Intervertebral disc degeneration identification method and device based on stepwise aggregation
By adopting a stepwise polymerization structure in the disc degeneration recognition method and polymerizing feature maps of different sizes, the problem of low accuracy of disc degeneration recognition in the prior art is solved, and more accurate identification and effective positioning of lumbar spine sequence numbers are achieved.
Patent Information
- Application Number
- CN202411396646.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-09
- Publication Date
- 2025-02-18
AI Technical Summary
The current recognition accuracy of intervertebral disc degeneration is low, and it is impossible to effectively locate the lumbar spine sequence number of lumbar spine abnormalities.
Using the degeneration recognition method based on stepwise polymerization, the degeneration recognition model of intervertebral discs is obtained by acquiring MRI medical images and inputting a pre-trained degeneration recognition model. Using multiple feature extraction structures and stepwise polymerization structures set in parallel, the output feature maps of different sizes are aggregated to improve the model. accuracy of state fusion.
Through the use of stepwise polymerization structures, efficient fusion of different size characteristics is achieved, the accuracy of identification of intervertebral disc degeneration is improved, and the lumbar spine number of lumbar spine abnormalities can be positioned at the same time.
Smart Images
Figure CN119478597A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical image processing technology, and more specifically, to a method and apparatus for identifying intervertebral disc degeneration based on stepwise aggregation. Background Technology
[0002] Intervertebral disc degeneration (IDD) refers to the degenerative changes that occur in the intervertebral disc due to age, mechanical load, genetic factors, or other causes. It is one of the common spinal diseases and is usually closely related to related diseases such as low back pain, neck pain, herniated discs, and spinal stenosis.
[0003] However, current methods for identifying intervertebral disc degeneration use simple deep learning models such as CNNs, which have low accuracy. Summary of the Invention
[0004] The problem addressed by this application is the low accuracy rate of current intervertebral disc degeneration identification.
[0005] To address the aforementioned problems, the first aspect of this application provides a method for identifying intervertebral disc degeneration based on stepwise aggregation, comprising:
[0006] Acquire MRI medical images, wherein the MRI medical images are sagittal spinal medical images of any object;
[0007] MRI medical images are input into a pre-trained degeneration recognition model to obtain degeneration recognition results;
[0008] The degradation recognition model includes multiple feature extraction structures set in parallel and a stepwise aggregation structure. The output feature maps of different feature extraction structures have different sizes, and the stepwise aggregation structure is used to aggregate output feature maps of different sizes.
[0009] A second aspect of this application provides a stepwise aggregation-based intervertebral disc degeneration identification device, comprising:
[0010] An image acquisition module is used to acquire MRI medical images, wherein the MRI medical images are sagittal spinal medical images of any object;
[0011] The degradation recognition module is used to input MRI medical images into a pre-trained degradation recognition model to obtain degradation recognition results. The degradation recognition model includes multiple feature extraction structures set in parallel and a stepwise aggregation structure. The output feature maps of different feature extraction structures have different sizes, and the stepwise aggregation structure is used to aggregate output feature maps of different sizes.
[0012] A third aspect of this application provides an electronic device comprising: a memory and a processor;
[0013] The memory is used to store programs;
[0014] The processor, coupled to the memory, is used to execute the program for:
[0015] Acquire MRI medical images, wherein the MRI medical images are sagittal spinal medical images of any object;
[0016] MRI medical images are input into a pre-trained degeneration recognition model to obtain degeneration recognition results;
[0017] The degradation recognition model includes multiple feature extraction structures set in parallel and a stepwise aggregation structure. The output feature maps of different feature extraction structures have different sizes, and the stepwise aggregation structure is used to aggregate output feature maps of different sizes.
[0018] The fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the above-described stepwise aggregation-based intervertebral disc degeneration identification method.
[0019] In this application, feature extraction structures of different sizes are set to extract features of different sizes; through stepwise aggregation structures, efficient fusion of features of different sizes is achieved, thereby enabling more accurate identification of intervertebral disc degeneration. Attached Figure Description
[0020] Figure 1 This is a model architecture diagram of the intervertebral disc degeneration identification method according to an embodiment of this application;
[0021] Figure 2 This is a flowchart of the intervertebral disc degeneration identification method according to an embodiment of this application;
[0022] Figure 3 This is a diagram illustrating the stepwise aggregation architecture of the intervertebral disc degeneration identification method according to an embodiment of this application.
