Chromosome karyotype structural abnormality detection method and system based on banding encoding

Abstract

The application provides a chromosome karyotype structural abnormality detection method and system based on banding coding, comprising the following steps: generating coding information according to the gray scale of the banding of a chromosome monomer image collected by microscopic imaging, forming a chromosome karyotype coding subset by coding all monomers in the karyotype; extracting the skeleton of the stained monomer, inputting the skeleton information and the stained monomer image into a Transform coding network to generate coding information of the chromosome monomer; matching the obtained chromosome karyotype coding set with a chromosome karyotype database to obtain an abnormal type, and adding the karyotype information to the database if there is no matching result. The application greatly improves the sensitivity, accuracy and fine granularity of the chromosome structural karyotype abnormality detection, and is suitable for the research and diagnosis of chromosome structural abnormalities in the fields of medical genetics, oncology and the like.

Description

Technical Field

[0001] This invention relates to the field of biological genetic chromosome information recognition technology, specifically to a method and system for detecting chromosome karyotype structural abnormalities based on banding encoding. Background Technology

[0002] Chromosomal structural abnormalities refer to changes in the physical structure of chromosomes, leading to abnormalities in the number or arrangement of genes, affecting an individual's development, health, and reproductive capacity. Common types of abnormalities include deletions, duplications, inversions, translocations, ring chromosomes, and insertions. Chromosomal structural abnormalities have significant clinical importance in genetic counseling, disease diagnosis, and prognostic assessment. The application of modern genetic techniques provides crucial evidence for understanding and treating related diseases.

[0003] Traditional methods for detecting chromosomal structural abnormalities largely rely on experienced cytogeneticists who manually analyze chromosome morphology and structure under a microscope. This approach is not only low in automation but also severely limits accuracy and efficiency, as it requires extensive expertise and experience to accurately identify chromosomal abnormalities. This human-intensive method is both time-consuming and susceptible to subjective factors, increasing the risk of misdiagnosis and missed detection.

[0004] With the rapid development of artificial intelligence technology, methods for detecting chromosome structural abnormalities based on machine learning and deep learning have gradually emerged. These methods, by training on a large amount of chromosome image data, can automatically learn and identify complex structural features of chromosomes, thereby significantly improving the automation and accuracy of detection.

[0005] Song Ning et al. proposed a method and apparatus for training a classifier model to detect chromosomal structural abnormalities. This method and apparatus train the classifier model by obtaining the difference matrix between real normal chromosomes and artificially created defective chromosomes to detect chromosomal structural abnormalities. Although this patent supplements the data by constructing artificially created defective chromosomes, these artificial defects may not fully simulate the complexity and diversity of real defective chromosomes, thus affecting the model's generalization ability and detection accuracy.

[0006] Cai Yufeng et al. proposed a method, system, and computer equipment for assisting in the identification of chromosome structural abnormalities. This method generates reconstructed chromosome images and calculates their similarity to real chromosome images to assist doctors in identifying chromosome structural abnormalities. However, this method uses a similarity abnormality warning threshold to determine whether a chromosome is abnormal, but the similarity threshold may vary between different samples, leading to insufficient stability and reliability of the detection results.

[0007] The structural abnormalities detected by the above methods all focus on the entire chromosome and do not reach the band information, resulting in very low granularity. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for detecting chromosome karyotype structural abnormalities based on banding encoding.

[0009] The method for detecting chromosome karyotype structural abnormalities based on banding encoding according to the present invention includes:

[0010] The steps for constructing a chromosome coding dataset are as follows: For chromosome monomorph images acquired by microscopic imaging, coding information is generated based on the gray levels of the banding. All monomorphs within the karyotype are coded to form a chromosome karyotype coding subset. The coding subsets of all types of chromosomes with normal karyotypes and structurally abnormal karyotypes are combined to form the training set of the coding network, and the coding network is trained.

[0011] Automatic chromosome coding generation steps: acquire the karyotype image to be analyzed and input it into the coding network; the coding network outputs the coding information of the entire chromosome karyotype.

