Training method of centrum position recognition model and centrum position recognition method

Abstract

The invention provides a training method of a centrum position recognition model and a centrum position recognition method, and the training method of the centrum position recognition model comprises the steps: obtaining a plurality of groups of sample data; any group of sample data comprises a sample image and sample labeling information corresponding to a centrum in the sample image; taking each group of sample images as the input of a preset model, and taking the corresponding sample labeling information as the output of the preset model, so as to obtain a centrum position recognition model for detecting centrum information in any image through training; the preset model comprises a feature reinforcing structure, and the feature reinforcing structure is used for reinforcing feature information in the sample image; the feature enhancement structure comprises N channel attention mechanism modules and N space attention mechanism modules, through the method provided by the disclosure, centrum information in the image can be quickly obtained, and the user experience is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of image recognition, and in particular to a training method for a vertebral position recognition model and a vertebral position recognition method. Background Art

[0002] In the context of computer-assisted spinal surgery, it is very important to know the exact shape of individual vertebrae, e.g. for spinal biopsy, implant or pedicle screw insertion, etc. However, in most cases, it is not only required to accurately segment the vertebrae but also to locate and identify them.

[0003] However, due to the similar appearance of different vertebrae, various pathological types and imaging artifacts, it is impossible to accurately identify the vertebrae in the image. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the defect in the prior art that vertebrae in images cannot be recognized, and to provide a training method for a vertebrae position recognition model and a vertebrae position recognition method.

[0005] The present invention solves the above technical problems through the following technical solutions:

[0006] According to a first aspect of the present disclosure, a training method for a vertebral position recognition model is provided, the training method comprising:

[0007] Acquire several groups of sample data; wherein any group of the sample data includes a sample image and sample annotation information corresponding to a vertebra in the sample image;

[0008] Taking each group of the sample images as input and the corresponding sample annotation information as output, the preset model is trained to obtain a vertebral position recognition model for detecting vertebral information in any image;

[0009] Among them, the preset model includes a feature enhancement structure, which is used to enhance the feature information in the sample image; the feature enhancement structure includes a channel attention mechanism structure and a spatial attention mechanism structure connected in sequence, the channel attention mechanism structure includes N channel attention mechanism modules connected in parallel, and the spatial attention mechanism structure includes N spatial attention mechanism modules connected in parallel, N ≥ 2 and is a positive integer.

[0010] Preferably, the preset model further comprises a double-flow characteristic tower structure provided between the model input end and the characteristic enhancement structure;

[0011] The dual-stream feature tower structure is used to obtain feature information in the sample image, and input the feature information into the feature enhancement structure for feature enhancement;

[0012] and / or,

[0013] The channel attention mechanism structure includes two channel attention mechanism modules, and the spatial attention mechanism structure includes two spatial attention mechanism modules.

[0014] Preferably, the preset model further includes a CSPDarknet53 network (a backbone network of a convolutional neural network (CNN)) with an added SPP (Spatial Pyramid Pooling) module; the CSPDarknet53 network is used to extract feature information from the sample image;

[0015] and / or,

[0016] The preset model also includes a plurality of decoding units, and different decoding units have different input sizes; the decoding units are used to output vertebral information in the sample image.

[0017] According to a second aspect of the present disclosure, a training system for a vertebral position recognition model is provided, the training system comprising:

[0018] A sample acquisition module, used to acquire several groups of sample data; wherein any group of the sample data includes a sample image and sample annotation information corresponding to a vertebra in the sample image;

[0019] A model training module, used to take each group of sample images as input and the corresponding sample annotation information as output, to train a preset model to obtain vertebral information for detecting any image;

[0020] Among them, the preset model includes a feature enhancement structure, which is used to enhance the feature information in the sample image; the feature enhancement structure includes a channel attention mechanism structure and a spatial attention mechanism structure connected in sequence, the channel attention mechanism structure includes N channel attention mechanism modules connected in parallel, and the spatial attention mechanism structure includes N spatial attention mechanism modules connected in parallel, N ≥ 2 and is a positive integer.

