An image recognition method and system for spatial transcriptomics sections

Abstract

The present invention discloses an image recognition method and system for spatial transcriptomics sections, including image acquisition, image processing, image recognition, region marking, and contour line memory. Compared with the prior art, which may only save the original data of the section image and does not precisely memorize and process the contour lines of key regions, resulting in the need to perform cumbersome image recognition operations again when the original section is deformed or damaged. This solution uses a contour tracking algorithm to extract the contour lines of the regions to be spliced, and performs feature extraction and coordinate calculation. Especially in the coordinate calculation stage, an exact transformation of the contour line coordinates is achieved by constructing a deformation matrix. It has the advantages of precise contour line memory, can effectively prevent subsequent re-recognition caused by the deformation of the original section, and improves the efficiency of image processing and recognition, providing more reliable and efficient technical support for the image recognition and analysis of spatial transcriptomics sections.

Description

Technical Field

[0001] The present invention relates to an image recognition method and system for spatial transcriptomics sections, belonging to the field of medical testing technology. Background Art

[0002] Spatial Transcriptomics is an emerging technology that can simultaneously obtain gene expression information and spatial location information in tissue sections, thereby revealing the spatial distribution and functional state of cells in tissues. This technology has broad application prospects in fields such as tumor research, developmental biology, and neuroscience. However, in the experimental process of spatial transcriptomics, the preparation of tissue sections is a key step. In particular, how to precisely cut the wax block to obtain high-quality tissue sections directly affects the subsequent experimental results.

[0003] The traditional wax block cutting process mainly relies on manual operation. Technicians need to manually draw lines on the wax block based on experience to determine the cutting position and direction. However, this method has the following problems:

[0004] Insufficient precision: Manually drawing lines is difficult to ensure the accuracy of the cutting line. Especially under complex tissue contours, it is easy to cause cutting deviations, affecting the quality of the sections.

[0005] Low efficiency: Manual operation takes a long time and is difficult to meet the requirements of high-throughput experiments.

[0006] Tissue deformation: During the preparation and sectioning of the wax block, the tissue may rotate or deform. Manual operation is difficult to accurately identify and correct these changes, resulting in the cutting line not matching the actual tissue contour.

[0007] In recent years, with the rapid development of artificial intelligence (AI) technology, the application of image recognition and computer vision technology in medical image processing has become increasingly widespread. Especially in the identification of tumor regions, AI can automatically identify tumor regions in tissue sections through deep learning algorithms and accurately label their boundaries. This provides new possibilities for the automation and intelligence of the wax block cutting process.

[0008] Based on this, the present invention proposes an image recognition method and system for spatial transcriptomics sections, aiming to automatically identify the tissue contour in the wax block through AI technology, and combine with the cutting line set by the user to intelligently calculate the cutting direction and depth, realizing the automation and precision of wax block cutting. This method can not only improve the cutting precision and efficiency, but also effectively cope with the challenges brought by tissue deformation, providing high-quality section samples for spatial transcriptomics experiments. Summary of the Invention

[0009] To overcome the problems proposed in the above background art, the present invention provides an image recognition method and system for spatial transcriptomics sections.

[0010] The technical solution of the present invention is: an image recognition method for spatial transcriptomics sections, comprising the following steps:

[0011] S11: Image acquisition, acquiring an image of the original section to obtain a high-definition image of the original section;

[0012] S12: Image processing, including correcting and stitching the high-definition image of the original section to obtain a complete section image;

[0013] S13: Image recognition, using AI technology to recognize the tumor region in the complete section image obtained after processing;

[0014] S14: Region marking, automatically marking the tumor region in the complete section region according to the tumor region recognized in the complete section image, then manually defining the region to be stitched according to the tumor region, and calculating the tumor percentage in this region;

[0015] S15: Contour line memory, recognizing and marking the contour line of the manually defined region to be stitched, and memorizing the characteristics of the contour line.

