Chromosome classification processing method and system based on self-supervised contrast learning

Abstract

The chromosome classification processing method and system based on self-supervised contrast learning relates to the technical field of image processing; the chromosome classification processing method and system comprises obtaining a preliminary chromosome classification result; according to a self-defined homologous chromosome loss function, self-supervised contrast learning is performed on all first chromosome images and second chromosome images to obtain chromosome similarity analysis results of all first chromosome images and second chromosome images; the homologous chromosome loss function introduces a homologous similarity constraint term, and the chromosome similarity analysis results include the chromosome similarity of the first chromosome image and each second chromosome image; according to the preliminary chromosome classification result, the chromosome similarity analysis result and a priority matching rule, a final chromosome classification result is output, so as to solve the problem that the feature similarity of homologous chromosomes is not captured by the existing pre-training classification model, resulting in classification errors and affecting the accuracy of chromosome karyotype analysis.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a chromosome classification processing method and system based on self-supervised contrastive learning. Background Technology

[0002] Homologous chromosomes are extremely similar in morphological parameters and banding patterns. In particular, the differences between chromosomes within the same group may only manifest as subtle band spacing or local grayscale changes. Although existing chromosome karyotype analysis uses deep learning-based image segmentation and classification models to automatically extract high-dimensional features, it is limited by the sample balance and annotation accuracy of homologous chromosomes in the training data. The image segmentation and classification models may over-rely on global features and ignore key local details, resulting in insufficient capture of the feature similarity of homologous chromosomes, leading to classification errors and inaccurate chromosome karyotype analysis results. Summary of the Invention

[0003] The purpose of this invention is to provide a chromosome classification method and system based on self-supervised contrastive learning, which solves the problem that existing pre-trained classification models are insufficient in capturing the feature similarity of homologous chromosomes, leading to classification errors and affecting the accuracy of chromosome karyotype analysis.

[0004] To achieve the above objectives, in a first aspect, the present invention provides a chromosome classification method based on self-supervised contrastive learning, the method comprising:

[0005] Obtain preliminary chromosome classification results; wherein, the preliminary chromosome classification results include single chromosome images labeled with preliminary classification information, with any single chromosome image as the first chromosome image, and any single chromosome image other than the first chromosome image as the second chromosome image;

[0006] Based on a custom homologous chromosome loss function, self-supervised comparative learning is performed on all first chromosome images and second chromosome images to obtain chromosome similarity analysis results between all first chromosome images and second chromosome images. The homologous chromosome loss function introduces a homologous similarity constraint term, and the chromosome similarity analysis results include the chromosome similarity between the first chromosome image and each second chromosome image.

[0007] Based on the preliminary chromosome classification results, chromosome similarity analysis results, and priority matching rules, the final chromosome classification results are output. The priority matching rules are designed based on chromosome similarity and chromosome classification confidence. The final chromosome classification results include a single chromosome image labeled with the final classification information. The first chromosome image and the second chromosome image with the same final classification information are the most similar chromosomes to each other.

[0008] As an optional implementation, the step of outputting the final chromosome classification result based on the preliminary chromosome classification result, chromosome similarity analysis result, and priority matching rules includes:

[0009] The chromosome similarity analysis results of the first chromosome image and the second chromosome image with the same preliminary classification information are obtained to determine whether the first chromosome image and the second chromosome image are the most similar chromosomes to each other;

[0010] If yes, output the final chromosome classification result for the first and second chromosome images; otherwise,

[0011] The first chromosome image and the second chromosome image are placed into the chromosome queue to be matched. The first chromosome image and the second chromosome image that are the most similar to each other are matched from the chromosomes to be matched. The final chromosome classification result of the first chromosome image and the second chromosome image is output.

[0012] As an optional implementation, the step of outputting the final chromosome classification result based on the preliminary chromosome classification result, chromosome similarity analysis result, and priority matching rules includes:

[0013] If there is a case where the first chromosome image is the default chromosome, then based on the chromosome classification confidence of the first chromosome image and the chromosome similarity between the first chromosome image and each second chromosome image to be matched, the best first chromosome image and second chromosome image are matched from the chromosome queue to be matched, and the final chromosome classification result of the first chromosome image and second chromosome image is output; where the default chromosome is used to refer to the first chromosome image and second chromosome image that do not have the most similar chromosomes in the chromosome queue to be matched.

