A method for calculating chromosome karyotype quality index based on deep learning technology

Abstract

The present invention provides a method for calculating a chromosome karyotype quality index based on deep learning technology, wherein the method comprises: obtaining a given set of chromosome karyotype images; obtaining chromosome category information corresponding to the chromosome karyotype images based on the given set of chromosome karyotype images; calculating a chromosome karyotype quality index based on the chromosome category information; and evaluating the quality of the chromosome karyotype image based on the chromosome karyotype quality index. The present invention can overcome the subjectivity and huge amount of annotation required for chromosome karyotype quality annotation and truly reflect the quality of chromosome detection and classification. It provides data support for the analysis of chromosome karyotype quality and improves the efficiency of obtaining and using chromosome karyotype quality.

Description

Technical Field

[0001] The present invention relates to the technical field of chromosome analysis, and in particular to a method for calculating a chromosome karyotype quality index based on deep learning technology. Background Art

[0002] Chromosome karyotype analysis is a common method for clinical screening of genetic diseases. This test allows for precise analysis of a patient's chromosomes and comparison with normal human chromosomes. The significance of karyotype analysis lies in screening for chromosomal abnormalities in patients and ruling out a range of genetic diseases caused by chromosomal mutations. Karyotype analysis is a fundamental method in cytogenetic research and an important tool for studying the evolution and classification of species and the relationship between chromosome structure, morphology, and function. After karyotype analysis, the cause of an organism's disease can be determined based on variations in chromosome structure and number, such as the missing gene that causes the disease. It can also accurately identify gene mutations within the chromosomal DNA chain based on differences in single bases.

[0003] The chromosome karyotype is an important data foundation for chromatin karyotype analysis. The chromosome karyotype can sometimes determine the structure of the chromatin karyotype analysis, so the quality of the chromosome karyotype is crucial. Currently, there is a lack of a method that can directly analyze the quality of chromosome karyotypes. Summary of the Invention

[0004] Based on this, it is necessary to provide a chromosome karyotype quality index calculation method based on deep learning technology to address the above technical problems.

[0005] A method for calculating a chromosome karyotype quality index based on deep learning technology comprises the following steps:

[0006] Get a given set of chromosome karyotype images;

[0007] According to the given chromosome karyotype image set, obtaining chromosome category information corresponding to the chromosome karyotype image;

[0008] Calculating a chromosome karyotype quality index based on the chromosome category information;

[0009] Performing chromosome karyotype quality evaluation according to the chromosome karyotype quality index;

[0010] The step of calculating the chromosome karyotype quality index based on the chromosome category information includes:

[0011] Obtaining a classification index according to the chromosome category information;

[0012] According to the chromosome annotation, a dispersion index is obtained;

[0013] Obtaining a chromosome length index according to the chromosome subgraph;

[0014] According to the classification index, the dispersion index and the chromosome length index, the chromosome karyotype quality index is obtained:

[0015] ;

[0016] ;

[0017] in, represents the chromosome karyotype quality index, represents the classification index, represents the classification index weight, represents the dispersion index, represents the dispersion index weight, represents the chromosome length index, represents the weight of chromosome length indicator;

[0018] Wherein, obtaining the classification index according to the chromosome category information includes:

[0019] Obtaining the number of chromosomes of each type for each species, and obtaining a precision index and a recall index based on the number of chromosomes of each type for each species and the chromosome category information;

[0020] Obtaining a classification index according to the precision index and the recall index;

[0021] Wherein, obtaining a chromosome length index according to the chromosome subgraph includes:

[0022] The length index is calculated using the following formula:

[0023] ;

[0024] in, Indicates the length indicator, represents the median length of all chromosomes in the current graph, represents the median of the image, represents the maximum median length of chromosomes in all graphs, is the total number of images to be tested.

[0025] In one embodiment, obtaining chromosome category information corresponding to the chromosome karyotype image according to the given chromosome karyotype image set includes:

[0026] Obtaining chromosome annotation according to the chromosome karyotype image and the chromosome detection model;

[0027] According to the chromosome annotation, a chromosome subgraph is obtained;

[0028] Chromosome category information is obtained according to the chromosome subgraph and the chromosome classification model.

