A method for preprocessing microscopic slice images and spatial multimodal data

By performing multi-dimensional quantitative quality inspection and constructing an adaptive joint loss function on microscopic slice images, the problems of low efficiency and large error in the registration of microscopic slice images with spatial multimodal data are solved, and high-precision, automated data registration and binding are achieved.

CN123115604APending Publication Date: 2026-11-17WUHAN XINYUNDAO DATA TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611625005.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-20
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, the registration of microscopic slice images with spatial multimodal data suffers from low efficiency and large errors. In particular, manual visual registration is inefficient and cannot effectively distinguish between high-resolution and flawed areas, leading to data analysis distortion.

Method used

By performing multi-dimensional quantitative quality inspection on microscopic slice images, a comprehensive image quality score and pixel-level defect mask map are generated to screen out invalid defect areas. An adaptive joint loss function is constructed for coordinate registration, thereby achieving accurate registration between the effective imaging area and structured multimodal data.

Benefits of technology

It achieves high-precision, automated, and accurate registration of microscopic slice images with spatial multimodal data, eliminating the interference of imaging defects on data analysis and improving the accuracy and efficiency of data analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115604A_ABST
    Figure CN123115604A_ABST
Patent Text Reader

Abstract

This invention provides a method for preprocessing microscopic section images and spatial multimodal data, comprising: performing multi-dimensional quantitative quality inspection on the microscopic section images to generate an overall image quality score and a pixel-level defect mask; filtering out invalid defect regions in the microscopic section images based on the pixel-level defect mask to extract effective imaging regions; adaptively configuring the weighting coefficients of multimodal molecular feature matching terms and spatial topological continuity constraint terms based on the overall image quality score to construct a joint loss function; performing coordinate registration between the pixel coordinate system of the effective imaging regions and the sequencing site coordinate system of the structured multimodal data based on the joint loss function to obtain a coordinate mapping relationship; and associating and binding the tissue morphology information of the effective imaging regions with the molecular expression information of the structured multimodal data based on the coordinate mapping relationship to generate a target structured multimodal dataset. This invention can achieve accurate registration between microscopic section images and spatial multimodal data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of spatial multi-omics analysis technology, and in particular to a method for preprocessing microscopic slice images and spatial multimodal data. Background Technology

[0002] In spatial multi-omics analysis, tissue microscopic section images and structured multimodal data are typically relied upon. Microscopic imaging is the core pre-acquisition process. Due to factors such as slide preparation process, equipment focusing, and lighting conditions, microscopic section images are highly susceptible to defects such as blurring, wrinkles, reflections, and artifacts. If the inherent errors at the imaging end are ignored and data analysis is carried out directly, it can easily lead to distortion in subsequent multimodal molecular feature analysis from the source.

[0003] To establish the association between microscopic slide images and spatial multimodal sequencing data, existing techniques typically rely on manual visual registration for calibration, or use fixed-weight algorithms to perform overall coordinate alignment and correction of the image and text data during the post-processing stage.

[0004] However, manual registration in batch processing is inefficient and prone to human error. Furthermore, existing technologies lack a quality control step for microscopic images, and the standardized processing logic fails to differentiate between high-resolution and defective areas, making it impossible to compensate for data errors caused by imaging defects and easily leading to feature matching mismatches. Therefore, to eliminate the interference of microscopic imaging defects on data analysis, achieving accurate registration between microscopic slice images and spatial multimodal data has become an urgent problem to be solved in this field. Summary of the Invention

[0005] This invention provides a method for preprocessing microscopic slice images and spatial multimodal data to solve the technical problem of how to achieve accurate registration of microscopic slice images and spatial multimodal data.

[0006] This invention provides a method for preprocessing microscopic slice images and spatial multimodal data, including: Multi-dimensional quantitative quality inspection is performed on the microscopic slice images to be processed to generate a comprehensive image quality score and a pixel-level defect mask map. Based on the pixel-level defect mask image, invalid defect areas in the microscopic section image are screened out, and effective imaging areas are extracted; Based on the comprehensive image quality score, the weighting coefficients of the multimodal molecular feature matching term and the spatial topological continuity constraint term are adaptively configured to construct a joint loss function; Based on the joint loss function, the pixel coordinate system of the effective imaging region is registered with the sequencing site coordinate system of the structured multimodal data to obtain the coordinate mapping relationship; Based on the coordinate mapping relationship, the tissue morphology information of the effective imaging region is associated and bound with the molecular expression information of the structured multimodal data to generate a target structured multimodal dataset.

[0007] According to the microscopic section image and spatial multimodal data preprocessing method provided by the present invention, multi-dimensional quantitative quality inspection is performed on the microscopic section image to be processed to generate a comprehensive image quality score and a pixel-level defect mask map, including: The microscopic slice images are subjected to clarity quantification, grayscale uniformity quantification, tissue integrity quantification, and artifact interference quantification respectively, generating four corresponding individual quality scores; The four individual quality scores are weighted and summed to generate the overall image quality score; Based on the quantitative detection results of the four dimensions, the blurry, reflective, blank, wrinkled and artifact defect areas are classified and labeled respectively, and the pixel-level defect mask map labeled with the invalid defect area and the effective imaging area is generated.

[0008] According to the microscopic section image and spatial multimodal data preprocessing method provided by the present invention, the sharpness measurement of the microscopic section image is calculated to generate a single-item sharpness quality score, including: The microscopic slice image is convolved with a Laplacian operator, and the pixel variance of the convolution result is calculated. The pixel variance is mapped to the sharpness single-item quality score according to the hierarchical mapping rule; The blurred defect areas of the microscopic section image are classified and labeled, including: Image regions with pixel variance below a set lower threshold are marked as blurry defect regions.

[0009] According to the microscopic section image and spatial multimodal data preprocessing method provided by the present invention, grayscale uniformity quantification calculation is performed on the microscopic section image to generate a single-item quality score for grayscale uniformity, including: The microscopic slice image is divided into image sub-blocks of fixed size and non-overlapping, and the mean gray value and standard deviation of the pixels in each image sub-block are calculated. Calculate the global dispersion of the mean gray level of all image sub-blocks in the entire image, and obtain the gray level uniformity quality score based on the global dispersion. The reflective defect areas of the microscopic section image are classified and labeled, including: Image sub-blocks whose local gray-scale mean deviates from the global gray-scale mean by more than a set deviation threshold are marked as reflective defect areas.