[0012] Chromosomal structural abnormality detection steps: Match the obtained chromosome karyotype encoding information with the chromosome karyotype database to obtain the abnormality type; if no matching result is found, add the information of the chromosome karyotype of that case to the database.

[0013] Preferably, the chromosome-encoded dataset construction step includes:

[0014] A dataset of metaphase images of chromosome division in normal human cells was obtained. This dataset includes chromosome images captured by photomicrography. The obtained chromosome metaphase image dataset was segmented to obtain chromatid images. The segmented chromosome images were straightened and aligned to ensure that the p-arms of the chromosomes were at the top of the image and the q-arms were at the bottom.

[0015] For each type of straightened chromosome dataset, all chromosome data are sampled from top to bottom to obtain the average pixel count and variance of the sampled pixel rows. For each sampled row, abnormal data that does not meet the preset requirements are removed, and then the sampling is performed again.

[0016] The obtained sampled data is divided into gray levels, and encoding information is generated based on the gray level division.

[0017] Preferably, the automatic chromosome encoding generation step includes:

[0018] The chromosome karyotype image was segmented into chromosome monocytes, and the skeleton of each chromosome monocyte was extracted.

[0019] For each chromatid, the input image is sliced ​​along its skeleton direction. Each square is sent as a small patch to the Embedding layer to flatten the input sequence.

[0020] The flattened sequence is input into a standard Transformer network to generate chromatid encoding information;

[0021] The encoding information output from all chromatids in the chromosome karyotype image after processing by Transformer is integrated to generate the overall encoding of the karyotype.

[0022] Preferably, the chromosome structural abnormality detection step includes:

[0023] The obtained chromosome karyotype codes are compared with the karyotype codes of normal chromosomes. A similarity threshold is set. When the similarity is less than the threshold, the chromosome is considered normal; when the similarity is greater than the threshold, the chromosome is considered abnormal.

[0024] Determine whether the abnormal chromosome involves two chromosomes, including chromosomal translocation and insertion; or one chromosome, including chromosomal deletion, duplication, inversion, and insertion, and look up the corresponding normal chromosome number.

[0025] Based on the obtained chromosome abnormality coding information, a search and matching process is performed with the abnormality coding information in the database. The specific location of the abnormality and the corresponding band information are determined based on the similarity of the matching.

[0026] Preferably, the obtained structural anomaly karyotype codes are matched with karyotype codes in the database based on similarity.

[0027] If the similarity is below the threshold, the doctor will determine the type of structural abnormality for this type of karyotype. If a structural abnormality is indeed found, the doctor's determination of the type of structural abnormality will be matched with the karyotype image.

[0028] The images of the structurally abnormal karyotypes, along with their matching chromosome codes and structural abnormality types, are stored in the database.

[0029] The chromosome karyotype structural abnormality detection system based on banding encoding provided by the present invention includes:

[0030] Chromosome coding dataset construction module: For chromosome monomorph images acquired by microscopic imaging, coding information is generated based on the gray level of the banding. All monomorphs within the karyotype are encoded to form a chromosome karyotype coding subset. The coding subsets of all types of chromosomes with normal karyotypes and structurally abnormal karyotypes are combined to form the training set of the coding network, and the coding network is trained.

[0031] Chromosome coding automatic generation module: acquires the karyotype image to be analyzed and inputs it into the coding network, and the coding network outputs the coding information of the entire chromosome karyotype;

[0032] Chromosomal structural abnormality detection module: Matches the obtained chromosome karyotype encoding information with the chromosome karyotype database to obtain the abnormality type; if no matching result is found, the information of the chromosome karyotype is added to the database.

[0033] Preferably, the chromosome coding dataset construction module includes:

[0034] A dataset of metaphase images of chromosome division in normal human cells was obtained. This dataset includes chromosome images captured by photomicrography. The obtained chromosome metaphase image dataset was segmented to obtain chromatid images. The segmented chromosome images were straightened and aligned to ensure that the p-arms of the chromosomes were at the top of the image and the q-arms were at the bottom.

