Weakly supervised chromosome karyotype classification method and device based on similar image matching

Abstract

The present invention discloses a weakly supervised chromosome karyotype classification method and device based on similar image matching. The method extracts characteristic information of homologous chromosomes to be classified, calculates characteristic distance values ​​between characteristic information and characteristics of labeled chromosomes in a labeling library, judges whether the characteristic distance value is greater than a characteristic distance threshold, and if the characteristic distance value is greater than or equal to the characteristic distance threshold, calls a general classification network to obtain the karyotype result of the chromosome to be predicted, and if the characteristic distance value is less than the characteristic distance threshold, uses the labeled chromosome corresponding to the characteristic distance value as auxiliary information to obtain the karyotype result of the chromosome to be predicted, thereby solving the problem that due to the habit of karyotype analysis workers in making films and the different sensors of image acquisition equipment, there are large differences in metaphase chromosome images, resulting in a significant decrease in model accuracy, which in turn causes karyotype analysis workers to need to make a large number of adjustments on the karyotype map, seriously reducing work efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of chromosome karyotype classification, and in particular to a weakly supervised chromosome karyotype classification method and device based on similar image matching. Background Art

[0002] In the human chromosome karyotype automatic analysis system, the recognition accuracy of the karyotype recognition task directly affects the final effect on the karyotype map. Higher karyotype recognition accuracy can reduce the number of adjustments made by karyotype analysts on the karyotype map. In the task of karyotype recognition, the data of a single point can often obtain higher accuracy by training network parameters. However, in untrained points, due to the habits of karyotype analysts in film production and the different sensors of image acquisition equipment, there are large differences in metaphase chromosome images, resulting in a significant decrease in model accuracy. As a result, karyotype analysts need to make a large number of adjustments on the karyotype map, which seriously reduces work efficiency. Therefore, it is necessary to propose a weakly supervised chromosome karyotype classification method and device based on similar image matching to solve the above problems. Summary of the invention

[0003] The purpose of the present invention is to provide a weakly supervised chromosome karyotype classification method and device based on similar image matching, so as to solve the problem in the prior art that in untrained local points, due to the habits of karyotype analysts in making films and the different sensors of image acquisition equipment, there are large differences in metaphase chromosome images, resulting in a significant decrease in model accuracy, which in turn requires karyotype analysts to make a large number of adjustments on the karyotype diagram, seriously reducing the work efficiency.

[0004] In a first aspect, the present invention provides a weakly supervised chromosome karyotype classification method based on similar image matching, comprising:

[0005] Extract characteristic information of homologous chromosomes to be classified;

[0006] Calculating a characteristic distance value between the characteristic information and the characteristic of the annotated chromosome in the annotation library;

[0007] Determine whether the feature distance value is greater than a feature distance threshold;

[0008] If the characteristic distance value is greater than or equal to the characteristic distance threshold, calling the universal classification network to obtain the karyotype result of the chromosome to be predicted;

[0009] If the characteristic distance value is less than the characteristic distance threshold, the annotated chromosome corresponding to the characteristic distance value is used as auxiliary information to obtain the karyotype result of the chromosome to be predicted.

[0010] Furthermore, extracting characteristic information of homologous chromosomes to be classified includes:

[0011] The morphological information of the homologous chromosomes to be classified, the mean value of the skeleton line length, the mean value of the chromosome aspect ratio, and the grayscale histogram features are extracted.

[0012] Furthermore, the annotated chromosome corresponding to the characteristic distance value is used as auxiliary information to obtain the karyotype result of the chromosome to be predicted, including:

[0013] 24 categories are extracted from the annotated chromosomes, with one chromosome in each category, and the selected chromosomes do not have any crossover or deep adhesion with other chromosomes;

[0014] Preprocess the extracted chromosomes and homologous chromosomes to be classified;

[0015] Input the preprocessed data into the classification network to obtain the category distance matrix;

[0016] By selecting the minimum value of the category distance matrix, the karyotype of the homologous chromosome to be classified is obtained.

[0017] Furthermore, the extracted chromosomes and the homologous chromosomes to be classified are preprocessed, including:

[0018] The average widths of the extracted chromosomes and the homologous chromosomes to be classified are uniformly scaled to the specified size, filled to 224x224 pixels, and normalized.