[0010] According to the microscopic section image and spatial multimodal data preprocessing method provided by the present invention, the tissue integrity of the microscopic section image is quantitatively calculated to generate a single quality score for tissue integrity, including: The Otsu threshold segmentation algorithm was used to perform binary segmentation on the microscopic section image to extract the connected components of the foreground tissue. The area of ​​the connected components of the foreground organization and the percentage of effective organization are statistically analyzed, and the organizational integrity single quality score is obtained by mapping the percentage of effective organization. The blank and wrinkled defect areas of the microscopic section image are classified and labeled, including: Regions whose local connected area is lower than a set area threshold and whose background proportion exceeds a set ratio threshold are marked as blank defect regions. Based on morphological opening and closing operations, narrow, twisted connected regions formed by tissue compression and folding are identified and marked as wrinkle defect regions.

[0011] According to the microscopic section image and spatial multimodal data preprocessing method provided by the present invention, the microscopic section image is subjected to artifact interference quantification calculation to generate a single-item quality score for artifact interference degree, including: The high-frequency noise component is extracted by high-pass filtering of the microscopic slice image, and the local high-frequency noise energy is statistically analyzed. Morphological filtering is used to extract strip and spot-like abnormal texture features from the image; The artifact interference level single-item quality score is generated based on the abnormal texture distribution mapping; The artifact defect regions of the microscopic section images are classified and labeled, including: Regions with local high-frequency noise energy exceeding a set energy threshold or containing the aforementioned striped and speckled abnormal texture features are marked as artifact defect regions.

[0012] According to the microscopic slice image and spatial multimodal data preprocessing method provided by the present invention, based on the comprehensive image quality score, the weighting coefficients of the multimodal molecular feature matching term and the spatial topological continuity constraint term are adaptively configured to construct a joint loss function, including: Construct a loss function structure that includes a weighted multimodal molecular feature matching term and a weighted spatial topological continuity constraint term; The molecular feature analysis weights corresponding to the multimodal molecular feature matching terms and the topological constraint strengths corresponding to the spatial topological continuity constraint terms are calculated based on the overall image quality score of the sub-region; the sum of the molecular feature analysis weights and the topological constraint strengths is equal to 1. When the overall image quality score increases, the weight of the molecular feature analysis is increased while the strength of the topological constraint is decreased simultaneously. When the overall image quality score decreases, the weight of the molecular feature analysis is reduced while the strength of the topological constraint is increased simultaneously to complete the construction of the joint loss function.

[0013] According to the microscopic slice image and spatial multimodal data preprocessing method provided by the present invention, the pixel coordinate system of the effective imaging region and the sequencing site coordinate system of the structured multimodal data are registered according to the joint loss function to obtain a coordinate mapping relationship, including: A global affine rigid registration algorithm is used to coarsely align the effective imaging region with the structured multimodal data, eliminating global translation, rotation, and scaling biases between the pixel coordinate system and the sequencing site coordinate system; Based on the tissue contour, the coarsely aligned effective imaging area is divided into multiple image sub-regions. The tissue morphology edges and high-expression marker gene sequencing sites in each image sub-region are extracted as dual anchor points, and spatial continuity constraints between adjacent image sub-regions are constructed. The local offset of each image sub-region is solved one by one based on the joint loss function, and the coordinate mapping between adjacent image sub-regions is smoothly fused to correct the imaging cropping deviation and local position offset, thereby obtaining the coordinate mapping relationship.

[0014] The microscopic section image and spatial multimodal data preprocessing method provided by the present invention further includes, before performing multidimensional quantitative quality inspection on the microscopic section image to be processed: The structured multimodal data and the microscopic slice images in lossless image format are accessed after the experiment. The structured multimodal data includes a standard spatial coordinate matrix and an expression matrix of at least one modality among transcriptome, protein, epigenetics and metabolome, and the accessed data does not include raw sequencing read data. The accessed microscopic section image is used as the microscopic section image to be processed.

[0015] According to the microscopic section image and spatial multimodal data preprocessing method provided by the present invention, the tissue morphology information of the effective imaging region is associated and bound with the molecular expression information of the structured multimodal data based on the coordinate mapping relationship to generate a target structured multimodal dataset, including: The tissue morphology information, after coordinate matching is completed, is written into the underlying data structure corresponding to the structured multimodal data; By maintaining the spatial neighborhood topology of each sequencing site, the target structured multimodal dataset containing complete spatial topology information and multimodal association features is generated.

[0016] The microscopic section image and spatial multimodal data preprocessing method provided by this invention performs multi-dimensional quantitative quality inspection on the microscopic section image to be processed, generating an image comprehensive quality score and a pixel-level defect mask map. Based on this, invalid defect areas are screened out and effective imaging areas are extracted. An additional pre-microscopic imaging quality control step is added, effectively avoiding the feature matching misalignment problem caused by the analysis of mixed high and low quality data due to the lack of quality inspection. By adaptively configuring the weighting coefficients of multimodal molecular feature matching terms and spatial topological continuity constraint terms according to the image comprehensive quality score, a joint loss function is constructed, introducing a quality-driven dynamic weighting mechanism, which effectively distinguishes high-definition effective areas from defective areas and compensates for data errors caused by imaging defects. Then, the pixel coordinate system of the effective imaging area is automatically registered with the sequencing site coordinate system through the joint loss function to obtain the coordinate mapping relationship, and realize the association and binding of morphological information and molecular expression information. This overcomes the shortcomings of traditional manual visual registration, which is inefficient and prone to human error, thereby eliminating the interference of microscopic imaging defects on data analysis and achieving high-precision, automated, accurate registration and reliable binding of microscopic section images and spatial multimodal data. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is one of the flowcharts illustrating the microscopic slice image and spatial multimodal data preprocessing method provided by the present invention.