[0021] Preferably, the preset model further comprises a double-flow characteristic tower structure provided between the model input end and the characteristic enhancement structure;

[0022] The dual-stream feature tower structure is used to obtain feature information in the sample image, and input the feature information into the feature enhancement structure for feature enhancement;

[0023] and / or,

[0024] The channel attention mechanism structure includes two channel attention mechanism modules, and the spatial attention mechanism structure includes two spatial attention mechanism modules.

[0025] Preferably, the preset model further includes a CSPDarknet53 network with an SPP module added thereto; the CSPDarknet53 network is used to extract feature information from the sample image;

[0026] and / or,

[0027] The preset model also includes a plurality of decoding units, and different decoding units have different input sizes; the decoding units are used to output vertebral information in the sample image.

[0028] According to a third aspect of the present disclosure, a vertebral position recognition method is provided, the vertebral position recognition method comprising:

[0029] Acquire a target vertebral image;

[0030] The target vertebral image is input into the vertebral position recognition model trained by the vertebral position recognition model training method described in the first aspect of the present disclosure to obtain vertebral information of different vertebrae in the target vertebral image.

[0031] Preferably, the identification method further comprises:

[0032] Before the step of inputting the target vertebral body image into the vertebral body position recognition model, the recognition method further comprises:

[0033] Based on the target vertebral image, determining whether a first target vertebral body exists in the target vertebral body image;

[0034] In response to the presence of the first target vertebra in the target vertebra image, the target vertebra image is identified in a first identification manner; wherein the first identification manner is determined based on the first target vertebra;

[0035] In response to the first target vertebra not existing in the target vertebra image, determining whether a second target vertebra exists in the target vertebra image;

[0036] In response to the presence of the second target vertebra in the target vertebra image, the target vertebra image is identified in a second identification manner; wherein the second identification manner is determined based on the second target vertebra;

[0037] In response to the second target vertebra not existing in the target vertebra image, the target vertebra image is input into the vertebra position recognition model for vertebra recognition.

[0038] Preferably, the first target vertebra is the T12 vertebra, and the second target vertebra is the coccyx;

[0039] The first identification method includes:

[0040] Taking the T12 vertebra as the starting point, successively from bottom to top and from top to bottom, respectively, identify the vertebrae in the target vertebrae image;

[0041] The second identification method includes:

[0042] Starting from the coccyx, the vertebrae in the target vertebrae image are identified sequentially from bottom to top.

[0043] According to a fourth aspect of the present disclosure, a vertebral body recognition system is provided, the vertebral body recognition system comprising:

[0044] A data acquisition module, used for acquiring a target vertebral body image;

[0045] The vertebral information acquisition module is used to input the target vertebral image into the vertebral position recognition model trained by the vertebral position recognition model training method described in the first aspect of the present disclosure to obtain vertebral information of different vertebrae in the target vertebral image.

[0046] Preferably, the vertebral body recognition system further comprises a pre-recognition module, the pre-recognition module being used to determine whether there is a first target vertebral body in the target vertebral body image based on the target vertebral body image before inputting the target vertebral body image into the vertebral body position recognition model;

[0047] In response to the presence of the first target vertebra in the target vertebra image, the target vertebra image is identified in a first identification manner; wherein the first identification manner is determined based on the first target vertebra;

[0048] In response to the first target vertebra not existing in the target vertebra image, determining whether a second target vertebra exists in the target vertebra image;

[0049] In response to the presence of the second target vertebra in the target vertebra image, the target vertebra image is identified in a second identification manner; wherein the second identification manner is determined based on the second target vertebra;

[0050] In response to the second target vertebra not existing in the target vertebra image, the target vertebra image is input into the vertebra position recognition model for vertebra recognition.

[0051] Preferably, the first target vertebra is the T12 vertebra, and the second target vertebra is the coccyx;

[0052] The first identification method includes:

[0053] Taking the T12 vertebra as the starting point, successively from bottom to top and from top to bottom, respectively, identify the vertebrae in the target vertebrae image;

[0054] The second identification method includes:

[0055] Starting from the coccyx, the vertebrae in the target vertebrae image are identified sequentially from bottom to top.