[0023] Figure 4 This is an architecture diagram of the attention extraction method for intervertebral disc degeneration identification according to an embodiment of this application;
[0024] Figure 5 This is a specially extracted architecture diagram of the intervertebral disc degeneration identification method according to the embodiments of this application;
[0025] Figure 6 This is a flowchart of the model training for the intervertebral disc degeneration identification method according to an embodiment of this application;
[0026] Figure 7 This is an architecture diagram of the detection module of the intervertebral disc degeneration identification method according to an embodiment of this application;
[0027] Figure 8 This is a structural block diagram of an intervertebral disc degeneration identification device according to an embodiment of this application;
[0028] Figure 9 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation
[0029] To make the above-mentioned objects, features, and advantages of this application more apparent and understandable, specific embodiments of this application will be described in detail below with reference to the accompanying drawings. Although exemplary embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of this application and to fully convey the scope of this application to those skilled in the art.
[0030] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains.
[0031] Images of the lumbar spine reflect the structure of the human lower back. When abnormalities occur in the lumbar spine, symptoms such as lower back pain may appear. Correspondingly, the images will show an incorrect shape in a certain part of the lumbar spine. For example, lumbar degeneration leading to joint hyperplasia and ligament hypertrophy can compress the normal volume of the spinal canal. This morphological change is called spinal stenosis, which is a type of lumbar spine abnormality.
[0032] However, the current identification of lumbar spine abnormalities is done through cross-sectional images of MRI images. Although this method can accurately identify lumbar spine abnormalities, it cannot determine the lumbar spine number of the abnormality. It is also necessary to combine the images of other lumbar spines for analysis and localization, which is quite cumbersome.
[0033] To address the aforementioned issues, this application provides a novel intervertebral disc degeneration identification scheme based on stepwise aggregation. This scheme improves the accuracy of modality fusion by cross-fusion of coarsely extracted feature maps, thus solving the problem that current methods cannot simultaneously locate the abnormal lumbar vertebrae during identification.
[0034] This application provides a method for identifying intervertebral disc degeneration based on stepwise aggregation. The specific scheme of this method is as follows: Figures 1-7 As shown, this method can be executed by a stepwise aggregation-based intervertebral disc degeneration identification device, which can be integrated into electronic devices such as computers, servers, computer clusters, and data centers. Combined with... Figure 1 , Figure 2 The diagram shows a flowchart of a stepwise aggregation-based intervertebral disc degeneration identification method according to an embodiment of this application; wherein the stepwise aggregation-based intervertebral disc degeneration identification method includes:
[0035] S101, acquire an MRI medical image, wherein the MRI medical image is a sagittal spinal medical image of any object;
[0036] S102, Input the MRI medical image into the pre-trained degeneration recognition model to obtain the degeneration recognition result;
[0037] The degradation recognition model includes multiple feature extraction structures set in parallel and a stepwise aggregation structure. The output feature maps of different feature extraction structures have different sizes, and the stepwise aggregation structure is used to aggregate output feature maps of different sizes.
[0038] In one specific implementation, combined with Figure 1 As shown,
[0039] The process of inputting MRI medical images into a pre-trained degeneration recognition model to obtain degeneration recognition results includes:
[0040] MRI medical images are input into multiple feature extraction structures set in parallel to obtain multiple output feature maps; the feature extraction structures include global extraction structures and local extraction structures, and the output feature maps include global feature maps and local feature maps.
[0041] By inputting global and local feature maps of the same size into the stepwise aggregation structure, an aggregated feature map of the corresponding size is obtained.
[0042] By inputting aggregated feature maps of different sizes into the upsampling structure, an upsampling feature map is obtained.
[0043] The upsampled feature map is input into a special extraction module and a detection module to obtain the degradation recognition result.
[0044] In one specific implementation,
[0045] The global extraction structure is an improved Transformer network structure, and the local extraction structure is a ConvNext network structure.