[0019] Furthermore, after obtaining the karyotype of the homologous chromosome to be classified by selecting the minimum value of the category distance matrix, the method further includes:

[0020] Count the number of chromosomes in each karyotype;

[0021] For karyotypes with more than three chromosomes, the chromosome with the largest characteristic distance is changed to the next smallest karyotype according to the category distance matrix.

[0022] In a second aspect, the present invention provides a weakly supervised chromosome karyotype classification device based on similar image matching, comprising:

[0023] An extraction unit, used for extracting characteristic information of homologous chromosomes to be classified;

[0024] A calculation unit, used to calculate a characteristic distance value between the characteristic information and the characteristic of the annotated chromosome in the annotation library;

[0025] A judging unit, used to judge whether the characteristic distance value is greater than a characteristic distance threshold;

[0026] A calling unit, configured to call a universal classification network to obtain a karyotype result of the chromosome to be predicted when the feature distance value is greater than or equal to a feature distance threshold;

[0027] The auxiliary unit is used to use the labeled chromosome corresponding to the characteristic distance value as auxiliary information to obtain the karyotype result of the chromosome to be predicted when the characteristic distance value is less than the characteristic distance threshold.

[0028] Furthermore, the extraction unit is used to extract the morphological information of the homologous chromosomes to be classified, the mean of the skeleton line length, the mean of the chromosome aspect ratio, and the grayscale histogram features.

[0029] Furthermore, the auxiliary unit comprises:

[0030] An extraction subunit is used to extract 24 categories from the labeled chromosomes, with one chromosome in each category, and the selected chromosomes do not have any crossover or deep adhesion with other chromosomes;

[0031] A preprocessing subunit, used for preprocessing the extracted chromosomes and homologous chromosomes to be classified;

[0032] An input subunit is used to input the preprocessed data into the classification network to obtain a category distance matrix;

[0033] The subunit is selected to obtain the karyotype of the homologous chromosome to be classified by selecting the minimum value of the category distance matrix.

[0034] Furthermore, the extraction subunit is used to uniformly scale the average width of the extracted chromosome and the homologous chromosome to be classified to a specified size, fill them to 224x224 pixels, and perform normalization processing.

[0035] Furthermore, the complex chromosome region segmentation device based on auxiliary information also includes:

[0036] A statistical subunit is used to obtain the karyotype of the homologous chromosomes to be classified by selecting the minimum value of the category distance matrix, and then count the number of chromosomes in each category of karyotype;

[0037] The subunit of change is used to change the karyotype of a karyotype with more than three chromosomes to the chromosome with the largest characteristic distance, and according to the category distance matrix, to the karyotype with the second smallest characteristic distance.

[0038] The beneficial effects of the present invention are as follows: the present invention provides a weakly supervised chromosome karyotype classification method and device based on similar image matching, which extracts feature information of homologous chromosomes to be classified, calculates the feature distance value between the feature information and the feature of the labeled chromosome in the label library, and judges whether the feature distance value is greater than the feature distance threshold. If the feature distance value is greater than or equal to the feature distance threshold, a general classification network is called to obtain the karyotype result of the chromosome to be predicted. If the feature distance value is less than the feature distance threshold, the labeled chromosome corresponding to the feature distance value is used as auxiliary information to obtain the karyotype result of the chromosome to be predicted. The present invention extracts morphological information of homologous chromosomes, searches for similar metaphase chromosome images in an established label library, and assists in classification tasks, thereby solving the problem in the prior art that at untrained points, due to the habits of karyotype analysts in making films and different sensors of image acquisition equipment, there are large differences in metaphase chromosome images, resulting in a significant decrease in model accuracy, which in turn causes karyotype analysts to make a large number of adjustments on the karyotype map, seriously reducing work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0040] Figure 1 The present invention is a flowchart of a weakly supervised chromosome karyotype classification method based on similar image matching.

[0041] Figure 2 It is a network structure diagram of chromosome karyotype recognition based on feature similarity.