[0035] For each type of straightened chromosome dataset, all chromosome data are sampled from top to bottom to obtain the average pixel count and variance of the sampled pixel rows. For each sampled row, abnormal data that does not meet the preset requirements are removed, and then the sampling is performed again.

[0036] The obtained sampled data is divided into gray levels, and encoding information is generated based on the gray level division.

[0037] Preferably, the chromosome encoding automatic generation module includes:

[0038] The chromosome karyotype image was segmented into chromosome monocytes, and the skeleton of each chromosome monocyte was extracted.

[0039] For each chromatid, the input image is sliced ​​along its skeleton direction. Each square is sent as a small patch to the Embedding layer to flatten the input sequence.

[0040] The flattened sequence is input into a standard Transformer network to generate chromatid encoding information;

[0041] The encoding information output from all chromatids in the chromosome karyotype image after processing by Transformer is integrated to generate the overall encoding of the karyotype.

[0042] Preferably, the chromosome structural abnormality detection module includes:

[0043] The obtained chromosome karyotype codes are compared with the karyotype codes of normal chromosomes. A similarity threshold is set. When the similarity is less than the threshold, the chromosome is considered normal; when the similarity is greater than the threshold, the chromosome is considered abnormal.

[0044] Determine whether the abnormal chromosome involves two chromosomes, including chromosomal translocation and insertion; or one chromosome, including chromosomal deletion, duplication, inversion, and insertion, and look up the corresponding normal chromosome number.

[0045] Based on the obtained chromosome abnormality coding information, a search and matching process is performed with the abnormality coding information in the database. The specific location of the abnormality and the corresponding band information are determined based on the similarity of the matching.

[0046] Preferably, the obtained structural anomaly karyotype codes are matched with karyotype codes in the database based on similarity.

[0047] If the similarity is below the threshold, the doctor will determine the type of structural abnormality for this type of karyotype. If a structural abnormality is indeed found, the doctor's determination of the type of structural abnormality will be matched with the karyotype image.

[0048] The images of the structurally abnormal karyotypes, along with their matching chromosome codes and structural abnormality types, are stored in the database.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] This invention proposes a method and system for detecting chromosomal karyotype structural abnormalities based on banding encoding. This method and system can automatically and efficiently identify minute structural abnormalities in chromosomes, thereby significantly improving the accuracy, sensitivity, and fine granularity of chromosomal karyotype structural abnormality detection. It is applicable to the research and diagnosis of chromosomal structural abnormalities in fields such as medical genetics and oncology. Attached Figure Description

[0051] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0052] Figure 1 A flowchart illustrating a method for detecting abnormal chromosome karyotype structures encoded by banding.

[0053] Figure 2 This is a schematic diagram of the grayscale distribution of chromosome bands;

[0054] Figure 3 A schematic diagram of the encoding network is generated;

[0055] Figure 4 This is a flowchart illustrating the process of determining the type of abnormality based on chromosome karyotype coding. Detailed Implementation

[0056] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the scope of protection of the present invention.

[0057] Example 1

[0058] like Figure 1 As shown in the figure, this embodiment provides a method for detecting chromosome karyotype structural abnormalities based on banding encoding, which includes the following steps:

[0059] S100, construction of chromosome coding dataset.

[0060] Specifically, the overall network structure is as follows: Figure 3 As shown, images of chromosome karyotype metaphase divisions were obtained from microscopes or image databases, and a dataset of chromatids was constructed containing normal and abnormal chromosomes.

[0061] The dataset includes chromosome karyotype images from multiple sources to ensure the robustness and broad applicability of the model;

[0062] The acquired chromosome karyotype images were preprocessed, including noise removal, image enhancement, and segmentation. Noise removal was performed using Gaussian filtering, image contrast was enhanced using histogram equalization, and a deep learning-based segmentation algorithm was used to separate chromatids from the image.

[0063] The pre-segmented dataset was classified. Normal chromosome karyotypes were divided into 24 categories, corresponding to the 22 pairs of human autosomes and the sex chromosomes X and Y. For abnormal chromosome karyotype images, abnormal chromosomes were additionally selected and labeled with their corresponding structural abnormality types.