[0029] In one embodiment, the training process of the chromosome detection model includes:

[0030] Obtaining a given chromosome karyotype image set, and dividing the chromosome karyotype image set into a training set for a chromosome detection model and a validation set for validating the chromosome detection model according to a preset ratio;

[0031] For each chromosome karyotype image in the training set, the chromosome annotation corresponding to the chromosome karyotype image is obtained, and each chromosome karyotype image is used as the source input of the model. The output of the model and the chromosome annotation corresponding to the chromosome karyotype image are used as the input of the loss function. Each chromosome karyotype image is used to train the chromosome detection model to obtain the optimized chromosome detection model.

[0032] In one embodiment, the training process of the chromosome classification model includes:

[0033] Obtaining a chromosome subgraph, and dividing the chromosome subgraph into a training set for a chromosome classification model and a validation set for validating the chromosome classification model according to a preset ratio;

[0034] For each chromosome subgraph in the training set, the chromosome category information corresponding to the chromosome subgraph is obtained, each chromosome subgraph is used as the source input of the model, the output of the model and the chromosome category information corresponding to the chromosome subgraph are used as the input of the loss function, and each chromosome subgraph is applied to train the chromosome classification model to obtain the optimized chromosome classification model.

[0035] In one embodiment, obtaining a classification index based on the precision index and the recall index includes:

[0036] The classification index is calculated according to the following formula:

[0037] ;

[0038] in, represents the classification index, represents the recall index, express The weight of Indicates the accuracy index.

[0039] In one embodiment, obtaining the dispersion index according to the chromosome annotation includes:

[0040] According to the chromosome annotation, obtaining the sum of the total area of ​​the chromosome annotation frame and the area of ​​the mask of the chromosome annotation frame on the original image;

[0041] The dispersion index is obtained according to the sum of the total area of ​​the chromosome annotation box and the area of ​​the mask of the chromosome annotation box on the original image:

[0042] ;

[0043] in, The denotes the dispersion index, Represents the sum of the areas of the masks of the chromosome annotation boxes on the original image, Represents the total area of ​​the chromosome annotation box.

[0044] Compared with existing technologies, the advantages and benefits of the present invention are that it overcomes the subjectivity and large amount of annotation required for karyotype quality annotation, truly reflects the quality of chromosome detection and classification, provides data support for karyotype quality analysis, and improves the efficiency of karyotype quality acquisition and use.

[0045] Compared with existing technologies, the advantages and benefits of the present invention are that it overcomes the subjectivity and large amount of annotation required for karyotype quality annotation, truly reflects the quality of chromosome detection and classification, provides data support for karyotype quality analysis, and improves the efficiency of karyotype quality acquisition and use. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 1 is a flowchart of a method for calculating a chromosome karyotype quality index based on deep learning technology in one embodiment;

[0047] Figure 2 A schematic diagram of a chromosome detection model detection process in one embodiment;

[0048] Figure 3 A schematic diagram of the classification process of a chromosome classification model in one embodiment;

[0049] Figure 4 is a schematic diagram of chromosome category information in one embodiment;

[0050] Figure 5 is a schematic diagram of a chromosome karyotype image in one embodiment;

[0051] Figure 6 A schematic diagram of annotated images in one embodiment;

[0052] Figure 7 A schematic diagram of a chromosome subgraph in one embodiment;

[0053] Figure 8 A schematic diagram of a chromosome labeling box in one embodiment;

[0054] Figure 9 Schematic diagram of the area of ​​the mask of the chromosome annotation frame on the original image in one embodiment;

[0055] Figure 10 is a schematic diagram of a chromosome karyotype with a quality index of 1 in one embodiment;

[0056] Figure 11 is a schematic diagram of chromosome category information of a chromosome karyotype map with a quality index of 1 in one embodiment;

[0057] Figure 12 A schematic diagram of a chromosome karyotype with a quality index of 0.92 in one embodiment;