[0016] Preferably, when performing image acquisition, the acquired image includes an overall image and partitioned high-definition images. Among them, the overall image is an image containing all regions of the original section, the partitioned high-definition images are magnified high-definition images of partial regions of the original section, and the number of partitioned high-definition images is 8 - 24.

[0017] Preferably, when performing image processing, it specifically includes:

[0018] S21: Image preprocessing, performing preprocessing operations on the high-definition image of the original section to eliminate errors in the image, including color correction, brightness adjustment, contrast enhancement, and denoising;

[0019] S22: Image merging, merging multiple groups of partitioned high-definition images into the overall image through image registration, stitching, and fusion technologies to obtain a complete section image;

[0020] S23: Image verification, performing quality inspection on the complete section image obtained after merging, including checking the accuracy, integrity, and consistency of the stitching.

[0021] Preferably, when performing image merging, it specifically includes:

[0022] S31: Image registration. First, use the feature extraction algorithm to extract the feature point data in the overall image and the partitioned high-definition image. Then, perform feature point matching through the feature matching algorithm. Finally, calculate the transformation matrix based on the matched feature points;

[0023] S32: Image stitching. Smoothly transition the overlapping area through methods such as weighted average, maximum selection, and minimum selection, and perform transformation and stitching on the image;

[0024] S33: Image fusion. First, perform image color correction by adjusting the brightness, contrast, and hue parameters of the image. Then, perform image fusion through the multi-resolution fusion method.

[0025] Preferably, when performing image verification, it specifically includes:

[0026] S41: Image quality inspection. First, use the feature point matching algorithm to check for stitching errors and image distortion in the complete image, and then use the color correction algorithm to adjust the color in the complete image;

[0027] S42: Image blurring processing. Perform blurring processing on the obtained complete image to obtain the first detection image, and perform blurring processing on the overall image to obtain the second detection image;

[0028] S43: Similarity detection. Extract feature points and perform feature point matching on the first detection image and the second detection image, calculate the similarity between the first detection image and the second detection image. When the similarity is greater than the set threshold, the image verification passes.

[0029] Preferably, when using AI technology to identify the tumor area in the processed complete slice image, the specific principle steps are as follows:

[0030] S51: Feature extraction. Use the feature extraction algorithm to extract feature data from the complete slice image;

[0031] S52: Region segmentation. Segment the complete slice image into multiple groups of regions according to the extracted feature data;

[0032] S53: Image recognition. First, screen the multiple groups of segmented regions, remove the irrelevant regions, and then perform separate feature extraction and recognition on each group of regions.

[0033] Preferably, when performing region marking, it specifically includes:

[0034] S61: Automatic marking. Automatically mark the tumor area in the complete slice image according to the result of image recognition;

[0035] S62: Manually demarcate. Based on the automatic marking, a staff member demarcates the area to be spliced according to actual needs;

[0036] S63: Calculate the percentage of the tumor area. Within the manually demarcated area to be spliced, calculate the percentage of the tumor area. Among them, one of the area ratio method and the pixel ratio method is used to calculate the percentage of the tumor area;

[0037] S64: Information recording. Record and save the marked tumor area, the area to be spliced and its contour line information.

[0038] Preferably, when performing contour line memory, it specifically includes:

[0039] S71: Image extraction. Extract the contour line of the area to be spliced and extract the image of the area where the contour line is located;

[0040] S72: Contour line feature extraction. Extract the features of the image of the area where the extracted contour line is located;

[0041] S73: Feature point selection. According to the size of the original slice, select the corresponding number of feature points, and select the corresponding feature points according to the result of feature extraction in step S72;

[0042] S74: Coordinate calculation. Convert and calculate the contour line coordinates and the feature point coordinates;

[0043] S75: Result saving. Save the feature point data and the coordinate calculation result.