[0046] In one specific implementation, combined with Figure 3 As shown,
[0047] The step of inputting global and local feature maps of the same size into a progressive aggregation structure to obtain an aggregated feature map of the corresponding size includes:
[0048] The global feature map is input into the attention extraction module to obtain the global extracted feature map;
[0049] The local feature maps are sequentially input into the special extraction module and the attention extraction module to obtain the local extracted feature maps;
[0050] The global extracted feature map and the local extracted feature map are fused to obtain the aggregated feature map.
[0051] In one specific implementation, combined with Figure 4 As shown,
[0052] The step of inputting the global feature map into the attention extraction module to obtain the global extracted feature map includes:
[0053] The global feature map is converted into a horizontal feature map, and then convolution and activation processing are performed to obtain the horizontal extraction map.
[0054] The global feature map is converted into a vertical feature map and then convolutional to obtain the vertical extraction map.
[0055] The global feature map is alternately processed by convolution and activation to obtain a local weight map;
[0056] The horizontal extraction map and the numerical extraction map are multiplied together and then fused with the local weight map to obtain the global extracted feature map.
[0057] In one specific implementation, combined with Figure 6 As shown,
[0058] The training process of the degradation recognition model includes:
[0059] S201, Obtain a sample medical image, wherein the sample medical image is labeled with the recognition result;
[0060] S202, input the sample medical image into multiple feature extraction structures set in parallel to obtain global sample feature maps and local sample feature maps of multiple sizes;
[0061] S203, input global sample feature maps and local sample feature maps of the same size into the stepwise aggregation structure to obtain aggregated sample feature maps of the corresponding size;
[0062] S204: Input aggregated sample feature maps of different sizes into the upsampling structure to obtain upsampled sample feature maps;
[0063] S205, input the feature map of the upsampled sample into the special extraction module and the detection module to obtain the prediction result;
[0064] S206, Calculate the overall loss of the degradation recognition model based on the prediction results and the labeled recognition results;
[0065] S207, iterate the degradation identification model according to the overall loss until the loss converges.
[0066] In one specific implementation, combined with Figure 7 As shown, the processing procedure of the detection module includes:
[0067] The input feature map is processed by the first convolution, the second convolution, and the third convolution to obtain the detection convolution map;
[0068] The feature map after the first convolution is processed by convolution, fully connected layers, and sigmoid to obtain the detection coefficients.
[0069] The detection convolutional image and the detection coefficients are multiplied together, and then convolution is performed to obtain the detection box result.
[0070] The input feature map is subjected to three consecutive convolutions to obtain a continuous convolution map;
[0071] The classification result is obtained by adding the continuous convolutional image and the feature map after one convolution.
[0072] The detection box result and the classification result are combined to form the degradation recognition result.
[0073] The special extraction process is as follows:
[0074] Use the output feature map of the attention class as the input feature map for special extraction;
[0075] The input feature map is fed into a parallel 1×1 convolutional layer to obtain the first extraction map, the second extraction map, and the third extraction map, respectively; wherein the first extraction map and the third extraction map are in HW×N format, and the second extraction map is in N×HW format;
[0076] Multiply the first extracted image and the second extracted image to obtain the multiplied feature map;
[0077] The multiplication feature maps are processed using a softmax-like method to obtain the multiplication coefficients;
[0078] Multiply the multiplication coefficients by the third extraction graph to obtain the coefficient multiplication graph;
[0079] The coefficient multiplication map and the input feature map are added together to obtain the specially extracted output feature map.
[0080] In this application, the input feature map is divided into three branches and processed by 1×1 convolution. The dimensions of the first and third branches are adjusted to HW×N format through this convolution, and the dimension of the second branch is N×HW format. Then, the first branch and the third branch are multiplied to obtain a feature map with a dimension of HW×HW, and the coefficients are obtained by processing it through a softmax-like process. The coefficients are multiplied by the third branch, and the multiplied feature map is added to the input feature map to obtain the output feature map.
