A microsatellite instability detection method and device based on multi-resolution multi-region

Abstract

The application discloses a microsatellite instability detection method and device based on multi-resolution multi-region, and the method comprises the following steps: S1, obtaining a multi-resolution WSI image; S2, performing pixel classification on the low-resolution image obtained in S1 according to different colors, and pixels with the same or similar color are classified into one category, and each category forms an image with staining information; S3, selecting the image with the most staining information, and dividing the image into multiple small regions; S4, selecting a small region with the most pixels from all the small regions; S5, obtaining a region corresponding to the small region selected in S4 from a high-resolution image, and dividing the region into multiple small image blocks; S6, extracting feature maps from all the small image blocks by using a pre-trained feature extraction model; and S7, sequentially classifying the feature maps by using a classification model, and determining whether the small image blocks corresponding to the feature maps are MSHI or nonMSIH. The application makes the detection of the microsatellite instability state more efficient, more reliable and lower in cost.

Description

Technical Field

[0001] This invention belongs to the field of gene sequencing, specifically relating to a method and device for detecting microsatellite instability based on multi-resolution and multi-regional sequencing. Background Technology

[0002] Colorectal cancer is the third most common cancer in the world and the second leading cause of cancer death, with approximately 900,000 people dying from it each year.

[0003] Microsatellite instability (MSI) refers to the phenomenon of altered microsatellite (MS) sequence lengths caused by insertion or deletion mutations during DNA replication, often due to defects in mismatch repair (MMR). MS sequences are short, repetitive DNA sequences, typically consisting of 1–6 nucleotides arranged in tandem repeats. Common types include dibasic CA / GA / GT sequences and monobasic A / T sequences. MS sequences can be located in important non-coding regions or coding regions of genes, exhibiting polymorphism throughout the genome and significant individual variability.

[0004] The MSI phenomenon was first discovered in colorectal cancer by Jacobs et al. in 1993. As research into MSI has deepened, it has been found that the MSI phenomenon is not limited to colorectal cancer, but also occurs in solid tumors such as endometrial cancer, gastric cancer, hepatocellular carcinoma, and breast cancer.

[0005] Clinical studies have confirmed a close relationship between microsatellite instability (MSI) and the prognosis of colorectal cancer. Patients with colorectal cancer exhibiting high MSI have a significant survival advantage compared to those with stable MSI, despite having poorer clinical presentations. Studies have demonstrated that, for stage II / III colorectal cancer patients, those with high MSI have significantly prolonged overall survival and disease-free survival. The National Comprehensive Cancer Network (NCCN) guidelines for colorectal cancer recommend that MSI testing be performed on all patients with a history of colorectal cancer.

[0006] Currently, when detecting MSI in cancer cells, it is possible to determine whether MSI has occurred by detecting MMR gene deletion, such as protein level detection relying on immunohistochemistry, or by directly detecting sequence changes of MSI, such as molecular level detection by PCR (polymerase chain reaction).

[0007] A common method is to use immunohistochemistry to detect the expression of mismatch repair genes MLH1, MSH2, MSH6, and PMS2 in tumor tissue. The absence of any one of these mismatch repair genes is defined as MSI; otherwise, it is considered microsatellite stable (MSS). The absence of one gene is termed low microsatellite instability (MSI-L), and the absence of two or more proteins is termed high microsatellite instability (MSI-H). This method is relatively simple and inexpensive for detecting MSI. However, it has some drawbacks, such as inconsistencies in the antibodies used and inconsistencies in pathologists' interpretation standards.

[0008] Molecular-level detection primarily employs multiplex fluorescent PCR combined with capillary electrophoresis. MSI status is determined by detecting specific microsatellite repeat sequence amplification using PCR, comparing the site mutation status of tumor patient specimens with normal tissues. The comparison sites are the five microsatellite sites recommended by the National Cancer Institute (NCI): BAT25, BAT26, D5S346, D2S123, and D17S250. Changes at ≥2 sites are classified as high microsatellite instability (MSI-H), changes at only one site as low microsatellite instability (MSI-L), and no changes as microsatellite stable (MSS). This method is currently the "gold standard" for MSI detection, but the procedure is complex and costly, and issues such as excessively strong or weak fluorescence, nonspecific peaks, insignificant peak size changes, and loss of heterozygosity can arise during result interpretation.

