High-generalization spine sagittal view centrum detection method based on multi-scale block data amplification

Abstract

The invention discloses a high-generalization spine sagittal view centrum detection method based on multi-scale block data amplification, and belongs to the technical field of image detection.The high-generalization spine sagittal view centrum detection method based on multi-scale block data amplification comprises the steps that a spine sagittal view X-ray picture to be detected is acquired; inputting a to-be-detected spine sagittal X-ray picture into the trained whole spine detection model to obtain a detection result of the to-be-detected spine sagittal X-ray picture; the whole spine detection model is obtained through training of multi-center training data, and the multi-center training data is obtained through a tag block segmentation method. According to the method, diversified training samples are dynamically generated through data processing modes such as style conversion, the overall quality of the multi-center spine X-ray image is improved, and then the performance and applicability of a vertebral body detection model are improved.

Description

Technical Field

[0001] This application belongs to the technical field of image detection, and specifically relates to a high-generalization spinal sagittal vertebral body detection method based on multi-scale block data augmentation. Background Art

[0002] In recent years, deep convolutional neural networks (CNNs) have achieved remarkable results in medical image detection. Although these models have been successful, their performance largely depends on the assumption that the training samples and test samples are independent and identically distributed. However, in medical applications, due to different scanning protocols, different device vendors, and different imaging modalities, etc., the distribution between the source (training) and target (test) data often changes. Directly applying the model trained on the source domain to an unknown target domain with a different distribution usually results in a significant performance decline.

[0003] Currently, style transfer technology has been widely used in medical image processing, especially in cross-domain image generation. However, existing style transfer methods have some limitations when dealing with spinal sagittal X-ray images. First, traditional style transfer methods usually perform coarse-grained style conversion at the image level, which may lead to blurred vertebral edges or inconsistent with the annotations, thus affecting the detection accuracy; second, the style transfer process is very sensitive to the image spatial information. When directly processing X-ray images, the quality of the generated images is reduced and not realistic enough. Summary of the Invention

[0004] The purpose of this application is to provide a high-generalization spinal sagittal vertebral body detection method based on multi-scale block data augmentation to solve the problem of poor cross-center generalization of the vertebral body detection model due to style differences between multiple centers, which helps the cross-center deployment of the whole spine detection model.

[0005] According to the first aspect of the embodiments of this application, a high-generalization spinal sagittal vertebral body detection method based on multi-scale block data augmentation is provided. The method may include:

[0006] Obtain the spinal sagittal X-ray image to be detected;

[0007] Input the spinal sagittal X-ray image to be detected into the trained whole spine detection model to obtain the detection result of the spinal sagittal X-ray image to be detected;

[0008] The whole spine detection model is trained with multi-center training data, and the multi-center training data is obtained by the block segmentation method of labels.

[0009] In some alternative embodiments of this application, the whole spine detection model is a style transfer model;

[0010] The block segmentation method of labels includes:

[0011] Obtain the first sagittal X-ray data of the spine with source center tags;

[0012] Generate the second sagittal X-ray data of the spine with pseudo-tags at the target center using the first sagittal X-ray data of the spine;

[0013] Perform multi-scale segmentation on the first sagittal X-ray data of the spine and the second sagittal X-ray data of the spine to obtain training data containing global and local features.

[0014] In some alternative embodiments of the present application, the multi-scale segmentation includes:

[0015] Large block segmentation, medium block segmentation, and small block segmentation;

[0016] The large block segmentation is to randomly select the bounding boxes of two vertebrae and merge them to form the boundary range of the large block segmentation;

[0017] The medium block segmentation is to generate a bounding box for the first eighteen vertebrae of the whole spine, and expand the height of half a vertebra in the up and down directions and expand 100 pixels on the left and right respectively to form the medium block segmentation;

[0018] The small block segmentation is to divide the first eighteen vertebrae into three groups, randomly select one group of vertebrae to generate a bounding box, and expand the height of half a vertebra in the up and down directions and expand 100 pixels on the left and right respectively to form the small block segmentation.