[0081] In this application, the dimensions of the first and second branches are reversed by 1×1 convolution, thereby realizing the multiplication between feature maps; through this multiplication, long-range dependencies are captured; by setting coefficients, the dependencies are embedded in the feature maps; and on this basis, the original input is added and residual connections are used to embed the special extraction into the model without destroying the parameters.
[0082] In this application, long-range dependencies between captured features are extracted in a special way, which enables the model to extract global features and makes up for the deficiency of the current model that mainly extracts local features.
[0083] In this application, the response at a location is calculated as a weighted sum of features at all locations, so that when processing information at each location, processing is performed based on all possible location information.
[0084] In one implementation, the activation function of the SoftMax-like operation is:
[0085]
[0086] Where SoftMax1(x) is the activation function, x i x j Let i and j be the elements in the input vector, where i and j are the element indices.
[0087] In this application, SoftMax is a mathematical function typically used to convert a set of arbitrary real numbers into real numbers representing a probability distribution. Essentially, it is a normalization function that transforms a set of arbitrary real values into probability values between [0,1]. Because SoftMax converts them to values between 0 and 1, they can be interpreted as probabilities. If one of the inputs is small or negative, SoftMax transforms it into a low probability; if the input is large, it transforms it into a high probability, but it will always remain between 0 and 1.
[0088] However, for the standard SoftMax function, since the inputs are mapped to the range of 0 to 1 and the sum of all output values is 1, this means that even some very small input values will have a non-zero output value after processing by the SoftMax function. This amplifies the noise, resulting in the final output being more affected by noise.
[0089] In this application, a 1 is added to the denominator of the SoftMax-like functions; this change means that when the input values are very small, their output values can be closer to zero. This allows the corresponding output to tend to zero when there is no valuable information to add, thus greatly reducing unnecessary noise.
[0090] This application provides a stepwise aggregation-based intervertebral disc degeneration identification device for executing the stepwise aggregation-based intervertebral disc degeneration identification method described above. The stepwise aggregation-based intervertebral disc degeneration identification device will be described in detail below.
[0091] like Figure 8 As shown, the intervertebral disc degeneration identification device based on stepwise aggregation includes:
[0092] Image acquisition module 101 is used to acquire MRI medical images, wherein the MRI medical images are sagittal spinal medical images of any object;
[0093] The degradation recognition module 102 is used to input MRI medical images into a pre-trained degradation recognition model to obtain degradation recognition results. The degradation recognition model includes multiple feature extraction structures set in parallel and a stepwise aggregation structure. The output feature maps of different feature extraction structures have different sizes, and the stepwise aggregation structure is used to aggregate output feature maps of different sizes.
[0094] In one embodiment, the degradation identification module 102 is further configured to:
[0095] MRI medical images are input into multiple parallel feature extraction structures to obtain multiple output feature maps. The feature extraction structures include global extraction structures and local extraction structures, and the output feature maps include global feature maps and local feature maps. Global feature maps and local feature maps of the same size are input into a stepwise aggregation structure to obtain aggregated feature maps of corresponding sizes. Aggregated feature maps of different sizes are input into an upsampling structure to obtain upsampled feature maps. The upsampled feature maps are input into a special extraction module and a detection module to obtain the degradation recognition result.
[0096] In one implementation, the global extraction structure is an improved Transformer network structure, and the local extraction structure is a ConvNext network structure.
[0097] In one embodiment, the degradation identification module 102 is further configured to:
[0098] The global feature map is input into the attention extraction module to obtain the global extracted feature map; the local feature map is sequentially input into the special extraction module and the attention extraction module to obtain the local extracted feature map; the global extracted feature map and the local extracted feature map are fused to obtain the aggregated feature map.
[0099] In one embodiment, the degradation identification module 102 is further configured to:
[0100] The global feature map is converted into a horizontal feature map and then subjected to convolution and activation processing to obtain a horizontal extraction map. The global feature map is converted into a vertical feature map and then subjected to convolution processing to obtain a vertical extraction map. The global feature map is then subjected to convolution and activation processing alternately to obtain a local weight map. The horizontal extraction map and the numerical extraction map are multiplied together and then fused with the local weight map to obtain the global extracted feature map.
