Lumbar vertebra lesion identification method and device, medium and electronic equipment
By processing T2-weighted MRI images using a target detection model and combining downsampling and pooling layers, the accuracy problem of lumbar spine lesion identification was solved, achieving more efficient lesion type identification.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANTONG INFECTIOUS DISEASE PREVENTION & CONTROL INST
- Filing Date
- 2025-08-28
- Publication Date
- 2026-08-04
AI Technical Summary
Existing methods for identifying lumbar spine lesions are not very accurate and are prone to errors. Both manual judgment and traditional machine learning methods have shortcomings.
A target detection model, including a downsampling layer and a pooling layer, is used to process T2-weighted MRI images. Feature data is extracted through the downsampling layer, and classification is performed through the pooling layer. Feature extraction and nonlinear mapping are performed using a downsampling patch embedding layer and a moving KAN layer. The model is optimized by combining a focusing loss function.
It improves the accuracy of lumbar spine lesion identification, reduces errors, simplifies computational complexity, and is suitable for real-time medical image classification and analysis.
Smart Images
Figure CN121033542B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of data processing technology, and more specifically, to methods, devices, media, and electronic equipment for identifying lumbar spine lesions. Background Technology
[0002] Lumbar spine disorders primarily refer to degenerative conditions such as spondylitis and lumbar osteophyte formation, which often lead to spinal canal and joint depression, as well as narrowing of the neural foramen. Lumbar spine disorders can cause nerve compression, persistent pain, limited mobility, and neurological dysfunction. Therefore, early diagnosis and appropriate intervention can slow the progression of the disease. Current techniques often rely on the user's experience or traditional machine learning methods to determine if a user has lumbar spine disorders; however, these methods are not very accurate and prone to errors. Summary of the Invention
[0003] The purpose of this disclosure is to provide methods, devices, media, and electronic equipment for identifying lumbar spine lesions in order to solve the technical problems existing in the related art.
[0004] To achieve the above objectives, in a first aspect, this disclosure provides a method for identifying lumbar spine lesions, comprising: Determine the user's lumbar spine image, which is a T2-weighted magnetic resonance imaging image; The lumbar spine image is input into a target detection model to obtain the lesion category of the user's lumbar spine. The target detection model includes a downsampling layer and a pooling layer. The downsampling layer is used to extract feature data from the lumbar spine image, and the pooling layer is used to classify the feature data to obtain the lesion category of the user's lumbar spine.
[0005] Optionally, inputting the lumbar spine image into the target detection model to obtain the lesion category of the user's lumbar spine includes: The lumbar spine image is input into the downsampling layer to obtain the feature data. The downsampling layer includes a downsampling patch embedding layer and a moving KAN layer. The downsampling patch embedding layer is used to process the lumbar spine image to obtain multiple sub-feature images. The moving KAN layer is used to extract the features of each sub-feature image from the multiple sub-feature images to obtain the feature data. The feature data is input into the pooling layer to obtain the user's lumbar spine lesion type.
[0006] Optionally, inputting the lumbar spine image into the downsampling layer to obtain the feature data includes: The lumbar spine image is segmented into multiple sub-lumbar spine images by the downsampling patch embedding layer, and each sub-lumbar spine image in the multiple sub-lumbar spine images is subjected to convolution and normalization processing to obtain multiple sub-feature images; The feature data is obtained by processing each of the multiple sub-feature images through the moving KAN layer.
[0007] Optionally, the step of processing each of the plurality of sub-feature images through the moving KAN layer to obtain the feature data includes: The multiple sub-feature images are input into the mobile KAN layer to obtain the feature data. The mobile KAN layer includes a KAN network module, a deep convolution module, and an activation module. The deep convolution module is used to perform deep convolution processing on each channel of the multiple sub-feature images to obtain sub-feature data. The KAN network module is used to perform nonlinear mapping modeling on the sub-feature data to obtain a feature image model. The activation module is used to perform nonlinear activation processing on the feature image model to obtain the feature data.