[0019] Figure 2 This is the second flowchart of the microscopic slice image and spatial multimodal data preprocessing method provided by the present invention.

[0020] Figure 3 This is a schematic diagram of the structure of the microscopic slice image and spatial multimodal data preprocessing device provided by the present invention.

[0021] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] The following is combined Figures 1 to 4 This invention describes a method for preprocessing microscopic slice images and spatial multimodal data.

[0024] Figure 1 This is a flowchart illustrating the microscopic section image and spatial multimodal data preprocessing method provided by the present invention, as shown below. Figure 1 As shown, the method includes, but is not limited to, steps S1, S2, S3, S4 and S5.

[0025] Step S1: Perform multi-dimensional quantitative quality inspection on the microscopic slice image to be processed, and generate an overall image quality score and a pixel-level defect mask map.

[0026] Microscopic section images are digital image data acquired by pre-positioning in-situ imaging of biological tissue sample sections using optical microscopic imaging equipment in the context of multi-omics analysis of biological tissues. The image format can specifically include lossless image files such as Tagged Image File Format (TIFF) or Portable Network Graphics (PNG).

[0027] Multi-dimensional quantitative quality inspection overcomes the shortcomings of traditional post-processing, which is disconnected from the front-end microscopic imaging stage. It quantitatively detects, calculates, and scores microscopic section images from multiple distinct physical imaging quality characterization dimensions. The overall image quality score characterizes the overall imaging quality of a local area or the entire microscopic section image. The pixel-level defect mask map is a two-dimensional mask data matrix with a resolution corresponding to the microscopic section image to be processed, assigning corresponding classification information to the region category (such as various defect regions or effective regions) of each pixel at the pixel matrix level.

[0028] Step S2: Based on the pixel-level defect mask image, invalid defect areas in the microscopic section image are screened out, and the effective imaging area is extracted.

[0029] Invalid defect areas are image sub-regions in microscopic section images that have poor imaging quality and cannot provide reliable biological tissue morphology comparison information due to factors such as slide preparation process, equipment focusing, lighting conditions, or external pollution. They can include one or more combinations of wrinkled areas, blurred areas, reflective areas, blank areas, and artifact areas.

[0030] The effective imaging region is the sub-region of a microscopic section image that, excluding invalid defect areas, has clear and reliable tissue morphology and optical characteristics and can be used for subsequent multimodal correlation analysis.

[0031] The pixel labels in the pixel-level defect mask generated in step S1 can be parsed, and the image space covered by all pixels marked as invalid defect categories (such as wrinkles, blur, reflection, blank, etc.) can be filtered, removed or masked, and only the high-quality pixel set not covered by defect labels can be retained, thereby extracting the effective imaging area for downstream spatial omics analysis and reducing abnormal data interference from the data source level.

[0032] Step S3: Based on the overall image quality score, adaptively configure the weighting coefficients of the multimodal molecular feature matching term and the spatial topological continuity constraint term to construct a joint loss function.

[0033] The multimodal molecular feature matching term is an objective function term used to characterize the degree of feature fitting deviation between the multimodal molecular expression features and image morphological features of sequencing sites (Spot, i.e., sequencing capture sites on spatial multi-omics sequencing chips) during the spatial matching process.

[0034] The spatial topological continuity constraint term is an objective function term used to characterize the degree of deviation of adjacent sequencing sites from the original spatial geometry and neighborhood topological relationships of biological tissues during the registration spatial transformation process.

[0035] Based on the calculated overall image quality score, the weighting coefficient of the multimodal molecular feature matching term can be increased for high-quality imaging regions to fully utilize high-fidelity molecular features. For regions with imaging defects or weak signals, the weighting coefficient of the spatial topological continuity constraint term can be increased to strengthen spatial geometric prior constraints, thereby constructing a joint loss function that takes into account both molecular feature matching and spatial topological fidelity.

[0036] Step S4: Based on the joint loss function, the pixel coordinate system of the effective imaging region is registered with the sequencing site coordinate system of the structured multimodal data to obtain the coordinate mapping relationship.

[0037] The pixel coordinate system is an image plane coordinate system established based on the row and column grid of the two-dimensional discrete pixel array of a microscopic slice image.

[0038] Structured multimodal data is a collection of data that is structured and encapsulated using the standard Annotated Data (AnnData) data structure, and includes spatial physical sequencing coordinates and multimodal omics molecular expression matrices.

[0039] The sequencing site coordinate system is a two-dimensional or three-dimensional geometric coordinate system in which each sequencing physical capture site (Spot) on a spatial multi-omics sequencing vector (such as a sequencing chip) is located in a microscopic physical array.

[0040] The coordinate mapping relationship is a mathematical function for spatial position transformation between the pixel coordinate system of the microscopic slice image and the coordinate system of the multimodal sequencing site, such as a mapping model that includes rotation, translation, scaling and local nonlinear displacement transformation.

[0041] The joint loss function can be minimized as the optimization objective. The spatial transformation parameters between the pixel coordinate system of the effective imaging region and the sequencing site coordinate system of the structured multimodal data are automatically calculated iteratively. Coordinate matching is completed region by region. Spatial coordinate offsets caused by inconsistent imaging cropping, field of view placement angles and scales are automatically corrected. The bidirectional coordinate mapping relationship between the two coordinate systems is output.

[0042] Step S5: Based on the coordinate mapping relationship, the tissue morphology information of the effective imaging area is associated and bound with the molecular expression information of the structured multimodal data to generate the target structured multimodal dataset.

[0043] The tissue morphology information consists of morphological visual feature data such as grayscale, edge contour, cell texture, and microscopic phenotype of the microscopic slice image extracted within the effective imaging area.

[0044] Molecular expression information refers to the quantitative expression profiles of multi-omics molecules, such as transcriptome, proteome, epigenome, or metabolome, carried by each sequencing site (Spot) in structured multimodal data.

[0045] The target structured multimodal dataset is a standardized output dataset that uniformly encapsulates and integrates spatial multimodal molecular expression matrices, spatial geometric coordinates, and corresponding registered and bound microscopic image tissue morphology features.