[0019] In some alternative embodiments of the present application, performing multi-scale segmentation on the first sagittal X-ray data of the spine and the second sagittal X-ray data of the spine to obtain training data containing global and local features includes:

[0020] Perform multi-scale segmentation on the first sagittal X-ray data of the spine and the second sagittal X-ray data of the spine in an iterative manner;

[0021] Each time when accessing the images in the dataset during iteration, segment the first sagittal X-ray data of the spine and the second sagittal X-ray data of the spine according to the preset scale number ratio to obtain training data containing global and local features.

[0022] In some alternative embodiments of the present application, in each training iteration, generate a random number for each image data and perform segmentation at different scales according to the random number, so that different segmentation results are generated for the same image in different rounds.

[0023] In some alternative embodiments of the present application, generating the second sagittal X-ray data of the spine with pseudo-tags at the target center using the first sagittal X-ray data of the spine includes:

[0024] Train a vertebral body detection model using the first sagittal X-ray data of the spine to obtain a trained vertebral body detection model;

[0025] Obtain the sagittal X-ray data of the spine at the target center, and input the sagittal X-ray data of the spine into the trained vertebral body detection model to obtain data with pseudo-labels.

[0026] In some alternative embodiments of the present application, after obtaining the sagittal X-ray data of the spine at the target center, inputting the sagittal X-ray data of the spine into the trained vertebral body detection model, and obtaining data with pseudo-labels, it further includes:

[0027] Perform a consistency comparison between the data with pseudo-labels and the sagittal X-ray data of the spine;

[0028] Determine the data with pseudo-labels that pass the consistency comparison as the second sagittal X-ray data of the spine with pseudo-labels at the target center.

[0029] According to a second aspect of the embodiments of the present application, there is provided a highly generalized spinal sagittal vertebral body detection device based on multi-scale block data augmentation, and the device may include:

[0030] An acquisition module, configured to acquire a sagittal X-ray image of the spine to be detected;

[0031] A detection module, configured to input the sagittal X-ray image of the spine to be detected into the trained full-spine detection model to obtain a detection result of the sagittal X-ray image of the spine to be detected;

[0032] The full-spine detection model is trained by multi-center training data, and the multi-center training data is obtained by a method of block segmentation of labels.

[0033] According to a third aspect of the embodiments of the present application, there is provided an electronic device, and the electronic device may include:

[0034] A processor;

[0035] A memory for storing instructions executable by the processor;

[0036] Wherein, the processor is configured to execute instructions to implement the highly generalized spinal sagittal vertebral body detection method as shown in any one of the embodiments of the first aspect.

[0037] According to a fourth aspect of the embodiments of the present application, there is provided a storage medium, and when the instructions in the storage medium are executed by a processor of an information processing device or a server, the information processing device or the server is enabled to implement the highly generalized spinal sagittal vertebral body detection method as shown in any one of the embodiments of the first aspect.

[0038] The above technical solution of the present application has the following beneficial technical effects:

[0039] By means of data processing such as style conversion, the method of the embodiment of the present application dynamically generates diverse training samples, improves the overall quality of multi-center spinal X-ray images, and further improves the performance and applicability of the vertebral body detection model. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic flowchart of a high-generalization spinal sagittal vertebral body detection method based on multi-scale block data augmentation in an exemplary embodiment of the present application;

[0041] Figure 2 It is a schematic structural diagram of a high-generalization spinal sagittal vertebral body detection device based on multi-scale block data augmentation in an exemplary embodiment of the present application;

[0042] Figure 3 It is a schematic structural diagram of an electronic device in an exemplary embodiment of the present application;

[0043] Figure 4 It is a schematic hardware structure diagram of an electronic device in an exemplary embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] To make the objectives, technical solutions, and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the specific embodiments and with reference to the accompanying drawings. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present application. In addition, in the following description, the descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts of the present application.

[0045] In the drawings, a schematic diagram of a layer structure according to an embodiment of the present application is shown. These figures are not drawn to scale, where for the purpose of clarity, some details are enlarged and some details may be omitted. The shapes of various regions and layers shown in the figures, as well as their relative sizes and positional relationships, are merely exemplary and may actually deviate due to manufacturing tolerances or technical limitations. Those skilled in the art can additionally design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0046] Obviously, the described embodiments are some but not all of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts fall within the scope of protection of the present application.