[0009] With the development of artificial intelligence, microsatellite state classification methods for colorectal cancer based on artificial intelligence technology have begun to emerge. By designing targeted network architectures and autonomously learning from gastric cancer pathological images, trained neural network models can classify the microsatellite state of colorectal cancer, which to some extent makes up for the shortcomings of traditional methods.

[0010] However, this method also has the following problems:

[0011] 1) The annotation cost is huge, requiring doctors to manually delineate tumor areas, or for classification models trained after doctors have manually delineated the areas to help with screening. This process can also introduce differences in the subjective delineation by pathologists, which can affect the results.

[0012] 2) Due to the large size of WSI images, for a single pathological slide image, each 1cm*1cm slide at a resolution of 40X has a pixel size of about 0.25um, and the resolution of the entire pathological slide is about 40000x40000. The image file size is about 1GB, which is difficult for ordinary computers to process directly. Therefore, it needs to be processed into 512x512 small tiles for calculation. However, after this processing, the number of tile images increases dramatically.

[0013] C. In most cases during training, the same weights were used for the MSI states of the same WSI-generated patches, ignoring the heterogeneity between patches and affecting the accuracy of the results to some extent. Summary of the Invention

[0014] To address the aforementioned technical problems, this invention proposes a method and device for detecting microsatellite instability based on multi-resolution and multi-regional approaches.

[0015] To achieve the above objectives, the technical solution of the present invention is as follows:

[0016] On one hand, this invention discloses a method for detecting microsatellite instability based on multi-resolution and multi-regional methods, comprising the following steps:

[0017] S1: Obtain the target pathological image, read the WSI image information in the target pathological image, and obtain at least a low-resolution image and a high-resolution image;

[0018] S2: Based on the different colors, classify the low-resolution image obtained in S1 into pixels, grouping pixels with the same or similar colors into one category, and forming an image with coloring information for each category;

[0019] S3: Select the image with the most staining information from all the images with staining information obtained from S2, and divide it into multiple small regions;

[0020] S4: Select the region with the most pixels from all the regions obtained from S3;

[0021] S5: Obtain the region corresponding to the small region selected in S4 from the high-resolution image, and divide the region into multiple small image blocks;

[0022] S6: Use the pre-trained feature extraction model to extract feature maps from all the small image patches obtained in S5;

[0023] S7: Use a classification model to classify the feature maps extracted in S5 sequentially, and determine whether the small image patches of the corresponding feature maps are MSIH or nonMSIH;

[0024] If it is MSIH, then the entire small region corresponding to the small image patch is determined to be MSIH;

[0025] Otherwise, continue classifying the next feature map. If all the small image patches corresponding to all the feature maps are nonMSIH, then the entire small region corresponding to the small image patch is determined to be nonMSIH.

[0026] Based on the above technical solution, the following improvements can be made:

[0027] As a preferred solution, S2 specifically includes the following steps:

[0028] S2.1: Convert the low-resolution image obtained in S1 from RGB space to L*a*b color space. The L*a*b* space contains a photometric layer 'L*', a chromaticity layer 'a*', and a chromaticity layer 'b*'.

[0029] S2.2: Use clustering methods to classify pixels, grouping pixels with the same or similar colors into one class, and forming an image with coloring information for each class.

[0030] As a preferred option, the clustering method is one of K-means clustering, hierarchical clustering, SOM clustering, or FCM clustering.

[0031] As a preferred solution, S3 specifically includes the following steps:

[0032] S3.1: Determine the sum of RGB flux for each class of images obtained in S2 on the RGB channels. The image with the largest sum of flux and the color closest to white is the background image.

[0033] S3.2: For each type of image except the background image, the image whose color is closest to purple is the image with the most coloring information.

[0034] S3.3: Divide the image with the most coloring information obtained from S3.2 into multiple small regions.

[0035] As a preferred option, S4 specifically includes the following:

[0036] First, sort all the divided small regions according to the number of pixels they contain;

[0037] Then, select the smallest region containing the most pixels in order.