[0044] Preferably, when performing coordinate calculation, it specifically includes:

[0045] S81: Obtain the original coordinates. From the complete slice image, extract the original coordinates of the contour line and the original coordinates of the feature points;

[0046] S82: Deformation matrix calculation. Construct a deformation matrix and obtain the deformation matrix by matching a certain number of known point pairs, where the known point pairs are the initial feature points and the preset deformed feature points;

[0047] S83: Contour line coordinate transformation. Use the calculated deformation matrix to transform the coordinates of each point on the contour line from the original coordinate system to the new coordinate system.

[0048] An image recognition system for spatial transcriptomics slices, including:

[0049] An image acquisition module, used to collect images of the original slice according to the setting;

[0050] An image processing module for performing various processes on images, including image segmentation, image merging, and image preprocessing;

[0051] An image recognition module for using AI technology to recognize tumor regions in the complete slice images obtained after processing;

[0052] A region marking module for automatically marking tumor regions in the complete slice region according to the tumor regions recognized in the complete slice images, then manually defining the regions to be spliced according to the tumor regions, and calculating the tumor percentage within the regions;

[0053] A contour line memory module for recognizing and marking the contour lines of the regions to be spliced manually defined, and memorizing the features of the contour lines.

[0054] Advantages of the present invention:

[0055] 1. Compared with the prior art where the overall slice images may be directly processed, there are disadvantages such as insufficient image accuracy and loss of detailed information. This solution first obtains data containing high-definition images of the whole and partitions, and through image preprocessing, image registration, splicing, and fusion technologies, accurately merges multiple groups of partitioned high-definition images into the overall image to obtain a high-precision complete slice image. This solution has the advantages of significantly improving image accuracy, retaining more detailed information, and increasing the accuracy of image recognition, thus ensuring the effectiveness of subsequent tumor region recognition and region marking, etc.;

[0056] 2. Compared with the prior art where the quality of the merged images may be verified only through intuitive visual inspection or simple algorithms, there are disadvantages such as incomplete detection and difficulty in detecting subtle splicing errors and image distortion. After eliminating errors through image preprocessing, this solution merges images through image registration, splicing, and fusion technologies, and further introduces an image blurring processing step to perform blurring processing on the complete slice image and its overall image respectively to obtain a first detection image and a second detection image, and then performs similarity detection. This solution has the advantages of more comprehensive detection and the ability to accurately identify subtle splicing errors and distortion regions in the images, thus ensuring high accuracy and consistency in image merging and providing a more reliable and high-quality basis for subsequent image analysis and applications;

[0057] 3. Compared with the prior art where only the original data of the sliced images may be saved and the contour lines of the key regions are not precisely memorized and processed, resulting in the need for cumbersome image recognition operations when the original slices are deformed or damaged, this solution uses a contour tracking algorithm to extract the contour lines of the regions to be spliced, and performs feature extraction and coordinate calculation. Especially in the coordinate calculation stage, an exact transformation of the contour line coordinates is achieved by constructing a deformation matrix, and the coordinates of each contour line are smoothed to improve the accuracy and stability of the coordinate data. This solution has the advantages of precise contour line memory, effectively preventing subsequent re-identification caused by deformation of the original slices, and improving the efficiency of image processing and recognition, providing more reliable and efficient technical support for the image recognition and analysis of spatial transcriptomics slices. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 FIG. shows a schematic workflow diagram of the image recognition method for spatial transcriptomics slices of the present invention;

[0059] Figure 2 FIG. shows a schematic structural diagram of the image recognition system for spatial transcriptomics slices of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] The present invention will be further described below with reference to the drawings and embodiments.