[0064] The classified chromosome images are straightened and aligned, ensuring that the p-arms of the chromosomes are at the top of the image and the q-arms are at the bottom.

[0065] For each type of straightened chromosome dataset, all chromosome data are sampled from top to bottom to obtain the average pixel count of the sampled pixel rows;

[0066] For each sampling row, the mean ε and variance σ of the average pixels in that sampling row are calculated based on all chromosome images in that class of chromosomes. Data with average pixels outside [ε-2σ, ε+2σ] are removed as outliers.

[0067] The dataset after removing outlier data is sampled again to obtain the average pixel count of the sampled pixel rows;

[0068] Plot the average pixels of each sampling row as a curve s, set the ordinate of the lowest point to 0 and the ordinate of the highest point to 1, and obtain the parameter δ corresponding to the average pixels;

[0069]

[0070] Using δ as a threshold, calculate the area of ​​each curve located both on and below the δ line, and calculate the mean line for each part. The mean line divides the area of ​​each part equally, such as... Figure 2 As shown, the curve shown is a processed grayscale curve;

[0071] Calculate the ordinate h of the mean lines for each part i and length l i Then l i h i This is the encoding of that part. Combining the encodings of all parts yields the total encoding of the chromosome for that class, and this encoding information is used as a subset of the dataset.

[0072] The training set of the coding network is composed of the coding subsets of all types of chromosomes with normal karyotypes and structurally abnormal karyotypes. The network is trained to enable it to encode chromosomes.

[0073] S200, chromosome code automatically generated.

[0074] Specifically, deep learning algorithms are applied to the training dataset to segment it, so that the chromosome clusters in each chromosome karyotype metaphase division image are transformed into chromatids;

[0075] Obtain the backbone of each chromatid;

[0076] The chromatid image is segmented into patches along the direction perpendicular to the tangent of the skeleton, with each patch having a width equal to the width of the chromatid image.

[0077] Each patch is input into the Linear Projection of Flattened Patches (Embedding layer) and flattened into a vector X. e ;

[0078] Treat each patch as a text unit and add positional information X. p At the same time, add special character text units, such as class tokens;

[0079] X e +X p Obtain X t Input to the Encoder section;

[0080] Xt1 =Embedding-Lookup(X)+Positional-Encoding

[0081] The encoder consists of a first layer (Layer Normalization), a multi-head attention layer, a second layer (Layer Normalization), and a fully connected MLP layer. t After inputting into the Encoder, the input first enters the Layer Normalization layer;

[0082] The function of the Layer Normalization layer is to normalize the input to a standard normal distribution, X. t1 After entering the LayerNormalization layer, X is calculated first. t1 The mean and variance of the elements in the middle;

[0083]

[0084]

[0085] For X t1 Each element X in ij The standardization operation is performed, where ∈ is added to the denominator to prevent the denominator from being zero. The output of Layer Normalization is X. t2 ;

[0086]

[0087] Define three matrices W Q W K W V These three matrices will be used to apply X respectively. t2 Perform three linear transformations to obtain three new vectors q. t k t v t ;

[0088] q t =W Q ×X t2

[0089] k t =W K ×X t2

[0090] v t =W V ×X t2

[0091] All qt The vectors are concatenated into a large matrix, denoted as the query matrix Q, which contains all the k vectors. t The vectors are concatenated into a large matrix, denoted as the key matrix K, which contains all the v vectors. t Vectors are combined to form a large matrix, denoted as the value matrix V;

[0092] Q = Linear q (X)=XW Q

[0093] K = Linear k (X)=XW K

[0094] V = Linear v (X)=XW V

[0095] Define multiple groups of W Q W K W V This leads to the calculation of multiple sets of Q, K, and V matrices, with each set of matrices representing a Head.

[0096] For each Head, obtain the attention weights of the first patch, multiply them by the key matrix K using q1, and then perform a softmax operation on the resulting values ​​so that their sum is 1.