[0038] As a preferred solution, S5 specifically includes the following steps:

[0039] S5.1: Obtain the region from the high-resolution image corresponding to the small region selected in S4;

[0040] S5.2: Normalize the region obtained in S5.1;

[0041] S5.3: Divide the normalized region into multiple small image blocks, each of the same size.

[0042] As a preferred approach, the region corresponding to the small region selected by S4 is obtained from the high-resolution image, and its four vertex coordinates are as follows:

[0043] (x1*scale, y1*scale), (x1*scale, y2*scale), (x2*scale, y11*scale), (x2*scale, y2*scale);

[0044] Where (x1, y1), (x1, y2), (x2, y1), and (x2, y2) are the coordinates of the four vertices corresponding to the small region selected by S4;

[0045] scale is the ratio of the resolution of a high-resolution image to the resolution of a low-resolution image.

[0046] Furthermore, this invention also discloses a microsatellite instability detection device based on multi-resolution and multi-regional methods, comprising:

[0047] One or more processors;

[0048] Memory;

[0049] And one or more programs, wherein the one or more programs are stored in memory and configured to be executed by one or more processors, and the one or more programs include instructions for performing any of the microsatellite instability detection methods described above.

[0050] The present invention provides a method and device for detecting microsatellite instability based on multi-resolution and multi-regional methods, which has the following advantages:

[0051] First, this invention employs artificial intelligence and image processing methods to solve the problems of traditional immunohistochemistry and molecular-level detection methods, which are complex to operate and highly susceptible to other factors. This invention reduces costs, eliminating the need to purchase expensive molecular detection equipment, enabling even small hospitals to perform microsatellite instability detection.

[0052] Second, without the need for doctors to make subjective delineations, the method uses low-resolution image segmentation and block image entropy thresholding to filter useful small image patches and extract regions rich in cellular information, which greatly reduces the number of small image patches and makes multi-instance learning possible.

[0053] Third, the multi-resolution, multi-region approach reduces the amount of computation, solves the problem of heterogeneity between patches, and eliminates the need to determine whether each example is a target region. During model training, there is no need to label the tumor region, reducing the cost of manual annotation and addressing the issues of massive data volume and high annotation costs faced by most deep learning methods.

[0054] In summary, this invention discloses a method and device for detecting microsatellite instability based on multi-resolution and multi-regional methods. This method significantly lowers the threshold for detecting microsatellite instability, making the detection of microsatellite instability states more efficient, reliable, and cost-effective. It helps doctors provide more targeted treatment to patients and has extremely high clinical significance. Attached Figure Description

[0055] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a flowchart of a microsatellite instability detection method provided in an embodiment of the present invention.

[0057] Figure 2 This is a schematic diagram of low-resolution image pixel classification provided in an embodiment of the present invention.

[0058] Figure 3 This is an example diagram of region segmentation provided in an embodiment of the present invention;

[0059] Figure 3 (a) is a partitioning diagram using a randomly generated block pattern;

[0060] Figure 3 (b) is a partition diagram of the pattern using rule-based partitioning.

[0061] Figure 4 This is an example diagram illustrating the generation of small tiles from a region, provided as an embodiment of the present invention.

[0062] Figure 5 This is an example diagram illustrating feature map extraction of all small image patches using a pre-trained feature extraction model, as provided in an embodiment of the present invention.

[0063] Figure 6 This is an example diagram illustrating the classification of feature maps using a classification model, as provided in an embodiment of the present invention. Detailed Implementation

[0064] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] Using ordinal numbers such as “first,” “second,” “third,” etc. to describe ordinary objects merely indicates different instances of similar objects and is not intended to imply that the objects being described must have a given order in time, space, sequence, or any other way.

[0067] Furthermore, the expression "includes" is an "open-ended" expression, which means only that there is a corresponding component or step, and should not be interpreted as excluding additional components or steps.

[0068] To achieve the objectives of this invention, some embodiments of a multi-resolution, multi-region microsatellite instability detection method and device are provided, such as... Figure 1 As shown, the microsatellite instability detection method includes the following steps:

[0069] S1: Obtain the target pathological image, read the WSI image information in the target pathological image, and obtain at least a low-resolution image LR-image and a high-resolution image HR-image;

[0070] S2: Based on the different colors, the low-resolution image LR-image obtained in S1 is classified into pixels, and pixels with the same or similar colors are grouped into one category. Each category forms an image with coloring information.