[0046] The calculated coordinate mapping relationship can be used to accurately map the pixel-level morphological information of the effective imaging area to the corresponding sequencing site (Spot), thereby completing the cross-modal data binding of microscopic image morphological features and omics molecular features, and generating a target structured multimodal dataset that conforms to the standard structured format.

[0047] As described above, this invention performs multi-dimensional quantitative quality inspection on the microscopic section images to be processed, generating a comprehensive image quality score and a pixel-level defect mask map. Based on this, invalid defect areas are screened out and effective imaging areas are extracted. This adds a pre-microscopic imaging quality control step, effectively avoiding the feature matching misalignment problem caused by the mixed analysis of high and low quality data due to the lack of quality inspection. By adaptively configuring the weighting coefficients of multimodal molecular feature matching terms and spatial topological continuity constraint terms according to the comprehensive image quality score to construct a joint loss function, a quality-driven dynamic weighting mechanism is introduced, effectively distinguishing between high-definition effective areas and defective areas, and compensating for data errors caused by imaging defects. Furthermore, the joint loss function automatically registers the pixel coordinate system of the effective imaging area with the sequencing site coordinate system to obtain the coordinate mapping relationship, and realizes the association and binding of morphological information and molecular expression information. This overcomes the shortcomings of traditional manual visual registration, which is characterized by low efficiency and large human error, thereby eliminating the interference of microscopic imaging defects on data analysis and achieving high-precision, automated, accurate registration and reliable binding of microscopic section images and spatial multimodal data.

[0048] In one embodiment, step S1 may further include: The image was subjected to three separate quantifications: clarity, grayscale uniformity, tissue integrity, and artifact interference. These quantifications were then used to generate four individual quality scores. The four individual quality scores are weighted and summed to generate a comprehensive image quality score. Based on the quantitative detection results of the four dimensions, the blurry, reflective, blank, wrinkled and artifact defect areas are classified and labeled, and pixel-level defect mask maps with invalid defect areas and valid imaging areas are generated.

[0049] Clarity quantification is used to quantitatively evaluate the focusing accuracy and edge texture detail richness of section microscopic imaging. Gray-scale uniformity quantification is used to quantitatively evaluate the illumination uniformity and local strong light reflection within the entire section's field of view. Tissue integrity quantification is used to quantitatively evaluate the effective biological tissue coverage ratio and physical deformation integrity of the section. Artifact interference quantification is used to quantitatively evaluate the interference degree of noise textures such as external impurities, contaminant particles, and mechanical scratches in the image.

[0050] After calculation for each dimension, the corresponding individual quality score is output, and the value range of each individual quality score is standardized and mapped to 0 to 100. Then, a preset weight coefficient is assigned to the sharpness score, grayscale uniformity score, tissue integrity score, and artifact interference score (the sum of the weights of each individual item is 1). The overall image quality score between 0 and 100 is calculated by weighted linear summation.

[0051] Based on the pixel-by-pixel or local block-by-local block judgment results of four dimensions—clarity, grayscale uniformity, tissue integrity, and artifact interference—a spatial multi-class mask matrix can be established. Different classification label codes are assigned to blur defects caused by defocus, reflection defects caused by strong light illumination, blank defects caused by lack of sample coverage, wrinkle defects formed by slicing compression, and artifact defects formed by impurity scratches. Areas not covered by any defect label are marked as valid imaging area labels, thereby generating a pixel-level defect mask map.

[0052] This invention performs comprehensive quantitative detection and weighted summation from four complementary dimensions: sharpness, grayscale uniformity, tissue integrity, and artifact interference. This enables a comprehensive, objective, and quantitative characterization of the overall imaging quality of microscopic section images. Simultaneously, by using multi-dimensional judgment results, it provides multi-category pixel-level fine-grained labeling for defects such as blur, reflection, blanking, wrinkles, and artifacts, achieving precise spatial positioning and refined classification of various physical defects in microscopic sections.

[0053] In one embodiment, performing a sharpness quantification calculation on a microscopic section image to generate a single-item sharpness quality score may include: Perform Laplacian convolution on the microscopic slice image and calculate the pixel variance of the convolution result; The pixel variance is mapped to a single quality score for sharpness according to the hierarchical mapping rule; Classifying and labeling blurred defect areas in microscopic section images can include: Image regions with pixel variance below a set lower threshold are marked as blurry defect regions.

[0054] The Laplacian variance algorithm can be used to analyze the sharpness of microscopic section images. Specifically, a convolution filtering operation is performed on the microscopic section image using the discrete Laplacian differential operator to extract the second-order high-frequency edge derivatives of the image. Then, the pixel variance of the convolution result for the entire image or a local sub-block is calculated. A higher pixel variance value indicates richer and sharper microscopic edge contours and higher focus sharpness in the image; conversely, a lower variance value indicates smoother and blurrier image edges and more severe defocusing. Based on a preset hierarchical threshold mapping rule, the calculated Laplacian pixel variance can be mapped to a range of 0 to 100 points, outputting a single-item sharpness quality score.

[0055] A lower limit threshold for sharpness variance can be preset. During pixel-level detection, the Laplacian convolution variance of a local image window is calculated. When the calculated pixel variance of a local area is lower than the preset lower limit threshold, the local area is determined to be out of focus or blurred, and the corresponding pixels in the pixel-level defect mask are uniformly marked as blur defect classification labels.

[0056] This invention utilizes the high sensitivity of the Laplacian operator to high-frequency edge information to achieve an objective quantitative assessment of the sharpness of microscopic slice images by calculating the variance of convolutional pixels. Furthermore, it uses a lower limit threshold of variance to achieve pixel-level marking of local out-of-focus and blurry defect areas, ensuring the reliability of filtering out blurry and low-quality areas.

