Chromosome classification method based on sequence cross correlation
By padding chromosome images to multiples of 8 and then cutting them into 8×8 image expansion data blocks, and combining local and global self-attention encoders, the problems of chromosome size information loss and feature relationship neglect in traditional methods are solved, achieving higher chromosome classification accuracy and generalization.
Patent Information
- Application Number
- CN202511003798.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-21
- Publication Date
- 2025-11-07
AI Technical Summary
Traditional convolutional neural networks require scaling or padding images to a fixed size for chromosome image classification, which leads to the loss of chromosome size information, affecting classification accuracy. Furthermore, they ignore the feature relationships between homologous chromosomes and the whole chromosome, resulting in classification errors.
A chromosome classification method based on sequence cross-correlation is adopted. The chromosome image is padded to a multiple of 8 and then cut into 8×8 image expansion data blocks to generate position information. Sequences are generated through linear embedding. The feature encoders of local self-attention layer and global self-attention layer are combined to preserve chromosome size information and model the relationship between global and local features.
It effectively improves the generalizability and accuracy of chromosome classification, can reliably handle complex situations such as chromosome crossover, contamination, and adhesion, adapts to differences in slide preparation by different personnel and scenarios with poor slide preparation, and provides better support for chromosome karyotype analysis.
Smart Images
Figure CN120913200A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of chromosome karyotype analysis, and particularly to a chromosome classification method based on sequence cross-correlation. BACKGROUND
[0002] In the classification of chromosome images, a traditional convolutional neural network (CNN) needs to scale or pad the images to a fixed size, resulting in the loss of the size information of the chromosomes themselves and affecting the classification accuracy. Meanwhile, existing models mostly focus on the features of a single chromosome, ignoring the feature relationships between homologous chromosomes and the overall chromosomes, and are prone to classification errors when the chromosomes are crossing, contaminated, adhered, or the quality of the prepared slides is poor. Therefore, there is an urgent need for a classification method that can retain the size information of the chromosomes and model the relationships between the overall and local features. SUMMARY
[0003] Therefore, it is necessary to provide a chromosome classification method based on sequence cross-correlation to solve the above technical problems.
[0004] A chromosome classification method based on sequence cross-correlation includes the following steps:
[0005] Obtain a chromosome cell image, and obtain chromosome images of all single chromosomes according to the chromosome cell image;
[0006] Fill the width and height of all the chromosome images to a multiple of 8, and then cut them into 8x8 image extension data blocks. Generate position information according to the cutting positions of the image extension data blocks;
[0007] Based on the image extension data blocks, generate sequences according to the processing time through linear embedding, and add the position information to the sequences to obtain image extension data sequences;
[0008] Input the image extension data sequences into a feature encoder, and output sequence features;
[0009] Pool the sequence features, and output an initial classification result through a classification head;
[0010] Calculate a class probability distribution based on the initial classification result through softmax, and take the index of the highest probability as the classification result.
[0011] In one embodiment, generating the position information according to the cutting positions of the image extension data blocks includes:
[0012] Generating the position information according to the cutting positions of the image extension data blocks through a sine-cosine position encoding algorithm.
[0013] In one embodiment, generating the sequences according to the processing time through linear embedding based on the image extension data blocks includes:
[0014] Based on the image expansion data block, each image expansion data block is mapped as a fixed dimension vector according to the processing time by linear embedding, and a sequence is generated.
[0015] In one of the embodiments, the feature encoder comprises a local self-attention layer, a global self-attention layer and a feature fusion layer.
[0016] The local self-attention layer extracts subsequence features according to the position and length of the sequence features of a single chromosome, performs multi-head self-attention calculation and feature encoding, and then processes through the FFN layer to obtain local features.
[0017] The global self-attention layer performs multi-head self-attention calculation and feature encoding according to all sequence features of the chromosome cell map, and then processes through the FFN layer to obtain global features.
[0018] The feature fusion layer fuses and adds the local features and the global features to obtain sequence features.