[0058] Figure 13 Schematic diagram of chromosome category information of a chromosome karyotype map with a quality index of 0.92 in one embodiment;

[0059] Figure 14 A schematic diagram of a chromosome karyotype with a quality index of 0.57 in one embodiment;

[0060] Figure 15 Schematic diagram of chromosome category information of a chromosome karyotype map with a quality index of 0.57 in one embodiment. DETAILED DESCRIPTION

[0061] Before describing the specific embodiments of the present invention, the overall concept of the present invention is described as follows:

[0062] The present invention is mainly developed based on the chromosome karyotype analysis process. Currently, there is no method that can calculate the quality index of chromosomes and analyze the quality of chromosome karyotypes.

[0063] Therefore, the present invention proposes a method for calculating the quality index of a chromosome karyotype map based on deep learning technology. After obtaining a chromosome karyotype image set, the chromosome category information corresponding to the chromosome karyotype image can be obtained according to the chromosome karyotype image set. Based on the chromosome category information corresponding to the chromosome karyotype image, the chromosome karyotype map quality index can be calculated. By using the chromosome karyotype map quality index, the quality evaluation of the chromosome karyotype map can be performed. The method provided by the present invention can overcome the subjectivity and huge amount of annotation of the quality annotation of the chromosome karyotype map, and truly reflects the quality of chromosome detection and classification. Provide data support for the analysis of the quality of the chromosome karyotype map, and improve the efficiency of the acquisition and use of the quality of the chromosome karyotype map.

[0064] After introducing the overall concept of the present invention, in order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below through specific implementation methods in conjunction with the accompanying drawings.

[0065] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of this specification should have the usual meanings understood by people with ordinary skills in the field to which the invention belongs. The words "first", "second" and similar terms used in one or more implementations of this specification do not indicate any order, quantity or importance, but are only used to distinguish different components. Words such as "include" or "comprise" mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Words such as "connect" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0066] In one embodiment, Figure 1 As shown, a method for calculating the quality index of a chromosome karyotype map based on deep learning technology is provided, comprising the following steps:

[0067] Step S101: obtaining a given chromosome karyotype image set.

[0068] Specifically, a given set of chromosome karyotype images requiring chromosome karyotype quality judgment is obtained.

[0069] Step S102: obtaining chromosome category information corresponding to the chromosome karyotype image according to the given chromosome karyotype image set.

[0070] Specifically, based on a given chromosome karyotype image set and according to a pre-designed model, the chromosome category information corresponding to the chromosome karyotype image is calculated.

[0071] On this basis, according to the given chromosome karyotype image set, the chromosome category information corresponding to the chromosome karyotype image is obtained, including:

[0072] Obtaining chromosome annotations according to the given chromosome karyotype image set and chromosome detection model;

[0073] According to the chromosome annotation, a chromosome subgraph is obtained;

[0074] Chromosome category information is obtained according to the chromosome subgraph and the chromosome classification model.

[0075] Specifically, two deep learning models were designed: one for detection and one for classification. The chromosome detection model has a classification head with 1 class, while the chromosome classification model has N classes, where N is the number of chromosome classes in the species. The chromosome detection model has two loss functions: one for determining whether an image is a chromosome and one for regressing the target detection bounding box coordinates. The chromosome classification model has a single loss function for distinguishing which chromosome class an input image belongs to.

[0076] like Figure 2 As shown in , chromosome karyotype images are input into the chromosome detection model to obtain chromosome annotations. Figure 3 As shown in , the chromosome subgraph is input into the chromosome classification model to obtain the chromosome category information. The specific chromosome category information is as follows Figure 4 shown.

[0077] On this basis, the training process of the chromosome detection model includes:

[0078] Obtaining a given chromosome karyotype image set, and dividing the chromosome karyotype image set into a training set for a chromosome detection model and a validation set for validating the chromosome detection model according to a preset ratio;

[0079] For each chromosome karyotype image in the training set, the chromosome annotation corresponding to the chromosome karyotype image is obtained. Each chromosome karyotype image is used as the source input of the model. The output of the model and the chromosome annotation corresponding to the chromosome karyotype image are used as the input of the loss function. Each chromosome karyotype image is used to train the chromosome detection model to obtain the optimized chromosome detection model.