[0047] In the description of the present application, it should be noted that the terms "first", "second", and "third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0048] In addition, the technical features involved in different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0049] The following will combine the accompanying drawings and, through specific embodiments and their application scenarios, provide a detailed description of the high-generalization spinal sagittal vertebral detection method based on multi-scale block data augmentation provided by the embodiments of the present application.

[0050] As Figure 1 shown, in the first aspect of the embodiment of the present application, a high-generalization spinal sagittal vertebral detection method based on multi-scale block data augmentation is provided, and the method may include:

[0051] S110: Obtain a spinal sagittal X-ray image to be detected;

[0052] S120: Input the spinal sagittal X-ray image to be detected into a trained full-spine detection model to obtain the detection result of the spinal sagittal X-ray image to be detected;

[0053] The full-spine detection model is trained through multi-center training data, and the multi-center training data is obtained through a block segmentation method of labels.

[0054] The method of this embodiment dynamically generates diverse training samples through data processing methods such as style conversion, improves the overall quality of multi-center spinal X-ray images, and further improves the performance and applicability of the vertebral detection model.

[0055] In this embodiment, the trained full-spine detection model is a CycleGAN model. Currently, the application of the CycleGAN model in the medical field mostly focuses on the style conversion of CT and MRI images. Since there are obvious differences in contrast, resolution, noise level, and displayed details between these two types of medical images, CycleGAN can better capture these differences for effective conversion. However, for spinal X-ray images from different medical centers, the style differences are often not as significant as those between CT and MRI. There are differences in X-ray imaging equipment, shooting techniques, and the operating habits of medical staff in different medical centers, and these differences are unstable, which makes it difficult for CycleGAN to effectively learn and capture style differences, thus affecting the effect of the model in the data conversion process.

[0056] In some embodiments, the full-spine detection model is a style transfer model;

[0057] The block segmentation method of labels includes:

[0058] Obtain the first spinal sagittal X-ray data with labels from the source center;

[0059] Generate the second sagittal X-ray data of the spine with pseudo-labels at the target center using the first sagittal X-ray data of the spine;

[0060] Perform multi-scale segmentation on the first sagittal X-ray data of the spine and the second sagittal X-ray data of the spine to obtain training data containing global and local features.

[0061] For the cross-center full-spine detection task, it is necessary to train a full-spine detection model using the labeled source center and unlabeled target center data. This embodiment can be used to solve the problem of poor cross-center generalization of the vertebral body detection model due to style differences between multiple centers in a multi-center scenario, and helps with the cross-center deployment of the full-spine detection model.

[0062] In some embodiments, the multi-scale segmentation includes:

[0063] Large-piece segmentation, medium-piece segmentation, and small-piece segmentation;

[0064] The large-piece segmentation is to randomly select the bounding boxes of two vertebrae and merge them to form the boundary range of the large-piece segmentation;

[0065] The medium-piece segmentation is to generate bounding boxes for the first eighteen vertebrae of the full spine, and expand the height of half a vertebra in the up and down directions and expand 100 pixels on the left and right respectively to form the medium-piece segmentation;

[0066] The small-piece segmentation is to divide the first eighteen vertebrae into three groups, randomly select one group of vertebrae to generate a bounding box, and expand the height of half a vertebra in the up and down directions and expand 100 pixels on the left and right to form the small-piece segmentation.

[0067] The style differences in multi-center data are mostly reflected in aspects such as color, texture, contrast, brightness, and artifacts. The style transfer model can quickly and effectively convert the images of the source center into the image style of the target center. In this embodiment, based on the CycleGAN style transfer model, a part of the source center images will be converted into the image style of the target center to narrow the data distributions of the source center and the target center.

[0068] Aiming at the problem that CycleGAN performs poorly in image details and small objects when generating pictures, this embodiment proposes a label-based Patch segmentation method. Patch refers to a local image with a fixed size extracted from the original image, which is used to concentrate on processing the features of this area. In this way, more precise style transfer can be performed in a smaller local area, avoiding information loss in details and small objects, thereby improving the overall quality of the generated images.