[0056] According to a fifth aspect of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, wherein when the processor executes the computer program, the training method of the vertebral position recognition model described in the first aspect of the present disclosure and / or the vertebral position recognition method described in the third aspect of the present disclosure is implemented.

[0057] According to a sixth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that when the computer program is executed by a processor, the training method of the vertebral position recognition model described in the first aspect of the present disclosure and / or the vertebral position recognition method described in the third aspect of the present disclosure is implemented.

[0058] According to a seventh aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the training method for the vertebral position recognition model described in the first aspect of the present disclosure and / or the vertebral position recognition method described in the third aspect of the present disclosure is implemented.

[0059] The positive and progressive effects of the present invention are:

[0060] In the training method of the vertebral position recognition model provided by the present invention, by introducing a feature enhancement structure, the trained vertebral position recognition model can more accurately distinguish the vertebra from other backgrounds or objects, reduce dependence on unimportant or redundant features, thereby reducing the negative impact of these features on the final recognition results and improving the recognition accuracy.

[0061] The vertebral position recognition method provided by the present disclosure not only improves work efficiency through automated, rapid and accurate recognition, but also reduces the error rate by reducing manual operations, thereby providing a strong foundation for subsequent pathological analysis and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A schematic diagram of a flow chart of a method for training a vertebral position recognition model provided in Embodiment 1 of the present disclosure;

[0063] Figure 2It is a structural schematic diagram of the feature enhancement structure provided in Example 1 of the present disclosure;

[0064] Figure 3 This is a schematic diagram of the structure of the CSPDarknet53 network provided in Example 1 of the present disclosure;

[0065] Figure 4 This is a schematic diagram of the structure of the vertebral position recognition model provided in Example 1 of the present disclosure;

[0066] Figure 5 A schematic diagram of the structure of a training system for a vertebral position recognition model provided in Embodiment 2 of the present disclosure;

[0067] Figure 6 A schematic diagram of the flow of the vertebral position identification method provided in Embodiment 3 of the present disclosure;

[0068] Figure 7 For X-rays that need to identify vertebrae;

[0069] Figure 8 For X-rays that have been identified;

[0070] Fig. 9 This is a flowchart of a specific implementation of the vertebral position identification method in Embodiment 2 of the present disclosure;

[0071] Fig.10 It is a structural schematic diagram of a vertebral body recognition system provided in Embodiment 4 of the present disclosure;

[0072] Fig.11 This is a schematic diagram of the structure of the electronic device provided in Example 5 of the present disclosure. DETAILED DESCRIPTION

[0073] The present disclosure is further described below by way of examples, but the present disclosure is not limited to the scope of the examples.

[0074] Prefixes such as "first" and "second" are used in the embodiments of the present disclosure only to distinguish different description objects, and have no limiting effect on the position, order, priority, quantity or content of the described objects. The use of prefixes such as ordinal numbers to distinguish description objects in the embodiments of the present disclosure does not constitute a limitation on the described objects. For the statement of the described objects, please refer to the description in the context of the claims or embodiments, and no unnecessary limitation should be constituted due to the use of such prefixes. In addition, in the description of the present embodiment, unless otherwise specified, the meaning of "plurality" is two or more.

[0075] In the embodiments of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0076] Example 1

[0077] like Figure 1 As shown, in this embodiment, a training method for a vertebral position recognition model is provided, and the training method includes:

[0078] S11: Acquire several groups of sample data; wherein any group of sample data includes a sample image and sample annotation information of a vertebra in the corresponding sample image.

[0079] S12: Taking each group of sample images as input and the corresponding sample annotation information as output, the preset model is trained to obtain a vertebral position recognition model for detecting vertebral information in any image.

[0080] Among them, the preset model includes a feature enhancement structure, which is used to enhance the feature information in the sample image; the feature enhancement structure includes a channel attention mechanism structure and a spatial attention mechanism structure connected in sequence, the channel attention mechanism structure includes N channel attention mechanism modules connected in parallel, and the spatial attention mechanism structure includes N spatial attention mechanism modules connected in parallel, N ≥ 2 and is a positive integer.

