Transformer-based methods, devices, equipment, and storage media for Down syndrome screening.

Abstract

This application belongs to the technical field of medical screening methods, specifically relating to a Transformer-based method, apparatus, device, and storage medium for Down syndrome screening. The method includes: inputting a chromosome image for screening into a preset Transformer segmentation model to segment individual chromosomes in the chromosome image, obtaining chromosome sub-blocks; the Transformer segmentation model is a Two-Stage structure model; aligning the chromosome sub-blocks to obtain individual chromosome images; inputting the individual chromosome images into a preset Transformer classification model to perform classification prediction on the individual chromosome images, obtaining predicted classification results; and performing chromosome screening based on the predicted classification results to obtain chromosome indication results. This addresses the problem of low screening efficiency.

Description

Technical Field

[0001] This application belongs to the field of medical screening methods technology, specifically relating to a Transformer-based method, device, equipment, and storage medium for Down syndrome screening. Background Technology

[0002] Down syndrome, also known as trisomy 21 or Down's Syndrome (DS), is the most common birth defect causing intellectual disability. Its occurrence is accidental and random, and the incidence increases with the maternal age. There is currently no effective treatment for the disease, therefore, prenatal screening during pregnancy is necessary.

[0003] Traditional screening methods for Down syndrome include: in routine clinical workflows, microscopy is typically used to observe chromosome number and morphology in order to screen for genetic disorders such as Down syndrome.

[0004] However, screening by observing chromosomes under a microscope requires a high level of expertise from medical staff, is complex and time-consuming, resulting in low screening efficiency. Summary of the Invention

[0005] This application provides a Transformer-based method, apparatus, device, and storage medium for Down syndrome screening, which can solve the problem of low screening efficiency. This application provides the following technical solution:

[0006] Firstly, a Transformer-based method for screening Down syndrome is provided. The method includes: inputting a chromosome image for screening into a preset Transformer segmentation model to segment individual chromosomes in the chromosome image, obtaining chromosome sub-blocks; the Transformer segmentation model is a Two-Stage structure model; aligning the chromosome sub-blocks to obtain individual chromosome images; inputting the individual chromosome images into a preset Transformer classification model to perform classification prediction on the individual chromosome images, obtaining predicted classification results; and performing chromosome screening based on the predicted classification results to obtain chromosome indication results.

[0007] Optionally, the step of segmenting a single chromosome in the chromosome image to obtain several chromosome sub-blocks includes: performing feature recognition on the chromosome image using the Transformer segmentation model to identify single chromosomes in the chromosome image; predicting the bounding box corresponding to each single chromosome; and performing masking processing according to the bounding box to obtain the chromosome sub-blocks.

[0008] Optionally, the step of aligning the chromosome sub-blocks using the Transformer segmentation model to obtain a single chromosome image includes: rotating the chromosome sub-blocks to align the chromosomes within them; and cropping the masked region of the chromosome sub-blocks based on the edge contours of the chromosomes within them to obtain the single chromosome image.

[0009] Optionally, before inputting the chromosome images to be screened into the preset Transformer segmentation model, the method further includes: obtaining pre-trained model parameters; loading the model parameters into the preset Transformer model; constructing a chromosome image segmentation training set; and training the preset Transformer model based on the chromosome image segmentation training set to obtain the Transformer segmentation model.

[0010] Optionally, the chromosome image segmentation training set includes a training set, a validation set, a test set, and a label file; constructing the chromosome image segmentation training set includes: obtaining a preset chromosome dataset, including a segmentation dataset and a classification dataset; the image samples in the segmentation dataset have the same label as the corresponding samples in the classification dataset; writing the bounding box information and mask region information corresponding to the sample images in the preset chromosome dataset into the label file; determining a preset proportion of samples in the classification dataset; generating T21 chromosome samples based on the preset proportion of samples; and dividing the segmentation dataset and the classification dataset into the training set, the validation set, and the test set.

[0011] Optionally, the preset chromosome dataset includes a G-Band dataset, a Q-Band dataset, or an M-FISH dataset.

[0012] Optionally, the predicted classification result includes 24 predicted chromosome classifications; the chromosome screening based on the predicted classification result to obtain the chromosome indication result includes: screening among the 24 chromosomes to determine the predicted number of chromosome 21; determining the chromosome indication result based on the number of chromosome 21; the chromosome indication result D is:

[0013]

[0014] Where, N 21 The predicted number of chromosome 21.