[0097] Multiply the obtained weights by their corresponding value vectors v. t ;

[0098] Sum the obtained fully weighted value vectors to get the output of the first patch. The above steps can be represented by a matrix. Where d k Let K be the dimension of the key vector.

[0099] Perform the same operation on the parameter matrices Q, K, and V in all Heads to obtain all outputs after self-attention.

[0100]

[0101] All Head Add to get X A ;

[0102]

[0103] X A With X t1 The summation is input into the second Layer Normalization layer;

[0104] X1 = X A +X t1

[0105] The previous layer's input and output values ​​are taken from the MLP fully connected layer to obtain output X2. Then, X1 and X2 are added together.

[0106] Repeat the above steps six times (N times) to obtain the final output, which is the total output of the final Encoder x3.

[0107] X3 is used as the parameter matrix K and V of the new multi-head attention mechanism and input into the Decoder;

[0108] The Decoder module consists of a first Layer Normalization layer, a Masked Multi-Head Self-Attention module, a second Layer Normalization layer, a Multi-Head Self-Attention module, a third Layer Normalization layer, and an MLP fully connected layer. The Multi-Head Self-Attention module, Layer Normalization layer, and MLP fully connected layer are the same as those in the Encoder module.

[0109] Before the Decoder, the unit encoding at time t is continuously input as the ground truth (the reference real value needed during training). This input goes directly into the internal Mask Multi-Head Attention module.

[0110] The ground truth, after passing through the Word Embedding and Positional Encoding modules, yields the output X. 4， Inputting X4 into the Layer Normalization layer yields X5;

[0111] Input X5 into the Masked Multi-Head Self-Attention module to perform the steps in the Encoder (i.e., calculate the...). This will produce the Scaled Scores output.

[0112] Perform a mask operation on the Scaled Scores, adding it to a matrix whose lower triangular values ​​are 0 and whose upper triangular values ​​are all negative infinity;

[0113]

[0114] Perform a softmax operation on the obtained Masked Scores, and the resulting matrix is ​​the weight of each encoding multiplied by 6.

[0115] The obtained weights X6 and X4 are added together to form the Q matrix, which is then input into the second Layer Normalization layer along with the K and V matrices output by the Encoder. The subsequent steps are the same as those for the Encoder part.

[0116] The above Decoder operation is performed 6 times, and the final output is X7;

[0117] Let X7 enter the fully connected layer and perform a softmax operation;

[0118] Fine-tune the network to enable it to encode abnormal chromosomes. Input the abnormal chromosomes and their corresponding abnormal codes into the network, freeze the first three layers of the Encoder part, load a new dataset for training, obtain the final output codes, and obtain the weight file of the trained network output after training is complete.

[0119] A set of chromosome karyotype metaphase images was selected as the test dataset, which included images of normal chromosomes and images of abnormal chromosomes.

[0120] Based on the weight file output by the training network, obtain the chromosome karyotype encoding corresponding to the test dataset.

[0121] S300, Chromosomal structural abnormality detection.

[0122] Specifically, such as Figure 4 As shown in the flowchart, the obtained chromosome karyotype code is compared with the chromosome karyotype code of normal structure. The Dynamic Time Warping (DTW) algorithm is used to calculate the similarity between the two, and a similarity threshold of 0.8 is set. When the similarity is less than this threshold, it is considered a normal chromosome; when the similarity is greater than this threshold, it is considered an abnormal chromosome.

[0123] For chromosomes identified as abnormal, the first step is to determine whether the abnormality involves two chromosomes (e.g., translocation, insertion) or a single chromosome (e.g., deletion, duplication, inversion, insertion). Based on the type of abnormality, the corresponding normal chromosome number is retrieved for further analysis.

[0124] The obtained abnormal coding information is compared with all the abnormal chromosome coding information in the dataset. All the abnormal chromosome coding information in the dataset is set as the left vertex set in the bipartite graph, and the obtained abnormal coding information to be matched is set as the right vertex set in the bipartite graph.

[0125] During the initialization phase, the matched state of all nodes in the bipartite graph is set to the unmatched state. This is achieved by constructing a bipartite graph to represent all possible encoded pairings.