[0081] In the training method of the vertebral position recognition model provided by the present invention, by introducing a feature enhancement structure, the trained vertebral position recognition model can more accurately distinguish the vertebra from other backgrounds or objects, reduce dependence on unimportant or redundant features, thereby reducing the negative impact of these features on the final recognition results and improving the recognition accuracy.

[0082] In one embodiment, if Figure 2 As shown in the figure, N is 2, that is, the feature enhancement structure includes two channel attention mechanisms and two spatial attention mechanism modules. First, the feature map is input to perform parallel multiplication and fusion of the results of the two channel attention mechanism modules, and then the output features are multiplied and fused in parallel with the results of the two spatial attention mechanism modules, and finally the final feature map is output.

[0083] By adding a self-developed multi-attention mechanism hybrid module structure and adding multiple attention mechanism fusion modules to the multi-scale network structure, feature enhancement is achieved on the feature map extracted by the image feature backbone network.

[0084] The preset model in this embodiment also includes a dual-stream feature tower structure arranged between the model input end and the feature enhancement structure; the dual-stream feature tower structure is used to obtain feature information in the sample image and input the feature information into the feature enhancement structure for feature enhancement.

[0085] By setting up a dual-stream feature tower structure, the generated functions contain multiple scales and multi-semantic level information, which solves the multi-scale problem in object detection. Through simple changes in network connections, the performance of object detection is greatly improved without increasing the computational complexity of the original model.

[0086] like Figure 3 As shown, the preset model in this embodiment also includes a CSPDarknet53 network with an SPP module added; the CSPDarknet53 network is used to extract feature information from the sample image.

[0087] By adding the SPP module, the CSPDarknet53 network is provided with greater flexibility and stronger feature extraction capabilities while maintaining efficient reasoning speed; by adopting the CSPDarknet53 network with the SPP module added, feature propagation and feature reuse are realized, the number of parameters and the computing power consumed by the model are reduced, while ensuring the real-time performance and effect of the network.

[0088] The preset model in this embodiment also includes a plurality of decoding units, and different decoding units have different input sizes, thereby ensuring the generalization of the model for detecting targets of different sizes; the decoding unit is used to output vertebral information in the sample image.

[0089] In this embodiment, the anchor-based multi-scale detection scheme allocates three prior anchor boxes of different proportions on each feature map grid in the multi-scale feature map. The detection head predicts the position deviation and the scaling of the height and width, as well as the probability corresponding to each category and the confidence of whether the target is included, so as to realize the recognition of vertebrae in the target image.

[0090] In addition, the loss function in this embodiment is L det ; where L det is the loss of the detection task, which is the weighted sum of classification, target confidence and target coordinate frame regression. The formula is as follows:

[0091] L det =α 1 *L class +α 2 *L obj +α 3 *L box

[0092] L class and L objBoth are focal loss (a loss function specially designed to solve the problem of class imbalance in target detection), which is used for classification and confidence calculation respectively. This loss not only reduces the weight of the background class, but also reduces the weight of easy positive / negative (easy to distinguish positive samples / negative samples);

[0093] L box For L CIOU The loss function is used to calculate the skeleton position coordinates. The loss takes into account the distance, overlap, scale and aspect ratio similarity between the predicted box and the groundtruth.

[0094] The following is a specific implementation method to illustrate the implementation principle of the vertebral position recognition model training method in the embodiment of the present disclosure:

[0095] The preset model in this embodiment is YOLO-XRAY (an end-to-end lumbar lateral fracture recognition network), which can realize bone position detection. YOLO-XRAY consists of an encoder for feature extraction and a decoder for processing position detection tasks.

[0096] like Figure 4 As shown, the deep learning network structure of the YOLO-XRAY disclosed in the present invention is shown. It is a target detection model for X-ray image analysis. YOLO-XRAY is based on the Darknet53 architecture, but the fully connected (FC) layer is removed to adapt to the characteristics of the convolutional neural network (CNN).

[0097] Input layer: The image is first input into the network.

