analysis of kidney pathology.

[0054] This application embodiment acquires the image set of the current fluorescent slide itself and uses its scanned images as training samples to train a model specific to the fluorescent slide. The model can directly identify the unique light intensity distribution information, background information, and noise information of the slide from the actual image of the slide. Then, based on these real-time acquired specific information, the scanned image is calibrated so that the correction process for background interference and uneven illumination is fully adapted to the imaging characteristics of the current slide itself. This eliminates the brightness unevenness and background interference caused by differences in material, staining degree, and background fluorescence between different slides, resulting in a fluorescent slide image that truly reflects the target signal and has uniform brightness.

[0055] Based on the same inventive concept, this application also provides a second embodiment, referring to... Figure 2 , Figure 2 This is a schematic flowchart of the second embodiment of the image processing method for fluorescent glass slides in this application.

[0056] In this embodiment, the image processing method for the fluorescent slide further includes steps S11 to S14: Step S11: Perform a pre-scan on the fluorescent slide to obtain the autofluorescence intensity distribution of the fluorescent slide under different excitation wavelengths; Step S12: Determine the autofluorescence excitation peak band of the fluorescent slide based on the autofluorescence intensity distribution; Step S13: Select the working excitation wavelength according to the autofluorescence excitation peak band; Step S14: Based on the working excitation wavelength, scan and excite the fluorescent slide.

[0057] In this embodiment, the image processing system performs a pre-scan of the fluorescent slide before performing the formal scan. By analyzing the pre-scan results, the system determines the excitation band with the strongest autofluorescence of the slide. Then, using a tunable white laser, the optimized working excitation wavelength is selected for the formal scan while avoiding this band. The pre-scan refers to a rapid spectral sampling of the fluorescent slide at a lower resolution before the formal scan, used to obtain the autofluorescence response intensity of the slide at different excitation wavelengths. The autofluorescence excitation peak band refers to the wavelength range in which the endogenous fluorescent material in the fluorescent slide is most sensitive to excitation light and produces the strongest autofluorescence. A tunable white laser is a laser source capable of continuously and selectively outputting a single excitation wavelength within a certain wavelength range.

[0058] Specifically, the image processing system controls a tunable white laser to change the excitation wavelength sequentially within a wavelength range of 440 nm to 790 nm at preset wavelength step intervals. Each wavelength switch illuminates the slide with excitation light at a power lower than the actual scanning power. The detector collects the fluorescence signal at the corresponding wavelength and generates a low-resolution preview image. The image processing system selects a background region from each preview image that is determined to be devoid of the target fluorescent dye, calculates the average grayscale value of all pixels within that region, and uses this as the autofluorescence intensity value at that excitation wavelength. After scanning all wavelengths, the image processing system plots the autofluorescence intensity values ​​at all wavelengths as a curve of autofluorescence intensity versus excitation wavelength, identifying the wavelength range corresponding to the peak position on this curve, i.e., the autofluorescence excitation peak band. Within the wavelength range of the tunable white laser, the image processing system searches for the excitation spectrum data of the target fluorescent dye and selects a wavelength that is both within the effective excitation range of the dye and does not overlap with the autofluorescence excitation peak band as the working excitation wavelength. During the formal scanning, the image processing system controls the tunable white laser to output the working excitation wavelength and performs full-field scanning imaging on the slide. The acquired image shows a significant reduction in autofluorescence components because the excitation wavelength has bypassed the autofluorescence sensitive range.

[0059] This embodiment obtains the autofluorescence excitation peak information of the current slide itself through pre-scanning and selects an excitation wavelength that avoids the peak accordingly. This reduces the amount of autofluorescence generated from the source of the fluorescence signal, which significantly reduces the background of autofluorescence interference in the input images faced by the subsequent dedicated model training and inference, and further improves the final calibration effect.

[0060] Since the system described in Embodiment 2 of this application is a system used to implement the method of Embodiment 1 of this application, those skilled in the art can understand the specific structure and variations of the system based on the method described in Embodiment 1 of this application, and therefore will not be described again here. All systems used in the method of Embodiment 1 of this application fall within the scope of protection of this application.