[0126] Iterate through all unmatched nodes. For each unmatched node, use Depth-First Search (DFS) to find an augmenting path. An augmenting path is defined as a path whose starting and ending points are both unmatched nodes, and whose edges alternate between currently matched and unmatched edges. If an augmenting path is found, add the edges from that path to the matching process and update the matching state.

[0127] If no augmenting path is found, the search ends. If an augmenting path exists, return to the previous step and continue searching. If no augmenting path exists, the current match is the maximum match, and the algorithm ends.

[0128] Output the maximum matching result, which is the number of matching edges in the bipartite graph as much as possible;

[0129] Determine the similarity between the maximum matching result and the abnormal karyotype of the structure, and set the similarity threshold to 0.8. When the similarity is less than this threshold, the match is considered to have failed.

[0130] When the similarity is greater than the threshold, the specific abnormal part and its corresponding normal chromosome number are determined by combining the results of the augmentation path.

[0131] Karyotype images that cannot be matched with abnormal chromosome karyotypes in the dataset are given to doctors to determine whether there is a structural abnormality in the karyotype. If the doctor determines that there is a structural abnormality in the karyotype and there is no such abnormality in the database, the coding information of the structural abnormality karyotype and its matching, as well as the structural abnormality type, are stored in the database.

[0132] Based on the matching results, the abnormality type and location of the chromosome are generated, and a detection report is output, including the abnormal chromosome number, abnormality type, and specific abnormal band location.

[0133] The system displays test results and images of abnormal chromosomes. Users can input images and view test results through the user interface.

[0134] Example 2

[0135] The present invention also provides a chromosome karyotype structural abnormality detection system based on banding encoding. The chromosome karyotype structural abnormality detection system based on banding encoding can be implemented by executing the process steps of the chromosome karyotype structural abnormality detection method based on banding encoding. That is, those skilled in the art can understand the chromosome karyotype structural abnormality detection method based on banding encoding as a preferred embodiment of the chromosome karyotype structural abnormality detection system based on banding encoding.

[0136] The chromosome karyotype structural abnormality detection system based on banding encoding provided by the present invention includes:

[0137] Chromosome coding dataset construction module: For chromosome monomorph images acquired by microscopic imaging, coding information is generated based on the gray level of the banding. All monomorphs within the karyotype are encoded to form a chromosome karyotype coding subset. The coding subsets of all types of chromosomes with normal karyotypes and structurally abnormal karyotypes are combined to form the training set of the coding network, and the coding network is trained.

[0138] Chromosome coding automatic generation module: acquires the karyotype image to be analyzed and inputs it into the coding network, and the coding network outputs the coding information of the entire chromosome karyotype;

[0139] Chromosomal structural abnormality detection module: Matches the obtained chromosome karyotype encoding information with the chromosome karyotype database to obtain the abnormality type; if no matching result is found, the information of the chromosome karyotype is added to the database.

[0140] The chromosome-encoded dataset construction module includes:

[0141] A dataset of metaphase images of chromosome division in normal human cells was obtained. This dataset includes chromosome images captured by photomicrography. The obtained chromosome metaphase image dataset was segmented to obtain chromatid images. The segmented chromosome images were straightened and aligned to ensure that the p-arms of the chromosomes were at the top of the image and the q-arms were at the bottom.

[0142] For each type of straightened chromosome dataset, all chromosome data are sampled from top to bottom to obtain the average pixel count and variance of the sampled pixel rows. For each sampled row, abnormal data that does not meet the preset requirements are removed, and then the sampling is performed again.

[0143] The obtained sampled data is divided into gray levels, and encoding information is generated based on the gray level division.

[0144] The chromosome encoding automatic generation module includes:

[0145] The chromosome karyotype image was segmented into chromosome monocytes, and the skeleton of each chromosome monocyte was extracted.

[0146] For each chromatid, the input image is sliced ​​along its skeleton direction. Each square is sent as a small patch to the Embedding layer to flatten the input sequence.

[0147] The flattened sequence is input into a standard Transformer network to generate chromatid encoding information;

[0148] The encoding information output from all chromatids in the chromosome karyotype image after processing by Transformer is integrated to generate the overall encoding of the karyotype.