[0098] Darknet53withoutFClayer: The first part of the network is the Darknet53 structure, which is a deep residual network and contains multiple residual units (Resunit). The Darknet53 structure in this embodiment is divided into several stages:

[0099] CBL: ConvolutionalLayer, BatchNormalization, and LeakyReLU activation function.

[0100] Res1, Res2, Res4, Res8: These are residual units, each of which contains two CBL modules and adds the input to the output through an Add operation to form a residual connection.

[0101] CBL*5: After the Darknet53 structure, there are five CBL modules, which are used to further extract features.

[0102] Self-developed attention mechanism: After CBL*5, the network introduced a self-developed attention mechanism to enhance the network's attention to important features.

[0103] The self-developed attention mechanism includes two channel attention mechanisms and two spatial attention mechanism modules. First, the input features are multiplied and fused in parallel with the results of the two channel attention mechanism modules. Then the output features are multiplied and fused in parallel with the results of the two spatial attention mechanism modules. Finally, the final feature map is output.

[0104] By adding a self-developed multi-attention mechanism hybrid module structure and adding multiple attention mechanism fusion modules to the multi-scale network structure, feature enhancement is achieved on the feature map extracted by the image feature backbone network.

[0105] Detection Head: The network has three detection heads, each of which contains CBL and convolutional layers (Conv) for object detection at different scales.

[0106] Feature fusion: Before the detection head, the network combines feature maps of different scales through upsampling (UP) and feature fusion (Concate) operations to improve the detection accuracy.

[0107] Segmentation head: At the bottom, the network also has a segmentation head, which contains multiple CBL, upsampling (UP) and convolutional layers (Conv) for the image segmentation task.

[0108] In the training method of the vertebral position recognition model provided by the present disclosure, by introducing a feature enhancement structure, the trained vertebral position recognition model can more accurately distinguish the vertebra from other backgrounds or objects, reduce the reliance on unimportant or redundant features, thereby reducing the negative impact of these features on the final recognition results and improving the recognition accuracy. Through multi-scale feature fusion and self-developed attention mechanism, the performance of target detection and segmentation in X-ray images is improved, thereby improving the recognition of vertebrae in X-ray images.

[0109] Example 2

[0110] like Figure 5 As shown, in this embodiment, a training system for a vertebral position recognition model is provided, characterized in that the training system comprises:

[0111] The sample acquisition module 100 is used to acquire several groups of sample data; wherein any group of the sample data includes a sample image and sample annotation information corresponding to a vertebra in the sample image;

[0112] A model training module 200, for taking each group of sample images as input and the corresponding sample annotation information as output, to train a preset model, so as to obtain vertebral information for detecting any image;

[0113] Among them, the preset model includes a feature enhancement structure, which is used to enhance the feature information in the sample image; the feature enhancement structure includes a channel attention mechanism structure and a spatial attention mechanism structure connected in sequence, the channel attention mechanism structure includes N channel attention mechanism modules connected in parallel, and the spatial attention mechanism structure includes N spatial attention mechanism modules connected in parallel, N≥2 and is a positive integer.

[0114] The preset model also includes a dual-stream feature tower structure disposed between the model input end and the feature enhancement structure; the dual-stream feature tower structure is used to obtain feature information in the sample image and input the feature information into the feature enhancement structure for feature enhancement;

[0115] The channel attention mechanism structure in this embodiment includes two channel attention mechanism modules, and the spatial attention mechanism structure includes two spatial attention mechanism modules.

[0116] The preset model in this embodiment also includes a CSPDarknet53 network with an SPP module added; the CSPDarknet53 network is used to extract feature information from the sample image;

[0117] The preset model also includes a plurality of decoding units, and different decoding units have different input sizes; the decoding units are used to output vertebral information in the sample image.

[0118] As for the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The system embodiment described above is only illustrative, in which the units described as separate components may or may not be physically separated, and the components as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution.

[0119] In the training system of the vertebral position recognition model provided by the present invention, by introducing a feature enhancement structure, the trained vertebral position recognition model can more accurately distinguish the vertebra from other backgrounds or objects, reduce dependence on unimportant or redundant features, thereby reducing the negative impact of these features on the final recognition results and improving the recognition accuracy.