[0061] Please refer to Figure 1 , the present invention provides an embodiment: an image recognition method for spatial transcriptomics slices, including the following steps:

[0062] S11: Image acquisition, acquiring an image of the original slice to obtain a high-definition image of the original slice; specifically, when performing image acquisition:

[0063] The acquired image includes an overall image and partitioned high-definition images. Among them, the overall image is an image containing all regions of the original slice, and the partitioned high-definition images are magnified high-definition images of partial regions of the original slice, and the number of partitioned high-definition images is 8 - 24;

[0064] S12: Image processing, including correcting and splicing the high-definition image of the original slice to obtain a complete sliced image;

[0065] S13: Image recognition, using AI technology to recognize the tumor region in the complete sliced image obtained after processing;

[0066] S14: Region marking, automatically marking the tumor region in the complete sliced region according to the tumor region recognized in the complete sliced image, then manually demarcating the region to be spliced according to the tumor region, and calculating the tumor percentage within this region;

[0067] S15: Outline memory, identify and mark the outline of the area to be spliced manually, and memorize the characteristics of the outline.

[0068] It should be noted that the above steps are only the preferred implementation order. In the specific implementation process, on the premise of not affecting the overall implementation effect, some steps can be swapped. To explain the technical solution of this application more clearly, the following content will explain this solution in a preferred way.

[0069] In another alternative embodiment of the present invention, the above step S12 includes:

[0070] S21: Image preprocessing, perform preprocessing operations on the high-definition image of the original slice to eliminate errors in the image, including color correction, brightness adjustment, contrast enhancement, and denoising;

[0071] S22: Image merging, merge multiple groups of partitioned high-definition images into an overall image through image registration, splicing, and fusion techniques to obtain a complete slice image; specifically, it may include:

[0072] Image registration, first extract the feature point data in the overall image and the partitioned high-definition images using a feature extraction algorithm, then perform feature point matching through a feature matching algorithm, and finally calculate the transformation matrix based on the matched feature points, that is, locate the partitioned high-definition images through the overall image;

[0073] Image splicing, smooth the overlapping area through methods such as weighted average, maximum value selection, and minimum value selection, and transform and splice the partitioned high-definition images;

[0074] Image fusion, first perform image color correction by adjusting the brightness, contrast, and hue parameters of the image, and then perform image fusion through a multi-resolution fusion method;

[0075] S23: Image verification, perform quality inspection on the complete slice image obtained after merging, including checking the accuracy, integrity, and consistency of the splicing.

[0076] Optionally, in the above step S22, through the formula:

[0077] Smooth the overlapping area;

[0078] Where represents the pixel value of the fused image at the coordinate (x, y), represents the pixel value of a group of partitioned high-definition images at the coordinate (x, y), represents the pixel value of another group of partitioned high-definition images at the coordinate (x, y), and is the weight coefficient.

[0079] In this embodiment, errors are eliminated through image preprocessing. Image registration, stitching, and fusion techniques are used to merge multiple groups of partitioned high-definition images, and the overlapping regions are smoothed through a formula. Finally, image verification is performed to ensure the quality of the complete sliced image. This process effectively improves the accuracy and consistency of image merging, optimizes the image quality, and provides a reliable basis for subsequent image analysis and applications.

[0080] Optionally, in the above step S23, quality inspection is performed on the complete sliced image obtained after merging, specifically including:

[0081] S41: Image quality inspection. First, the feature point matching algorithm is used to check for stitching errors and image distortion in the complete image, and then the color correction algorithm is used to adjust the color in the complete image;

[0082] S42: Image blurring processing. The obtained complete image is blurred to obtain a first detection image, and the overall image is blurred to obtain a second detection image;

[0083] S43: Similarity detection. Feature points are extracted and feature point matching is performed on the first detection image and the second detection image, and the similarity between the first detection image and the second detection image is calculated. When the similarity is greater than the set threshold, the image verification passes.

[0084] Optionally, the above step S42 is performed through the formula:

[0085] to blur the image;

[0086] where is the blurred pixel value, is the pixel value in the original image, and N is the total number of pixels in the window.