[0061] Based on the same inventive concept, this application also provides a third embodiment, referring to... Figure 3 , Figure 3 This is a schematic flowchart of the third embodiment of the image processing method for fluorescent glass slides in this application.

[0062] In this embodiment, the image processing method for the fluorescent slide further includes steps S21 to S23: Step S21: Determine the radius of the rolling ball based on the maximum diameter of the target signal in the scanned image and the rolling ball algorithm; Step S22: Using the grayscale value of each pixel in the scanned image as the height, simulate the sphere rolling on the grayscale surface of the scanned image according to the radius of the rolling ball; Step S23: Identify the pixel areas that the sphere cannot reach as uniformly diffused background areas, and remove the uniformly diffused background areas from the scanned image.

[0063] In this embodiment, before inputting the scanned image into the dedicated model, the image processing system first uses a morphological rolling ball algorithm to preliminarily subtract the uniformly diffused autofluorescent background in the image, thus removing large-scale background interference from the image input into the dedicated model. Based on this, the dedicated model further identifies residual local complex background and light intensity noise information. The rolling ball radius refers to the radius parameter of the simulated sphere, which determines the maximum radius of curvature of the concave region that the sphere can enter when rolling on the grayscale surface. Concave regions larger than this radius cannot be reached by the sphere and are identified as background. The uniformly diffused background region refers to the fluorescent background region in the image that appears as a large area with a smooth grayscale change and does not contain details of the target signal, exhibiting autofluorescence.

[0064] Specifically, the image processing system iterates through the completed dedicated model training steps in the image set and executes the rolling ball algorithm before inputting the scanned image into the dedicated model. The image processing system obtains the typical size of the target signal in the scanned image and sets the radius of the rolling ball to 1.5 to 2 times this typical size, ensuring that the ball cannot enter the target signal region and can only roll within the background region. The image processing system uses the grayscale matrix of the scanned image as a three-dimensional surface. The horizontal and vertical coordinates of each pixel on the surface correspond to its position in the image, and the height coordinate corresponds to the grayscale value of that pixel. The algorithm simulates a ball with a radius equal to the rolling ball radius contacting the surface from below and rolling along the surface. The ball remains tangent to the surface during rolling. The surface positions that the ball can contact are considered signal regions, while the concave areas of the surface that the ball cannot contact due to its large size are considered uniformly diffused background regions. The image processing system uniformly replaces the grayscale values ​​of pixels identified as uniformly diffused background regions with the local background mean of that region, performing background subtraction, and outputs the preprocessed image for subsequent recognition by the dedicated model.

[0065] This embodiment removes large-scale uniformly diffused backgrounds using the rolling ball algorithm before processing by the dedicated model, reducing the dynamic range of the background that the dedicated model needs to process. This allows the model's feature extraction to focus on the fine distinction between the target signal and the local complex background. It also reduces the dependence of the dedicated model on uniformly diffused background samples during training, thus improving the efficiency of the overall processing flow.

[0066] Since the system described in Embodiment 3 of this application is a system used to implement the method of Embodiment 1 of this application, those skilled in the art can understand the specific structure and variations of the system based on the method described in Embodiment 1 of this application, and therefore will not be described again here. All systems used in the method of Embodiment 1 of this application fall within the scope of protection of this application.

[0067] Based on the same inventive concept, this application also provides a fourth embodiment, referring to... Figure 4 , Figure 4 This is a flowchart illustrating the fourth embodiment of the image processing method for fluorescent glass slides in this application.

[0068] In this embodiment, the image processing method for the fluorescent slide further includes steps S24-S25: Step S24: Obtain the point spread function corresponding to the scanned image; Step S25: Using the point spread function as input, perform iterative deconvolution operations on the scanned image using a blind deconvolution algorithm.

[0069] In this embodiment, before inputting the scanned image into the dedicated model, the image processing system performs a blind deconvolution operation on the image to restore the image blur caused by diffraction and scattering of the optical system. This amplifies the difference in sharpness between the target signal and the diffuse autofluorescent background, allowing the dedicated model to more accurately distinguish between the two in subsequent processing. The point spread function (PSF) refers to the two-dimensional intensity distribution of an image formed by an ideal point light source from an optical system; it describes the blur kernel by which the system diffuses the point source into a light spot. The blind deconvolution algorithm is an image restoration algorithm that simultaneously estimates the sharp image and the PSF through an iterative process when the PSF is not fully known.