[0069] Exemplarily, pseudo-labels are generated using the data from source center A, and Patch-level data augmentation is implemented on target center B to improve the style transfer ability of CycleGAN for spinal X-ray images. Since the data of target center B lacks annotations, first, based on the vertebral detection model (Mask R-CNN) trained on source center A, pseudo-labels of vertebral bounding boxes (Bounding Box, BBox) are generated on the images of target center B. Based on the pseudo-label information, multi-scale Patch slicing is performed on the datasets of source center A and target center B to generate training samples containing global and local features. Patch slicing can include four scales, namely the original image, large Patch, medium Patch, and small Patch, to capture multi-level feature information.

[0070] For large Patch slicing, by randomly selecting the Bounding Boxes of two vertebrae and merging them into an overall enclosed area, and randomly expanding a certain boundary range around it, a larger-scale Patch is generated. This method enables CycleGAN to learn the macroscopic structural features of the entire spine by capturing a large range of context information, while retaining the important anatomical information of the original image (such as the vertebral arrangement pattern and overall contour). The design of the large Patch ensures that the model can take into account global information during style transfer, rather than being limited to local features.

[0071] For medium Patch slicing, Bounding Boxes are generated for the first eighteen vertebrae of the entire spine (covering the cervical and thoracic regions), and the height of half a vertebra is expanded in the up and down directions, and the boundary is expanded by 100 pixels on both the left and right to form a medium-scale Patch. Medium Patch focuses on solving the problem of poor style transfer effect in the cervical and thoracic regions. By expanding a certain context range, the model can capture the detailed features of adjacent regions while focusing on local information (such as the texture and structure between vertebrae), which helps to improve the accuracy and consistency of generation in the cervical and thoracic regions.

[0072] Small Patch slicing further refines the feature capture ability. Specifically, the first eighteen vertebrae are divided into three groups (6 vertebrae in each group), and Bounding Boxes are randomly generated for one of the groups of vertebrae, and the height of half a vertebra is expanded in the up and down directions, and 100 pixels are expanded on both the left and right to form a small-scale Patch.

[0073] Small Patch slicing aims to capture the local details of the cervical, thoracic, and lumbar vertebrae, especially the changes in the shape, size, and subtle anatomical features of the vertebrae. By reducing the Patch range, the model can learn key local features at a smaller scale, which helps to achieve higher-quality style transfer in terms of details and ensures that the generated images are consistent with the original images in anatomical details.

[0074] In some embodiments, the first spinal sagittal X-ray data and the second spinal sagittal X-ray data are segmented at multiple scales to obtain training data containing global and local features, including:

[0075] Segment the first spinal sagittal X-ray data and the second spinal sagittal X-ray data at multiple scales in an iterative manner;

[0076] Each time the images in the dataset are accessed during iteration, the first spinal sagittal X-ray data and the second spinal sagittal X-ray data are segmented according to a preset scale number ratio to obtain training data containing global and local features.

[0077] In some embodiments, in each training iteration, a random number is generated for each image data and segmented at different scales according to the random number, so that different segmentation results are generated for the same image in different rounds.

[0078] In order to make the data distribution of the training set closer to the data distribution of the target center, each time the images in the dataset are accessed during iteration, according to the set ratio (exemplarily, original image: large Patch: medium Patch: small Patch = 5:2:3:4), the images are segmented and processed by randomly extracting probability values. According to the pseudo-label information, the segmentation operations of large, medium, and small Patches are performed on the images to ensure that the samples of different scale Patches are reasonably distributed in the dataset. In each training iteration, a random number is generated for each image and different segmentation strategies are triggered according to the random number, so that different segmentation results are generated for the same image in different rounds. In this way, the model can learn different feature information in each round of training, thereby significantly enhancing the generalization ability.

[0079] This enhancement method enables CycleGAN to more comprehensively learn the features under different scales and domain differences by dynamically generating diverse training samples, improving the style transfer performance and the application effect of the generalization ability of the model in multi-center scenarios.

[0080] In some embodiments, using the first spinal sagittal X-ray data to generate the second spinal sagittal X-ray data with pseudo-labels at the target center, including:

[0081] Train a vertebral body detection model using the first spinal sagittal X-ray data to obtain a trained vertebral body detection model;

[0082] Obtain the spinal sagittal X-ray data of the target center and input the spinal sagittal X-ray data into the trained vertebral body detection model to obtain data with pseudo-labels.