[0008] Optionally, the number of downsampling layers in the target detection model is less than or equal to a preset number, and the last downsampling layer is connected to the pooling layer.
[0009] Optionally, the KAN network module is expressed by the following calculation formula; in, For the l-th sub-feature image, All of these are trainable parameters. Let be the sub-feature image located in the j-th row and i-th column of the matrix function. is the sigma linear unit activation function, and x is the sub-feature image.
[0010] Optionally, the loss function of the target detection model is a focusing loss function.
[0011] Secondly, this disclosure provides a device for identifying lumbar spine lesions, including a determination module and an identification module; The determining module is used to determine the user's lumbar spine image, which is a T2-weighted magnetic resonance imaging image; The recognition module is used to input the lumbar spine image into a target detection model to obtain the lesion category of the user's lumbar spine. The target detection model includes a downsampling layer and a pooling layer. The downsampling layer is used to extract feature data from the lumbar spine image, and the pooling layer is used to classify the feature data to obtain the lesion category of the user's lumbar spine.
[0012] Optionally, the identification module includes: The processing module is used to input the lumbar spine image into the downsampling layer to obtain the feature data. The downsampling layer includes a downsampling patch embedding layer and a moving KAN layer. The downsampling patch embedding layer is used to process the lumbar spine image to obtain multiple sub-feature images. The moving KAN layer is used to extract the features of each sub-feature image from the multiple sub-feature images to obtain the feature data. The classification module is used to input the feature data into the pooling layer to obtain the user's lumbar spine lesion type.
[0013] Optionally, the classification module is used for: The lumbar spine image is segmented into multiple sub-lumbar spine images by the downsampling patch embedding layer, and each sub-lumbar spine image in the multiple sub-lumbar spine images is subjected to convolution and normalization processing to obtain multiple sub-feature images; The feature data is obtained by processing each of the multiple sub-feature images through the moving KAN layer.
[0014] Optionally, the classification module is used for: The multiple sub-feature images are input into the mobile KAN layer to obtain the feature data. The mobile KAN layer includes a KAN network module, a deep convolution module, and an activation module. The deep convolution module is used to perform deep convolution processing on each channel of the multiple sub-feature images to obtain sub-feature data. The KAN network module is used to perform nonlinear mapping modeling on the sub-feature data to obtain a feature image model. The activation module is used to perform nonlinear activation processing on the feature image model to obtain the feature data.
[0015] Optionally, the number of downsampling layers in the target detection model is less than or equal to a preset number, and the last downsampling layer is connected to the pooling layer.
[0016] Optionally, the KAN network module is expressed by the following calculation formula; in, For the l-th sub-feature image, All of these are trainable parameters. Let be the sub-feature image located in the j-th row and i-th column of the matrix function. is the sigma linear unit activation function, and x is the sub-feature image.
[0017] Optionally, the loss function of the target detection model is a focusing loss function.
[0018] Thirdly, this disclosure provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of any of the methods provided in the first aspect of this disclosure.
[0019] Fourthly, this disclosure provides an electronic device, comprising: A memory on which computer programs are stored; A processor is configured to execute the computer program in the memory to implement the steps of any of the methods provided in the first aspect of this disclosure. The above technical solution inputs the user's T2-weighted MRI image into a target detection model to obtain the user's lumbar spine lesion category. The target detection model includes a downsampling layer and a pooling layer. The downsampling layer extracts feature data from the lumbar spine image, and the pooling layer classifies the feature data to obtain the user's lumbar spine lesion category. By using the downsampling and pooling layers in the target detection model to identify the input T2-weighted MRI image and determine the user's lumbar spine lesion type, compared to related technologies that rely on manual judgment or machine learning methods to identify lumbar spine lesions, this approach reduces errors in lesion type identification and improves the accuracy of lumbar spine lesion recognition.
[0020] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description
[0021] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram illustrating a method for identifying lumbar spine lesions according to an exemplary embodiment of the present disclosure.
[0022] Figure 2 This is a schematic diagram illustrating an image of the lumbar spine according to an exemplary embodiment of the present disclosure.