[0019] In one of the embodiments, the classification head is a single-layer linear layer, and the output dimension is the total number of chromosome categories 24.
[0020] In one of the embodiments, based on the initial classification result, a category probability distribution is calculated by softmax, and the highest probability index is taken as the classification result.
[0021] The formula of softmax calculation is as follows:
[0022]
[0023] Wherein, softmax represents the probability of the current chromosome belonging to a certain category, C represents the total number of classification categories, j represents the category index, z i represents the i-th category score output by the chromosome classification head linear layer, e represents the exponential score, and z j represents the original confidence of judging that the chromosome belongs to the j-th category.
[0024] A chromosome classification system based on sequence cross-correlation is used to implement the chromosome classification method based on sequence cross-correlation as described above, comprising:
[0025] An image acquisition module is configured to acquire a chromosome cell map and acquire chromosome images of all single chromosomes according to the chromosome cell map.
[0026] An image segmentation module is configured to fill the width and height of all chromosome images to a multiple of 8, cut them into 8x8 image expansion data blocks, and generate position information according to the cutting position of the image expansion data blocks.
[0027] The sequence generation module is configured to generate a sequence by linear embedding according to a processing time based on the image extension data block, and add the position information to the sequence to obtain an image extension data sequence.
[0028] The feature acquisition module is configured to input the image extension data sequence into a feature encoder to output sequence features.
[0029] The initial classification module is configured to pool the sequence features and output an initial classification result through a classification head.
[0030] The result acquisition module is configured to calculate a class probability distribution through a softmax based on the initial classification result, and take a highest probability index as a classification result.
[0031] An apparatus includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the steps of a chromosome classification method based on sequence mutual correlation described in each of the above embodiments when executing the program.
[0032] A storage medium has a computer program stored thereon, and the program implements the steps of a chromosome classification method based on sequence mutual correlation described in each of the above embodiments when executed by a processor.
[0033] Compared with the prior art, the advantages and beneficial effects of the present application are that the present application avoids the damage to the original size information of chromosomes caused by the traditional scaling or padding operation by filling the chromosome image to a multiple of 8 and then cutting it into 8x8 image extension data blocks, so that the key classification clue of the size difference of different types of chromosomes is retained; at the same time, the feature encoder containing global and local self-attention is innovatively used, which focuses on the strip details of a single chromosome through local attention, and models the overall distribution relationship between chromosomes and the similarity of homologous chromosomes through global attention, effectively making up for the limitations of traditional methods that only focus on the features of a single chromosome. This makes the method have stronger generalization and classification accuracy when facing complex situations such as chromosome crossing, contamination, and adhesion, as well as different personnel preparation differences, poor preparation, and non-obvious chromosome features, and can more reliably complete the classification of 24 types of chromosomes, providing better technical support for chromosome karyotype analysis. BRIEF DESCRIPTION OF DRAWINGS
[0034] Figure 1 It is a whole framework diagram of a chromosome classification method based on sequence mutual correlation in one embodiment;
[0035] Figure 2 It is a flowchart of a chromosome classification method based on sequence mutual correlation in one embodiment;
[0036] Figure 3 A model framework structure diagram of a sequence cross-correlation-based chromosome classification method in an embodiment;
[0037] Figure 4 A structural schematic diagram of a sequence cross-correlation-based chromosome classification system in an embodiment;
[0038] Figure 5 An internal structural schematic diagram of an apparatus in an embodiment. DETAILED DESCRIPTION
[0039] Before the specific embodiment of the present application is described, the overall concept of the present application is described as follows:
[0040] The present application is mainly developed to improve the generalization of chromosome classification in the presence of complex situations such as crossing, contamination, adhesion, and poor or unsatisfactory preparation and unclear features of the preparation. The traditional chromosome classification method is insufficient in classification accuracy and generalization due to the loss of size information and the neglect of overall feature relationships when facing these situations.