[0083] In some embodiments, after acquiring the sagittal X-ray data of the spine at the target center and inputting the sagittal X-ray data of the spine into the trained vertebral body detection model to obtain the data with pseudo-labels, it further includes:

[0084] Performing a consistency comparison between the data with pseudo-labels and the sagittal X-ray data of the spine;

[0085] Determining the second sagittal X-ray data of the spine with pseudo-labels at the target center for the data with pseudo-labels that pass the consistency comparison.

[0086] Due to the limitations of the CycleGAN network structure, the length and width of the input and output images must be multiples of 4. This requirement may cause slight stretching (usually one or two pixels) of the image during generation, which may potentially affect the generated image and its corresponding label. Therefore, in this embodiment, to ensure the effectiveness and consistency of the augmented dataset, the labels of the augmented data are verified before using the generated images for the model.

[0087] To ensure the consistency of the labels between the images generated by CycleGAN and the original images, in this embodiment, corresponding red bounding boxes and segmentation regions are drawn on the original images and the generated images respectively according to the labels; the bounding boxes are used to mark the range of each vertebral body, and the segmentation regions are used to display the exact shape of the vertebral body. The original images and the generated images are compared one by one, and the consistency of the labels of the generated images and the original images is evaluated by comparing the positions, shapes, and coverage ranges of the bounding boxes and the segmentation regions. During the comparison process, check whether the shape and size of the vertebral body change, whether the center points of the bounding boxes remain consistent, and whether the contours of the segmentation regions deviate significantly. The comparison results can be displayed in a graphical form, and the consistency of each vertebral body label between the original image and the generated image is recorded.

[0088] The method of the embodiment of the present application aims to achieve style conversion through an improved CycleGAN while maintaining the details of the images. Specifically, aiming at the differences in imaging acquisition devices, shooting techniques, and operating habits between the source center and the target center, the generalization ability of style conversion is improved. In addition, based on the label-based patch segmentation method, the problem that CycleGAN performs poorly in terms of details and small objects during style conversion is solved. By generating pseudo-labels from the labeled data at the source center and implementing patch-level enhancement on the data at the target center. In addition, adapting the training strategy of multi-scale patch segmentation improves the image generation quality. Dynamic adjustment is performed during model training to generate different data samples, thereby improving the diversity of training data and enhancing the generalization performance and generation quality of the CycleGAN model.

[0089] In summary, the cross-center highly generalized spinal sagittal vertebral detection method based on multi-scale block data augmentation helps improve the cross-center generalization of the full-spine vertebral detection model and enables the vertebral detection model to be applied across different centers in orthopedic clinics.

[0090] It should be noted that for the highly generalized spinal sagittal vertebral detection method based on multi-scale block data augmentation provided in the embodiments of the present application, the execution subject can be a highly generalized spinal sagittal vertebral detection device based on multi-scale block data augmentation, or a control module in the highly generalized spinal sagittal vertebral detection device for executing the method of highly generalized spinal sagittal vertebral detection based on multi-scale block data augmentation. In the embodiments of the present application, the highly generalized spinal sagittal vertebral detection device based on multi-scale block data augmentation is taken as an example for executing the method of highly generalized spinal sagittal vertebral detection based on multi-scale block data augmentation to illustrate the highly generalized spinal sagittal vertebral detection device provided in the embodiments of the present application.

[0091] As Figure 2 shown, in the second aspect of the embodiments of the present application, a highly generalized spinal sagittal vertebral detection device based on multi-scale block data augmentation is provided. The device may include:

[0092] An acquisition module 210, configured to acquire a spinal sagittal X-ray image to be detected;

[0093] A detection module 220, configured to input the spinal sagittal X-ray image to be detected into a trained full-spine detection model to obtain a detection result of the spinal sagittal X-ray image to be detected;

[0094] The full-spine detection model is trained with multi-center training data, and the multi-center training data is obtained through a method of block segmentation of labels.

[0095] The high-generalization spinal sagittal vertebral detection device based on multi-scale block data augmentation in the embodiments of the present application can be a device, or a component, an integrated circuit, or a chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. Exemplarily, the mobile electronic device can be a mobile phone, a tablet computer, a laptop computer, a handheld computer, a vehicle-mounted electronic device, a wearable device, an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA), etc., and the non-mobile electronic device can be a server, a Network Attached Storage (NAS), a personal computer (PC), a television (TV), a teller machine, or a self-service machine, etc. The embodiments of the present application do not make specific limitations.