[0120] Example 3

[0121] like Figure 6 As shown, in this embodiment, a vertebral position recognition method is provided, and the vertebral position recognition method includes:

[0122] S61: Acquire a target vertebral body image;

[0123] S62: Inputting the target vertebral body image into the vertebral body position recognition model trained by the vertebral body position recognition model training method to obtain vertebral body information of different vertebrae in the target vertebral body image.

[0124] The target vertebral image is as follows: Figure 7 As shown in the figure, the vertebral information of different vertebrae in the target vertebral image is obtained as follows Figure 8 shown.

[0125] The vertebral position recognition method provided by the present disclosure not only improves work efficiency through automated, rapid and accurate recognition, but also reduces the error rate by reducing manual operations, thereby providing a strong foundation for subsequent pathological analysis and treatment.

[0126] The identification method in this embodiment also includes:

[0127] Before the step of inputting the target vertebral body image into the vertebral body position recognition model, the recognition method further comprises:

[0128] Based on the target vertebral image, determining whether a first target vertebral body exists in the target vertebral body image;

[0129] In response to the presence of a first target vertebra in the target vertebra image, the target vertebra image is identified in a first identification manner; wherein the first identification manner is determined based on the first target vertebra;

[0130] In response to the first target vertebra not existing in the target vertebra image, determining whether a second target vertebra exists in the target vertebra image;

[0131] In response to the presence of a second target vertebra in the target vertebra image, the target vertebra image is identified in a second identification manner; wherein the second identification manner is determined based on the second target vertebra;

[0132] In response to the absence of the second target vertebra in the target vertebra image, the target vertebra image is input into the vertebra position recognition model for vertebra recognition.

[0133] In this embodiment, the first target vertebra is the T12 vertebra, and the second target vertebra is the coccyx.

[0134] The first identification method includes:

[0135] Starting from the T12 vertebra, identify the vertebrae in the target vertebra image from bottom to top and from top to bottom;

[0136] The second identification method includes:

[0137] Starting from the coccyx, the vertebrae in the target vertebrae image are identified from bottom to top.

[0138] The vertebra in the target image can be quickly identified by using the first target vertebra and the second target vertebra.

[0139] The following is a specific implementation method to illustrate the implementation principle of the vertebral position recognition method in the embodiment of the present disclosure:

[0140] like Fig. 9 As shown, the position recognition logic starts, inputs the X-ray - thoracic and lumbar lateral view, and obtains all vertebrae (including coccyx) in the image through the image detection algorithm. The obtained result is as follows Figure 7 shown.

[0141] Identify the coccyx and the last vertebra with ribs T12 in the image (the last one from top to bottom); determine whether there is vertebra T12 in the image. If there is T12, take T12 as the starting point and mark the vertebra positions T11-T4 from bottom to top (based on the actual number of vertebrae detected), and take T12 as the starting point and mark the vertebra positions L1-L4 from top to bottom (based on the actual number of vertebrae detected), and complete the image recognition. The recognition results are as follows: Figure 8 As shown;

[0142] If T12 does not exist, it is further determined whether there is a coccyx in the image. If there is a coccyx, the vertebral positions are marked from bottom to top as L5-L1 (based on the actual number of vertebrae detected) according to the position of the coccyx, and the image is recognized. The recognition result is as follows: Figure 8 shown.

[0143] If the coccyx does not exist, the image is directly input into the vertebral position recognition model to recognize the position result through the algorithm and directly output it. Figure 8 shown.

[0144] The vertebral position recognition method provided by the present disclosure can quickly recognize different vertebrae in an image, providing a strong foundation for subsequent pathological analysis and treatment.

[0145] Example 4

[0146] like Fig.10 As shown, in this embodiment, a vertebral body recognition system is provided, and the vertebral body recognition system includes:

[0147] A data acquisition module 300 is used to acquire a target vertebral body image;

[0148] The vertebral information acquisition module 400 is used to input the target vertebral image into the vertebral position recognition model trained by the vertebral position recognition model training method to obtain the vertebral information of different vertebrae in the target vertebral image.