[0042] Figure 3 It is an information diagram of chromosome morphology. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. The technical solutions provided by the embodiments of the present invention are described in detail below in conjunction with the drawings.

[0044] See also Figure 1 The present invention provides a weakly supervised chromosome karyotype classification method based on similar image matching, comprising:

[0045] S101, extracting characteristic information of homologous chromosomes to be classified.

[0046] like Figure 3 As shown, in this embodiment, extracting characteristic information of homologous chromosomes to be classified includes: extracting morphological information of homologous chromosomes to be classified, the mean of skeleton line length, the mean of chromosome aspect ratio, and grayscale histogram features.

[0047] S102, calculating the feature distance value between the feature information and the features of the annotated chromosome in the annotation library.

[0048] Specifically, the chromosomes to be matched are selected from the annotation library with a ratio of 1.0±0.15 to the mean length of the skeleton line of the chromosome to be classified and a ratio of 1.0±0.2 to the mean length-to-width ratio. The grayscale histogram feature distance between the chromosome to be classified and the chromosome to be matched is calculated by DTW (dynamic time programming).

[0049] S103: Determine whether the feature distance value is greater than a feature distance threshold.

[0050] Specifically, the mean of the histogram feature distance calculated by the initial data (label data of the deployment point) and the chromosome data of the annotation library is set as the feature distance threshold.

[0051] S104: If the feature distance value is greater than or equal to the feature distance threshold, a general classification network is called to obtain a karyotype result of the chromosome to be predicted.

[0052] S105: If the characteristic distance value is less than the characteristic distance threshold, the annotated chromosome corresponding to the characteristic distance value is used as auxiliary information to obtain the karyotype result of the chromosome to be predicted.

[0053] See also Figure 2 In this embodiment, the annotated chromosome corresponding to the characteristic distance value is used as auxiliary information to obtain the karyotype result of the chromosome to be predicted, including: extracting 24 categories from the annotated chromosomes, one chromosome in each category, and the selected chromosomes do not have crossover or deep adhesion with other chromosomes; preprocessing the extracted chromosomes and the homologous chromosomes to be classified; inputting the preprocessed data into the classification network to obtain the category distance matrix; and obtaining the karyotype of the homologous chromosomes to be classified by selecting the minimum value of the category distance matrix. The label data of the general classification network in the training stage is to set the distance between the chromosome to be predicted and the same channel to zero. In the prediction stage, the minimum value of the distance matrix indicates that the predicted chromosome is most similar to the channel of this category.

[0054] In this embodiment, the extracted chromosomes and the homologous chromosomes to be classified are preprocessed, including: the average widths of the extracted chromosomes and the homologous chromosomes to be classified are uniformly scaled to a specified size, filled to 224x224 pixels, and normalized.

[0055] Specifically, a two-way network is used, and features are extracted through one of the networks, with a feature size of 24xD. The other network inputs the homologous chromosomes to be predicted, the number of which is about 46, recorded as B, and the output feature is BxD. The feature matrices obtained by the two-way network are multiplied, and finally the distance matrix from the chromosome to be predicted to the 24 categories is obtained, and the karyotype of the chromosome to be predicted is obtained by obtaining the minimum value.

[0056] In this embodiment, after obtaining the karyotype of the homologous chromosome to be classified by selecting the minimum value of the category distance matrix, the following steps are further included:

[0057] The number of chromosomes in each type of karyotype is counted; for karyotypes with more than three chromosomes, the chromosome with the largest characteristic distance is changed to the second smallest karyotype according to the category distance matrix. According to medical prior knowledge, the number of chromosomes in most chromosome images is 46, and all karyotypes except the sex chromosome karyotype have two chromosomes. Abnormal chromosome numbers are mostly manifested as three chromosomes of the same type. Therefore, when there are more than three chromosomes in a karyotype, the karyotype of the chromosome with the maximum characteristic distance is adjusted to the second smallest karyotype in terms of characteristic distance.