[0023] Figure 3 This is a schematic diagram illustrating the structure of a target detection model according to an exemplary embodiment of the present disclosure.
[0024] Figure 4 This is a schematic diagram illustrating downsampling layer processing according to an exemplary embodiment of the present disclosure.
[0025] Figure 5 This is a schematic diagram illustrating the structure of a mobile KAN layer according to an exemplary embodiment of the present disclosure.
[0026] Figure 6 This is a schematic diagram illustrating the structure of downsampling processing according to an exemplary embodiment of the present disclosure.
[0027] Figure 7 This is a schematic diagram illustrating a device for identifying lumbar spine lesions according to an exemplary embodiment of the present disclosure.
[0028] Figure 8 This is a block diagram illustrating an electronic device according to an exemplary embodiment. Detailed Implementation
[0029] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.
[0030] In related technologies, relevant personnel can analyze the corresponding MRI images of the lumbar spine to determine the lumbar spine lesion of the user. Alternatively, traditional machine learning methods, such as the VGG (Visual Geometry Group) model and the ResNet (Residual Network) model, can be used to identify the type of lumbar spine lesion.
[0031] However, the inventors discovered that when relevant personnel determine the type of lumbar spine lesions, different personnel may make different judgments based on their different experiences, leading to errors. Furthermore, the accuracy of traditional machine learning methods is relatively low.
[0032] In view of this, this disclosure provides a method, device, medium and electronic equipment for identifying lumbar spine lesions in order to solve the technical problems existing in the related art.
[0033] like Figure 1 As shown, Figure 1 This is a schematic diagram illustrating a method for identifying lumbar spine lesions according to an exemplary embodiment of the present disclosure, with reference to... Figure 1 ,include: S101: Determine the user's lumbar spine image, wherein the lumbar spine image is a T2-weighted magnetic resonance imaging image; S102: Input the lumbar spine image into the target detection model to obtain the lesion category of the user's lumbar spine. The target detection model includes a downsampling layer and a pooling layer. The downsampling layer is used to extract feature data from the lumbar spine image, and the pooling layer is used to classify the feature data to obtain the lesion category of the user's lumbar spine.
[0034] The above technical solution inputs the user's T2-weighted MRI image into a target detection model to obtain the user's lumbar spine lesion category. The target detection model includes a downsampling layer and a pooling layer. The downsampling layer extracts feature data from the lumbar spine image, and the pooling layer classifies the feature data to obtain the user's lumbar spine lesion category. By using the downsampling and pooling layers in the target detection model to identify the input T2-weighted MRI image and determine the user's lumbar spine lesion type, compared to related technologies that rely on manual judgment or machine learning methods to identify lumbar spine lesions, this approach reduces errors in lesion type identification and improves the accuracy of lumbar spine lesion recognition.
[0035] To enable those skilled in the art to better understand the method for identifying lumbar spine lesions provided in this disclosure, the above steps are illustrated in detail below.
[0036] For example, the lumbar spine image can be a T2-weighted MRI image. T2-weighted MRI images can be used for non-invasive and high-resolution observation of subtle changes in a patient's condition, including loss of intervertebral disc height and elasticity, reduction of proteoglycan content, and formation of bone hyperplasia. In this embodiment of the disclosure, the lumbar spine image can be a T2-weighted MRI image; however, this embodiment of the disclosure does not specifically limit this method. For example, Figure 2 As shown, in T2-weighted MRI images, areas filled with fluid appear as bright areas and are easily distinguishable; fat has lower brightness, while the brightness of muscles and other soft tissues is between dark and light. Therefore, when using this T2-weighted MRI image to identify the type of lumbar spine lesions, the accuracy of identifying the type of lumbar spine lesions can be improved. In this respect, the embodiments disclosed herein do not make specific limitations.