[0015] Secondly, a Transformer-based Down syndrome screening device is provided, characterized in that the device comprises:

[0016] The segmentation module is used to input the chromosome image to be screened into a preset Transformer segmentation model, and to segment the single chromosome in the chromosome image to obtain a preset number of chromosome sub-blocks; the Transformer segmentation model is a Two-Stage structure model;

[0017] An alignment module is used to align the chromosome sub-blocks to obtain a preset number of single chromosome images;

[0018] The classification module inputs the preset number of single chromosome images into a preset Transformer classification model, performs classification prediction on the single chromosome images, and obtains the predicted classification result;

[0019] The indicator module is used to perform chromosome screening based on the predicted classification results to obtain chromosome indicator results.

[0020] Thirdly, an electronic device is provided, including a memory, a controller, and a computer program stored in the memory and executable on the controller, wherein the controller executes the computer program to implement the steps of the Transformer-based Down syndrome screening method described above.

[0021] Fourthly, a computer-readable storage medium is provided, wherein a program is stored therein, which, when executed by a processor, is used to implement the Transformer-based Down syndrome screening and control method provided in the first aspect.

[0022] The beneficial effects of this application include at least the following: by inputting the chromosome image used for screening into a preset Transformer segmentation model, individual chromosomes in the chromosome image are segmented to obtain chromosome sub-blocks; the Transformer segmentation model is a two-stage structure model; the chromosome sub-blocks are aligned to obtain individual chromosome images; the individual chromosome images are input into a preset Transformer classification model to perform classification prediction on the individual chromosome images to obtain predicted classification results; and chromosome screening is performed based on the predicted classification results to obtain chromosome indication results. This addresses the problem of low screening efficiency. By inputting the chromosome image into the Transformer segmentation model to segment individual chromosomes in the chromosome image, inputting the segmentation results into the Transformer classification model to classify the segmentation results, and then indicating the chromosomes based on the classification results, there is no need to observe the chromosomes under a microscope, thus improving screening efficiency.

[0023] Furthermore, the Transformer segmentation model and the Transformer classification model are models obtained through transfer learning, enabling Transformer-based networks to share weights learned from the source domain (e.g., the ImageNet dataset) and optimize the weights using limited labeled data from the target domain (e.g., the BioImLab dataset). By training the Transformer model on a chromosome image segmentation training set, the Transformer segmentation model can be obtained, thus improving the accuracy of screening.

[0024] In addition, the Transformer segmentation model is a two-stage model. During the segmentation process, after extracting the features of the image, it predicts and corrects the candidate bounding boxes of objects in the image in a separate process, instead of directly segmenting after extracting the image features. This can improve the accuracy of segmentation. Attached Figure Description

[0025] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 This is a flowchart of a Transformer-based Down syndrome screening method provided in one embodiment of this application;

[0027] Figure 2 This is a schematic diagram of a chromosome image to be screened provided in one embodiment of this application;

[0028] Figure 3 This is a schematic diagram of a chromosome image after a predicted bounding box, provided in one embodiment of this application;

[0029] Figure 4 This is a schematic diagram of a model structure provided in one embodiment of this application;

[0030] Figure 5 This is a schematic diagram of the classification result of the Transformer classification model provided in one embodiment of this application;

[0031] Figure 6 This is a confusion matrix of the prediction results provided in one embodiment of this application;

[0032] Figure 7 This is a block diagram of a Transformer-based Down syndrome screening method apparatus provided in one embodiment of this application;

[0033] Figure 8 This is a block diagram of an electronic device provided in one embodiment of this application. Detailed Implementation

[0034] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. The application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0035] It should be noted that the terms "first," "second," etc., in the specification, claims, and drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0036] In this application, unless otherwise stated, directional terms such as "upper," "lower," "top," and "bottom" are generally used in relation to the direction shown in the accompanying drawings, or in relation to the vertical, perpendicular, or gravitational direction of the component itself; similarly, for ease of understanding and description, "inner" and "outer" refer to the inner and outer contours of each component itself, but the above directional terms are not intended to limit the invention.

[0037] The following is an explanation of the terms used in this application.