[0087] In another optional embodiment of the present invention, the specific principle steps of the above step S13 are:

[0088] S51: Feature extraction. The feature extraction algorithm is used to extract feature data from the complete sliced image;

[0089] S52: Region segmentation. The complete sliced image is segmented into multiple groups of regions according to the extracted feature data;

[0090] S53: Image recognition. First, the multiple groups of regions segmented are screened to eliminate irrelevant regions, and then separate feature extraction and recognition are performed on each group of regions.

[0091] Optionally, the above step S51 can use a convolutional neural network for image feature extraction. The principle formula for feature extraction by the convolutional neural network is as follows:

[0092] ;

[0093] where X is the input image, W is the convolutional kernel, and Y is the output feature map.

[0094] Optionally, the above step S52 performs image segmentation through the formula:

[0095] where is the pixel value of the original image, T is the threshold, and is the binary image after segmentation.

[0096] Optionally, the above step S53 can classify the image regions through a classification algorithm based on a random forest. The classification principle is as follows: First, perform separate screening on multiple groups of regions, screen out the background images and images with an area smaller than the set threshold, and then perform feature extraction and recognition on the remaining regions, and classify the identified tumor regions and regions that cannot be recognized as tumor regions.

[0097] In this embodiment, compared with the prior art where it may directly perform overall recognition on the complete slice image, it is difficult to accurately distinguish the tumor region from the irrelevant regions, and there may be a risk of missed detection for tumors with abnormal morphology. After using the feature extraction algorithm to extract the feature data of the complete slice image, this solution uses image segmentation technology to divide the image into multiple groups of regions and screen out the irrelevant regions. In the image recognition stage, this solution not only performs separate feature extraction and recognition on the screened regions, but also classifies the regions that cannot be recognized as tumor regions to prevent errors in the detection results due to abnormal tumor morphology. This solution has the advantages of more accurate recognition, lower missed detection rate, and better detection effect for tumors with abnormal morphology, providing strong support for the accurate recognition of tumor regions and subsequent treatment.

[0098] In another optional embodiment of the present invention, the above step S14 specifically includes:

[0099] S61: Automatic marking. According to the result of image recognition, automatically mark the tumor region in the complete slice image;

[0100] S62: Manual delimitation. On the basis of automatic marking, the staff delimits the area to be stitched according to actual needs;

[0101] S63: Calculate the percentage of the tumor region. In the area to be stitched delimited manually, calculate the percentage of the tumor region. Among them, one of the area ratio method and the pixel ratio method is used to calculate the percentage of the tumor region.

[0102] S64: Information recording, record and save the marked tumor region, the region to be stitched, and their contour line information.

[0103] Optionally, in the above step S61, through the formula:

[0104] , perform segmentation detection on the tumor region;

[0105] where M is the image gradient matrix, Det(M) is the determinant of M, Tr(M) is the trace of M, and c is a constant.

[0106] In this embodiment, first, use image recognition technology to automatically mark the tumor region in the complete slice image. Subsequently, based on the automatic marking result, the staff manually delimits the region to be stitched according to actual needs. Then, within the delimited region to be stitched, accurately calculate the percentage of the tumor region occupied by the area ratio method or the pixel ratio method. Finally, record and save in detail the key information such as the marked tumor region, the region to be stitched, and their contour lines for subsequent analysis and reference.

[0107] In another optional embodiment of the present invention, the above step S15 specifically includes:

[0108] S71: Image extraction, extract the contour line of the region to be stitched, and extract the image of the region where the contour line is located, where the contour tracking algorithm is used for image extraction;

[0109] S72: Contour line feature extraction, extract features from the image of the region where the extracted contour line is located;

[0110] S73: Feature point selection, select the corresponding number of feature points according to the size of the original slice, and select the corresponding feature points according to the feature extraction result in step S72;

[0111] S74: Coordinate calculation, convert and calculate the contour line coordinates and the feature point coordinates;

[0112] S75: Result saving, save the feature point data and the coordinate calculation result.