[0058] It can be seen from the above embodiments that by extracting the characteristic information of the homologous chromosomes to be classified, calculating the characteristic distance value between the characteristic information and the characteristics of the labeled chromosomes in the label library, judging whether the characteristic distance value is greater than the characteristic distance threshold, if the characteristic distance value is greater than or equal to the characteristic distance threshold, calling the general classification network to obtain the karyotype result of the chromosome to be predicted, if the characteristic distance value is less than the characteristic distance threshold, using the labeled chromosome corresponding to the characteristic distance value as auxiliary information to obtain the karyotype result of the chromosome to be predicted, the present invention extracts the morphological information of homologous chromosomes, searches for similar metaphase chromosome images in the established label library, and assists in the classification task, thereby solving the problem in the prior art that in the untrained local point, due to the habits of karyotype analysts in making films and the different sensors of image acquisition equipment, there are large differences in metaphase chromosome images, resulting in a significant decrease in model accuracy, which in turn causes the karyotype analyst to make a large number of adjustments on the karyotype map, which seriously reduces the work efficiency.

[0059] The present invention can directly obtain better recognition effect through a small amount of markers on untrained data sets, solving the problem of reduced recognition accuracy due to model robustness when classifying chromosome images. By extracting morphological information of homologous chromosomes, similar metaphase chromosome images are found in the established label library to assist in the classification task.

[0060] The present invention provides a weakly supervised chromosome karyotype classification device based on similar image matching, comprising:

[0061] An extraction unit, used for extracting characteristic information of homologous chromosomes to be classified;

[0062] A calculation unit, used to calculate a characteristic distance value between the characteristic information and the characteristic of the annotated chromosome in the annotation library;

[0063] A judging unit, used to judge whether the characteristic distance value is greater than a characteristic distance threshold;

[0064] A calling unit, configured to call a universal classification network to obtain a karyotype result of the chromosome to be predicted when the feature distance value is greater than or equal to a feature distance threshold;

[0065] The auxiliary unit is used to use the labeled chromosome corresponding to the characteristic distance value as auxiliary information to obtain the karyotype result of the chromosome to be predicted when the characteristic distance value is less than the characteristic distance threshold.

[0066] Furthermore, the extraction unit is used to extract the morphological information of the homologous chromosomes to be classified, the mean of the skeleton line length, the mean of the chromosome aspect ratio, and the grayscale histogram features.

[0067] Furthermore, the auxiliary unit comprises:

[0068] An extraction subunit is used to extract 24 categories from the labeled chromosomes, with one chromosome in each category, and the selected chromosomes do not have any crossover or deep adhesion with other chromosomes;

[0069] A preprocessing subunit, used for preprocessing the extracted chromosomes and homologous chromosomes to be classified;

[0070] An input subunit is used to input the preprocessed data into the classification network to obtain a category distance matrix;

[0071] The subunit is selected to obtain the karyotype of the homologous chromosome to be classified by selecting the minimum value of the category distance matrix.

[0072] Furthermore, the extraction subunit is used to uniformly scale the average width of the extracted chromosome and the homologous chromosome to be classified to a specified size, fill them to 224x224 pixels, and perform normalization processing.

[0073] Furthermore, the complex chromosome region segmentation device based on auxiliary information also includes:

[0074] A statistical subunit is used to obtain the karyotype of the homologous chromosomes to be classified by selecting the minimum value of the category distance matrix, and then count the number of chromosomes in each category of karyotype;

[0075] The subunit of change is used to change the karyotype of a karyotype with more than three chromosomes to the chromosome with the largest characteristic distance, and according to the category distance matrix, to the karyotype with the second smallest characteristic distance.

[0076] The embodiment of the present invention further provides a storage medium, and the embodiment of the present invention further provides a storage medium, wherein a computer program is stored in the storage medium, and when the computer program is executed by a processor, some or all of the steps in each embodiment of the complex chromosome region segmentation method based on auxiliary information provided by the present invention are implemented. The storage medium can be a magnetic disk, an optical disk, a read-only memory (English: Read-Only Memory, abbreviated: ROM) or a random access memory (English: Random Access Memory, abbreviated: RAM), etc.

[0077] Those skilled in the art can clearly understand that the technology in the embodiments of the present invention can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution in the embodiments of the present invention can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present invention or certain parts of the embodiments.