[0057] In one embodiment, performing grayscale uniformity quantification calculation on a microscopic section image to generate a single-item quality score for grayscale uniformity may include: The microscopic slice image is divided into image sub-blocks of fixed size and non-overlapping, and the mean gray value and standard deviation of the pixels in each image sub-block are calculated. Calculate the global dispersion of the mean gray level of all image sub-blocks in the entire image, and obtain the single quality score of gray level uniformity based on the global dispersion. Classify and label reflective defect areas in microscopic section images, including: Image sub-blocks whose local gray-scale mean deviates from the global gray-scale mean by more than a set deviation threshold are marked as reflective defect areas.

[0058] A block-based grayscale statistical and standard deviation calculation method can be used. First, the entire microscopic section image is uniformly divided into multiple image sub-blocks of a preset fixed size (e.g., N×N pixels) that do not overlap spatially. Each sub-block is traversed, and the mean grayscale value and standard deviation of each pixel are statistically analyzed. The dispersion of the mean grayscale value of all sub-blocks relative to the global mean grayscale value is calculated, such as the variance or standard deviation of the sub-block mean. A greater degree of grayscale dispersion between sub-blocks indicates uneven microscope illumination or the presence of strong local reflections. Based on this global dispersion, it is inversely mapped to a grayscale uniformity quality score from 0 to 100; the smaller the dispersion, the closer the score is to 100.

[0059] Local grayscale deviation thresholds can be set. In block-by-block detection, when the absolute value of the difference between the internal grayscale mean of a certain local image sub-block and the global grayscale mean of the whole image exceeds the set deviation threshold (manifested as an abnormally bright area), it is determined that the sub-block is affected by strong light reflection or optical overexposure interference. In the pixel-level defect mask map, all pixels covered by the image sub-block are marked as reflective defect classification labels.

[0060] This invention combines local sub-block gridding statistics with full-image grayscale mean dispersion analysis to objectively quantify the uniformity of optical illumination across the entire field of view. It can also accurately detect areas of strong light reflection caused by microscope light source reflection or local tissue specular reflection, achieving pixel-level precise calibration of reflective defects.

[0061] In one embodiment, performing tissue integrity quantification calculations on microscopic slice images to generate a single-item tissue integrity quality score may include: The Otsu threshold segmentation algorithm was used to perform binary segmentation on microscopic slide images and extract connected components of foreground tissue. The area of ​​the connected domains of the statistical prospective organization and the percentage of effective organizations are calculated, and the organizational integrity score is obtained by mapping the percentage of effective organizations. Classifying and labeling blank and wrinkled defect areas in microscopic section images can include: Regions whose local connected area is lower than a set area threshold and whose background proportion exceeds a set ratio threshold are marked as blank defect regions. Based on morphological opening and closing operations, narrow, twisted connected regions formed by tissue compression and folding are identified and marked as wrinkle defect regions.

[0062] The Otsu thresholding algorithm (also known as the maximum inter-class variance adaptive binarization segmentation method) can be used to binarize microscopic section images, adaptively segmenting the image into foreground tissue regions and background blank regions, and extracting connected components of the foreground tissue. The geometric area of ​​the extracted foreground tissue connected components and the effective tissue area ratio within the entire section's field of view are then statistically analyzed. Based on the effective tissue ratio, a single-item quality score of 0 to 100 for tissue integrity is output according to a mapping function. The higher the effective tissue ratio conforms to the experimental standards, the higher the score.

[0063] A connected component area threshold and a background ratio threshold can be set. When the area of ​​the tissue connected component in a local field of view of a slice is lower than the set connected component area threshold, and the corresponding binary background ratio exceeds the set background ratio threshold, the area is determined to be an invalid background area lacking biological sample coverage, and is marked as a blank defect classification label in the pixel-level defect mask image.

[0064] Morphological opening operations (erosion followed by dilation to eliminate fine burrs) and closing operations (dilation followed by erosion to fill pores) can be performed on the extracted foreground tissue connected regions. This allows for the identification and extraction of narrow, twisted morphological connected regions formed by tissue compression and overlapping of sections during slide preparation. The corresponding pixels are then marked as wrinkle defect classification labels in the pixel-level defect mask image.

[0065] This invention utilizes Otsu threshold binary segmentation to adaptively separate effective foreground tissue from blank background. Combined with connected component morphology analysis and morphological opening and closing operations, it can not only quantify the effective tissue ratio of the slice and accurately remove blank background without samples, but also effectively identify and mark narrow, twisted, and wrinkled defect areas caused by slide extrusion, thereby improving the accuracy of physical morphological quality screening of the slice.

[0066] In one embodiment, the process of quantifying artifact interference in a microscopic slice image and generating a single-item quality score for artifact interference may include: High-pass filtering was applied to the microscopic slice image to extract high-frequency noise components, and the local high-frequency noise energy was statistically analyzed. Morphological filtering is used to extract strip and spot-like abnormal texture features from the image; A single quality score for artifact interference is generated based on the mapping of abnormal texture distribution. Classifying and labeling artifact defect regions in microscopic section images can include: Regions with local high-frequency noise energy exceeding a set energy threshold or containing striped or speckled abnormal texture features are marked as artifact defect regions.

[0067] High-pass filtering and convolution operations can be performed on microscopic section images to remove low-frequency tissue background, extract high-frequency noise components, and statistically analyze the distribution intensity of high-frequency noise energy according to local windows. Simultaneously, morphological filtering is performed on the image using specific directional linear or circular structuring elements to extract abnormal non-inherent tissue texture features that appear as elongated stripes (such as slicing marks or slide scratches) and spots (such as impurities, dust, or dirt). Based on the local high-frequency energy intensity and the density of the extracted abnormal texture distribution, a single-item quality score of artifact interference from 0 to 100 is generated. The smaller the artifact interference, the closer the score is to 100.

[0068] A local high-frequency noise energy threshold can be set. When the high-frequency noise energy in a local image area exceeds the set energy threshold, or when abnormal textures such as long strip-shaped scratches and spot-like stains extracted by morphological filtering are detected, it is determined that there is an external physical artifact in the area, and the pixels in the area are marked as artifact defect classification labels in the pixel-level defect mask map.