[0014] As an optional implementation, the step of outputting the final chromosome classification result based on the preliminary chromosome classification result, chromosome similarity analysis result, and priority matching rules includes:

[0015] Determine if there are any abnormalities in the number of chromosomes in the preliminary chromosome classification results;

[0016] If so, the chromosome similarity analysis results of the first chromosome image and the second chromosome image that obtained the same preliminary classification information are executed to determine whether the first chromosome image and the second chromosome image are the most similar chromosomes to each other; otherwise,

[0017] The preliminary chromosome classification result is used as the final chromosome classification result, and the final chromosome classification result is output.

[0018] As an optional implementation, the step of performing self-supervised contrastive learning on all first chromosome images and second chromosome images according to a custom homologous chromosome loss function to obtain chromosome similarity analysis results between all first chromosome images and second chromosome images, including:

[0019] The first chromosome image is used as the positive sample input image, and the second chromosome image is used as the negative sample input image. They are respectively input into an encoder composed of a SimCLR deep neural network model for feature vector extraction, so as to obtain the chromosome similarity between all the first chromosome images and each second chromosome image, and then obtain the chromosome similarity analysis results between all the first chromosome images and the second chromosome images.

[0020] As an optional implementation method, the method for establishing the homologous chromosome loss function includes:

[0021] Based on the input anchor sample, homologous chromosome positive sample, and non-homologous chromosome negative sample, calculate the homology similarity constraint term;

[0022] By combining the homology similarity constraint, cosine similarity, and distance metric, the homology chromosome loss function is obtained.

[0023] As an optional implementation, the priority matching rule is as follows:

[0024] Prioritize matching the first and second chromosome images based on chromosome similarity, where the chromosomes are the most similar to each other.

[0025] Secondly, based on chromosome similarity and chromosome classification confidence, the best first chromosome image and second chromosome image are matched.

[0026] Secondly, the present invention provides a chromosome classification system based on self-supervised contrastive learning, the system comprising:

[0027] The preliminary classification data acquisition module is used to acquire preliminary chromosome classification results; wherein, the preliminary chromosome classification results include single chromosome images labeled with preliminary classification information, with any single chromosome image as the first chromosome image, and any single chromosome image other than the first chromosome image as the second chromosome image;

[0028] The self-supervised contrastive learning model is used to perform self-supervised contrastive learning on all first chromosome images and second chromosome images according to a custom homologous chromosome loss function, and obtain chromosome similarity analysis results between all first chromosome images and second chromosome images. The homologous chromosome loss function introduces a homologous similarity constraint term, and the chromosome similarity analysis results include the chromosome similarity between the first chromosome image and each second chromosome image.

[0029] The classification and matching module is used to output the final chromosome classification result based on the preliminary chromosome classification result, chromosome similarity analysis result and priority matching rule. The priority matching rule is designed based on chromosome similarity and chromosome classification confidence. The final chromosome classification result includes a single chromosome image labeled with the final classification information. The first chromosome image and the second chromosome image with the same final classification information are the most similar chromosomes to each other.

[0030] Thirdly, the present invention provides an electronic device, including a memory and a processor;

[0031] The memory is used to store computer programs;

[0032] The processor is configured to execute the computer program and, when executing the computer program, implement the steps of the method as described in any of the preceding descriptions.