[0038] Swin Transformer is a simple network based entirely on an attention mechanism, consisting of an encoder and a decoder. The attention mechanism allows the network to add more weights to the interest context, achieving higher performance and lower parameter costs.

[0039] Down syndrome (DS), also known as trisomy 21, is a serious genetic disorder caused by an extra chromosome 21. Typical symptoms include intellectual disability, physical abnormalities, and heart disease. Chromosomal karyotype analysis is widely used for prenatal screening and diagnosis of genetic diseases such as Down syndrome.

[0040] The following section provides a detailed description of the Transformer-based Down syndrome screening method provided in this application.

[0041] This embodiment provides a Transformer-based method for Down syndrome screening, such as... Figure 1 As shown. This method includes at least the following steps:

[0042] Step 101: Input the chromosome image to be screened into the preset Transformer segmentation model to segment the single chromosome in the chromosome image to obtain chromosome sub-blocks.

[0043] The Transformer segmentation model is a two-stage model. This two-stage model first performs feature recognition on the input chromosome image to obtain individual chromosomes, and then predicts the bounding box of each chromosome.

[0044] Specifically, the process involves segmenting individual chromosomes in a chromosome image to obtain a predetermined number of chromosome sub-blocks, including: using a Transformer segmentation model to identify features in the chromosome image and identify single chromosomes in the chromosome image; predicting the bounding box corresponding to each individual chromosome; and performing masking processing based on the bounding boxes to obtain chromosome sub-blocks.

[0045] For example: Reference Figures 2 to 3 ,Will Figure 2 The chromosome image shown is input into the Transformer segmentation model, which predicts the bounding box corresponding to each chromosome (see reference). Figure 3 Each chromosome is individually masked according to the predicted bounding box to obtain chromosome sub-blocks.

[0046] Deep learning-based image segmentation techniques can be divided into one-stage and two-stage approaches. One-stage refers to directly predicting the bounding boxes (including the position and edges of the predicted objects) of the image after extracting its features. Two-stage, on the other hand, involves predicting and refining the bounding boxes of the objects in the image in a separate process after extracting the image features.

[0047] Based on this, in this embodiment, two modules, segmentation and classification, are designed based on model-level strategies and transfer learning mechanisms. This enables the Transformer-based network to share weights learned from the source domain (e.g., the ImageNet dataset) and optimize the weights using limited labeled data from the target domain (e.g., the BioImLab dataset). The Transformer model is then trained by constructing a chromosome image segmentation training set to obtain the Transformer segmentation model.

[0048] Specifically, before inputting the chromosome images to be screened into the preset Transformer segmentation model, the process includes: obtaining pre-trained model parameters; loading the model parameters into the preset Transformer model; constructing a chromosome image segmentation training set; and training the preset Transformer model based on the chromosome image segmentation training set to obtain the Transformer segmentation model.

[0049] The chromosome image segmentation training set includes a training set, a validation set, a test set, and a label file.

[0050] In this embodiment, the chromosome image segmentation training set can be divided into a training set, a validation set, and a test set according to a certain ratio. The proportions of the training set, validation set, and test set in the chromosome image segmentation training set can be the same or different.

[0051] In one example, the training set accounts for 85% of the chromosome image segmentation training set, while the validation set accounts for 5% and the test set accounts for 15%.

[0052] In another example, the training set accounts for 80% of the chromosome image segmentation training set, and correspondingly, the validation set accounts for 10% and the test set accounts for 10%.

[0053] This embodiment does not limit the proportion of the training set, validation set, and test set in the chromosome image segmentation training set.

[0054] Specifically, constructing a chromosome image segmentation training set includes: obtaining a preset chromosome dataset, including a segmentation dataset and a classification dataset; ensuring that the image samples in the segmentation dataset have the same label as the corresponding samples in the classification dataset; writing the bounding box information and mask region information corresponding to the sample images in the preset chromosome dataset into a label file; determining a preset proportion of samples in the classification dataset; generating T21 chromosome samples based on the preset proportion of samples; and dividing the segmentation dataset and the classification dataset into the training set, validation set, and test set.

[0055] The preset chromosome datasets include the G-Band dataset, the Q-Band dataset, or the M-FISH dataset.