[0037] For example, the lesion categories may include left foraminal stenosis, right foraminal stenosis, left subarticular stenosis, right subarticular stenosis, and spinal canal stenosis, as well as the severity of each lesion category. The severity of the lesion category is categorized as normal / mild, moderate, and severe; however, this disclosure does not specifically limit the severity in this regard. A target detection model can be used to detect lesion types in the lumbar spine. This target detection model includes a downsampling layer and a pooling layer. The downsampling layer extracts feature data from the lumbar spine image, and the pooling layer classifies the feature data to obtain the lesion type of the lumbar spine. The downsampling layer can be used to reduce the spatial dimensionality of the feature image. The pooling layer can perform aggregation operations within local regions to reduce the spatial dimension of the feature map.
[0038] In this embodiment, a user's lumbar spine image can be input into a target detection model. The target detection model processes the lumbar spine image to obtain the user's lumbar spine lesion category. By using downsampling and pooling layers in the target detection model to identify the input T2-weighted MRI image, the user's lumbar spine lesion type can be determined. Compared to related technologies that rely on manual judgment or machine learning methods to identify lumbar spine lesions, this approach reduces errors in lesion type identification and improves the accuracy of lumbar spine lesion recognition.
[0039] In one possible manner, inputting the lumbar spine image into a target detection model to obtain the lesion category of the user's lumbar spine includes: The lumbar spine image is input into the downsampling layer to obtain the feature data. The downsampling layer includes a downsampling patch embedding layer and a moving KAN layer. The downsampling patch embedding layer is used to process the lumbar spine image to obtain multiple sub-feature images. The moving KAN layer is used to extract the features of each sub-feature image from the multiple sub-feature images to obtain the feature data. The feature data is input into the pooling layer to obtain the user's lumbar spine lesion type.
[0040] It should be understood that the downsampling layer may include a downsampling patch embedding layer and a moving KAN layer. The downsampling patch embedding layer can reduce the spatial dimensionality of the feature map and map it to a higher-dimensional feature space, thereby improving computational efficiency while preserving important image features. The moving KAN layer can improve the model's efficiency and feature extraction capabilities on mobile devices. In the embodiments of this disclosure, as... Figure 3 As shown, when the lumbar spine image is input into the downsampling layer, it can be processed sequentially through the downsampling patch embedding layer and the moving KAN layer.
[0041] In one possible manner, inputting the lumbar spine image into the downsampling layer to obtain the feature data includes: The lumbar spine image is segmented into multiple sub-lumbar spine images by the downsampling patch embedding layer, and each sub-lumbar spine image in the multiple sub-lumbar spine images is subjected to convolution and normalization processing to obtain multiple sub-feature images; The feature data is obtained by processing each of the multiple sub-feature images through the moving KAN layer.
[0042] It should be understood that, such as Figure 4 As shown, within the downsampling patch embedding layer, processing can be performed using the following formula.
[0043] PathEmbed(x)=Conv(x;Kernel-size=7, stride=4) Here, x represents the lumbar spine image, and the convolutional layer uses a kernel size of 7*7 with a span of 4. This allows each sub-feature image in the lumbar spine image to be converted into a one-dimensional labeled sequence, effectively reducing dimensionality while preserving basic spatial information.
[0044] It can then be processed using the following formula.
[0045] in, This is the average value. This represents the standard deviation.
[0046] In one possible manner, processing each of the plurality of sub-feature images through the moving KAN layer to obtain the feature data includes: The multiple sub-feature images are input into the mobile KAN layer to obtain the feature data. The mobile KAN layer includes a KAN network module, a deep convolution module, and an activation module. The deep convolution module is used to perform deep convolution processing on each channel of the multiple sub-feature images to obtain sub-feature data. The KAN network module is used to perform nonlinear mapping modeling on the sub-feature data to obtain a feature image model. The activation module is used to perform nonlinear activation processing on the feature image model to obtain the feature data.
[0047] It should be understood that, such as Figure 5 As shown, the feature data obtained by processing each sub-feature image in multiple sub-feature images using a moving KAN layer can be expressed by the following calculation formula.
[0048] Y=ReLU(DwConv(KAN(x))) Here, KAN can be the transformation formed by the KAN layer, DwConv (.) represents the depthwise convolution operation, and ReLU (.) applies the rectified linear unit activation function. This formula simplifies the computational complexity and improves computational efficiency.