[0096] The high-generalization spinal sagittal vertebral detection device based on multi-scale block data augmentation in the embodiments of the present application can be a device with an operating system. The operating system can be the Android operating system, the iOS operating system, or other possible operating systems. The embodiments of the present application do not make specific limitations.

[0097] The high-generalization spinal sagittal vertebral detection device provided in the embodiments of the present application can implement Figure 1 each process implemented by the method embodiments. To avoid repetition, it will not be elaborated here.

[0098] Optionally, as Figure 3 shown, the embodiments of the present application further provide an electronic device 300, including a first processor 301, a first memory 302, a program or instruction stored on the first memory 302 and executable on the first processor 301. When the program or instruction is executed by the first processor 301, it implements each process of the above-mentioned high-generalization spinal sagittal vertebral detection method embodiment based on multi-scale block data augmentation, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0099] It should be noted that the electronic device in the embodiments of the present application includes the above-mentioned mobile electronic devices and non-mobile electronic devices.

[0100] Figure 4 It is a schematic diagram of the hardware structure of an electronic device for implementing the embodiments of the present application.

[0101] The hardware structure 400 of the electronic device includes, but is not limited to, components such as a radio frequency unit 401, a network module 402, an audio output unit 403, an input unit 404, a sensor 405, a display unit 406, a user input unit 407, an interface unit 408, a second memory 409, and a second processor 410.

[0102] Those skilled in the art can understand that the hardware structure 400 of the electronic device may further include a power source (such as a battery) for supplying power to each component. The power source can be logically connected to the second processor 410 through a power management system, so as to implement functions such as management of charging, discharging, and power consumption management through the power management system. Figure 4 The structure of the electronic device shown does not limit the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements, which will not be elaborated here.

[0103] It should be understood that in the embodiments of the present application, the input unit 404 may include a Graphics Processing Unit (GPU) 4041 and a microphone 4042. The graphics processor 4041 processes the image data of static pictures or videos obtained by an image capture device (such as a camera) in a video capture mode or an image capture mode. The display unit 406 may include a display panel 4061, and the display panel 4061 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc. The user input unit 407 includes a touch panel 4071 and other input devices 4072. The touch panel 4071 is also called a touch screen. The touch panel 4071 may include two parts: a touch detection device and a touch controller. The other input devices 4072 may include, but are not limited to, a physical keyboard, function keys (such as volume control keys, switch keys, etc.), a trackball, a mouse, and a joystick, which will not be elaborated here. The second memory 409 may be used to store software programs and various data, including but not limited to application programs and operating systems. The second processor 410 may integrate an application processor and a modulation and demodulation processor. Among them, the application processor mainly processes the operating system, user interface, and application programs, and the modulation and demodulation processor mainly processes wireless communication. It can be understood that the above modulation and demodulation processor may not be integrated into the second processor 410.

[0104] The embodiments of the present application also provide a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, it implements each process of the above-mentioned embodiment of the high-generalization spinal sagittal vertebral detection method based on multi-scale block data augmentation, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0105] Among them, the processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media such as computer read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs, etc.

[0106] Another embodiment of the present application provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement each process of the above-mentioned embodiment of the high-generalization spinal sagittal vertebral detection method based on multi-scale block data augmentation, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0107] It should be understood that the chip mentioned in the embodiments of the present application may also be referred to as a system-on-chip, system chip, chip system, or system-on-chip, etc.

[0108] It should be noted that in this article, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including that element. In addition, it should be pointed out that the methods and devices in the embodiments of the present application are not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in a reverse order according to the functions involved. For example, the described methods may be performed in an order different from that described, and various steps may be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0109] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disc), and includes several instructions to enable a terminal (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present application.

[0110] The embodiments of the present application have been described above in conjunction with the accompanying drawings. However, the present application is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present application, those of ordinary skill in the art can also make many forms without departing from the purpose of the present application and the scope protected by the claims, and all of them fall within the protection scope of the present application.