[0101] In one implementation, a model training module is also included, which is used for:
[0102] A sample medical image is acquired, and the sample medical image is labeled with the recognition result. The sample medical image is input into multiple feature extraction structures set in parallel to obtain global sample feature maps and local sample feature maps of multiple sizes. Global sample feature maps and local sample feature maps of the same size are input into a stepwise aggregation structure to obtain aggregated sample feature maps of corresponding sizes. Aggregated sample feature maps of different sizes are input into an upsampling structure to obtain upsampled sample feature maps. Upsampled sample feature maps are input into a special extraction module and a detection module to obtain prediction results. Based on the prediction results and the labeled recognition results, the overall loss of the degradation recognition model is calculated. The degradation recognition model is iterated according to the overall loss until the loss converges.
[0103] In one embodiment, the degradation identification module 102 is further configured to:
[0104] The input feature map is processed by a first convolution, a second convolution, and a third convolution to obtain a detection convolution map. The feature map after the first convolution is processed by convolution, fully connected layers, and a sigmoid function to obtain detection coefficients. The detection convolution map and the detection coefficients are multiplied and then convolved to obtain the detection box result. The input feature map is processed by three consecutive convolutions to obtain a continuous convolution map. The continuous convolution map is added to the feature map after one convolution and then convolved to obtain the classification result. The detection box result and the classification result are combined to obtain the degradation recognition result.
[0105] The intervertebral disc degeneration identification device based on stepwise aggregation provided in the above embodiments of this application corresponds to the intervertebral disc degeneration identification method based on stepwise aggregation provided in the embodiments of this application. Therefore, the specific contents of the device correspond to the intervertebral disc degeneration identification method. The specific contents can be referred to the records in the intervertebral disc degeneration identification method, which will not be repeated in this application.
[0106] The intervertebral disc degeneration identification device based on stepwise aggregation provided in the above embodiments of this application and the intervertebral disc degeneration identification method based on stepwise aggregation provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.
[0107] The above describes the internal function and structure of the stepwise aggregation-based intervertebral disc degeneration identification device, such as... Figure 9 As shown, in practice, the intervertebral disc degeneration identification device based on stepwise aggregation can be implemented as an electronic device, including: a memory 301 and a processor 303.
[0108] Memory 301 can be configured to store a program.
[0109] Additionally, memory 301 can also be configured to store various other data to support operation on the electronic device. Examples of this data include instructions for any application or method used to operate on the electronic device, contact data, phonebook data, messages, pictures, videos, etc.
[0110] The memory 301 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.
[0111] Processor 303, coupled to memory 301, is used to execute programs in memory 301 for:
[0112] Acquire MRI medical images, wherein the MRI medical images are sagittal spinal medical images of any object;
[0113] MRI medical images are input into a pre-trained degeneration recognition model to obtain degeneration recognition results;
[0114] The degradation recognition model includes multiple feature extraction structures set in parallel and a stepwise aggregation structure. The output feature maps of different feature extraction structures have different sizes, and the stepwise aggregation structure is used to aggregate output feature maps of different sizes.
[0115] In one implementation, the processor 303 is further configured to:
[0116] MRI medical images are input into multiple parallel feature extraction structures to obtain multiple output feature maps. The feature extraction structures include global extraction structures and local extraction structures, and the output feature maps include global feature maps and local feature maps. Global feature maps and local feature maps of the same size are input into a stepwise aggregation structure to obtain aggregated feature maps of corresponding sizes. Aggregated feature maps of different sizes are input into an upsampling structure to obtain upsampled feature maps. The upsampled feature maps are input into a special extraction module and a detection module to obtain the degradation recognition result.
[0117] In one implementation, the global extraction structure is an improved Transformer network structure, and the local extraction structure is a ConvNext network structure.
[0118] In one implementation, the processor 303 is further configured to:
[0119] The global feature map is input into the attention extraction module to obtain the global extracted feature map; the local feature map is sequentially input into the special extraction module and the attention extraction module to obtain the local extracted feature map; the global extracted feature map and the local extracted feature map are fused to obtain the aggregated feature map.