[0071] S3: Select the image with the most staining information from all the images with staining information obtained from S2, and divide it into multiple small regions;

[0072] S4: Select the region with the most pixels from all the regions obtained from S3;

[0073] S5: Obtain the region corresponding to the small region selected in S4 from the high-resolution image HR-image, and divide the region into multiple small image blocks;

[0074] S6: Use the pre-trained feature extraction model to extract feature maps from all the small image patches obtained in S5;

[0075] S7: Use a classification model to classify the feature maps extracted in S5 sequentially, and determine whether the small image patches of the corresponding feature maps are MSIH or nonMSIH;

[0076] If it is MSIH, then the entire small region corresponding to the small image patch is determined to be MSIH;

[0077] Otherwise, continue classifying the next feature map. If all the small image patches corresponding to all the feature maps are nonMSIH, then the entire small region corresponding to the small image patch is determined to be nonMSIH.

[0078] In S1, the target pathological image is a tissue area image obtained after staining cells in a patient's tissue sample.

[0079] To further optimize the implementation effect of the present invention, in some other embodiments, the remaining technical features are the same, except that S2 specifically includes the following steps:

[0080] S2.1: Convert the low-resolution image LR-image obtained in S1 from RGB space to L*a*b* color space. The L*a*b* space contains a photometric layer 'L*', a chromaticity layer 'a*', and a chromaticity layer 'b*'.

[0081] S2.2: Use clustering methods to classify pixels, grouping pixels with the same or similar colors into one class, and forming an image with coloring information for each class.

[0082] In this context, the chroma layer 'a*' represents the position of the color along the red-green axis, and the chroma layer 'b*' represents the position of the color along the blue-yellow axis. All color information is contained in the 'a*' and 'b*' layers.

[0083] Furthermore, the clustering method is one of K-means clustering, hierarchical clustering, SOM clustering, or FCM clustering.

[0084] To further optimize the implementation effect of the present invention, in some other embodiments, the remaining technical features are the same, except that S3 specifically includes the following steps:

[0085] S3.1: Determine the sum of RGB flux for each class of images obtained in S2 on the RGB channels. The image with the largest sum of flux and the color closest to white is the background image.

[0086] S3.2: For each type of image except the background image, the image whose color is closest to purple is the image with the most coloring information.

[0087] S3.3: Divide the image with the most coloring information obtained in S3.2 into multiple small regions. In S3.3, the image with the most coloring information obtained in S3.2 can be divided into multiple small regions using either a random block generation mode or a regular division mode.

[0088] To further optimize the implementation effect of the present invention, in some other embodiments, the remaining technical features are the same, except that S4 specifically includes the following:

[0089] First, sort all the divided small regions according to the number of pixels they contain;

[0090] Then, select the smallest region containing the most pixels in order.

[0091] To further optimize the implementation effect of the present invention, in some other embodiments, the remaining technical features are the same, except that S5 specifically includes the following steps:

[0092] S5.1: Obtain the region corresponding to the small region selected in S4 from the high-resolution image HR-image;

[0093] S5.2: Normalize the region obtained in S5.1;

[0094] S5.3: Divide the normalized region into multiple small image blocks, each of the same size.

[0095] In step S5.1, the region corresponding to the small region selected in S4 is obtained from the high-resolution image HR-image, and its four vertex coordinates are as follows:

[0096] (x1*scale, y1*scale), (x1*scale, y2*scale), (x2*scale, y11*scale), (x2*scale, y2*scale);

[0097] Where (x1, y1), (x1, y2), (x2, y1), and (x2, y2) are the coordinates of the four vertices corresponding to the small region selected by S4;

[0098] scale is the ratio of the resolution of the high-resolution image (HR-image) to the resolution of the low-resolution image (LR-image).

[0099] The above-mentioned implementation methods can be carried out in parallel or in a cross-cutting manner.

[0100] Furthermore, embodiments of the present invention also disclose a microsatellite instability detection device based on multi-resolution and multi-regional methods, comprising:

[0101] One or more processors;

[0102] Memory;

[0103] And one or more programs, wherein the one or more programs are stored in memory and configured to be executed by one or more processors, and the one or more programs include instructions for performing the microsatellite instability detection method disclosed in any of the above embodiments.