[0080] Specifically, a given chromosome karyotype image set is obtained and divided into N folds. Each fold is divided into a training set for the chromosome detection model and a validation set for validating the chromosome detection model according to a preset ratio. In actual applications, based on the size of the entire chromosome karyotype image set data, the division ratio of the training set data to the validation set data can be 6:4, 7:3, or 8:2. For large amounts of data, 9:1 or even 99:1 can be used. The specific method of dividing the validation set varies.

[0081] In one embodiment, a given chromosome karyotype image set is divided into five folds, each of which is divided into a training set and a validation set (the total number of the training set and the validation set is equal to the given chromosome karyotype image set). Figure 5 As shown in the figure, it is the chromosome karyotype image in the training set. Each fold is trained, and the chromosome annotation corresponding to the chromosome karyotype image is obtained for the chromosome karyotype image in the training set. The annotated image is as follows Figure 6As shown in the figure, each chromosome karyotype image is used as the model source input, and the model output and the chromosome annotation corresponding to the chromosome karyotype image are used as the input of the loss function. The loss function is minimized through optimization. Each chromosome karyotype image is used to train the chromosome detection model. The parameters with the highest validation set index are selected as the optimal parameters, and the validation set detection results are output at the same time. A total of five sets of parameters are obtained from the training.

[0082] On this basis, the training process of the chromosome classification model includes:

[0083] Obtaining a chromosome subgraph, and dividing the chromosome subgraph into a training set for a chromosome classification model and a validation set for validating the chromosome classification model according to a preset ratio;

[0084] For each chromosome subgraph in the training set, the chromosome category information corresponding to the chromosome subgraph is obtained, each chromosome subgraph is used as the source input of the model, the output of the model and the chromosome category information corresponding to the chromosome subgraph are used as the input of the loss function, and each chromosome subgraph is applied to train the chromosome classification model to obtain the optimized chromosome classification model.

[0085] Specifically, first combine the chromosome detection model 5-fold validation set test results (chromosome annotations) together, and then Figure 7 As shown, each chromosome annotation is cut out of the chromosome subgraph according to the annotation box of the chromosome annotation.

[0086] The chromosome subgraph is divided into a training set for a chromosome classification model and a validation set for validating the chromosome classification model according to a preset ratio.

[0087] For each chromosome subgraph in the training set, the chromosome annotation corresponding to the chromosome subgraph is obtained, each chromosome subgraph is used as the source input of the model, the output of the model and the chromosome category information corresponding to the chromosome subgraph are used as the input of the loss function, and each chromosome subgraph is applied to train the chromosome classification model. The training minimizes the loss function and takes the weight of the maximum value of the validation set indicator as the final weight to obtain the optimized chromosome classification model.

[0088] Step S103: Calculate the chromosome karyotype quality index based on the chromosome category information.

[0089] The chromosome karyotype quality index calculated according to the chromosome category information includes:

[0090] Obtaining a classification index according to the chromosome category information;

[0091] According to the chromosome annotation, a dispersion index is obtained;

[0092] Obtaining a chromosome length index according to the chromosome subgraph;

[0093] According to the classification index, the dispersion index and the chromosome length index, the chromosome karyotype quality index is obtained:

[0094] ;

[0095] ;

[0096] in, represents the chromosome karyotype quality index, represents the classification index, represents the classification index weight, represents the dispersion index, represents the dispersion index weight, represents the chromosome length index, Represents the chromosome length indicator weight.

[0097] Specifically, the classification index, dispersion index and chromosome length index of the chromosome karyotype map can be calculated based on the chromosome category information, chromosome annotation and chromosome submap, and then the chromosome karyotype map quality index can be calculated based on the classification index, dispersion index and chromosome length index.

[0098] In one embodiment, Classification index weight, Diversification Index Weights and The chromosome length index weight is an artificially set value. The value of is 0.8, The value of is 0.1, The value of is 0.1, .