[0120] In one implementation, the processor 303 is further configured to:
[0121] The global feature map is converted into a horizontal feature map and then subjected to convolution and activation processing to obtain a horizontal extraction map. The global feature map is converted into a vertical feature map and then subjected to convolution processing to obtain a vertical extraction map. The global feature map is then subjected to convolution and activation processing alternately to obtain a local weight map. The horizontal extraction map and the numerical extraction map are multiplied together and then fused with the local weight map to obtain the global extracted feature map.
[0122] In one implementation, the processor 303 is further configured to:
[0123] A sample medical image is acquired, and the sample medical image is labeled with the recognition result. The sample medical image is input into multiple feature extraction structures set in parallel to obtain global sample feature maps and local sample feature maps of multiple sizes. Global sample feature maps and local sample feature maps of the same size are input into a stepwise aggregation structure to obtain aggregated sample feature maps of corresponding sizes. Aggregated sample feature maps of different sizes are input into an upsampling structure to obtain upsampled sample feature maps. Upsampled sample feature maps are input into a special extraction module and a detection module to obtain prediction results. Based on the prediction results and the labeled recognition results, the overall loss of the degradation recognition model is calculated. The degradation recognition model is iterated according to the overall loss until the loss converges.
[0124] In one implementation, the processor 303 is further configured to:
[0125] The input feature map is processed by a first convolution, a second convolution, and a third convolution to obtain a detection convolution map. The feature map after the first convolution is processed by convolution, fully connected layers, and a sigmoid function to obtain detection coefficients. The detection convolution map and the detection coefficients are multiplied and then convolved to obtain the detection box result. The input feature map is processed by three consecutive convolutions to obtain a continuous convolution map. The continuous convolution map is added to the feature map after one convolution and then convolved to obtain the classification result. The detection box result and the classification result are combined to obtain the degradation recognition result.
[0126] In this application, the processor is also specifically used to execute all the processes and steps of the above-mentioned intervertebral disc degeneration identification method based on stepwise aggregation. For details, please refer to the records in the intervertebral disc degeneration identification method. This application will not repeat them here.
[0127] In this application, Figure 9 The diagram only shows some components and does not mean that the electronic device includes only these components. Figure 9 The components shown.
[0128] The electronic device provided in this embodiment is based on the same inventive concept as the intervertebral disc degeneration identification method based on stepwise aggregation provided in this application embodiment, and has the same beneficial effects as the methods adopted, run or implemented by the application stored therein.
[0129] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0130] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0131] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.
[0132] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0133] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0134] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0135] This application also provides a computer-readable storage medium corresponding to the stepwise aggregation-based intervertebral disc degeneration identification method provided in the foregoing embodiments, wherein a computer program (i.e., a program product) is stored thereon, and the computer program, when run by a processor, executes the stepwise aggregation-based intervertebral disc degeneration identification method provided in any of the foregoing embodiments.
[0136] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0137] The computer-readable storage medium provided in the above embodiments of this application and the intervertebral disc degeneration identification method based on stepwise aggregation provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.
[0138] It should be noted that numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this application may be practiced without these specific details. In some instances, well-known structures and techniques have not been shown in detail so as not to obscure the understanding of this specification.
[0139] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0140] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.
Claims
1. A method for identifying intervertebral disc degeneration based on stepwise aggregation, characterized in that: include: Acquire an MRI medical image, wherein the MRI medical image is a sagittal spinal medical image of any object; Inputting the MRI medical image into the pre-trained degradation recognition model to obtain the degradation recognition result; The degradation recognition model includes a plurality of feature extraction structures and a step-by-step aggregation structure arranged in parallel, the output feature maps of different feature extraction structures have different sizes, and the step-by-step aggregation structure is used to aggregate output feature maps of different sizes.
2. The method for identifying intervertebral disc degeneration according to claim 1, characterized in that: The step of inputting the MRI medical image into the pre-trained degradation recognition model to obtain the degradation recognition result includes: Inputting the MRI medical image into a plurality of feature extraction structures arranged in parallel to obtain a plurality of output feature maps; the feature extraction structure comprises a global extraction structure and a local extraction structure, and the output feature map comprises a global feature map and a local feature map; Input the global feature map and local feature map of the same size into the step-by-step aggregation structure to obtain an aggregated feature map of the corresponding size; Input the aggregated feature maps of different sizes into the upsampling structure to obtain upsampled feature maps; The upsampled feature map is input into the special extraction module and the detection module to obtain the degradation recognition result.