[0104] To facilitate understanding of the present invention, a specific embodiment is described below. The method for detecting microsatellite instability in colorectal cancer using the present invention includes the following steps:

[0105] S1: Obtain the target pathological image, read the WSI image information in the target pathological image, and obtain at least a low-resolution image LR-image and a high-resolution image HR-image;

[0106] S2: Based on the different colors, the low-resolution image LR-image obtained in S1 is classified into pixels, and pixels with the same or similar colors are grouped into one category. Each category forms an image with coloring information.

[0107] S3: Select the image with the most staining information from all the images with staining information obtained from S2, and divide it into multiple small regions;

[0108] S4: Select the region with the most pixels from all the regions obtained from S3;

[0109] S5: Obtain the region corresponding to the small region selected in S4 from the high-resolution image HR-image, and divide the region into multiple small image blocks;

[0110] S6: Use the pre-trained feature extraction model to extract feature maps from all the small image patches obtained in S5;

[0111] S7: Use a classification model to classify the feature maps extracted in S5 sequentially, and determine whether the small image patches of the corresponding feature maps are MSIH or nonMSIH;

[0112] If it is MSIH, then the entire small region corresponding to the small image patch is determined to be MSIH;

[0113] Otherwise, continue classifying the next feature map. If all the small image patches corresponding to all the feature maps are nonMSIH, then the entire small region corresponding to the small image patch is determined to be nonMSIH.

[0114] Each step is described in detail below:

[0115] S1 specifically refers to:

[0116] Read multi-resolution images from the original WSI image information.

[0117] Taking a 2cm x 2cm pathological section as an example: at 40X magnification and with MPP (micrometers per pixel) of 0.25, the original image is a high-resolution image (HR-image) with a resolution of approximately 80000*80000, while the low-resolution image (LR-image) has a resolution of approximately 1024*1024.

[0118] It is worth noting that, in a sense, the low-resolution image (LR-image) can be viewed as a feature set of the pixel mean of the high-resolution image (HR-image). Subsequent processing of the low-resolution image (LR-image) reduces the resolution while still preserving the overall characteristics of the high-resolution image (HR-image).

[0119] S2 specifically includes the following steps:

[0120] S2.1: Convert the low-resolution image LR-image obtained in S1 from RGB space to L*a*b color space. The L*a*b* space contains a luminance layer 'L*', a chrominance layer 'a*' (indicating the position of the color along the red-green axis), and a chrominance layer 'b*' (indicating the position of the color along the blue-yellow axis); all color information is in the 'a*' and 'b*' layers.

[0121] S2.2: Use K-means clustering to classify pixels, grouping pixels with the same or similar colors into one class, and forming an image with color information for each class.

[0122] Euclidean distance is used to measure the difference between the two colors. Since K-means clustering is an unsupervised algorithm, it groups pixels that are close together into one class, but the number of classes needs to be predetermined. In this embodiment, the number of classes is predetermined to be 4. K-means clustering can effectively extract the tumor region completely, such as... Figure 2 As shown.

[0123] S3 specifically includes the following steps:

[0124] S3.1: Determine the sum of RGB flux for each class of images obtained in S2 on the RGB channel. The image with the largest sum of flux and the color closest to white is the background image. The background image contains fat information.

[0125] S3.2: Judge the other three types of images besides the background image. The image with the color closest to purple is the image with the most staining information, denoted as LR-MASK. The LR-MASK image contains tumor cells, inflammatory cells and part of the colorectal mucosa.

[0126] S3.3: In order to further reduce the amount of computation, the image with the most coloring information obtained in S3.2 is denoted as LR-MASK and divided into multiple small regions.

[0127] In S3.3, the division can be done using a randomized block generation method, such as... Figure 3 As shown in (a).

[0128] Alternatively, a rule-based partitioning model can be used for partitioning, such as... Figure 3 As shown in (b).

[0129] S4 specifically includes the following:

[0130] First, sort all the divided regions according to the number of pixels they contain;

[0131] Then, in order, select the small region containing the most pixels, denoted as LR-ROI, and the coordinates of the four vertices of LR-ROI are (x1, y1), (x1, y2), (x2, y1), (x2, y2).