[0033] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the steps of the method as described in any of the preceding claims.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The chromosome classification method and system based on self-supervised contrastive learning provided by this invention calculates the similarity between individual chromosomes in the preliminary chromosome classification results through self-supervised contrastive learning. The preliminary chromosome classification results are then verified and matched based on the similarity and priority matching rules between individual chromosomes, optimizing the final chromosome classification results. This invention introduces a homologous chromosome loss function on the basis of self-supervised contrastive learning, significantly improving the similarity between homologous chromosomes of the same type while effectively reducing the similarity between non-homologous chromosomes. This allows for more accurate capture of the feature similarity between individual input chromosomes, significantly improving the accuracy of chromosome classification. It addresses the problem of existing pre-trained classification models failing to capture the feature similarity of homologous chromosomes, leading to classification errors and affecting the accuracy of chromosome karyotype analysis, thus providing more accurate classification results for subsequent chromosome analysis tasks. Attached Figure Description

[0036] Figure 1 The diagram shown is a schematic flowchart of a chromosome classification processing method according to an embodiment of the present invention;

[0037] Figure 2 The diagram shown is a schematic representation of the implementation process of chromosome pairing and matching according to an embodiment of the present invention.

[0038] Figure 3The diagram shown is a schematic representation of the implementation process of self-supervised contrastive learning training according to an embodiment of the present invention.

[0039] Figure 4 The diagram shown is a schematic frame of a chromosome classification and processing system according to an embodiment of the present invention.

[0040] Figure 5 The diagram shown is a structural schematic of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0041] Various exemplary embodiments, features, and aspects of the present invention will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0042] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0043] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0044] It should be understood that the terms "first," "second," and "third," etc., in the claims, specification, and drawings of this invention are used to distinguish different objects, rather than to describe a specific order. The terms "comprising" and "including" used in the specification and claims of this invention indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0045] Furthermore, to better illustrate the present invention, numerous specific details are set forth in the following detailed embodiments. Those skilled in the art will understand that the present invention can be practiced without certain specific details. In some instances, methods, means, elements, and circuits well known to those skilled in the art have not been described in detail in order to highlight the spirit of the invention.

[0046] For example, such as Figure 1As shown, embodiments of the present invention provide a chromosome classification method based on self-supervised contrastive learning, the method comprising the following steps S100~S300:

[0047] Step S100: Obtain preliminary chromosome classification results; wherein, the preliminary chromosome classification results include single chromosome images labeled with preliminary classification information, with any single chromosome image as the first chromosome image, and any single chromosome image other than the first chromosome image as the second chromosome image;

[0048] It is important to understand that chromosome image data captured under a microscope, after being processed by existing pre-trained segmentation models, yields segmented individual chromosome images. These images, along with preliminary classification labels, can then be generated using existing pre-trained classification models. However, homologous chromosomes are extremely similar in morphological parameters and banding patterns. In particular, differences among chromosomes within the same group may only manifest as subtle band spacing or local grayscale variations. Pre-trained classification models, limited by the sample balance and labeling accuracy of homologous chromosomes in the training data, may over-rely on global features while ignoring key local details, resulting in insufficient capture of the feature similarity of homologous chromosomes and leading to classification errors. The chromosome classification processing method of this invention can embed corresponding process modules to execute method steps based on existing pre-trained classification models, or it can set up process modules after the pre-trained classification model to verify and perform secondary classification matching on the output preliminary chromosome classification results.

[0049] Step S200: Based on the custom homologous chromosome loss function, perform self-supervised contrastive learning on all first chromosome images and second chromosome images to obtain chromosome similarity analysis results between all first chromosome images and second chromosome images; wherein, the homologous chromosome loss function introduces a homologous similarity constraint term, and the chromosome similarity analysis results include the chromosome similarity between the first chromosome image and each second chromosome image.

[0050] This invention employs self-supervised contrastive learning to mine the intrinsic structure of input chromosome images. Self-supervised contrastive learning utilizes data augmentation to generate positive sample pairs (e.g., cropping, rotation), forcing attention to semantic content rather than irrelevant changes. Furthermore, by bringing positive samples closer and pushing negative samples further away, it learns a highly discriminative feature space and calculates the similarity between individual input chromosome images and other individual chromosome images. A homology similarity constraint term is introduced into the homology chromosome loss function. By explicitly constraining the similarity relationship between samples, it guides the self-supervised contrastive learning to produce discriminative feature representations, resulting in higher accuracy of the obtained chromosome similarity.