[0041] The inventors found that the main reason for the above problems is that the existing method needs to scale or pad the chromosome image to a fixed size, which destroys the key size information, and only focuses on the features of a single chromosome while ignoring the feature distribution relationship between homologous chromosomes and the overall chromosome. If the original size information of the chromosome is preserved and the relationship between the single chromosome feature and the overall feature is modeled, the above problems can be avoided.
[0042] Therefore, the present application proposes a sequence cross-correlation-based chromosome classification method, which cuts the chromosome image into 8x8 image expansion data blocks after padding it to a multiple of 8 and generates a sequence, combines linear embedding, position coding, and feature encoder with global and local attention, realizes accurate classification of chromosomes, and effectively improves the generalization in complex scenarios.
[0043] After introducing the overall concept of the present application, in order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application will be further described in detail through specific embodiments combined with the drawings.
[0044] It should be noted that, unless otherwise defined, technical or scientific terms used in one or more embodiments of the present specification shall have the meanings commonly understood by one of ordinary skill in the art to which the present application pertains. The terms "first", "second", and similar terms used in one or more embodiments of the present specification do not denote any order, quantity, or importance, but are used to distinguish different components. The terms "include" or "contain" and similar terms mean that the elements or objects before the terms encompass the elements or objects listed after the terms and their equivalents, and do not exclude other elements or objects. The terms "connected" or "connected" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right", and the like are only used to represent relative positional relationships, and when the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0045] For the convenience of understanding, the following explains the terms involved in the embodiments of the present application:
[0046] Softmax: Normalized exponential function, which converts a real number vector into a probability distribution, so that each element in the vector ranges from 0 to 1, and the sum of all elements is 1.
[0047] FFN layer: Feedforward neural network layer, used for nonlinear transformation and information enhancement of the features output by the self-attention mechanism.
[0048] patch block: A fixed-size pixel block (8x8 pixels) formed after cutting the chromosome image, which is the basic unit for converting a two-dimensional image into a one-dimensional sequence. Its essence is to discretize the continuous chromosome image into multiple local pixel sets, both preserving the local visual features of the image (such as strip texture, gray scale variation) and reflecting the original size information of the chromosome through the number difference.
[0049] Linear embedding: Linear embedding, which maps the 8x8 pixel patch block to a fixed-dimensional vector through linear transformation, so that the two-dimensional image information is converted into a one-dimensional "text-like" sequence, providing input for the subsequent self-attention mechanism.
[0050] Classified head: Classification head, which maps the encoded sequence features to specific categories
[0051] As shown in Figure 1 , it is the overall framework diagram of a chromosome classification method based on sequence cross-correlation. First, all chromosome images in the entire chromosome cell image are obtained. Fill all the chromosome images to the multiples of 8, and cut them into 8*8 patch blocks (image expansion data blocks), and then generate position information p according to the slice position of each imagesi ;
[0052] All patch blocks are spliced into a patch sequence in order, and then linear embedding is performed on the patch sequence to obtain a sequence seq;
[0053] The position information of the patch blocks is embedded in the sequence seq;
[0054] The sequence seq is input into a feature encoder encoder to obtain sequence features seq of the sequence seq i
[0055] Each sequence feature seq i is subjected to a pooling operation, and then classified by a classified head;
[0056] The classification result is calculated by a softmax to obtain a class probability distribution, and finally the index with the highest output probability value is taken as the class c of the current chromosome image, c∈[1,2,3,...,22,X,Y], a total of 24 classes.
[0057] In one embodiment, as shown in Figure 2 , a sequence cross-correlation-based chromosome classification method is provided, comprising the following steps:
[0058] Step S201, obtaining a chromosome cell map, and obtaining all single-chromosome chromosome images according to the chromosome cell map.
[0059] Specifically, all chromosome images in the whole chromosome cell map are obtained.
[0060] Step S202, filling the width and height of all chromosome images to a multiple of 8, cutting them into 8x8 image expansion data blocks, and generating position information according to the cutting position of the image expansion data blocks.