[0056] For example, taking the preset chromosome dataset as the Q-Band dataset, which contains 163 images but does not include Down syndrome cases, we can randomly select 25% of the samples from the classification dataset, copy one more chromosome 21 to make it a T21 sample, and select one chromosome 21 from the corresponding chromosome karyotype image and paste it into the microscope image for use as the input image during segmentation training.

[0057] In actual implementation, the preset chromosome dataset can be other publicly available chromosome datasets or user-defined, unpublished chromosome datasets. This embodiment does not limit the implementation method of the preset chromosome dataset.

[0058] In this embodiment, the preset ratio is the proportion of T21 chromosome samples in the classification dataset that is set in advance. This ratio can be 25% or 30%, etc. This embodiment does not limit the value of the preset ratio.

[0059] Step 102: Align the chromosome sub-blocks to obtain a preset number of monochromatic chromosome images.

[0060] In this embodiment, by aligning the chromosome sub-blocks, the edges of the bounding box can be aligned to more closely match the edge contours of the chromosomes within the chromosome sub-blocks based on pixel dimensions.

[0061] Specifically, aligning the chromosome sub-blocks to obtain a preset number of single chromosome images includes: rotating the chromosome sub-blocks to align the chromosomes within them; and cropping the chromosome sub-blocks based on the edge contours of the chromosomes within them to obtain the single chromosome image.

[0062] Furthermore, even if a segmentation error occurs during chromosome image segmentation, resulting in two chromosome segments being taken as one, it is still treated as a single chromosome.

[0063] Step 103: Input a preset number of single chromosome images into a preset Transformer classification model to perform classification prediction on the single chromosome images and obtain the predicted classification results.

[0064] like Figure 4 As shown, in this embodiment, classification prediction of a single chromosome image is performed between local windows to accelerate performance. Layer normalization is applied before the Attn1, Attn2, and FFN modules, and residual connections follow Attn1, Attn2, and FFN. Attn1 is a multi-head self-attention algorithm that is computed only within a patched (4x4) window. Attn2 expands Attn1 by moving the window one pixel to the bottom right corner. The FFN module consists of two linear layers. The output channels for the four stages are 96, 192, 384, and 768, respectively. Finally, the output of the fourth stage is achieved by a normalization layer, an average pooling layer, and a fully connected layer, outputting the predicted class from 1 to 24.

[0065] In this embodiment, since the structure and training process of the Transformer classification model are the same as those of the Transformer segmentation model, they will not be described again here.

[0066] In this embodiment, a single chromosome image is input into the Transformer classification model to extract features from the single chromosome image, obtain the image features of each single chromosome image, and perform classification prediction based on the image features.

[0067] After inputting a predetermined number of single chromosome images into the Transformer classification model, the classification prediction obtained by the Transformer classification model is as follows: Figure 5 The classification prediction results are shown.

[0068] Step 104: Based on the predicted classification results, chromosome screening is performed to obtain chromosome indicator results.

[0069] The predicted classification results include 24 predicted chromosome classifications.

[0070] In this embodiment, based on the segmentation of chromosome images using the Transformer segmentation model, the segmented individual chromosome images are further processed. Specifically, the segmented individual chromosomes are classified using the Transformer classification model, and the classified individual chromosomes are screened to determine the number of chromosome 21. Down syndrome is then screened based on the number of chromosome 21.

[0071] Specifically, chromosome screening based on the predicted classification results yields chromosome indicator results including: screening 24 chromosome types to determine the predicted number of chromosome 21; determining the chromosome indicator result based on the number of chromosome 21; and the chromosome indicator result D is:

[0072]

[0073] Where, N 21 The predicted number of chromosome 21.

[0074] As can be seen from the formula corresponding to chromosome indicator result D, in this embodiment, when the number of chromosomes classified as 21 is greater than 2, the value of chromosome indicator result D is 1; when the number of chromosomes classified as 21 is less than or equal to 2, the value of chromosome indicator result D is 0.

[0075] In this embodiment, if the value of chromosome indicator result D is 1, then chromosome indicator result D indicates that the patient has Down syndrome; if the value of chromosome indicator result D is 0, then chromosome indicator result D indicates that the patient does not have Down syndrome.