[0069] This invention combines high-pass filtering with high-frequency energy detection and morphological anomalous texture filtering to sensitively capture microscopic scratches, dust contamination, and inherent impurities of non-biological tissues on microscopic sections, achieving quantitative evaluation of the degree of artifact interference and automated pixel-level precise calibration of artifact defect areas.

[0070] In one embodiment, step S3 may include: Construct a loss function structure that includes a weighted multimodal molecular feature matching term and a weighted spatial topological continuity constraint term; The molecular feature analysis weights corresponding to the multimodal molecular feature matching terms and the topological constraint strengths corresponding to the spatial topological continuity constraint terms are calculated based on the overall image quality score of the sub-regions; the sum of the molecular feature analysis weights and the topological constraint strengths equals 1. When the overall image quality score increases, the weight of molecular feature analysis is increased while the strength of topological constraints is simultaneously reduced. When the overall image quality score decreases, the weight of molecular feature analysis is reduced while the strength of topological constraints is increased simultaneously to complete the construction of the joint loss function.

[0071] The joint loss function (total loss function), which includes a weighted multimodal molecular feature matching term and a weighted spatial topological continuity constraint term, is mathematically expressed as follows: L total =w f (Q)·L feat +w t (Q)·L topo ; The algorithm minimizes L through iterative optimization. total Solve for the registration parameters.

[0072] L feat This represents a multimodal molecular feature matching term, used to quantify the cross-modal feature matching bias between the multimodal molecular expression features and image morphological features of a sequencing site (Spot); w f L represents the weights for multimodal molecular feature analysis, used to control the proportion of the optimization contribution of multimodal molecular features in the registration objective function; topo This represents a spatial topological continuity constraint term, used to quantify the degree of deviation and distortion of adjacent sequencing sites (Spots) from their original spatial topological geometry after registration deformation; w t (Q) represents the spatial topological continuity constraint strength, which is used to control the constraint strength of the spatial topological prior geometry on the registration result; Q represents the overall image quality score calculated from the current sub-region of the image to be registered, with a score between 0 and 100.

[0073] In registration optimization, the weighting coefficient w f (Q) and w t (Q) Instead of using a fixed constant for the entire image, it is dynamically calculated based on the overall image quality score Q of each sub-region, and both strictly satisfy the normalization constraint at any time: w f (Q)+w t (Q)=1.

[0074] When the sub-region imaging quality is good and the overall image quality score Q increases, the weight of molecular feature analysis can be increased. f (Q), the preferred calculation result is between 0.7 and 0.8; at the same time, the topological constraint strength w t (Q) Synchronous reduction, preferably with a calculation result of 0.2 to 0.3, so that the registration process preferentially depends on high-confidence molecular features.

[0075] When a sub-region has minor imaging defects or a weak signal, resulting in a lower overall image quality score Q, the weight of molecular feature analysis can be reduced. f (Q), for example, gradually decreasing from 0.7 to 0.3; and simultaneously increasing the spatial topological constraint strength w.t (Q), for example, gradually increases from 0.3 to 0.7, effectively suppressing the propagation of analysis errors caused by image quality defects by strengthening the prior constraints of spatial neighborhood topology geometry.

[0076] This invention achieves an adaptive dynamic balance between molecular feature fitting and spatial topological continuity by constructing a joint loss function that satisfies normalization constraints and dynamically adjusts the molecular feature weights and topological constraint strengths inversely with the local imaging comprehensive quality score. In high-quality areas, the molecular feature weights are increased to fully leverage the guiding role of high-fidelity expression spectra, while in defective areas, the rigid constraints of spatial topology are strengthened to suppress fitting distortion caused by imaging defects, thus significantly improving the robustness and accuracy of image-text registration.

[0077] In one embodiment, step S4 may include: A global affine rigid registration algorithm is used to coarsely align the effective imaging region with the structured multimodal data, eliminating global translation, rotation and scaling biases between the pixel coordinate system and the sequencing site coordinate system; Based on the tissue contour, the coarsely aligned effective imaging area is divided into multiple image sub-regions. The tissue morphology edges and high-expression marker gene sequencing sites in each image sub-region are extracted as dual anchor points, and spatial continuity constraints between adjacent image sub-regions are constructed. The local offset of each image sub-region is solved one by one by relying on the joint loss function, and the coordinate mapping between adjacent image sub-regions is smoothly fused to correct the imaging cropping deviation and local position offset, thus obtaining the coordinate mapping relationship.

[0078] For the extracted effective imaging region image and structured multimodal data, a global affine rigid registration algorithm can be used to perform macroscopic scale spatial transformation calculations; solve the global translation vector, global rotation angle and global scaling factor between the microscopic image and the spatial multimodal sequencing chip grid, eliminate the global translation, rotation and scaling deviations caused by the tilt of the slice placement angle, macroscopic translation and field of view scale differences, and achieve coarse alignment between the pixel coordinate system and the sequencing site coordinate system.

[0079] Based on the macroscopic tissue contour boundaries in the microscopic section images, the coarsely aligned effective imaging area is divided into multiple local image sub-regions by meshing. Within each image sub-region, the microscopic morphological edge features of the tissue in the section image and the sequencing sites (Spots) corresponding to the highly expressed marker genes in the multimodal data are extracted as spatial dual anchor points. Spatial neighborhood coordinate continuity constraints are constructed between spatially adjacent image sub-regions.

[0080] Based on the joint loss function, the local spatial nonlinear offset is solved independently and iteratively for each divided image sub-region to correct field-of-view cropping deviation, local distortion and local position misalignment in the microscopic imaging process; coordinate mapping smoothing and fusion calculation is performed in the boundary and overlapping transition area of ​​adjacent image sub-regions to eliminate splicing jumps between sub-regions and obtain a bidirectional coordinate mapping relationship between the globally continuous and locally accurate pixel coordinate system and the sequencing site coordinate system.

[0081] This invention eliminates large-scale global rigid translation, rotation and scaling errors by using global affine coarse alignment and sub-region fine alignment based on dual anchor points of tissue edges and highly expressed marker genes. It can also finely correct local cropping and field of view misalignment deviations in microscopic imaging by relying on dual anchor points and neighborhood space continuity constraints. With the help of sub-region smooth fusion, it significantly improves the accuracy and continuity of cross-modal coordinate mapping of images and text.