3. The method for identifying intervertebral disc degeneration according to claim 2, characterized in that: The global extraction structure is an improved Transformer network structure, and the local extraction structure is a ConvNext network structure.
4. The method for identifying intervertebral disc degeneration according to claim 2, characterized in that: The step of inputting the global feature map and the local feature map of the same size into the stepwise aggregation structure to obtain an aggregated feature map of the corresponding size includes: Input the global feature map into the attention extraction module to obtain a global extracted feature map; The local feature map is sequentially input into the special extraction module and the attention extraction module to obtain a local extracted feature map; The global extracted feature map and the local extracted feature map are fused to obtain the aggregated feature map.
5. The method for identifying intervertebral disc degeneration according to claim 4, characterized in that: The step of inputting the global feature map into the attention extraction module to obtain the global extracted feature map includes: The global feature map is converted into a horizontal feature map, and convolution and activation are performed to obtain a horizontal extraction map; Convert the global feature map into a vertical feature map and perform convolution processing to obtain a vertical extraction map; The global feature map is alternately convolved and activated to obtain a local weight map; The horizontal extraction map and the value extraction map are multiplied and then fused with the local weight map to obtain the global extraction feature map.
6. The method for identifying intervertebral disc degeneration according to claim 1, characterized in that: The training process of the degradation recognition model includes: Acquire a sample medical image, wherein the sample medical image is annotated with a recognition result; Inputting the sample medical image into multiple feature extraction structures arranged in parallel to obtain global sample feature maps and local sample feature maps of multiple sizes; Input the global sample feature map and the local sample feature map of the same size into the step-by-step aggregation structure to obtain an aggregated sample feature map of the corresponding size; Inputting aggregated sample feature maps of different sizes into the upsampling structure to obtain upsampled sample feature maps; The upsampled sample feature map is input into the special extraction module and the detection module to obtain the prediction result; Based on the prediction results and the annotated recognition results, the overall loss of the degradation recognition model is calculated; The degradation identification model is iterated according to the overall loss until the loss converges.
7. The method for identifying intervertebral disc degeneration according to claim 2, characterized in that: The processing process of the detection module includes: Perform the first convolution, the second convolution and the third convolution on the input feature map to obtain a detection convolution map; The feature map after the first convolution process is convolved, fully connected and sigmoid processed to obtain the detection coefficient; After multiplying the detection convolution map and the detection coefficient, convolution is performed to obtain the detection frame result; Perform three consecutive convolutions on the input feature map to obtain a continuous convolution map; After adding the continuous convolution map and the feature map after one convolution, convolution is performed to obtain the classification result; The detection frame result and the classification result are combined into a degeneration recognition result.
8. A device for identifying intervertebral disc degeneration based on stepwise aggregation, characterized in that: include: An image acquisition module, which is used to acquire an MRI medical image, wherein the MRI medical image is a sagittal spinal medical image of any object; A degradation recognition module is used to input an MRI medical image into a pre-trained degradation recognition model to obtain a degradation recognition result; the degradation recognition model includes a plurality of feature extraction structures and a step-by-step aggregation structure arranged in parallel, the output feature graphs of different feature extraction structures have different sizes, and the step-by-step aggregation structure is used to aggregate output feature graphs of different sizes.
9. An electronic device, characterized in that: include: Memory and processor; The memory is used to store programs; The processor, coupled to the memory, is configured to execute the program to: Acquire an MRI medical image, wherein the MRI medical image is a sagittal spinal medical image of any object; Inputting the MRI medical image into the pre-trained degradation recognition model to obtain the degradation recognition result; The degradation recognition model includes a plurality of feature extraction structures and a step-by-step aggregation structure arranged in parallel, the output feature maps of different feature extraction structures have different sizes, and the step-by-step aggregation structure is used to aggregate output feature maps of different sizes.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the intervertebral disc degeneration identification method based on stepwise aggregation as described in any one of claims 1 to 7.