[0099] On this basis, according to the chromosome category information, the classification index includes:

[0100] Obtaining the number of chromosomes of each type for each species, and obtaining a precision index and a recall index based on the number of chromosomes of each type for each species and the chromosome category information;

[0101] A classification index is obtained according to the precision index and the recall index.

[0102] On this basis, according to the precision index and the recall index, the classification index includes:

[0103] The classification index is calculated according to the following formula:

[0104] ;

[0105] in, represents the classification index, represents the recall index, express The weight of Indicates the accuracy index.

[0106] Specifically, The calculation formula is as follows:

[0107] ;

[0108] in, represents the recall index, Represents the accuracy index, express The weight is an artificially set value of 0.5. The calculation process is as follows:

[0109] On this basis, according to the chromosome annotation, the dispersion index includes:

[0110] According to the chromosome annotation, obtaining the sum of the total area of ​​the chromosome annotation frame and the area of ​​the mask of the chromosome annotation frame on the original image;

[0111] The dispersion index is obtained according to the sum of the total area of ​​the chromosome annotation box and the area of ​​the mask of the chromosome annotation box on the original image:

[0112] ;

[0113] in, The denotes the dispersion index, Represents the sum of the areas of the masks of the chromosome annotation boxes on the original image, Represents the total area of ​​the chromosome annotation box.

[0114] Specifically, The calculation formula is as follows:

[0115] ;

[0116] in, is the total area of ​​all chromosome labeled boxes, is the sum of the areas of the masks of all chromosome annotation boxes on the original image. The total area of ​​the chromosome annotation box is Figure 8 The sum of the areas of all rectangular boxes shown and the areas of the masks of the chromosome annotation boxes on the original image is Figure 9 The black area is shown.

[0117] On this basis, according to the chromosome subgraph, the chromosome length index is obtained including:

[0118] The length index is calculated using the following formula:

[0119] ;

[0120] in, Indicates the length indicator, represents the median length of all chromosomes in the current graph, represents the median of the image, represents the maximum value of the median length of all graphs, is the total number of images to be tested.

[0121] Specifically, The calculation formula is as follows:

[0122] ;

[0123] in, is the median length of all chromosomes in the current graph, is the median of the image, That is, the maximum value of the median length of all graphs, is the total number of images to be tested.

[0124] Step S104: evaluating the quality of the chromosome karyotype map according to the chromosome karyotype map quality index.

[0125] Specifically, the chromosome karyotype quality index ranges from 1.5 to 2.5, with a larger value indicating a higher quality of the chromosome karyotype.

[0126] Figure 10 is a chromosome karyotype map with a quality index of 1, Figure 11 is the chromosome category information of the chromosome karyotype map with a quality index of 1, Figure 12 The chromosome karyotype map has a quality index of 0.92. Figure 13 is the chromosome category information of the chromosome karyotype map with a quality index of 0.92, Figure 14 The chromosome karyotype map has a quality index of 0.57. Figure 15 This is the chromosome category information of the chromosome karyotype map with a quality index of 0.57.

[0127] This paper provides a method for calculating the karyotype quality index based on deep learning technology. It can overcome the subjectivity and large amount of annotation required for karyotype quality annotation and truly reflect the quality of chromosome detection and classification. It provides data support for karyotype quality analysis and improves the efficiency of karyotype quality acquisition and use.

[0128] It should be noted that the method of the embodiment of the present invention can be performed by a single device, such as a computer or server. The method of this embodiment can also be applied in a distributed scenario, where multiple devices cooperate to perform the method. In such a distributed scenario, one of the multiple devices may only perform one or more steps of the method of the embodiment of the present invention, and the multiple devices will interact with each other to complete the method.