[0049] In the depthwise convolution module, depthwise convolution processing can be performed on each channel of each input sub-feature image, thus enabling independent convolution operations on each channel. Compared to standard convolution processing, this reduces the number of parameters and computational complexity. Specifically, the depthwise convolution process can be expressed by the following formula.
[0050] DwConv(x)=Conv(x;kernel-size=3, stride=1, groups-C) Where C is the number of input channels. This depthwise convolution process ensures that each channel in each sub-feature image is convolved separately, while maintaining the integrity of spatial features and minimizing computational load.
[0051] Inspired by the Colmogorod-Arnold theorem, the KAN network module can decompose multivariate functions into unit functions, enabling it to model complex nonlinear mappings. This allows it to learn complex dependencies in image data, improving its ability to capture high-level semantic information. Specifically, the KAN network module is expressed using the following formula; in, For the l-th sub-feature image, All of these are trainable parameters. Let be the sub-feature image located in the j-th row and i-th column of the matrix function. is the sigma linear unit activation function, and x is the sub-feature image.
[0052] The activation module can be processed by introducing nonlinearity, specifically, it can be represented by the following calculation formula.
[0053] ReLU(x) = max(0, x) This nonlinear activation ensures that the network captures the intricate patterns and dependencies in the feature maps, thereby enhancing the overall performance of the model.
[0054] In one possible manner, the number of downsampling layers in the target detection model is less than or equal to a preset number, and the last downsampling layer is connected to the pooling layer.
[0055] It should be understood that in related technologies, feature data can be obtained by processing the input lumbar spine image through multiple downsampling operations. However, this method increases the complexity and computational tedium of processing lumbar spine images. Therefore, in the embodiments of this disclosure, as... Figure 6 As shown, the number of downsampling layers is less than or equal to a preset number, where the preset number is less than the number of downsampling processes in related technologies. Processing lumbar spine images with a smaller number of downsampling layers to obtain corresponding feature data can reduce the complexity of processing and simplify the tediousness of calculation.
[0056] By stacking multiple downsampling layers, with the number of downsampling layers less than or equal to a preset number, hierarchical and multi-scale representations of input lumbar spine images can be achieved, simultaneously capturing local patterns and global textual information. This hierarchical feature extraction is crucial for accurately modeling the inherent spatial and structural complexity of medical images, thereby improving the overall diagnostic performance of the downsampling layers. Simultaneously, it ensures that the downsampling layers maintain high feature extraction capabilities while operating efficiently, making them ideal for real-time medical image classification and analysis tasks.
[0057] In one possible manner, the loss function of the target detection model is a focusing loss function.
[0058] It should be understood that by setting the focusing loss function as the loss function of the object detection model, the detail performance of the object detection model in processing lumbar spine images can be improved. Given that different regions in different lumbar spine images have different levels of importance, a spatial attention mechanism can be introduced into the object detection model. This spatial attention mechanism can assign higher weights to image pixels in the region of interest, thereby allowing the object detection model to adaptively focus its efforts on the feature data.
[0059] in, Attention atlas for lumbar spine images. This is a scaling factor that can be used to adjust the contribution of the attention mechanism. To represent the distribution of real data samples Take the expected value, which is the average operation over the entire training dataset. To represent the predicted output of an object detection model for an input lumbar spine image on a corresponding task, such as bounding box parameters and class probabilities, or segmentation probability maps, To represent the true annotations corresponding to the input lumbar spine image, including true bounding boxes / category or segmentation annotations, To represent the focus loss term, it is used to constrain the difference between the prediction and the true label within the region of interest, thereby enhancing the model's learning of small but crucial regions. To represent the weighting coefficients of the focusing loss, used to adjust the relative importance of this term in the total loss, in the embodiments of this disclosure... , To represent the attention-weighted loss term, i.e. in the attention map The error between the predicted and the true annotation is calculated under weighted conditions and used to adaptively increase the contribution of pixels in the region of interest.