[0051] Step S300: Based on the preliminary chromosome classification results, chromosome similarity analysis results, and priority matching rules, output the final chromosome classification results; wherein, the priority matching rules are designed based on chromosome similarity and chromosome classification confidence, and the final chromosome classification results include a single chromosome image labeled with the final classification information, and the first chromosome image and the second chromosome image with the same final classification information are the most similar chromosomes to each other.

[0052] Compared with existing technologies, the chromosome classification processing method of this invention calculates the similarity between individual chromosomes in the preliminary chromosome classification results through self-supervised contrastive learning. Based on the similarity between individual chromosomes and priority matching rules, the preliminary chromosome classification results are verified and matched to optimize the final chromosome classification results. This invention, based on self-supervised contrastive learning, introduces a homologous chromosome loss function, significantly improving the similarity between homologous chromosomes while effectively reducing the similarity between non-homologous chromosomes. This allows for more accurate capture of the feature similarity between individual input chromosomes, significantly improving the accuracy of chromosome classification. It solves the problem of insufficient capture of feature similarity between homologous chromosomes in existing pre-trained classification models, leading to classification errors and affecting the accuracy of chromosome karyotype analysis, thus providing more accurate classification results for subsequent chromosome analysis tasks.

[0053] For example, in step S300, the step of outputting the final chromosome classification result based on the preliminary chromosome classification result, chromosome similarity analysis result, and priority matching rules includes steps S310 to S320, wherein:

[0054] Step S310: Determine whether there is an abnormality in the number of chromosomes in the preliminary chromosome classification result; if so, proceed to step S320 below to obtain the chromosome similarity analysis results of the first chromosome image and the second chromosome image with the same preliminary classification information, and determine whether the first chromosome image and the second chromosome image are the most similar chromosomes to each other; otherwise, proceed to step S310b to take the preliminary chromosome classification result as the final chromosome classification result and output the final chromosome classification result.

[0055] It is understood that the number and type of chromosomes in normal cells are fixed. This invention first determines whether there are any abnormalities in the number of chromosomes in the preliminary chromosome classification result. By comprehensively and accurately analyzing and judging abnormalities in chromosome number, it can effectively distinguish between true and false cases of chromosome number abnormalities, further improving the accuracy and reliability of chromosome classification. If the chromosome type and number in the preliminary chromosome classification result are correct, it can be determined that the preliminary chromosome classification result output by the previously pre-trained classification model is relatively accurate. Therefore, the preliminary chromosome classification result can be used as the final chromosome classification result and output, reducing the system's computational load.

[0056] The priority matching rule in this embodiment of the invention is a pre-established priority ranking mechanism. The priority matching rule is as follows: First, based on chromosome similarity, match the first chromosome image and the second chromosome image that are the most similar to each other; second, based on chromosome similarity and chromosome classification confidence, match the best-matched first chromosome image and the second chromosome image. It can be understood that the most similar chromosomes refer to the pair of chromosomes with the highest similarity. For example, chromosome A's maximum chromosome similarity matches chromosome B, and chromosome B's maximum chromosome similarity matches chromosome A; therefore, chromosomes A and B are the most similar chromosomes. If chromosomes A and B are not the most similar chromosomes, then based on the chromosome similarity and chromosome classification confidence of chromosomes A and B, and other chromosomes, if chromosome C has the highest similarity to chromosome A and the most relevant chromosome classification confidence, then chromosomes A and C are matched as the most similar chromosomes. The chromosome classification confidence of a single chromosome image can be calculated through self-supervised contrastive learning, or it can be initially calculated by the pre-trained classification model in the previous step, used to reflect the probability estimate of which chromosome category the chromosome image belongs to. Based on the aforementioned priority sorting mechanism, the final chromosome classification result is output, and the first chromosome image and the second chromosome image with the same final classification information are the most similar chromosomes to each other.