[0132] S5 specifically includes the following steps:

[0133] S5.1: Based on the scaling ratio (scale), obtain the region corresponding to the small region LR-ROI selected in S4 from the high-resolution image HR-image, denoted as HR-ROI. The coordinates of the four vertices of HR-ROI are as follows:

[0134] (x1*scale, y1*scale), (x1*scale, y2*scale), (x2*scale, y11*scale), (x2*scale, y2*scale);

[0135] S5.2: Scale the HR-ROI obtained in S5.1 to the target pixel size, take the size of each pixel as 0.5um, denote the pixel-normalized image as HR-nROI, and normalize the color of HR-nROI;

[0136] S5.3: Divide the normalized region HR-nROI into multiple small image tiles, each with the same size of 512*512, such as... Figure 4 As shown.

[0137] The scaling ratio (scale) is the ratio of the resolution of the high-resolution image (HR-image) to the resolution of the low-resolution image (LR-image). Each small image tile is encoded according to its position.

[0138] S6 specifically involves using a pre-trained feature extraction model, shuffle Net, to extract feature maps from all the small image tiles obtained in S5, such as... Figure 5 As shown.

[0139] The Shuffle Net feature extraction model is used to train small image tiles. For the trained model, the fully connected layers are discarded, and then a pre-trained Feature Map can be output.

[0140] S7 specifically involves using the ResNet50 classification model to classify the feature maps extracted in S5 sequentially, and determining whether the small image patches of the corresponding feature maps are MSIH or nonMSIH.

[0141] If it is MSIH, then the entire small region corresponding to the small image patch is determined to be MSIH;

[0142] Otherwise, continue classifying the next feature map. If all the small image patches corresponding to all feature maps are nonMSIH, then the entire small region corresponding to the small image patch is determined to be nonMSIH. Figure 6 As shown.

[0143] For this classification model ResNet50, the feature map is divided into a training set and a validation set for classification training.

[0144] The model is optimized using the training set, and the model weights are saved when the loss value is minimized. Training can be stopped when the number of training epochs exceeds a preset number of epochs (e.g., 50 epochs) or when the loss value on the validation set no longer decreases for a preset number of consecutive epochs.

[0145] Furthermore, this specific embodiment also discloses a multi-resolution, multi-region microsatellite instability detection device for colorectal cancer, comprising:

[0146] One or more processors;

[0147] Memory;

[0148] And one or more programs, wherein the one or more programs are stored in memory and configured to be executed by one or more processors, and the one or more programs include instructions for performing the microsatellite instability detection method disclosed in the above specific embodiments.

[0149] The present invention provides a method and device for detecting microsatellite instability based on multi-resolution and multi-regional methods, which has the following advantages:

[0150] First, this invention employs artificial intelligence and image processing methods to solve the problems of traditional immunohistochemistry and molecular-level detection methods, which are complex to operate and highly susceptible to other factors. This invention reduces costs, eliminating the need to purchase expensive molecular detection equipment, enabling even small hospitals to perform microsatellite instability detection.

[0151] Second, without the need for doctors to make subjective delineations, the method uses low-resolution image segmentation and block image entropy thresholding to filter useful small image patches and extract regions rich in cellular information, which greatly reduces the number of small image patches and makes multi-instance learning possible.

[0152] Third, the multi-resolution, multi-region approach reduces the amount of computation, solves the problem of heterogeneity between patches, and eliminates the need to determine whether each example is a target region. During model training, there is no need to label the tumor region, reducing the cost of manual annotation and addressing the issues of massive data volume and high annotation costs faced by most deep learning methods.

[0153] In summary, this invention discloses a method and device for detecting microsatellite instability based on multi-resolution and multi-regional methods. This method significantly lowers the threshold for detecting microsatellite instability, making the detection of microsatellite instability states more efficient, reliable, and cost-effective. It helps doctors provide more targeted treatment to patients and has extremely high clinical significance.

[0154] It should be understood that the various techniques described herein can be implemented in combination with hardware or software, or a combination thereof. Thus, the methods and apparatus of the present invention, or certain aspects or portions thereof, can take the form of program code (i.e., instructions) embedded in a tangible medium, such as a floppy disk, CD-ROM, hard disk, or any other machine-readable storage medium, wherein when the program is loaded into and executed by a machine such as a computer, that machine becomes an apparatus for practicing the present invention.