[0070] Specifically, the image processing system obtains the point spread function (DFD) corresponding to the scanned image. This is achieved either by calculating the theoretical DFD based on optical parameters such as the numerical aperture of the microscope objective, excitation wavelength, and emission wavelength using diffraction theory, or by averaging the images of sub-resolution fluorescent microspheres to obtain the experimentally measured DFD. The image processing system performs blind deconvolution on the scanned image using a Richardson-Lucy algorithm, with 20 to 50 iterations. In each iteration, the algorithm deconvolves the current image based on the currently estimated DFD to obtain a clearer image estimate. This clearer image is then used to update the DFD, progressively approximating the true clear image and DFD. After the iteration terminates, the image processing system replaces the original scanned image with the clear image obtained from the deconvolution, using it as input to a dedicated model for subsequent processing. Because deconvolution restores the high-frequency details of the image, the boundaries of target signals that were previously blurred and mixed with the background become sharper. The diffuse autofluorescent background, lacking a clear boundary structure, retains its diffuse characteristics after deconvolution, significantly enhancing the morphological difference between the two.

[0071] This embodiment restores image sharpness by blindly deconvolutioning before processing with a dedicated model, making the boundary between the target signal and the diffuse autofluorescent background more distinct. The morphological features on which the dedicated model is based when performing foreground and background segmentation are more obvious, thereby improving the accuracy of the background mask and the precision of the light intensity distribution information in the signal edge region.

[0072] Since the system described in Embodiment 4 of this application is a system used to implement the method of Embodiment 1 of this application, those skilled in the art can understand the specific structure and variations of the system based on the method described in Embodiment 1 of this application, and therefore will not be described again here. All systems used in the method of Embodiment 1 of this application fall within the scope of protection of this application.

[0073] Based on the same inventive concept, this application also provides a fifth embodiment, referring to... Figure 5 , Figure 5This is a flowchart illustrating the fifth embodiment of the image processing method for fluorescent glass slides in this application.

[0074] In this embodiment, the image processing method for the fluorescent slide further includes steps S26-S29: Step S26: Acquire the fluorescence intensity values ​​of each pixel position on the fluorescent slide in multiple wavelength channels using a multispectral imaging system to obtain the complete emission spectrum curve of each pixel; Step S27: Obtain the reference emission spectrum and autofluorescence reference spectrum of the pre-built target fluorescent dye; Step S28: For each pixel, establish a linear mixing model, and solve the linear mixing model for each pixel using the least squares method to obtain the target signal intensity value and autofluorescence intensity value of the pixel; Step S29: Reconstruct the scanned image based on the target signal intensity value of the pixel, and input the reconstructed scanned image into the dedicated model.

[0075] In this embodiment, before inputting the scanned image into the dedicated model, the image processing system first separates the target fluorescence signal and the autofluorescence signal at each pixel location from the spectral dimension using multispectral imaging and a spectral linear unmixing algorithm. The reconstructed clean target signal image is then used to replace the original scanned image input into the dedicated model, significantly reducing the spectral cross-interference faced by the dedicated model. A multispectral imaging system refers to an imaging device capable of simultaneously or sequentially acquiring fluorescence signals in multiple narrow-band channels to generate a complete emission spectrum for each pixel. The reference emission spectrum refers to the normalized fluorescence emission spectrum curve measured under standard conditions for the target fluorescent dye, while the reference spectrum for autofluorescence refers to the normalized autofluorescence spectrum curve extracted from an unlabeled control sample or from a known target signal-free region of the current slide. The linear mixing model represents the measured spectrum of each pixel as a linear superposition model of the product of the target dye reference spectrum and the target signal intensity, plus the product of the autofluorescence reference spectrum and the autofluorescence intensity.