[0082] In one embodiment, before performing multi-dimensional quantitative quality inspection on the microscopic section image to be processed, the present invention may further include: The accessed data consisted of structured multimodal data and microscopic slice images in lossless image format. The structured multimodal data included a standard spatial coordinate matrix and an expression matrix of at least one modality from the transcriptome, protein, epigenetics, and metabolome. The accessed data did not include raw sequencing reads. The received microscopic section image is used as the microscopic section image to be processed.

[0083] Before performing multi-dimensional quantitative quality inspection on the microscopic slide images to be processed, this invention accesses the standardized dataset generated in the post-experiment stage. The overall preprocessing process only processes the standardized multimodal structured data and non-destructive microscopic slide images generated in the post-experiment stage, without accessing or processing the raw sequencing read data generated after the sequencer is removed, i.e., FASTQ format files. FASTQ is a basic text format for storing biological sequencing sequence information and corresponding sequencing quality scores.

[0084] The accessed structured multimodal data is encapsulated in the standard AnnData format. Its internal structure includes a standard spatial coordinate matrix for recording the physical location of each sequencing site, as well as an expression quantification matrix selected from at least one molecular modality among transcriptomics, proteomics, epigenetics, and metabolomics. The accessed microscopic slide images are in lossless image formats such as TIFF or PNG and are used as microscopic slide images to be processed in subsequent quality control processes.

[0085] This invention significantly reduces the computational and storage resource overhead of the data preprocessing stage by directly accessing standardized structured multimodal data and non-destructive microscopic slice images after the experiment, and strictly does not rely on the large and redundant raw FASTQ sequencing data.

[0086] In one embodiment, step S5 may include: The tissue morphology information after coordinate matching is written into the underlying data structure corresponding to the structured multimodal data; By preserving the spatial neighborhood topology of each sequencing site, a target structured multimodal dataset containing complete spatial topology information and multimodal association features is generated.

[0087] Based on the calculated coordinate mapping relationship, the morphological feature information of the microscopic slice tissue (including the microscopic contour, grayscale features, histological mask and morphological phenotypic parameters of the effective imaging area) that has completed spatial coordinate matching can be directly written and stored into the underlying observation attribute matrix or unstructured metadata annotation area corresponding to the structured multimodal data (AnnData).

[0088] During the writing process, the inherent neighborhood geometric topological connections of each sequencing site (Spot) in the in situ microscopic space of biological tissue are fully preserved, and the target structured multimodal dataset file is output, which fully encapsulates image tissue morphology information, multi-omics molecular expression matrix and spatial topological relationships.

[0089] This invention generates a target multimodal dataset that highly integrates morphology and omics and is compatible with standard data structures by directly writing the spatially matched tissue morphology information into the underlying data structure of structured multimodal data and fully preserving the spatial neighborhood topology of sequencing sites during output. This provides a high-fidelity data carrier for downstream spatial topology fault-tolerant computation.

[0090] Based on the preceding text, we can conclude that... Figure 2 The complete process of the technical solution shown.

[0091] The microscopic slice image and spatial multimodal data preprocessing apparatus provided by the present invention will be described below. The microscopic slice image and spatial multimodal data preprocessing apparatus described below can be referred to in correspondence with the microscopic slice image and spatial multimodal data preprocessing method described above.

[0092] like Figure 3 As shown, the microscopic slice image and spatial multimodal data preprocessing apparatus provided by the present invention includes: The quality inspection module is used to perform multi-dimensional quantitative quality inspection on the microscopic slice images to be processed, and generate a comprehensive image quality score and a pixel-level defect mask map. The filtering module is used to filter out invalid defect areas in the microscopic section image based on the pixel-level defect mask image and extract the effective imaging area. The module is used to adaptively configure the weighting coefficients of the multimodal molecular feature matching term and the spatial topological continuity constraint term based on the overall image quality score, and to construct the joint loss function; The registration module is used to register the pixel coordinate system of the effective imaging region with the sequencing site coordinate system of the structured multimodal data according to the joint loss function, so as to obtain the coordinate mapping relationship. The association module is used to associate and bind the tissue morphology information of the effective imaging area with the molecular expression information of the structured multimodal data according to the coordinate mapping relationship, so as to generate the target structured multimodal dataset.

[0093] Figure 4 A schematic diagram of the physical structure of an electronic device is provided. This device may include a processor, a communications interface, memory, and a communication bus, wherein the processor, communications interface, and memory communicate with each other via the communication bus. The processor can invoke logical instructions from the memory to execute preprocessing methods for microscopic slice images and spatial multimodal data.

[0094] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0095] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the microscopic slice image and spatial multimodal data preprocessing methods provided by the above methods.

[0096] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the microscopic slice image and spatial multimodal data preprocessing methods provided by the methods described above.

[0097] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0098] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for preprocessing microscopic slice images and spatial multimodal data, characterized in that, include: Multi-dimensional quantitative quality inspection is performed on the microscopic slice images to be processed to generate a comprehensive image quality score and a pixel-level defect mask map. Based on the pixel-level defect mask image, invalid defect areas in the microscopic section image are screened out, and effective imaging areas are extracted; Based on the comprehensive image quality score, the weighting coefficients of the multimodal molecular feature matching term and the spatial topological continuity constraint term are adaptively configured to construct a joint loss function; Based on the joint loss function, the pixel coordinate system of the effective imaging region is registered with the sequencing site coordinate system of the structured multimodal data to obtain the coordinate mapping relationship; Based on the coordinate mapping relationship, the tissue morphology information of the effective imaging region is associated and bound with the molecular expression information of the structured multimodal data to generate a target structured multimodal dataset.