[0149] The vertebral body recognition system in this embodiment further includes a pre-recognition module, which is used to determine whether there is a first target vertebral body in the target vertebral body image based on the target vertebral body image before inputting the target vertebral body image into the vertebral body position recognition model;

[0150] In response to the presence of a first target vertebra in the target vertebra image, the target vertebra image is identified in a first identification manner; wherein the first identification manner is determined based on the first target vertebra;

[0151] In response to the first target vertebra not existing in the target vertebra image, determining whether a second target vertebra exists in the target vertebra image;

[0152] In response to the presence of a second target vertebra in the target vertebra image, the target vertebra image is identified in a second identification manner; wherein the second identification manner is determined based on the second target vertebra;

[0153] In response to the absence of the second target vertebra in the target vertebra image, the target vertebra image is input into the vertebra position recognition model for vertebra recognition.

[0154] In this embodiment, the first target vertebra is the T12 vertebra, and the second target vertebra is the coccyx;

[0155] The first identification method includes:

[0156] Starting from the T12 vertebra, identify the vertebrae in the target vertebra image from bottom to top and from top to bottom;

[0157] The second identification method includes:

[0158] Starting from the coccyx, the vertebrae in the target vertebrae image are identified from bottom to top.

[0159] As for the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The system embodiment described above is only illustrative, wherein the units described as separate components may or may not be physically separated, and the components as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the disclosed solution.

[0160] The vertebral body recognition system provided by the present disclosure not only improves work efficiency through automated, rapid and accurate recognition, but also reduces the error rate by reducing manual operations, thereby providing a strong foundation for subsequent pathological analysis and treatment.

[0161] Example 5

[0162] Fig.11 This is a schematic diagram of the structure of an electronic device provided in Embodiment 5 of the present disclosure. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the method in the above embodiment is implemented when the processor executes the program. Fig.11 The electronic device 30 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0163] like Fig.11 As shown, the electronic device 30 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 30 may include, but are not limited to: at least one processor 31, at least one memory 32, and a bus 33 connecting different system components (including the memory 32 and the processor 31).

[0164] The bus 33 includes a data bus, an address bus, and a control bus.

[0165] The memory 32 may include a volatile memory, such as a random access memory (RAM) 321 and / or a cache memory 322 , and may further include a read-only memory (ROM) 323 .

[0166] The memory 32 may also include a program / utility base 325 having a set (at least one) of program modules 324, such program modules 324 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include an implementation of a network environment.

[0167] The processor 31 executes various functional applications and data processing by running the computer programs stored in the memory 32, such as the method in the above-mentioned embodiment of the present disclosure.

[0168] The electronic device 30 may also communicate with one or more external devices 34 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 35. Furthermore, the model generating device 30 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 36. Fig.11As shown, the network adapter 36 communicates with other modules of the model-generated device 30 via the bus 33. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the model-generated device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

[0169] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0170] Example 6

[0171] The embodiments of the present disclosure also provide a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method provided in the above-mentioned embodiment 1 or embodiment 3 is implemented.

[0172] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.

[0173] Example 7

[0174] The embodiments of the present disclosure further provide a computer program product, including a computer program, which implements the method provided in the above-mentioned embodiment 1 or embodiment 3 when executed by a processor.

[0175] Among them, the program code for executing the computer program product of the present disclosure can be written in any combination of one or more programming languages, and the program code can be executed completely on the user device, partially on the user device, as an independent software package, partially on the user device and partially on a remote device, or completely on the remote device.

[0176] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that this is only for illustration and the protection scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. A training method for a vertebral position recognition model, characterized in that: The training method comprises: Acquire several groups of sample data; wherein any group of the sample data includes a sample image and sample annotation information corresponding to a vertebra in the sample image; Taking each group of the sample images as input and the corresponding sample annotation information as output, the preset model is trained to obtain a vertebral position recognition model for detecting vertebral information in any image; Among them, the preset model includes a feature enhancement structure, which is used to enhance the feature information in the sample image; the feature enhancement structure includes a channel attention mechanism structure and a spatial attention mechanism structure connected in sequence, the channel attention mechanism structure includes N channel attention mechanism modules connected in parallel, and the spatial attention mechanism structure includes N spatial attention mechanism modules connected in parallel, N ≥ 2 and is a positive integer.