[0076] In summary, the Transformer-based Down syndrome screening method provided in this embodiment inputs the chromosome image for screening into a preset Transformer segmentation model to segment individual chromosomes in the chromosome image, obtaining chromosome sub-blocks. The Transformer segmentation model is a two-stage model. The chromosome sub-blocks are aligned to obtain individual chromosome images. These individual chromosome images are then input into a preset Transformer classification model for classification prediction, yielding predicted classification results. Chromosome screening is performed based on the predicted classification results to obtain chromosome indication results. This addresses the problem of low screening efficiency. By inputting the chromosome image into the Transformer segmentation model to segment individual chromosomes, and then inputting the segmentation results into the Transformer classification model for classification, the chromosomes are indicated based on the classification results. This eliminates the need for microscopic observation of the chromosomes, thus improving screening efficiency.

[0077] Furthermore, the Transformer segmentation model and the Transformer classification model are models obtained through transfer learning, enabling Transformer-based networks to share weights learned from the source domain (e.g., the ImageNet dataset) and optimize the weights using limited labeled data from the target domain (e.g., the BioImLab dataset). By training the Transformer model on a chromosome image segmentation training set, the Transformer segmentation model can be obtained, thus improving the accuracy of screening.

[0078] In addition, the Transformer segmentation model is a two-stage model. During the segmentation process, after extracting the features of the image, it predicts and corrects the candidate bounding boxes of objects in the image in a separate process, instead of directly segmenting after extracting the image features. This can improve the accuracy of segmentation.

[0079] In one example, a simulation experiment was conducted on a Transformer-based Down syndrome screening method. In the simulation experiment of this embodiment, the number of positive cases was 36, that is, the number of chromosome 21 in the 16 samples was 36; the number of negative cases was 704, that is, the number of chromosomes other than 21 was 704; the total number of chromosomes was 740.

[0080] In this embodiment, two other existing models, namely the ResNet-50 model (referred to as Res in Table 1) and the SE-ResNeXt-50 model (referred to as SE in Table 1), are selected and compared with the Two-Stage Transformer segmentation model and Transformer classification model (referred to as Swin in Table 1) provided in this embodiment through simulation experiments. Figure 6 , Figure 6 This is the confusion matrix of the prediction results. The actual comparison results are shown in Table 1.

[0081] Table 1:

[0082]

[0083] In Table 1, "on classification Test set" refers to the prediction result on the test set of the classification dataset; "on segment output" refers to the prediction result on the segmentation result of the test set of the segmentation dataset.

[0084] As can be seen from Table 1, the Transformer segmentation model and Transformer classification model with the Two-Stage structure after transfer learning in this embodiment have the best performance, which can effectively determine whether the case is normal, thereby reducing the possibility of false negative cases.

[0085] This embodiment provides a Transformer-based Down syndrome screening device, such as... Figure 7 As shown, the device includes at least the following modules: a segmentation module 710, an alignment module 720, a classification module 730, and an indication module 740.

[0086] The segmentation module 710 is used to input the chromosome image to be screened into a preset Transformer segmentation model, segment the single chromosome in the chromosome image, and obtain a preset number of chromosome sub-blocks; the Transformer segmentation model is a two-stage structure model;

[0087] Alignment module 720 is used to align chromosome sub-blocks to obtain a preset number of single chromosome images;

[0088] The classification module 730 inputs a preset number of single chromosome images into a preset Transformer classification model to perform classification prediction on the single chromosome images and obtain the predicted classification results;

[0089] The indicator module 740 is used to perform chromosome screening based on the predicted classification results and obtain chromosome indicator results.

[0090] For relevant details, please refer to the above-described method and equipment embodiments.

[0091] It should be noted that the Transformer-based Down syndrome screening device provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the Transformer-based Down syndrome screening device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the Transformer-based Down syndrome screening device and the Transformer-based Down syndrome screening method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0092] This embodiment provides an electronic device, such as... Figure 8 As shown. The electronic device includes at least a processor 801 and a memory 802.

[0093] Processor 801 may include one or more processing cores, such as a quad-core processor or an octa-core processor. Processor 801 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 801 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 801 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 801 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0094] Memory 802 may include one or more computer-readable storage media, which may be non-transitory. Memory 802 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in memory 802 is used to store at least one instruction, which is executed by processor 801 to implement the Transformer-based Down syndrome screening method provided in the method embodiments of this application.