[0060] The total loss of the target detection model can be achieved by adding the two losses mentioned above together with the standard loss, which can improve the accuracy of the target detection model.
[0061] The above technical solution uses downsampling and pooling layers in the target detection model to identify the input T2-weighted MRI image and determine the user's lumbar spine lesion type. Compared to related technologies that rely on manual judgment or machine learning methods to identify lumbar spine lesions, this approach reduces errors in lesion type identification and improves accuracy. Using fewer than a preset number of downsampling layers to process the lumbar spine image simplifies computation and reduces computational complexity.
[0062] Based on the same concept, this embodiment also discloses an image recognition device for lumbar spine lesions, such as... Figure 7 As shown, Figure 7 This is a schematic diagram illustrating a lumbar spine lesion identification device 700 according to an exemplary embodiment of the present disclosure, with reference to... Figure 7 It includes a determining module 701 and an identification module 702; The determining module 701 is used to determine the user's lumbar spine image, wherein the lumbar spine image is a T2-weighted magnetic resonance image; The recognition module 702 is used to input the lumbar spine image into the target detection model to obtain the lesion category of the user's lumbar spine. The target detection model includes a downsampling layer and a pooling layer. The downsampling layer is used to extract feature data from the lumbar spine image, and the pooling layer is used to classify the feature data to obtain the lesion category of the user's lumbar spine.
[0063] Optionally, the identification module 702 includes: The processing module is used to input the lumbar spine image into the downsampling layer to obtain the feature data. The downsampling layer includes a downsampling patch embedding layer and a moving KAN layer. The downsampling patch embedding layer is used to process the lumbar spine image to obtain multiple sub-feature images. The moving KAN layer is used to extract the features of each sub-feature image from the multiple sub-feature images to obtain the feature data. The classification module is used to input the feature data into the pooling layer to obtain the user's lumbar spine lesion type.
[0064] Optionally, the classification module is used for: The lumbar spine image is segmented into multiple sub-lumbar spine images by the downsampling patch embedding layer, and each sub-lumbar spine image in the multiple sub-lumbar spine images is subjected to convolution and normalization processing to obtain multiple sub-feature images; The feature data is obtained by processing each of the multiple sub-feature images through the moving KAN layer.
[0065] Optionally, the classification module is used for: The multiple sub-feature images are input into the mobile KAN layer to obtain the feature data. The mobile KAN layer includes a KAN network module, a deep convolution module, and an activation module. The deep convolution module is used to perform deep convolution processing on each channel of the multiple sub-feature images to obtain sub-feature data. The KAN network module is used to perform nonlinear mapping modeling on the sub-feature data to obtain a feature image model. The activation module is used to perform nonlinear activation processing on the feature image model to obtain the feature data.
[0066] Optionally, the number of downsampling layers in the target detection model is less than or equal to a preset number, and the last downsampling layer is connected to the pooling layer.
[0067] Optionally, the KAN network module is expressed by the following calculation formula; in, For the l-th sub-feature image, All of these are trainable parameters. Let be the sub-feature image located in the j-th row and i-th column of the matrix function. is the sigma linear unit activation function, and x is the sub-feature image.
[0068] Optionally, the loss function of the target detection model is a focusing loss function.
[0069] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.
[0070] Based on the same concept, this embodiment also discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the lumbar spine lesion identification method disclosed in this embodiment.
[0071] Based on the same concept, this embodiment also discloses an electronic device, including: A memory on which computer programs are stored; A processor is configured to execute the computer program stored in the memory to implement the steps of the lumbar spine lesion identification method disclosed in this embodiment. Figure 8 This is a block diagram illustrating an electronic device 800 according to an exemplary embodiment. For example... Figure 8 As shown, the electronic device 800 may include a processor 801 and a memory 802. The electronic device 800 may also include one or more of a multimedia component 803, an input / output (I / O) interface 804, and a communication component 805.