[0061] Specifically, in karyotype analysis, different sizes of chromosomes belong to different class groups. If the width and height of all chromosome images are adjusted to a fixed size or a uniform width-height ratio, this regulation will be destroyed, which will affect the classification result. In the present application, the width and height of all chromosome images obtained in step S201 are filled to a multiple of 8. Then the filled images are cut into 8x8 image expansion data blocks, and the length of the image expansion data blocks is obtained i , i∈[1,n], n is the number of chromosome images.
[0062] On this basis, the position information is generated according to the cutting position of the image expansion data blocks, which includes:
[0063] According to the image extension data block cutting position, position information is generated by a sinusoidal positional encoding algorithm.
[0064] Specifically, the image extension data blocks are unfolded into sub-sequences sub i in the order of left-to-right positions. i According to a sinusoidal positional encoding (Sinusoidal Positional Encoding) algorithm, the position information p si of the sub-sequences sub i is encoded.
[0065] In step S203, based on the image extension data blocks, a sequence is generated by linear embedding according to processing time, and the position information is added to the sequence to obtain an image extension data sequence.
[0066] On this basis, based on the image extension data blocks, a sequence is generated by linear embedding according to processing time, including:
[0067] Based on the image extension data blocks, each image extension data block is mapped to a fixed-dimensional vector by linear embedding according to processing time to generate a sequence.
[0068] Specifically, n sub-sequences sub i are spliced into a sequence in the order of processing, and the position p i of each sub-sequence in the sequence is recorded. Then linear embedding is performed, so that each patch block (image extension data block) is mapped to a sequence token, and finally a sequence seq similar to text is obtained.
[0069] The token is a feature vector formed after linear embedding, and is a basic data unit for connecting the chromosome image and the Transformer model. Its essence is to map the image patch (pixel block) to a fixed-dimensional vector through linear transformation, so that the two-dimensional image information can be serialized into one-dimensional "text-like" form for processing by the self-attention mechanism of the Transformer.
[0070] According to the position p i and the length l i , the position information p si is embedded into the sequence seq to obtain an image extension data sequence.
[0071] In step S204, the image extension data sequence is input into a feature encoder to output a sequence feature.
[0072] Specifically, the image expansion data sequence is input into the feature encoder designed by the application, wherein the feature encoder is composed of 12 identical global and local self-attention blocks.
[0073] On this basis, the feature encoder comprises a local self-attention layer, a global self-attention layer and a feature fusion layer.
[0074] The local self-attention layer extracts subsequence features according to the position and length of the sequence features of a single chromosome, performs multi-head self-attention calculation and feature encoding, and then obtains local features through FFN layer processing.
[0075] The global self-attention layer performs multi-head self-attention calculation and feature encoding according to all sequence features of the chromosome cell map, and then obtains global features through FFN layer processing.
[0076] The feature fusion layer fuses and adds the local features and the global features to obtain sequence features.
[0077] Specifically, the global and local self-attention block comprises two parts: a local self-attention layer and a global self-attention layer in parallel, as shown in Figure 3 The fusion results are connected through a residual connection.(1) The local self-attention layer, which mainly focuses on the encoding of single chromosome features, inputs each chromosome subsequence sub i According to the position p i and length l i of each chromosome subsequence sub i in the image expansion data sequence, subsequence features seq i ∈{seq1,seq2,seq3,seq4,...,seq n} are obtained, and seq iThe multi-head self-attention calculation and feature encoding are performed to capture the local features (such as band texture and edge continuity) in a single chromosome. Then, the feature encoding of this part is completed through the enhancement of the non-linear expression by the FFN layer (consisting of two linear layers); (2) the global self-attention layer, which mainly focuses on the similarity and discrimination encoding between all chromosome features under the chromosome cell graph. The global image expansion data sequence of all chromosomes is input, and the multi-head self-attention calculation is directly performed on the image expansion data sequence to calculate the correlation between all sequences (whether belonging to the same chromosome or not). Then, the feature encoding of this part is completed through the FFN layer. Finally, the results of the two parts are fused and added to obtain the encoding result of this module, that is, the sequence feature. Such a feature encoder not only considers the feature distribution of a single chromosome, but also combines the feature distribution of other same and different types in the whole cell graph. Compared with the feature encoder for single chromosome image classification, the class feature of the feature encoder is more accurate and the generalization is more robust.