[0113] Optionally, in the above step S74, it includes through the formula:

[0114] , calculate the Euclidean distance between the feature point and the contour line;

[0115] where, is the coordinate of the feature point, is the coordinate of a point on the contour line, and d is the Euclidean distance between the feature point and a point on the contour line.

[0116] Optionally, when performing coordinate calculation, it specifically includes:

[0117] S81: Obtain the original coordinates, and extract the original coordinates of the contour line and the original coordinates of the feature points from the complete slice image;

[0118] S82: Deformation matrix calculation, construct a deformation matrix, and obtain the deformation matrix by matching a certain number of known point pairs, where the known point pairs are the initial feature points and the preset deformed feature points;

[0119] S83: Contour line coordinate transformation, use the calculated deformation matrix to transform the coordinates of each point on the contour line from the original coordinate system to the new coordinate system.

[0120] Optionally, after the contour line coordinate transformation, through the formula:

[0121] , perform smoothing processing on each contour line coordinate;

[0122] Among them, is the pixel value of the filtered image at the coordinate (x, y), is the standard deviation of the Gaussian function, k is the size of the filter, and i and j are loop variables used to traverse all elements of the filter, is the pixel value of the input image at the coordinate i.e., an element in the filter.

[0123] As Figure 2 shown, the present invention also provides an image recognition system for spatial transcriptomics slices, including:

[0124] An image acquisition module for collecting images of the original slice according to the setting;

[0125] An image processing module for performing various processes on the image, including image segmentation, image merging, and image preprocessing;

[0126] An image recognition module for using AI technology to recognize the tumor region in the complete slice image obtained after processing;

[0127] A region marking module for automatically marking the tumor region in the complete slice region according to the tumor region recognized in the complete slice image, then manually defining the splicing region within the tumor region, and calculating the tumor percentage within this region;

[0128] A contour line memory module for recognizing and marking the contour line of the manually defined splicing region and memorizing the features of the contour line.

[0129] The embodiments provided above are not intended to limit the scope covered by the present invention, nor are the described steps intended to limit the order of their execution. Obvious improvements made by those skilled in the art to the present invention in combination with the existing well-known general knowledge also fall within the scope of protection defined by the claims of the present invention.

Claims

1. An image recognition method for spatial transcriptomics slices, characterized in that: The following steps are included: S11: image acquisition, acquiring the image of the original slice to obtain a high-definition image of the original slice; S12: Image processing, including correction and stitching of the high-definition images of the original slices to obtain complete slice images; S13: Image recognition, using AI technology to identify the tumor area on the processed complete slice image; S14: region marking: according to the tumor region identified in the complete slice image, the tumor region is automatically marked in the complete slice region, and then the proposed splicing region is manually delineated according to the tumor region, and the tumor percentage in the region is calculated; S15: Contour line memory, identifying and marking the contour lines of the manually delineated area to be spliced, and memorizing the features of the contour lines, including: S71: Image extraction, extracting the contour line of the area to be spliced, and extracting the image of the area where the contour line is located; S72: extracting contour features, performing feature extraction on the image of the area where the extracted contour is located; S73: feature point selection, selecting a corresponding number of feature points according to the size of the original slice, and selecting corresponding feature points according to the result of feature extraction in step S72; S74: coordinate calculation, converting and calculating the contour line coordinates and the feature point coordinates; S75: Save the result, save the feature point data and coordinate calculation results; When performing coordinate calculation, it specifically includes: S81: Obtaining original coordinates, extracting original coordinates of contour lines and original coordinates of feature points from the complete slice image; S82: deformation matrix calculation, constructing a deformation matrix, and obtaining the deformation matrix by matching a certain number of known point pairs, wherein the known point pairs are initial feature points and feature points after preset deformation; S83: Contour line coordinate transformation: using the calculated deformation matrix, transform the coordinates of each point on the contour line from the original coordinate system to the new coordinate system.