[0057] like Figure 2 As shown, for example, in conjunction with the aforementioned priority matching rules, step S300 includes:

[0058] Step S320: Obtain the chromosome similarity analysis results of the first chromosome image and the second chromosome image with the same preliminary classification information, and determine whether the first chromosome image and the second chromosome image are the most similar chromosomes to each other; if so, execute step S320a and output the final chromosome classification results of the first chromosome image and the second chromosome image; otherwise, execute step S320b, put the first chromosome image and the second chromosome image into the chromosome queue to be matched, match the first chromosome image and the second chromosome image that are the most similar chromosomes to each other from the chromosomes to be matched, and output the final chromosome classification results of the first chromosome image and the second chromosome image.

[0059] For example, after step S310b, the method further includes:

[0060] In step S320c, if there is a case where the first chromosome image is a default chromosome, then based on the chromosome classification confidence of the first chromosome image and the chromosome similarity between the first chromosome image and each second chromosome image to be matched, the best first chromosome image and second chromosome image to be matched are paired from the chromosome queue to be matched, and the final chromosome classification result of the first chromosome image and the second chromosome image is output; wherein, the default chromosome is used to refer to the first chromosome image and the second chromosome image that do not have the most similar chromosomes in the chromosome queue to be matched.

[0061] It is understood that the embodiments of the present invention can classify chromosomes in cellular chromosome images of paired chromosomes, polyploid chromosomes, and haploid chromosomes. Homologous chromosomes exist in both paired and polyploid chromosomes. In cases of cellular mutation, the number of chromosomes may increase or decrease compared to normal somatic cells. An increase in chromosomes may also involve homologous chromosomes, while a decrease in chromosome number may result in haploid chromosomes. Paired chromosomes refer to two chromosomes that pair up during meiosis, are similar in morphology and size, and carry the same gene loci. For example, human females have 22 pairs of homologous autosomes and one pair of homologous X chromosomes (XX), while males have 22 pairs of homologous autosomes and one pair of heterologous sex chromosomes (XY). Some heterologous sex chromosomes are homologous (pseudoautosomal regions), while the remaining regions are key areas for sex differentiation. Polyploid chromosomes refer to the situation where an organism's cells have more than two sets of chromosomes; chromosomes in the same chromosome set are homologous chromosomes. In this embodiment of the invention, if chromosomes A and B of the same chromosome category are the most similar chromosomes, the chromosome classification results for chromosomes A and B are output. If the maximum chromosome similarity of chromosome A matches chromosome B, and the maximum chromosome similarity of chromosome B matches chromosome C, then chromosomes A and B are placed in the chromosome queue to be matched. The chromosomes in the chromosome queue to be matched are traversed, and chromosome pairs with the highest similarity are matched. For example, if the maximum chromosome similarity of chromosome B matches chromosome D, and the maximum chromosome similarity of chromosome D matches chromosome B, then the chromosome classification results for chromosomes B and D are output. If there are still default chromosomes in the chromosome queue to be matched, that is, if there are no chromosome pairs with the highest similarity in the chromosome queue to be matched, then the chromosome pair with the highest mutual similarity and the most relevant chromosome classification confidence is matched from the chromosome queue to be matched as the most similar chromosome, until the optimal pairing of all chromosomes is completed. If there are still default chromosomes, the chromosome classification results can be output based on the chromosome classification confidence.

[0062] like Figure 3 As shown in this embodiment of the invention, the step of performing self-supervised contrastive learning on all first chromosome images and second chromosome images according to a custom homologous chromosome loss function to obtain chromosome similarity analysis results between all first chromosome images and second chromosome images includes:

[0063] Using the first chromosome image as the positive sample input image and the second chromosome image as the negative sample input image, they are respectively input into two branches of encoders composed of SimCLR deep neural network models for feature vector extraction, to obtain the chromosome similarity between all the first chromosome images and each of the second chromosome images, and then to obtain the chromosome similarity analysis results between all the first chromosome images and the second chromosome images.