[0149] The chromosome structural abnormality detection module includes:

[0150] The obtained chromosome karyotype codes are compared with the karyotype codes of normal chromosomes. A similarity threshold is set. When the similarity is less than the threshold, the chromosome is considered normal; when the similarity is greater than the threshold, the chromosome is considered abnormal.

[0151] Determine whether the abnormal chromosome involves two chromosomes, including chromosomal translocation and insertion; or one chromosome, including chromosomal deletion, duplication, inversion, and insertion, and look up the corresponding normal chromosome number.

[0152] Based on the obtained chromosome abnormality coding information, a search and matching process is performed with the abnormality coding information in the database. The specific location of the abnormality and the corresponding band information are determined based on the similarity of the matching.

[0153] The obtained structural anomaly kernel codes are matched with the kernel codes in the database for similarity.

[0154] If the similarity is below the threshold, the doctor will determine the type of structural abnormality for this type of karyotype. If a structural abnormality is indeed found, the doctor's determination of the type of structural abnormality will be matched with the karyotype image.

[0155] The images of the structurally abnormal karyotypes, along with their matching chromosome codes and structural abnormality types, are stored in the database.

[0156] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0157] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for detecting chromosome karyotype structural abnormalities based on banding encoding, characterized in that, include: The steps for constructing a chromosome coding dataset are as follows: For chromosome monomorph images acquired by microscopic imaging, coding information is generated based on the gray levels of the banding. All monomorphs within the karyotype are coded to form a chromosome karyotype coding subset. The coding subsets of all types of chromosomes with normal karyotypes and structurally abnormal karyotypes are combined to form the training set of the coding network, and the coding network is trained. Automatic chromosome coding generation steps: acquire the karyotype image to be analyzed and input it into the coding network; the coding network outputs the coding information of the entire chromosome karyotype. Chromosomal structural abnormality detection steps: Match the obtained chromosome karyotype encoding information with the chromosome karyotype database to obtain the abnormality type; If no match is found, the karyotype information for that example will be added to the database. The automatic chromosome coding generation step includes: The chromosome karyotype image was segmented into chromosome monocytes, and the skeleton of each chromosome monocyte was extracted. For each chromatid, the input image is sliced ​​along its skeleton direction. Each square is sent as a small patch to the Embedding layer to flatten the input sequence. The flattened sequence is input into a standard Transformer network to generate chromatid encoding information; The encoding information output from all chromatids in the chromosome karyotype image after processing by Transformer is integrated to generate the overall encoding of the karyotype.

2. The method for detecting chromosome karyotype structural abnormalities based on banding encoding according to claim 1, characterized in that, The steps for constructing the chromosome-encoded dataset include: A dataset of metaphase images of chromosome division in normal human cells was obtained. This dataset includes chromosome images captured by photomicrography. The obtained chromosome metaphase image dataset was segmented to obtain chromatid images. The segmented chromosome images were straightened and aligned to ensure that the p-arms of the chromosomes were at the top of the image and the q-arms were at the bottom. For each type of straightened chromosome dataset, all chromosome data are sampled from top to bottom to obtain the average pixel count and variance of the sampled pixel rows. For each sampled row, abnormal data that does not meet the preset requirements are removed, and then the sampling is performed again. The obtained sampled data is divided into gray levels, and encoding information is generated based on the gray level division.

3. The method for detecting chromosome karyotype structural abnormalities based on banding encoding according to claim 1, characterized in that, The chromosome structural abnormality detection steps include: The obtained chromosome karyotype codes are compared with the karyotype codes of normal chromosomes. A similarity threshold is set. When the similarity is less than the threshold, the chromosome is considered normal; when the similarity is greater than the threshold, the chromosome is considered abnormal. Determine whether the abnormal chromosome involves two chromosomes, including chromosomal translocation and insertion; or one chromosome, including chromosomal deletion, duplication, inversion, and insertion, and look up the corresponding normal chromosome number. Based on the obtained chromosome abnormality coding information, a search and matching process is performed with the abnormality coding information in the database. The specific location of the abnormality and the corresponding band information are determined based on the similarity of the matching.