2. The method for preprocessing microscopic section images and spatial multimodal data according to claim 1, characterized in that, Multi-dimensional quantitative quality inspection is performed on the microscopic section images to be processed, generating a comprehensive image quality score and a pixel-level defect mask map, including: The microscopic slice images are subjected to clarity quantification, grayscale uniformity quantification, tissue integrity quantification, and artifact interference quantification respectively, generating four corresponding individual quality scores; The four individual quality scores are weighted and summed to generate the overall image quality score; Based on the quantitative detection results of the four dimensions, the blurry, reflective, blank, wrinkled and artifact defect areas are classified and labeled respectively, and the pixel-level defect mask map labeled with the invalid defect area and the effective imaging area is generated.

3. The method for preprocessing microscopic section images and spatial multimodal data according to claim 2, characterized in that, The sharpness quantification of the microscopic slice image is performed to generate a single-item sharpness quality score, including: The microscopic slice image is convolved with a Laplacian operator, and the pixel variance of the convolution result is calculated. The pixel variance is mapped to the sharpness single-item quality score according to the hierarchical mapping rule; The blurred defect areas of the microscopic section image are classified and labeled, including: Image regions with pixel variance below a set lower threshold are marked as blurry defect regions.

4. The method for preprocessing microscopic section images and spatial multimodal data according to claim 2, characterized in that, The grayscale uniformity of the microscopic slice image is quantitatively calculated to generate a single-item quality score for grayscale uniformity, including: The microscopic slice image is divided into image sub-blocks of fixed size and non-overlapping, and the mean gray value and standard deviation of the pixels in each image sub-block are calculated. Calculate the global dispersion of the mean gray level of all image sub-blocks in the entire image, and obtain the gray level uniformity quality score based on the global dispersion. The reflective defect areas of the microscopic section image are classified and labeled, including: Image sub-blocks whose local gray-scale mean deviates from the global gray-scale mean by more than a set deviation threshold are marked as reflective defect areas.

5. The method for preprocessing microscopic section images and spatial multimodal data according to claim 2, characterized in that, The tissue integrity of the microscopic section image is quantified to generate a single quality score for tissue integrity, including: The Otsu threshold segmentation algorithm was used to perform binary segmentation on the microscopic section image to extract the connected components of the foreground tissue. The area of ​​the connected components of the foreground organization and the percentage of effective organization are statistically analyzed, and the organizational integrity single quality score is obtained by mapping the percentage of effective organization. The blank and wrinkled defect areas of the microscopic section image are classified and labeled, including: Regions whose local connected area is lower than a set area threshold and whose background proportion exceeds a set ratio threshold are marked as blank defect regions. Based on morphological opening and closing operations, narrow, twisted connected regions formed by tissue compression and folding are identified and marked as wrinkle defect regions.

6. The method for preprocessing microscopic slice images and spatial multimodal data according to claim 2, characterized in that, The artifact interference quantification of the microscopic slice image is performed to generate a single quality score for artifact interference, including: The high-frequency noise component is extracted by high-pass filtering of the microscopic slice image, and the local high-frequency noise energy is statistically analyzed. Morphological filtering is used to extract strip and spot-like abnormal texture features from the image; The artifact interference level single-item quality score is generated based on the abnormal texture distribution mapping; The artifact defect regions of the microscopic section images are classified and labeled, including: Regions with local high-frequency noise energy exceeding a set energy threshold or containing the aforementioned striped and speckled abnormal texture features are marked as artifact defect regions.

7. The method for preprocessing microscopic section images and spatial multimodal data according to claim 1, characterized in that, Based on the comprehensive image quality score, the weighting coefficients of the multimodal molecular feature matching term and the spatial topological continuity constraint term are adaptively configured to construct a joint loss function, including: Construct a loss function structure that includes a weighted multimodal molecular feature matching term and a weighted spatial topological continuity constraint term; The molecular feature analysis weights corresponding to the multimodal molecular feature matching terms and the topological constraint strengths corresponding to the spatial topological continuity constraint terms are calculated based on the overall image quality score of the sub-region; the sum of the molecular feature analysis weights and the topological constraint strengths is equal to 1. When the overall image quality score increases, the weight of the molecular feature analysis is increased while the strength of the topological constraint is decreased simultaneously. When the overall image quality score decreases, the weight of the molecular feature analysis is reduced while the strength of the topological constraint is increased simultaneously to complete the construction of the joint loss function.

8. The method for preprocessing microscopic section images and spatial multimodal data according to claim 1, characterized in that, Based on the joint loss function, the pixel coordinate system of the effective imaging region is registered with the sequencing site coordinate system of the structured multimodal data to obtain the coordinate mapping relationship, including: A global affine rigid registration algorithm is used to coarsely align the effective imaging region with the structured multimodal data, eliminating global translation, rotation, and scaling biases between the pixel coordinate system and the sequencing site coordinate system; Based on the tissue contour, the coarsely aligned effective imaging area is divided into multiple image sub-regions. The tissue morphology edges and high-expression marker gene sequencing sites in each image sub-region are extracted as dual anchor points, and spatial continuity constraints between adjacent image sub-regions are constructed. The local offset of each image sub-region is solved one by one based on the joint loss function, and the coordinate mapping between adjacent image sub-regions is smoothly fused to correct the imaging cropping deviation and local position offset, thereby obtaining the coordinate mapping relationship.

9. The method for preprocessing microscopic section images and spatial multimodal data according to claim 1, characterized in that, Before performing multi-dimensional quantitative quality inspection on the microscopic section images to be processed, the following steps are also included: The structured multimodal data and the microscopic slice images in lossless image format are accessed after the experiment. The structured multimodal data includes a standard spatial coordinate matrix and an expression matrix of at least one modality among transcriptome, protein, epigenetics and metabolome, and the accessed data does not include raw sequencing read data. The accessed microscopic section image is used as the microscopic section image to be processed.

10. The method for preprocessing microscopic section images and spatial multimodal data according to claim 1, characterized in that, Based on the coordinate mapping relationship, the tissue morphology information of the effective imaging region is associated and bound with the molecular expression information of the structured multimodal data to generate a target structured multimodal dataset, including: The tissue morphology information, after coordinate matching is completed, is written into the underlying data structure corresponding to the structured multimodal data; By maintaining the spatial neighborhood topology of each sequencing site, the target structured multimodal dataset containing complete spatial topology information and multimodal association features is generated.