2. The image recognition method for spatial transcriptomics slices according to claim 1, characterized in that: When acquiring images, the acquired images include overall images and partitioned high-definition images, wherein the overall image is an image containing the entire area of ​​the original slice, and the partitioned high-definition image is an enlarged high-definition image of a partial area of ​​the original slice, and the number of partitioned high-definition images is 8-24.

3. The image recognition method for spatial transcriptomics slices according to claim 2, characterized in that: When performing image processing, it specifically includes: S21: Image preprocessing, preprocessing the high-definition images of the original slices to eliminate errors in the images, including color correction, brightness adjustment, contrast enhancement and denoising; S22: Image merging: Through image registration, stitching and fusion technology, multiple sets of partitioned high-definition images are merged into the overall image to obtain a complete slice image; S23: Image verification, quality check of the merged complete slice image, including checking the accuracy, completeness and consistency of the stitching.

4. The image recognition method for spatial transcriptomics slices according to claim 3, characterized in that: When merging images, it specifically includes: S31: Image registration, firstly, feature point data in the overall image and the partitioned high-definition image are extracted using a feature extraction algorithm, then feature point matching is performed using a feature matching algorithm, and finally, a transformation matrix is ​​calculated based on the matched feature points; S32: Image stitching, through weighted average, maximum selection and minimum selection methods, smooth transition overlapping areas, and transform and stitch images; S33: Image fusion, firstly, image color correction is performed by adjusting the brightness, contrast and hue parameters of the image, and then image fusion is performed by a multi-resolution fusion method.

5. The image recognition method for spatial transcriptomics slices according to claim 4, characterized in that: When performing image verification, it specifically includes: S41: image quality check, first using a feature point matching algorithm to check the complete image for stitching errors and image distortion, and then using a color correction algorithm to adjust the color in the complete image; S42: performing image fuzzification processing, performing fuzzification processing on the obtained complete image to obtain a first detection image, and performing fuzzification processing on the entire image to obtain a second detection image; S43: Similarity detection, extracting and matching feature points of the first detection image and the second detection image, and calculating the similarity between the first detection image and the second detection image. When the similarity is greater than a set threshold, the image verification passes.

6. The image recognition method for spatial transcriptomics slices according to claim 5, characterized in that: When using AI technology to identify tumor areas on the processed complete slice image, the specific principles and steps are as follows: S51: feature extraction, extracting feature data from the complete slice image using a feature extraction algorithm; S52: region segmentation, segmenting the complete slice image into multiple groups of regions according to the extracted feature data; S53: Image recognition, firstly screening the multiple groups of segmented areas, removing irrelevant areas, and then performing separate feature extraction and recognition on each group of areas.

7. The image recognition method for spatial transcriptomics slices according to claim 6, characterized in that: When marking an area, it includes: S61: Automatic labeling: automatically labeling the tumor area in the complete slice image based on the image recognition result; S62: Manual demarcation: On the basis of automatic marking, staff will demarcate the area to be spliced ​​according to actual needs; S63: calculating the percentage of the tumor area, within the manually demarcated proposed stitching area, calculating the percentage of the tumor area, wherein the percentage of the tumor area is calculated by using one of an area ratio method and a pixel ratio method; S64: Information recording, recording and saving the marked tumor area, the area to be spliced, and its contour information.

8. An image recognition system for spatial transcriptomics slices, used in the image recognition method for spatial transcriptomics slices according to claims 1-7, characterized in that: Included are: An image acquisition module is used to acquire images of original slices according to settings; Image processing module, used to perform various image processing, including image segmentation, image merging and image preprocessing; An image recognition module is used to identify tumor areas in the processed complete slice images using AI technology; A region marking module is used to automatically mark the tumor region in the complete slice region according to the tumor region identified in the complete slice image, and then manually define the proposed splicing region according to the tumor region, and calculate the tumor percentage in the region; The contour line memory module is used to identify and mark the contour lines of the manually delineated area to be spliced, and to memorize the features of the contour lines.

Images