[0155] In the description of this invention, it should be understood that the terms "coaxial," "bottom," "one end," "top," "middle," "other end," "upper," "side," "top," "inner," "front," "center," "both ends," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0156] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "setting," "connection," "fixing," "screw connection," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal connection of two components or the interaction between two components. Unless otherwise explicitly limited, those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0157] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the present invention. Various changes and modifications can be made to the present invention without departing from its spirit and scope. All such changes and modifications fall within the scope of the present invention as claimed, which is defined by the appended claims and their equivalents.

[0158] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it. They should not be used to limit the scope of protection of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for detecting microsatellite instability based on multi-resolution and multi-regional methods, characterized in that, Includes the following steps: S1: Obtain the target pathological image, read the multi-resolution image information of the target pathological image, and obtain at least a low-resolution image and a high-resolution image; S2: Based on the different colors, classify the low-resolution image obtained in S1 into pixels, grouping pixels with the same or similar colors into one category, and forming an image with coloring information for each category; S3: Select the image with the most staining information from all the images with staining information obtained from S2, and divide it into multiple small regions; S4: Select the region with the most pixels from all the regions obtained from S3; S5: Obtain the region corresponding to the small region selected in S4 from the high-resolution image, and divide the region into multiple small image blocks; S6: Use the pre-trained feature extraction model to extract feature maps from all the small image patches obtained in S5; S7: Use a classification model to classify the feature maps extracted in S5 sequentially, and determine whether the small image patches of the corresponding feature maps are MSIH or nonMSIH; If it is MSIH, then the entire small region corresponding to the small image patch is determined to be MSIH; Otherwise, continue classifying the next feature map. If all the small image patches corresponding to all the feature maps are nonMSIH, then the entire small region corresponding to the small image patch is determined to be nonMSIH.

2. The microsatellite instability detection method according to claim 1, characterized in that, S2 specifically includes the following steps: S2.1: Convert the low-resolution image obtained in S1 from RGB space to L*a*b color space. The L*a*b* space contains a photometric layer 'L*', a chromaticity layer 'a*', and a chromaticity layer 'b*'. S2.2: Use clustering methods to classify pixels, grouping pixels with the same or similar colors into one class, and forming an image with coloring information for each class.

3. The microsatellite instability detection method according to claim 2, characterized in that, The clustering method is one of K-means clustering, hierarchical clustering, SOM clustering, or FCM clustering.

4. The microsatellite instability detection method according to claim 1, characterized in that, S3 specifically includes the following steps: S3.1: Determine the sum of RGB flux for each class of images obtained in S2 on the RGB channels. The image with the largest sum of flux and the color closest to white is the background image. S3.2: For each type of image except the background image, the image whose color is closest to purple is the image with the most coloring information. S3.3: Divide the image with the most coloring information obtained from S3.2 into multiple small regions.

5. The microsatellite instability detection method according to claim 1, characterized in that, S4 specifically includes the following: First, sort all the divided small regions according to the number of pixels they contain; Then, select the smallest region containing the most pixels in order.

6. The microsatellite instability detection method according to claim 1, characterized in that, S5 specifically includes the following steps: S5.1: Obtain the region from the high-resolution image corresponding to the small region selected in S4; S5.2: Normalize the region obtained in S5.1; S5.3: Divide the normalized region into multiple small image blocks, each of the same size.

7. The microsatellite instability detection method according to claim 6, characterized in that, In S5.1, the region corresponding to the small region selected in S4 is obtained from the high-resolution image, and its four vertex coordinates are as follows: (x1*scale, y1*scale), (x1*scale, y2*scale), (x2*scale, y11*scale), (x2*scale, y2*scale); Where (x1, y1), (x1, y2), (x2, y1), and (x2, y2) are the coordinates of the four vertices corresponding to the small region selected by S4; scale is the ratio of the resolution of a high-resolution image to the resolution of a low-resolution image.

8. A microsatellite instability detection device based on multi-resolution and multi-regional methods, characterized in that, include: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors, the one or more programs including instructions for performing the microsatellite instability detection method according to any one of claims 1-7.

Images