[0064] It is understood that this embodiment of the invention uses the SimCLR deep neural network model to construct a self-supervised contrastive learning model, abandoning the complex structures such as repositories and momentum encoders relied upon by other contrastive learning methods. It only requires a basic encoder and a lightweight projection head to achieve efficient learning, and has stronger transfer learning capabilities. By inputting chromosome A images and chromosome B images into the feature encoder for feature vector extraction, the chromosome similarity between chromosome A and chromosome B images can be calculated. Furthermore, the method for establishing the homologous chromosome loss function includes:

[0065] Based on the input anchor sample, homologous chromosome positive sample, and non-homologous chromosome negative sample, calculate the homology similarity constraint term;

[0066] By combining the homology similarity constraint, cosine similarity, and distance metric, the homology chromosome loss function is obtained.

[0067] In this embodiment of the invention, the common-origin similarity constraint term guides the model to learn more discriminative and robust feature representations by explicitly constraining the similarity relationship between samples. The expression for the common-origin similarity constraint term can be:

[0068] L_homo =∑(s(a,p) - s(a,n));

[0069] Where a is the anchor sample, p is the homologous positive sample, and n is the non-homologous negative sample; it can be understood that homologous positive samples refer to chromosomes of the same category within the same cell, and non-homologous negative samples refer to chromosomes of the same category within different cells as well as chromosomes of other different categories. The chromosome cell affiliation and category are the labels given when training the model. s(a,p) refers to the similarity calculation between the anchor sample and the homologous positive sample, and s(a,n) refers to the similarity calculation between the anchor sample and the non-homologous negative sample. The similarity calculation is shown below.

[0070] The expression combining cosine similarity and distance metric can be:

[0071]

[0072] in For cosine similarity, For normalized Euclidean distance, This is the similarity weighting coefficient. This is the distance scaling factor. x i Refers to anchor samples; x j It refers to either homologous positive samples or non-homologous negative samples.

[0073] like Figure 4As shown, this embodiment of the invention also provides a chromosome classification system based on self-supervised contrastive learning. The system includes a preliminary classification data acquisition module, a self-supervised contrastive learning model, and a classification pairing module, wherein:

[0074] The preliminary classification data acquisition module 410 is used to perform the above step S100 to obtain preliminary chromosome classification results; wherein, the preliminary chromosome classification results include single chromosome images labeled with preliminary classification information, with any single chromosome image as the first chromosome image, and any single chromosome image other than the first chromosome image as the second chromosome image.

[0075] The self-supervised contrastive learning model 420 is used to perform the above step S200. Based on the custom homologous chromosome loss function, it performs self-supervised contrastive learning on all first chromosome images and second chromosome images to obtain chromosome similarity analysis results between all first chromosome images and second chromosome images. The homologous chromosome loss function introduces a homologous similarity constraint term, and the chromosome similarity analysis results include the chromosome similarity between the first chromosome image and each second chromosome image.

[0076] The classification and matching module 430 is used to perform the above step S300, and output the final chromosome classification result based on the preliminary chromosome classification result, chromosome similarity analysis result and priority matching rule; wherein, the priority matching rule is designed based on chromosome similarity and chromosome classification confidence, and the final chromosome classification result includes a single chromosome image labeled with the final classification information, and the first chromosome image and the second chromosome image with the same final classification information are the most similar chromosomes to each other.

[0077] like Figure 4 As shown, the classification and matching module 430 includes a classification anomaly detection module 431 and a matching module 432, which are connected. In conjunction with the chromosome classification processing method of the above embodiment, when executing step S200, wherein:

[0078] The classification anomaly detection module 431 is used to execute step S310 to determine whether there is an abnormality in the number of chromosomes in the preliminary chromosome classification result; if the number corresponding to the chromosome category is normal, the preliminary chromosome classification result is used as the final chromosome classification result and the final chromosome classification result is output; otherwise, the data is transferred to the pairing module.