[0076] Specifically, the image processing system controls the multispectral imaging system to acquire fluorescence signals in the wavelength range of 400 nm to 700 nm, with each channel representing 5 nm. This yields a set of discrete emission spectral data containing 61 channels at each pixel location. The image processing system reads the reference emission spectrum of the target fluorescent dye from a pre-defined spectral library. This reference spectrum is calibrated against the spectral response functions of each channel of the multispectral system. Simultaneously, it extracts the autofluorescence reference spectrum from a manually selected target-signal-free region in the current slide image, or selects a reference spectrum matching the current tissue type from a pre-built autofluorescence spectral library. The image processing system establishes a linear mixing equation for the spectral data of each pixel, representing the measured spectral vector as the target reference spectral vector multiplied by the target signal intensity scalar plus the autofluorescence reference spectral vector multiplied by the autofluorescence intensity scalar. The equation is solved using a least squares method with non-negative constraints to obtain the target signal intensity value and autofluorescence intensity value for that pixel. After all pixels are solved, the image processing system combines the target signal intensity values ​​of all pixels into a reconstructed image of the same size as the original image. The spectral components of autofluorescence in this image have been stripped, and only the spatial distribution of the target fluorescence signal is retained. The image processing system uses this reconstructed image to replace the original scanned image and inputs it into a dedicated model for the identification and correction of light intensity distribution, residual background, and noise.

[0077] For example, after the operator places the fluorescent slide on the scanning stage, the system first performs a rapid pre-scan of the slide with low-power light to analyze the excitation band with the strongest tissue autofluorescence. Based on the characteristics of the target fluorescent dye, a wavelength is selected on the tunable white laser that avoids the autofluorescence peak but is still within the effective excitation range of the dye, reducing the generation of autofluorescence at the source. A high-quality narrowband filter is installed in the scanning optical path, with the bandwidth controlled between 20 and 50 nanometers to precisely match the emission peak of the target dye. The filter itself uses an ultra-low autofluorescence glass substrate to avoid generating background fluorescence, blocking unwanted wavelengths before the light enters the detector. If the device is equipped with a pulsed laser and a TCSPC module, the system waits 2 to 5 nanoseconds after the excitation pulse ends, allowing the short-lived autofluorescence to decay completely before opening the detection window to collect the remaining long-lived target fluorescence signal, further separating the signal in the time dimension.

[0078] The scanning device performs position-by-position scanning imaging of the entire fluorescent slide. After each position is captured, the system names the image according to the row number, column number, and filter color format and directly stores it into the image set. After storage, the stage immediately moves to the next position to continue capturing images without waiting for the current image to complete processing, thus achieving parallel execution of acquisition and processing. After acquiring the raw image, the system first uses the rolling ball algorithm to process the uniformly diffused autofluorescent background. This simulates a ball rolling on the image brightness surface and identifies areas that the ball cannot reach as background, which is then directly subtracted. The ball radius is set to be larger than the target signal diameter to avoid missubtracting signals, mainly addressing the overall background fluorescence caused by the fixative. If the image is blurry and the background is diffuse, the Richardson-Lucy blind deconvolution algorithm is used to restore sharpness. By inputting the theoretically or experimentally measured point spread function and iterating 20 to 50 times, the signal becomes clearer while the diffuse background is distinguished. If the device supports multispectral acquisition, each pixel acquires a complete emission spectrum. The system uses a pre-built reference spectrum library to establish a linear mixing model, and calculates the proportion of target signal and autofluorescence in each pixel using the least squares method, retaining only the target signal component to solve the spectral overlap problem.

[0079] The processor extracts the image of the current slide from the image set as a training sample and trains a convolutional neural network model on the spot. The training data, the trained model, and the objects to be processed are all images of the same slide. The model learns the specific characteristics of the light intensity distribution, background, and noise of the slide.

[0080] The trained model performs inference on each image, outputting a mask image to separate the target signal from the background region. The mask is used to subtract the autofluorescence of the background region from the original image to preserve the target signal and handle irregularly distributed autofluorescence. Simultaneously, the model outputs light intensity distribution information during inference. The system generates an adjustment layer from this information and applies it to the background-removed image. This adjustment layer reflects the true brightness distribution caused by the light source spot, and after fusion with the image, it evens out the brightness without calibration plate deviation. The model also outputs noise distribution information, using noise features learned by the adversarial network to remove sensor and environmental noise from the image while preserving image details. After all images are processed, the system stitches them together image by image according to the row and column numbers in the naming rules. Because the brightness has been evened out and the background and noise have been removed before stitching, feature matching is more accurate and the stitching yield is higher, resulting in a more complete final output panoramic image of the fluorescent slide.