4. The method for detecting chromosome karyotype structural abnormalities based on banding encoding according to claim 3, characterized in that, The obtained structural anomaly kernel codes are matched with the kernel codes in the database for similarity. If the similarity is below the threshold, the doctor will determine the type of structural abnormality for this type of karyotype. If a structural abnormality is indeed found, the doctor's determination of the type of structural abnormality will be matched with the karyotype image. The images of the structurally abnormal karyotypes, along with their matching chromosome codes and structural abnormality types, are stored in the database.

5. A chromosome karyotype structural abnormality detection system based on banding encoding, characterized in that, include: Chromosome coding dataset construction module: For chromosome monomorph images acquired by microscopic imaging, coding information is generated based on the gray level of the banding. All monomorphs within the karyotype are encoded to form a chromosome karyotype coding subset. The coding subsets of all types of chromosomes with normal karyotypes and structurally abnormal karyotypes are combined to form the training set of the coding network, and the coding network is trained. Chromosome coding automatic generation module: acquires the karyotype image to be analyzed and inputs it into the coding network, and the coding network outputs the coding information of the entire chromosome karyotype; Chromosomal structural abnormality detection module: Matches the obtained chromosome karyotype encoding information with the chromosome karyotype database to obtain the abnormality type; If no match is found, the karyotype information for that example will be added to the database. The chromosome encoding automatic generation module includes: The chromosome karyotype image was segmented into chromosome monocytes, and the skeleton of each chromosome monocyte was extracted. For each chromatid, the input image is sliced ​​along its skeleton direction. Each square is sent as a small patch to the Embedding layer to flatten the input sequence. The flattened sequence is input into a standard Transformer network to generate chromatid encoding information; The encoding information output from all chromatids in the chromosome karyotype image after processing by Transformer is integrated to generate the overall encoding of the karyotype.

6. The chromosome karyotype structural abnormality detection system based on banding encoding according to claim 5, characterized in that, The chromosome-encoded dataset construction module includes: A dataset of metaphase images of chromosome division in normal human cells was obtained. This dataset includes chromosome images captured by photomicrography. The obtained chromosome metaphase image dataset was segmented to obtain chromatid images. The segmented chromosome images were straightened and aligned to ensure that the p-arms of the chromosomes were at the top of the image and the q-arms were at the bottom. For each type of straightened chromosome dataset, all chromosome data are sampled from top to bottom to obtain the average pixel count and variance of the sampled pixel rows. For each sampled row, abnormal data that does not meet the preset requirements are removed, and then the sampling is performed again. The obtained sampled data is divided into gray levels, and encoding information is generated based on the gray level division.

7. The chromosome karyotype structural abnormality detection system based on banding encoding according to claim 5, characterized in that, The chromosome structural abnormality detection module includes: The obtained chromosome karyotype codes are compared with the karyotype codes of normal chromosomes. A similarity threshold is set. When the similarity is less than the threshold, the chromosome is considered normal; when the similarity is greater than the threshold, the chromosome is considered abnormal. Determine whether the abnormal chromosome involves two chromosomes, including chromosomal translocation and insertion; or one chromosome, including chromosomal deletion, duplication, inversion, and insertion, and look up the corresponding normal chromosome number. Based on the obtained chromosome abnormality coding information, a search and matching process is performed with the abnormality coding information in the database. The specific location of the abnormality and the corresponding band information are determined based on the similarity of the matching.

8. The chromosome karyotype structural abnormality detection system based on banding encoding according to claim 7, characterized in that, The obtained structural anomaly kernel codes are matched with the kernel codes in the database for similarity. If the similarity is below the threshold, the doctor will determine the type of structural abnormality for this type of karyotype. If a structural abnormality is indeed found, the doctor's determination of the type of structural abnormality will be matched with the karyotype image. The images of the structurally abnormal karyotypes, along with their matching chromosome codes and structural abnormality types, are stored in the database.

Images