[0079] The pairing module 432 is used to execute step S320, obtain the chromosome similarity analysis results of the first chromosome image and the second chromosome image with the same preliminary classification information, and determine whether the first chromosome image and the second chromosome image are the most similar chromosomes to each other. If so, step S320a is executed, and the final chromosome classification result of the first chromosome image and the second chromosome image is output; otherwise, step S320b is executed, the first chromosome image and the second chromosome image are placed in the chromosome queue to be matched, and the first chromosome image and the second chromosome image that are the most similar to each other are matched from the chromosome queue to be matched, and the final chromosome classification result of the first chromosome image and the second chromosome image is output. After step S310b, step S320 also includes: step S320c, if there is a case where the first chromosome image is a default chromosome, then based on the chromosome classification confidence of the first chromosome image and the chromosome similarity between the first chromosome image and each second chromosome image to be matched, the best first chromosome image and the second chromosome image to be matched are matched from the chromosome queue to be matched, and the final chromosome classification result of the first chromosome image and the second chromosome image is output; wherein, the default chromosome is used to refer to the first chromosome image and the second chromosome image that are not the most similar to each other in the chromosome queue to be matched.

[0080] In this embodiment of the invention, a self-supervised contrastive learning model 420 is constructed using the SimCLR deep neural network model. The chromosome image is input into the feature encoder to extract the feature vector and calculate the chromosome similarity between chromosome images. Then, the pairing module 432 uses the chromosome similarity to verify and classify each chromosome image, so as to fully capture the feature similarity of homologous chromosomes.

[0081] Compared with existing technologies, the chromosome classification processing system of this invention calculates the similarity between individual chromosomes in the preliminary chromosome classification results through self-supervised contrastive learning. Based on the similarity between individual chromosomes and priority matching rules, the preliminary chromosome classification results are verified and matched to optimize the final chromosome classification results. This invention introduces a homologous chromosome loss function on the basis of self-supervised contrastive learning, significantly improving the similarity between homologous chromosomes of the same type while effectively reducing the similarity between non-homologous chromosomes. This allows for more accurate capture of the feature similarity between individual input chromosomes, significantly improving the accuracy of chromosome classification. It solves the problem of insufficient capture of feature similarity between homologous chromosomes in existing pre-trained classification models, leading to classification errors and affecting the accuracy of chromosome karyotype analysis, thus providing more accurate classification results for subsequent chromosome analysis tasks.

[0082] It should be noted that those skilled in the art will understand that Figure 4The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the chromosome classification and processing system applied thereto. A specific chromosome classification and processing system may include more or fewer components than those shown in the figure to achieve the chromosome classification and processing method and effects of the above-described embodiments of the invention.

[0083] Thirdly, embodiments of the present invention also provide an electronic device, which may be a terminal, and its internal structure diagram may be as follows. Figure 5 As shown, the electronic device includes a processor, memory, communication interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements the method described in the above embodiments. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0084] Those skilled in the art will understand that Figure 5 The structures shown are merely block diagrams of some structures related to the present invention and do not constitute a limitation on the electronic devices to which the present invention is applied. Specific electronic devices may include more or fewer components than those shown in the figures, or combine certain components, or have different component arrangements.

[0085] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0086] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions created by the present invention, and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions created by the present invention without departing from the essence and scope of the technical solutions created by the present invention.

Claims

1. A chromosome classification method based on self-supervised contrastive learning, characterized in that, The method includes: Obtain preliminary chromosome classification results; wherein, the preliminary chromosome classification results include single chromosome images labeled with preliminary classification information, with any single chromosome image as the first chromosome image, and any single chromosome image other than the first chromosome image as the second chromosome image; Based on a custom homologous chromosome loss function, self-supervised comparative learning is performed on all first chromosome images and second chromosome images to obtain chromosome similarity analysis results between all first chromosome images and second chromosome images. The homologous chromosome loss function introduces a homologous similarity constraint term, and the chromosome similarity analysis results include the chromosome similarity between the first chromosome image and each second chromosome image. Based on the preliminary chromosome classification results, chromosome similarity analysis results, and priority matching rules, the final chromosome classification results are output. The priority matching rules are designed based on chromosome similarity and chromosome classification confidence. The final chromosome classification results include a single chromosome image labeled with the final classification information. The first chromosome image and the second chromosome image with the same final classification information are the most similar chromosomes to each other.