[0078] In the present embodiment, the local branch is dynamically divided by the position and length of the chromosome sub-sequence, rather than a fixed window. This design preserves the original size difference of the chromosome (the large chromosome sub-sequence is longer), which meets the size-dependent requirements of clinical classification.
[0079] The local and global branches are calculated synchronously to avoid the time-consuming serial of the traditional method of first local and then global. Furthermore, the position information of the chromosome sequence is combined to strengthen the size perception.
[0080] The module is repeatedly stacked for 12 layers, and the interaction between the local and global features is refined in each layer. Compared with a shallow model, a deeper level is more suitable for learning the complex band pattern of the chromosome.
[0081] In step S205, the sequence feature is pooled, and the initial classification result is output through the classification head.
[0082] Specifically, according to each chromosome sub-sequence sub i The position p i and the length l i of the image expansion data sequence are obtained in the output of the last layer of the feature encoder encoder. i The sub-sequence feature seq i is pooled and classified through the classified head (a linear layer).
[0083] On this basis, the classification head is a single-layer linear layer, and the output dimension is the total number of chromosome categories, which is 24.
[0084] Specifically, the sequence features after the pooling operation (the key features of each chromosome subsequence are aggregated into a fixed dimension vector) are mapped to a 24-dimensional output vector (corresponding to 24 categories of chromosomes) through linear transformation (matrix multiplication), and the output is the original category score (initial classification result, logits without normalization).
[0085] In step S206, the category probability distribution is calculated based on the initial classification result by using softmax, and the index with the highest probability is taken as the classification result.
[0086] Specifically, the classification result is calculated by using softmax to obtain the probability distribution of the category, and finally the index with the highest probability value is taken as the category c of the current chromosome image, c∈[1, 2, 3,..., 22, X, Y], which is divided into 24 categories.
[0087] On this basis, the category probability distribution is calculated based on the initial classification result by using softmax, and the index with the highest probability is taken as the classification result, including:
[0088] The formula for calculating softmax is as follows:
[0089]
[0090] Wherein, softmax represents the probability that the current chromosome belongs to a certain category, C represents the total number of classification categories, j represents the category index, z i represents the i-th category score output by the chromosome classification head linear layer, e represents the exponential score, and z j represents the original confidence that the chromosome belongs to the j-th category.
[0091] The application provides a chromosome classification method based on sequence mutual correlation, which avoids damage to the original proportion of chromosomes caused by fixed size scaling, and converts the difference in the number of patches of chromosomes of different sizes into a sequence length difference, so that the model can directly use the key rule of "chromosome size grouping" in clinical karyotype analysis.
[0092] It should be noted that the method of the embodiments of the application can be executed by a single device, such as a computer or a server. The method of the embodiments of the application can also be applied to a distributed scenario and completed by multiple devices in cooperation. In the distributed scenario, one of the multiple devices can only execute one or more steps in the method of the embodiments of the application, and the multiple devices can interact with each other to complete the method.
[0093] It should be noted that some embodiments of the application have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than the order described above and still achieve the desired result. In addition, the processes depicted in the figures do not necessarily require the particular order shown or sequential order to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous or necessary.
[0094] Based on the same inventive concept, the application also provides a chromosome classification system based on sequence mutual correlation corresponding to any of the above-mentioned embodiment methods.
[0095] Reference Figure 4 The chromosome classification system based on sequence mutual correlation comprises:
[0096] The image acquisition module 401 is configured to acquire a chromosome cell image and acquire chromosome images of all single chromosomes according to the chromosome cell image.