[0078] In this specification, the same or similar parts between the various embodiments can be referred to each other. In particular, for the embodiment of the complex chromosome region segmentation device based on auxiliary information, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description in the method embodiment.

[0079] The above-described embodiments of the present invention do not limit the protection scope of the present invention.

Claims

1. A weakly supervised chromosome karyotype classification method based on similar image matching, characterized in that: include: Extract characteristic information of homologous chromosomes to be classified; Calculating a characteristic distance value between the characteristic information and the characteristic of the annotated chromosome in the annotation library; Determine whether the feature distance value is greater than a feature distance threshold; If the characteristic distance value is greater than or equal to the characteristic distance threshold, calling the universal classification network to obtain the karyotype result of the chromosome to be predicted; If the characteristic distance value is less than the characteristic distance threshold, the annotated chromosome corresponding to the characteristic distance value is used as auxiliary information to obtain the karyotype result of the chromosome to be predicted; Extracting characteristic information of homologous chromosomes to be classified includes: Extract the morphological information of the homologous chromosomes to be classified, the mean value of the skeleton line length, the mean value of the chromosome length-width ratio, and the grayscale histogram features; The annotated chromosome corresponding to the characteristic distance value is used as auxiliary information to obtain the karyotype result of the chromosome to be predicted, including: 24 categories are extracted from the annotated chromosomes, with one chromosome in each category, and the selected chromosomes do not have any crossover or deep adhesion with other chromosomes; Preprocess the extracted chromosomes and homologous chromosomes to be classified; Input the preprocessed data into the classification network to obtain the category distance matrix; By selecting the minimum value of the category distance matrix, the karyotype of the homologous chromosome to be classified is obtained; Preprocess the extracted chromosomes and homologous chromosomes to be classified, including: The average widths of the extracted chromosomes and the homologous chromosomes to be classified are uniformly scaled to the specified size, filled to 224x224 pixels, and normalized; After obtaining the karyotype of the homologous chromosome to be classified by selecting the minimum value of the category distance matrix, it also includes: Count the number of chromosomes in each karyotype; For karyotypes with more than three chromosomes, the chromosome with the largest characteristic distance is changed to the next smallest karyotype according to the category distance matrix.

2. A weakly supervised chromosome karyotype classification device based on similar image matching, characterized in that: include: An extraction unit, used for extracting characteristic information of homologous chromosomes to be classified; A calculation unit, used to calculate a characteristic distance value between the characteristic information and the characteristic of the annotated chromosome in the annotation library; A judging unit, used to judge whether the characteristic distance value is greater than a characteristic distance threshold; A calling unit, configured to call a universal classification network to obtain a karyotype result of the chromosome to be predicted when the feature distance value is greater than or equal to a feature distance threshold; An auxiliary unit, used to obtain a karyotype result of the chromosome to be predicted by using the labeled chromosome corresponding to the characteristic distance value as auxiliary information when the characteristic distance value is less than the characteristic distance threshold; The extraction unit is used to extract the morphological information of the homologous chromosomes to be classified, the mean value of the skeleton line length, the mean value of the chromosome length-width ratio, and the grayscale histogram features; The auxiliary unit comprises: An extraction subunit is used to extract 24 categories from the labeled chromosomes, with one chromosome in each category, and the selected chromosomes do not have any crossover or deep adhesion with other chromosomes; A preprocessing subunit, used for preprocessing the extracted chromosomes and homologous chromosomes to be classified; An input subunit is used to input the preprocessed data into the classification network to obtain a category distance matrix; A subunit is selected to obtain the karyotype of the homologous chromosome to be classified by selecting the minimum value of the category distance matrix; The extraction subunit is used to uniformly scale the average width of the extracted chromosomes and the homologous chromosomes to be classified to a specified size, fill them to 224x224 pixels, and perform normalization processing; Also includes: A statistical subunit is used to obtain the karyotype of the homologous chromosomes to be classified by selecting the minimum value of the category distance matrix, and then count the number of chromosomes in each category of karyotype; The subunit of change is used to change the karyotype of a karyotype with more than three chromosomes to the chromosome with the largest characteristic distance, and according to the category distance matrix, to the karyotype with the second smallest characteristic distance.

Images