2. The method according to claim 1, characterized in that, Based on the preliminary chromosome classification results, chromosome similarity analysis results, and priority matching rules, the final chromosome classification results are output, including: The chromosome similarity analysis results of the first chromosome image and the second chromosome image with the same preliminary classification information are obtained to determine whether the first chromosome image and the second chromosome image are the most similar chromosomes to each other; If yes, output the final chromosome classification result for the first and second chromosome images; otherwise, The first chromosome image and the second chromosome image are placed into the chromosome queue to be matched. The first chromosome image and the second chromosome image that are the most similar to each other are matched from the chromosomes to be matched. The final chromosome classification result of the first chromosome image and the second chromosome image is output.

3. The method according to claim 2, characterized in that, Based on the preliminary chromosome classification results, chromosome similarity analysis results, and priority matching rules, the final chromosome classification results are output, including: If there is a case where the first chromosome image is the default chromosome, then based on the chromosome classification confidence of the first chromosome image and the chromosome similarity between the first chromosome image and each second chromosome image to be matched, the best first chromosome image and second chromosome image are matched from the chromosome queue to be matched, and the final chromosome classification result of the first chromosome image and second chromosome image is output; where the default chromosome is used to refer to the first chromosome image and second chromosome image that do not have the most similar chromosomes in the chromosome queue to be matched.

4. The method according to claim 2, characterized in that, Based on the preliminary chromosome classification results, chromosome similarity analysis results, and priority matching rules, the final chromosome classification results are output, including: Determine if there are any abnormalities in the number of chromosomes in the preliminary chromosome classification results; If so, the chromosome similarity analysis results of the first chromosome image and the second chromosome image that obtained the same preliminary classification information are executed to determine whether the first chromosome image and the second chromosome image are the most similar chromosomes to each other; otherwise, The preliminary chromosome classification result is used as the final chromosome classification result, and the final chromosome classification result is output.

5. The method according to claim 1, characterized in that, The process involves performing self-supervised contrastive learning on all first and second chromosome images based on a custom homologous chromosome loss function, yielding chromosome similarity analysis results between all first and second chromosome images, including: The first chromosome image is used as the positive sample input image, and the second chromosome image is used as the negative sample input image. They are respectively input into an encoder composed of a SimCLR deep neural network model for feature vector extraction, so as to obtain the chromosome similarity between all the first chromosome images and each second chromosome image, and then obtain the chromosome similarity analysis results between all the first chromosome images and the second chromosome images.

6. The method according to claim 1 or 5, characterized in that, Methods for establishing homologous chromosome loss functions include: Based on the input anchor samples, positive homologous chromosome samples, and negative non-homologous chromosome samples, calculate the homology similarity constraint term; By combining the homology similarity constraint, cosine similarity, and distance metric, the homology chromosome loss function is obtained.

7. The method according to claim 1, characterized in that, The priority matching rule is: Prioritize matching the first and second chromosome images based on chromosome similarity, where the chromosomes are the most similar to each other. Secondly, based on chromosome similarity and chromosome classification confidence, the best first chromosome image and second chromosome image are matched.

8. A chromosome classification system based on self-supervised contrastive learning, characterized in that, The system includes: The preliminary classification data acquisition module is used to acquire preliminary chromosome classification results; wherein, the preliminary chromosome classification results include single chromosome images labeled with preliminary classification information, with any single chromosome image as the first chromosome image, and any single chromosome image other than the first chromosome image as the second chromosome image; The self-supervised contrastive learning model is used to perform self-supervised contrastive learning on all first chromosome images and second chromosome images according to a custom homologous chromosome loss function, and obtain chromosome similarity analysis results between all first chromosome images and second chromosome images. The homologous chromosome loss function introduces a homologous similarity constraint term, and the chromosome similarity analysis results include the chromosome similarity between the first chromosome image and each second chromosome image. The classification and matching module is used to output the final chromosome classification result based on the preliminary chromosome classification result, chromosome similarity analysis result and priority matching rule. The priority matching rule is designed based on chromosome similarity and chromosome classification confidence. The final chromosome classification result includes a single chromosome image labeled with the final classification information. The first chromosome image and the second chromosome image with the same final classification information are the most similar chromosomes to each other.

9. An electronic device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

Images