[0097] An image segmentation module 402 is configured to fill the width and height of all chromosome images to a multiple of 8, and then cut the filled images into image extension data blocks of 8x8, and generate position information according to the cutting positions of the image extension data blocks;
[0098] A sequence generation module 403 is configured to generate a sequence by linear embedding according to a processing time based on the image extension data blocks, and add the position information to the sequence to obtain an image extension data sequence.
[0099] A feature acquisition module 404 is configured to input the image extension data sequence into a feature encoder, and output sequence features.
[0100] An initial classification module 405 is configured to pool the sequence features, and output an initial classification result through a classification head.
[0101] A result acquisition module 406 is configured to calculate a class probability distribution by softmax based on the initial classification result, and take the index of the highest probability as a classification result.
[0102] For the convenience of description, the above system is described in various modules in terms of functions. Of course, the functions of the modules can be implemented in one or more software and / or hardware when the present application is implemented.
[0103] The system of the above embodiment is used to implement the chromosome classification method based on sequence mutual correlation in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be described here.
[0104] Based on the same inventive concept, the present application also provides an electronic device corresponding to the method of any of the above embodiments, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the chromosome classification method based on sequence mutual correlation according to any of the above embodiments when executing the program.
[0105] Figure 5 A more specific hardware structure of an electronic device according to the present embodiment is shown in the figure. The device can include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are connected to each other through the bus 1050 for communication within the device.
[0106] The processor 1010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits, etc., for executing relevant programs to implement the technical solutions provided by the embodiments of the present specification.
[0107] The memory 1020 can be implemented by a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, etc. The memory 1020 can store an operating system and other application programs, and when the technical solutions provided by the embodiments of the present specification are implemented by software or firmware, the relevant program codes are stored in the memory 1020 and called and executed by the processor 1010.
[0108] The input / output interface 1030 is configured to connect input / output modules to implement information input and output. The input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. The input devices can include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output devices can include a display, a speaker, a vibrator, an indicator light, etc.
[0109] The communication interface 1040 is configured to connect a communication module (not shown in the figure) to implement the communication interaction between the device and other devices. The communication module can realize communication through a wired manner (such as USB, network cable, etc.) or through a wireless manner (such as mobile network, WIFI, Bluetooth, etc.).
[0110] The bus 1050 includes a channel for transmitting information between various components (such as the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040) of the device.
[0111] It should be noted that although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040, and the bus 1050, in the specific implementation process, the device can also include other components necessary for normal operation. In addition, those skilled in the art can understand that the above device can also only include the components necessary to implement the solutions of the embodiments of the present specification, and does not have to include all the components shown in the figure.
[0112] The electronic device of the above embodiment is used to implement the sequence cross-correlation-based chromosome classification method of any one of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.
[0113] Based on the same inventive concept, the present application also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the sequence cross-correlation-based chromosome classification method of any one of the above embodiments.
[0114] The computer-readable medium of the present embodiment includes permanent and non-permanent, removable and non-removable media, which can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device.
[0115] The computer instructions stored in the storage medium of the above embodiment are used to cause the computer to perform the sequence cross-correlation-based chromosome classification method of any one of the above embodiments, and have the beneficial effects of the corresponding method embodiment, which will not be repeated here.
[0116] Those skilled in the art should understand that the discussion of any of the above embodiments is only exemplary and is not intended to imply that the scope (including claims) of the present application is limited to these examples; under the idea of the present application, the above embodiments or technical features between different embodiments can also be combined, the steps can be implemented in any order, and there are many other changes of different aspects of the embodiments of the present application as described above. In order to be brief, they are not provided in detail.
[0117] Any process or method described in a flowchart or otherwise described herein can be understood as representing code modules, segments, or portions of code that include one or more executable instructions for implementing specific logic functions (or steps) of the process, and alternate implementations are possible. In this description and in the claims, each of the processes or methods that is described can be understood as comprising code stored in a computer-readable medium that, when executed by a computer, can cause the computer to perform a process. In this description and in the claims, each of the processes or methods that is described can be understood as comprising code stored in a computer-readable medium that, when executed by a computer, can cause the computer to perform a process.