[0129] It should be noted that the above description is limited to some embodiments of the present invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in an order different from that described in the above embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for calculating the quality index of a chromosome karyotype map based on deep learning technology, characterized in that: include: Get a given set of chromosome karyotype images; According to the given chromosome karyotype image set, obtaining chromosome category information corresponding to the chromosome karyotype image; Calculating a chromosome karyotype quality index based on the chromosome category information; Performing chromosome karyotype quality evaluation according to the chromosome karyotype quality index; The step of calculating the chromosome karyotype quality index based on the chromosome category information includes: Obtaining a classification index according to the chromosome category information; According to the chromosome annotation, a dispersion index is obtained; Obtaining a chromosome length index according to the chromosome subgraph; According to the classification index, the dispersion index and the chromosome length index, the chromosome karyotype quality index is obtained: ； ； in, represents the chromosome karyotype quality index, represents the classification index, represents the classification index weight, represents the dispersion index, represents the dispersion index weight, represents the chromosome length index, represents the weight of chromosome length indicator; Wherein, obtaining the classification index according to the chromosome category information includes: Obtaining the number of chromosomes of each type for each species, and obtaining a precision index and a recall index based on the number of chromosomes of each type for each species and the chromosome category information; Obtaining a classification index according to the precision index and the recall index; Wherein, obtaining a chromosome length index according to the chromosome subgraph includes: The length index is calculated using the following formula: ； in, Indicates the length indicator, represents the median length of all chromosomes in the current graph, represents the median of the image, represents the maximum median length of chromosomes in all graphs, is the total number of images to be tested.

2. The method for calculating the quality index of a chromosome karyotype map based on deep learning technology according to claim 1, characterized in that: The step of obtaining chromosome category information corresponding to the chromosome karyotype image according to the given chromosome karyotype image set includes: Obtaining chromosome annotations according to the given chromosome karyotype image set and chromosome detection model; According to the chromosome annotation, a chromosome subgraph is obtained; Chromosome category information is obtained according to the chromosome subgraph and the chromosome classification model.

3. The method for calculating the quality index of a chromosome karyotype map based on deep learning technology according to claim 2, characterized in that: The training process of the chromosome detection model includes: Obtaining a given chromosome karyotype image set, and dividing the chromosome karyotype image set into a training set for a chromosome detection model and a validation set for validating the chromosome detection model according to a preset ratio; For each chromosome karyotype image in the training set, the chromosome annotation corresponding to the chromosome karyotype image is obtained. Each chromosome karyotype image is used as the source input of the model. The output of the model and the chromosome annotation corresponding to the chromosome karyotype image are used as the input of the loss function. Each chromosome karyotype image is used to train the chromosome detection model to obtain the optimized chromosome detection model.

4. The method for calculating the quality index of a chromosome karyotype map based on deep learning technology according to claim 2, characterized in that: The training process of the chromosome classification model includes: Obtaining a chromosome subgraph, and dividing the chromosome subgraph into a training set for a chromosome classification model and a validation set for validating the chromosome classification model according to a preset ratio; For each chromosome subgraph in the training set, the chromosome category information corresponding to the chromosome subgraph is obtained, each chromosome subgraph is used as the source input of the model, the output of the model and the chromosome category information corresponding to the chromosome subgraph are used as the input of the loss function, and each chromosome subgraph is applied to train the chromosome classification model to obtain the optimized chromosome classification model.

5. The method for calculating the quality index of a chromosome karyotype map based on deep learning technology according to claim 1, characterized in that: The classification index obtained according to the precision index and the recall index includes: The classification index is calculated according to the following formula: ； in, represents the classification index, represents the recall index, The weight represented by Indicates the accuracy index.

6. The method for calculating the quality index of a chromosome karyotype map based on deep learning technology according to claim 1, characterized in that: Obtaining the dispersion index according to the chromosome annotation includes: According to the chromosome annotation, obtaining the sum of the total area of ​​the chromosome annotation frame and the area of ​​the mask of the chromosome annotation frame on the original image; The dispersion index is obtained according to the sum of the total area of ​​the chromosome annotation box and the area of ​​the mask of the chromosome annotation box on the original image: ； in, The denotes the dispersion index, Represents the sum of the areas of the masks of the chromosome annotation boxes on the original image, Represents the total area of ​​the chromosome annotation box.

Images