[0081] This embodiment uses spectral linear unmixing to initially separate the target signal from autofluorescence in the spectral dimension before processing with a dedicated model. This significantly reduces the remaining proportion of autofluorescence in the image input to the dedicated model. The background removal capabilities required by the dedicated model can then be focused more on processing other non-spectrally overlapping interferences. The extraction of light intensity distribution information and noise information is also less affected by spectral aliasing regions. As a result, the final output panoramic image achieves further improvement in signal purity.

[0082] Since the system described in Embodiment 5 of this application is a system used to implement the method of Embodiment 1 of this application, those skilled in the art can understand the specific structure and variations of the system based on the method described in Embodiment 1 of this application, and therefore will not be described again here. All systems used in the method of Embodiment 1 of this application fall within the scope of protection of this application.

[0083] This application provides an image processing apparatus for a fluorescent slide, the apparatus comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the image processing method for the fluorescent slide described in Embodiment 1 above.

[0084] The following is for reference. Figure 6 This document illustrates a schematic diagram of an image processing device suitable for implementing the embodiments of this application using a fluorescent glass slide. The image processing device for the fluorescent glass slide in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 6 The image processing device for the fluorescent slide shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0085] like Figure 6As shown, the image processing device for a fluorescent slide may include a processing unit 1001 (e.g., a core processor, a graphics processor, etc.) that can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the fluorescent slide image processing device. The processing unit 1001, the ROM 1002, and the RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the image processing device for the fluorescent slide to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows an image processing device for a fluorescent slide with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0086] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0087] The image processing device for fluorescent slides provided in this application, employing the image processing method for fluorescent slides in the above embodiments, can solve the technical problem of unstable processing results due to autofluorescence interference of fluorescent slides across different slides. Compared with the prior art, the beneficial effects of the image processing device for fluorescent slides provided in this application are the same as those of the image processing method for fluorescent slides provided in the above embodiments, and other technical features in this image processing device for fluorescent slides are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0088] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0089] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0090] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the image processing method for the fluorescent slide in the above embodiments.

[0091] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More 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 or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, radio frequency (RF), etc., or any suitable combination thereof.

[0092] The aforementioned computer-readable storage medium may be included in an image processing device for a fluorescent slide; or it may exist independently and not be assembled into an image processing device for a fluorescent slide.

[0093] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the image processing device for the fluorescent slide, the image processing device for the fluorescent slide causes the following: it acquires an image set of the fluorescent slide, uses the scanned images in the image set as training samples to train a preset model to obtain a dedicated model for the fluorescent slide, identifies the light intensity distribution information, background information, and noise information corresponding to the scanned images in the image set through the dedicated model, performs image calibration on the scanned images based on the light intensity distribution information, background information, and noise information to obtain a single field-of-view image, and stitches the single field-of-view images to generate a panoramic image of the fluorescent slide.

[0094] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0095] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation that may be implemented in systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0096] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0097] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described image processing method for fluorescent slides. This solves the technical problem of unstable processing results due to autofluorescence interference from fluorescent slides across different slides. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the image processing method for fluorescent slides provided in the above embodiments, and will not be repeated here.