Claims

1. A highly generalized spinal sagittal vertebra detection method based on multi-scale block data augmentation, characterized in that: include: Obtain a sagittal X-ray image of the spine to be tested; Inputting the spinal sagittal X-ray image to be detected into the trained full spine detection model to obtain the detection result of the spinal sagittal X-ray image to be detected; The whole spine detection model is trained with multi-center training data, and the multi-center training data is obtained by a label block segmentation method.

2. The highly generalized spinal sagittal vertebra detection method based on multi-scale block data augmentation according to claim 1, characterized in that: The full spine detection model is a style transfer model; The tag block segmentation method includes: Obtain sagittal X-ray data of the first spine with a label at the source center; Using the first spinal sagittal X-ray data, generate second spinal sagittal X-ray data with a pseudo label at the target center; The first spinal sagittal X-ray data and the second spinal sagittal X-ray data are segmented at multiple scales to obtain training data containing global and local features.

3. The highly generalized spinal sagittal vertebra detection method based on multi-scale block data augmentation according to claim 1, characterized in that: The multi-scale segmentation includes: Large block cutting, medium block cutting and small block cutting; The large block segmentation is formed by merging the boundary boxes of two randomly selected vertebrae to form a boundary range of the large block segmentation; The middle block segmentation is to generate a bounding box for the first eighteen vertebrae of the entire spine, and expand the height of half of the vertebra in the upper and lower directions, and expand the boundary of 100 pixels in the left and right directions to form a middle block segmentation; The small-block segmentation is to divide the first eighteen vertebrae into three groups, randomly select one of the groups of vertebrae to generate a bounding box, and expand the height of half a vertebra up and down, and expand 100 pixels on the left and right to form a small-block segmentation.

4. The highly generalized spinal sagittal vertebra detection method based on multi-scale block data augmentation according to claim 2, characterized in that: The multi-scale segmentation of the first spinal sagittal X-ray data and the second spinal sagittal X-ray data to obtain training data containing global and local features includes: Performing multi-scale segmentation on the first spinal sagittal X-ray data and the second spinal sagittal X-ray data in an iterative manner; Each time an image in the data set is iterated, the first spinal sagittal X-ray data and the second spinal sagittal X-ray data are segmented according to a preset scale quantity ratio to obtain training data containing global and local features.

5. The highly generalized spinal sagittal vertebra detection method based on multi-scale block data augmentation according to claim 4, characterized in that: In each training iteration, a random number is generated for each image data, and segmentation of different scales is performed according to the random number, so that the same image generates different segmentation results in different rounds.

6. The highly generalized spinal sagittal vertebra detection method based on multi-scale block data augmentation according to claim 2, characterized in that: The method of using the first spinal sagittal X-ray data to generate second spinal sagittal X-ray data with a pseudo label at the target center includes: Using the first spinal sagittal X-ray data to train a vertebral body detection model to obtain a trained vertebral body detection model; The sagittal X-ray data of the spine at the target center is obtained, and the sagittal X-ray data of the spine is input into the trained vertebral body detection model to obtain data with pseudo labels.

7. The highly generalized spinal sagittal vertebra detection method based on multi-scale block data augmentation according to claim 6, characterized in that: After acquiring the sagittal X-ray data of the spine at the target center and inputting the sagittal X-ray data of the spine into the trained vertebral body detection model to obtain the data with pseudo labels, the method further includes: Performing consistency comparison between the pseudo-labeled data and the spinal sagittal X-ray data; The pseudo-labeled data that has passed the consistency comparison is determined as the second spinal sagittal X-ray data with the pseudo-label at the target center.

8. A highly generalized spinal sagittal vertebra detection device based on multi-scale block data amplification, characterized in that: include: An acquisition module, used for acquiring a sagittal X-ray image of the spine to be detected; A detection module, used for inputting the spinal sagittal X-ray image to be detected into the trained full spine detection model to obtain the detection result of the spinal sagittal X-ray image to be detected; The whole spine detection model is trained with multi-center training data, and the multi-center training data is obtained by a label block segmentation method.

9. An electronic device, characterized in that: include: A processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of a highly generalized spinal sagittal vertebra detection method based on multi-scale block data amplification as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the highly generalized spinal sagittal vertebra detection method based on multi-scale block data amplification as described in any one of claims 1-7 are implemented.

Images