[0095] In some embodiments, the electronic device may also optionally include: a peripheral device interface and at least one peripheral device. The processor 801, memory 802, and peripheral device interface can be connected via a bus or signal line. Each peripheral device can be connected to the peripheral device interface via a bus, signal line, or circuit board. Indicatively, peripheral devices include, but are not limited to: radio frequency circuitry, a touch display screen, audio circuitry, and a power supply.

[0096] Of course, electronic devices may also include fewer or more components, and this embodiment does not limit this.

[0097] Optionally, this application also provides a computer-readable storage medium storing a program that is loaded and executed by a processor to implement the Transformer-based Down syndrome screening method of the above method embodiments.

[0098] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0099] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A Transformer-based Down syndrome screening method, characterized in that, The method comprises: inputting a chromosome image for screening into a preset Transformer segmentation model, segmenting a single chromosome in the chromosome image to obtain a chromosome sub-block; the Transformer segmentation model is a model of a Two-Stage structure; aligning the chromosome sub-block to obtain a single chromosome image; inputting the single chromosome image into a preset Transformer classification model, classifying and predicting the single chromosome image to obtain a predicted classification result; performing chromosome screening based on the predicted classification result to obtain a chromosome indication result; the aligning of the chromosome sub-block to obtain a single chromosome image comprises: rotating the chromosome sub-block to align the chromosome in the chromosome sub-block; trimming a mask region of the chromosome sub-block based on an edge contour of the chromosome in the chromosome sub-block to obtain the single chromosome image; the predicted classification result comprises a predicted classification of 24 chromosomes; the chromosome screening based on the predicted classification result to obtain a chromosome indication result comprises: screening and determining a predicted number of chromosome 21 from the 24 chromosomes; determining the chromosome indication result based on the number of chromosome 21; the chromosome indication result D is: wherein, is the predicted chromosome 21 number.

2. The method of claim 1, wherein, the segmenting of a single chromosome in the chromosome image to obtain a plurality of chromosome sub-blocks comprises: performing feature recognition on the chromosome image by the Transformer segmentation model to recognize a single chromosome in the chromosome image; predicting a bounding box corresponding to each single chromosome; performing mask processing according to the bounding box to obtain the chromosome sub-block.

3. The method of claim 1, wherein, Before the inputting of the chromosome image for screening into the preset Transformer segmentation model, the method further comprises: obtaining pre-trained model parameters; loading the model parameters into a preset Transformer model; constructing a chromosome image segmentation training set; training the preset Transformer model based on the chromosome image segmentation training set to obtain the Transformer segmentation model.

4. The method of claim 3, wherein, The chromosome image segmentation training set comprises a training set, a validation set, a test set and a label file; the constructing of the chromosome image segmentation training set comprises: obtaining a preset chromosome data set, including a segmentation data set and a classification data set; the image samples in the segmentation data set have the same label as the corresponding samples in the classification data set; writing the bounding box information and the mask region information corresponding to the sample images in the preset chromosome data set into the label file; determining a preset proportion of samples in the classification data set; generating T21 chromosome samples based on the preset proportion of samples; dividing the segmentation data set and the classification data set into the training set, the validation set and the test set.

5. The method of claim 4, wherein, The preset chromosome data set comprises a G-Band data set, a Q-Band data set or an M-FISH data set. 6.A Down syndrome screening device employing the Down syndrome screening method based on the Transformer according to any one of claims 1 to 5, characterized in that, The device comprises: The segmentation module is configured to input a chromosome image used for screening into a preset Transformer segmentation model, segment a single chromosome in the chromosome image, and obtain a chromosome sub-block; the Transformer segmentation model is a model with a Two-Stage structure. The alignment module is configured to perform alignment processing on the chromosome sub-block, and obtain a single chromosome image. The classification module is configured to input the single chromosome image into a preset Transformer classification model, perform classification prediction on the single chromosome image, and obtain a predicted classification result. The indication module is configured to perform chromosome screening based on the predicted classification result, and obtain a chromosome indication result.

7. An electronic device, comprising: The electronic device comprises a processor and a memory; the memory stores a program, which is loaded and executed by the processor to implement the Transformer-based Down's syndrome screening method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The storage medium stores a program, which is executed by the processor to implement the Transformer-based Down's syndrome screening method according to any one of claims 1 to 5.

Images