[0098] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. An image processing method for a fluorescent glass slide, characterized in that, The method includes the following steps: The fluorescence emitted by the fluorescent slide after excitation is filtered by a narrow-band filter, and the filtered fluorescence is collected by a detector to obtain a scanned image, which is then stored in the image set of the fluorescent slide. Using the scanned images in the image set as training samples, a preset model is trained to obtain a dedicated model for the fluorescent slide. Specifically, the scanned images in the image set are input into the preset model in batches, and the predicted values ​​of light intensity distribution, background mask, and noise distribution output by the preset model are obtained. The error values ​​of the predicted values ​​of light intensity distribution, background mask, and noise distribution are calculated using a loss function. The weight parameters of each layer of the preset model are updated using a backpropagation algorithm to obtain the dedicated model. The dedicated model identifies the light intensity distribution, background information, and noise information corresponding to the scanned images in the image set. Specifically, a multispectral imaging system acquires the fluorescence intensity values ​​of each pixel on the fluorescent slide across multiple wavelength channels to obtain the complete emission spectrum curve of each pixel. A pre-built reference emission spectrum and autofluorescence reference spectrum of the target fluorescent dye are obtained. For each pixel, a linear mixture model is established, and the least squares method is used to solve the linear mixture model for each pixel to obtain the target signal intensity value and autofluorescence intensity value of that pixel. Based on the target signal intensity value of the pixel, the scanned image is reconstructed, and the reconstructed scanned image is input into the dedicated model. Based on the light intensity distribution information, the background information, and the noise information, the scanned image is calibrated to obtain a single field-of-view image; The single-view images are stitched together to generate a panoramic image of the fluorescent slide.

2. The image processing method for fluorescent slides as described in claim 1, characterized in that, Before the steps of filtering the fluorescence emitted by the fluorescent slide after excitation using a narrow-band filter, collecting the filtered fluorescence with a detector to obtain a scanned image, and storing the scanned image into the image set of the fluorescent slide, the method further includes: The fluorescent slide was pre-scanned to obtain the distribution of autofluorescence intensity of the fluorescent slide under different excitation wavelengths; Based on the autofluorescence intensity distribution, the autofluorescence excitation peak band of the fluorescent glass slide is determined; Select the working excitation wavelength based on the described spontaneous fluorescence excitation peak band; The fluorescent slide is scanned and excited based on the operating excitation wavelength.

3. The image processing method for a fluorescent slide as described in claim 1, characterized in that, Before the step of identifying the light intensity distribution information, background information, and noise information corresponding to the scanned images in the image set using the dedicated model, the method further includes: The radius of the rolling ball is determined based on the maximum diameter of the target signal in the scanned image and the rolling ball algorithm. Using the grayscale value of each pixel in the scanned image as the height, and based on the radius of the rolling ball, the sphere is simulated to roll on the grayscale surface of the scanned image. The pixel areas that the sphere cannot reach are identified as uniformly diffused background areas, and these uniformly diffused background areas are removed from the scanned image.

4. The image processing method for fluorescent slides as described in claim 1, characterized in that, Before the step of identifying the light intensity distribution information, background information, and noise information corresponding to the scanned images in the image set using the dedicated model, the method further includes: Obtain the point spread function corresponding to the scanned image; Using the point spread function as input, the scanned image is subjected to iterative deconvolution operations through a blind deconvolution algorithm.

5. The image processing method for a fluorescent slide as described in claim 1, characterized in that, The step of identifying the light intensity distribution information, background information, and noise information corresponding to the scanned images in the image set using the dedicated model includes: The scanned image is input into the dedicated model, which classifies each pixel of the scanned image and outputs a target signal mask. The identification information corresponding to each pixel in the target signal mask is used to identify whether the pixel belongs to the target signal region or the background region. The pixels in the target signal mask that are identified as belonging to the background region are used as the background information; The illumination intensity value corresponding to each pixel position of the scanned image output by the dedicated model is used as the illumination intensity distribution information, and the noise feature map of the scanned image is used as the noise information.

6. The image processing method for a fluorescent slide as described in claim 5, characterized in that, The step of performing image calibration on the scanned image based on the light intensity distribution information, the background information, and the noise information to obtain a single-field-of-view image includes: Based on the background information, the values ​​of pixels belonging to the background region in the scanned image are replaced with preset background values ​​to obtain an image after background removal; The light intensity distribution information is used to generate a light intensity adjustment layer. The light intensity adjustment layer is then fused pixel by pixel with the background-removed image to correct the brightness unevenness caused by light spot in the background-removed image, thus obtaining a light intensity-corrected image. Based on the noise information, noise stripping is performed on the light intensity corrected image to obtain the single-view image.

7. An image processing device for a fluorescent glass slide, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the image processing method for a fluorescent slide as described in any one of claims 1 to 6.

8. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the image processing method for a fluorescent slide as described in any one of claims 1 to 6.

Images