2. The training method of the vertebral position recognition model according to claim 1, characterized in that: The preset model also includes a double-flow characteristic tower structure disposed between the model input end and the characteristic enhancement structure; The dual-stream feature tower structure is used to obtain feature information in the sample image, and input the feature information into the feature enhancement structure for feature enhancement; and / or, The channel attention mechanism structure includes two channel attention mechanism modules, and the spatial attention mechanism structure includes two spatial attention mechanism modules.

3. The training method of the vertebral position recognition model according to claim 1 or 2, characterized in that: The preset model also includes a CSPDarknet53 network with an SPP module added; the CSPDarknet53 network is used to extract feature information from the sample image; and / or, The preset model also includes a plurality of decoding units, and different decoding units have different input sizes; the decoding units are used to output vertebral information in the sample image.

4. A training system for a vertebral position recognition model, characterized in that: The training system comprises: A sample acquisition module, used to acquire several groups of sample data; wherein any group of the sample data includes a sample image and sample annotation information corresponding to a vertebra in the sample image; A model training module, used to take each group of sample images as input and the corresponding sample annotation information as output, to train a preset model to obtain vertebral information for detecting any image; Among them, the preset model includes a feature enhancement structure, which is used to enhance the feature information in the sample image; the feature enhancement structure includes a channel attention mechanism structure and a spatial attention mechanism structure connected in sequence, the channel attention mechanism structure includes N channel attention mechanism modules connected in parallel, and the spatial attention mechanism structure includes N spatial attention mechanism modules connected in parallel, N ≥ 2 and is a positive integer.

5. A method for identifying vertebral position, characterized in that: The vertebral position recognition method comprises: Acquire a target vertebral image; The target vertebral image is input into a vertebral position recognition model trained by the vertebral position recognition model training method according to any one of claims 1 to 3, so as to obtain vertebral information of different vertebrae in the target vertebral image.

6. The vertebral position recognition method according to claim 5, characterized in that: The identification method further comprises: Before the step of inputting the target vertebral body image into the vertebral body position recognition model, the recognition method further comprises: Based on the target vertebral image, determining whether a first target vertebral body exists in the target vertebral body image; In response to the presence of the first target vertebra in the target vertebra image, the target vertebra image is identified in a first identification manner; wherein the first identification manner is determined based on the first target vertebra; In response to the first target vertebra not existing in the target vertebra image, determining whether a second target vertebra exists in the target vertebra image; In response to the presence of the second target vertebra in the target vertebra image, the target vertebra image is identified in a second identification manner; wherein the second identification manner is determined based on the second target vertebra; In response to the second target vertebra not existing in the target vertebra image, the target vertebra image is input into the vertebra position recognition model for vertebra recognition.

7. The vertebral position recognition method according to claim 6, characterized in that: The first target vertebra is the T12 vertebra, and the second target vertebra is the coccyx; The first identification method includes: Taking the T12 vertebra as the starting point, successively from bottom to top and from top to bottom, respectively, identify the vertebrae in the target vertebrae image; The second identification method includes: Starting from the coccyx, the vertebrae in the target vertebrae image are identified from bottom to top.

8. A vertebral body recognition system, characterized in that: The vertebral body recognition system comprises: A data acquisition module, used for acquiring a target vertebral body image; The vertebral information acquisition module is used to input the target vertebral image into the vertebral position recognition model trained by the vertebral position recognition model training method according to any one of claims 1 to 3 to obtain vertebral information of different vertebrae in the target vertebral image.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, characterized in that: When the processor executes the computer program, the processor implements the training method of the vertebral position recognition model described in any one of claims 1-3 and / or the vertebral position recognition method described in any one of claims 5-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the training method for a vertebral position recognition model described in any one of claims 1 to 3 and / or the vertebral position recognition method described in any one of claims 5 to 7 are implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the training method for a vertebral position recognition model described in any one of claims 1 to 3 and / or the vertebral position recognition method described in any one of claims 5 to 7 are implemented.

Images