[0118] While specific embodiments of the application have been described in detail, it will be appreciated by those skilled in the art that various modifications and alternatives to those details could be developed in light of the overall teachings of the disclosure. Accordingly, this description is to be construed as illustrative only and is for the purpose of teaching those skilled in the art the general manner of carrying out the application. It is to be understood that the forms of the application herein shown and described are to be taken as the presently preferred embodiments. Elements and materials can be substituted for one another
[0119] Embodiments of the application are intended to cover all alternatives, modifications, and equivalents falling within the scope of the appended claims. Accordingly, any and all such alternatives, modifications, and equivalents are intended to be encompassed by the appended claims.
Claims
1. A sequence cross-correlation based chromosome classification method, characterized by, The method comprises the following steps: obtaining a chromosome cell image, and obtaining chromosome images of all single chromosomes according to the chromosome cell image; filling the width and height of all chromosome images to a multiple of 8, and cutting the filled images into 8*8 image extension data blocks, and generating position information according to the cutting positions of the image extension data blocks; generating a sequence by linear embedding according to the processing time based on the image extension data blocks, adding the position information to the sequence, and obtaining an image extension data sequence; inputting the image extension data sequence into a feature encoder, and outputting sequence features; pooling the sequence features, and outputting an initial classification result through a classification head; calculating a class probability distribution through softmax based on the initial classification result, and taking the index of the highest probability as the classification result.
2. The method of claim 1, wherein the step of classifying the chromosomes based on the sequence cross-correlation is performed by a computer program. The position information is generated according to the cutting positions of the image extension data blocks through a sine-cosine position encoding algorithm. The sequence is generated by linear embedding according to the processing time based on the image extension data blocks.
3. The method of claim 1, wherein the method is based on sequence cross-correlation. The feature encoder comprises a local self-attention layer, a global self-attention layer and a feature fusion layer. The local self-attention layer extracts subsequence features according to the position and length of the sequence features of a single chromosome, performs multi-head self-attention calculation and feature encoding, and then processes through an FFN layer to obtain local features.
4. The method of claim 1, wherein the step of classifying the chromosomes based on the sequence cross-correlation is characterized by, The global self-attention layer performs multi-head self-attention calculation and feature encoding according to all sequence features of the chromosome cell image, and then processes through an FFN layer to obtain global features. The feature fusion layer fuses and adds the local features and the global features to obtain sequence features. The classification head is a single-layer linear layer, and the output dimension is the total number of chromosome classes, which is 24. The softmax calculation formula is as follows:
5. The method of claim 1, wherein the step of classifying the chromosomes based on the sequence cross-correlation is characterized by, The method comprises the following steps:
6. The method of claim 1, wherein the step of classifying the chromosomes based on the sequence cross-correlation is characterized by, An image acquisition module is configured to obtain a chromosome cell image, and obtain chromosome images of all single chromosomes according to the chromosome cell image; An image segmentation module is configured to fill the width and height of all chromosome images to a multiple of 8, cut the filled images into 8*8 image extension data blocks, and generate position information according to the cutting positions of the image extension data blocks; wherein softmax represents the probability of the current chromosome belonging to a certain class I, C represents the total number of classification classes, j represents the class index, z i represents the i-th class score of the chromosome classification head linear layer output, e represents the exponential score, z j represents the original confidence of judging that the chromosome belongs to the j-th class.
7. A sequence cross-correlation based chromosome classification system characterized by, A sequence generation module is configured to generate a sequence by linear embedding according to the processing time based on the image extension data blocks, and add the position information to the sequence to obtain an image extension data sequence; A feature acquisition module is configured to input the image extension data sequence into a feature encoder, and output sequence features; An initial classification module is configured to pool the sequence features, and output an initial classification result through a classification head; A result acquisition module is configured to calculate a class probability distribution through softmax based on the initial classification result, and take the index of the highest probability as the classification result. 8. An apparatus comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.
9. A storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.