[0072] The processor 801 controls the overall operation of the electronic device 800 to complete all or part of the steps in the aforementioned method for identifying lumbar spine lesions. The memory 802 stores various types of data to support the operation of the electronic device 800. This data may include, for example, instructions for any application or method operating on the electronic device 800, and application-related data such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 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. The multimedia component 803 may include a screen and audio components. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in memory 802 or transmitted via communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as a keyboard, mouse, buttons, etc. These buttons may be virtual or physical buttons. Communication component 805 is used for wired or wireless communication between the electronic device 800 and other devices. Wireless communication may include Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof; therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, or an NFC module.
[0073] In an exemplary embodiment, the electronic device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the aforementioned method for identifying lumbar spine lesions.
[0074] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the above-described method for identifying lumbar spine lesions. For example, the computer-readable storage medium may be the memory 802 including the program instructions, which may be executed by the processor 801 of the electronic device 800 to complete the above-described method for identifying lumbar spine lesions.
[0075] In another exemplary embodiment, a computer program product is also provided, which includes a computer program executable by a processor, which, when executed by the processor, implements the steps of the above-described method for identifying lumbar spine lesions.
[0076] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.
[0077] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.
[0078] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.
Claims
1. A method for identifying lumbar spine lesions, characterized in that, include: Determine the user's lumbar spine image, which is a T2-weighted magnetic resonance imaging image; The lumbar spine image is input into a target detection model to obtain the lesion category of the user's lumbar spine. The target detection model includes a downsampling layer and a pooling layer. The downsampling layer extracts feature data from the lumbar spine image, and the pooling layer performs classification processing on the feature data to obtain the lesion category of the user's lumbar spine. The process of inputting the lumbar spine image into the target detection model to obtain the lesion category of the user's lumbar spine includes: The lumbar spine image is input into the downsampling layer to obtain the feature data. The downsampling layer includes a downsampling patch embedding layer and a moving KAN layer. The downsampling patch embedding layer is used to process the lumbar spine image to obtain multiple sub-feature images. The moving KAN layer is used to extract the features of each sub-feature image from the multiple sub-feature images to obtain the feature data. The feature data is input into the pooling layer to obtain the user's lumbar spine lesion type; The step of inputting the lumbar spine image into the downsampling layer to obtain the feature data includes: The lumbar spine image is segmented into multiple sub-lumbar spine images by the downsampling patch embedding layer, and each sub-lumbar spine image in the multiple sub-lumbar spine images is subjected to convolution and normalization processing to obtain multiple sub-feature images; The feature data is obtained by processing each of the multiple sub-feature images through the moving KAN layer; The process of processing each sub-feature image among the plurality of sub-feature images through the moving KAN layer to obtain the feature data includes: The multiple sub-feature images are input into the mobile KAN layer to obtain the feature data. The mobile KAN layer includes a KAN network module, a deep convolution module, and an activation module. The deep convolution module is used to perform deep convolution processing on each channel of the multiple sub-feature images to obtain sub-feature data. The KAN network module is used to perform nonlinear mapping modeling on the sub-feature data to obtain a feature image model. The activation module is used to perform nonlinear activation processing on the feature image model to obtain the feature data. The number of downsampling layers in the target detection model is less than or equal to a preset number, and the last downsampling layer is connected to the pooling layer; The KAN network module is expressed by the following calculation formula: ； ； wherein, is the ith sub-feature image, are trainable parameters, is the sub-feature image located at the jth row and the ith column in the matrix function, is a sigmoid linear unit activation function, and x is a sub-feature image.
2. The method of identifying a lumbar pathology according to claim 1, wherein, The loss function of the target detection model is the focusing loss function.
3. An apparatus for identifying a lumbar pathology, comprising: A method for identifying lumbar spine lesions according to any one of claims 1-2 includes a determination module and an identification module; The determining module is used to determine the user's lumbar spine image, which is a T2-weighted magnetic resonance imaging image; The recognition module is used to input the lumbar spine image into a target detection model to obtain the lesion category of the user's lumbar spine. The target detection model includes a downsampling layer and a pooling layer. The downsampling layer is used to extract feature data from the lumbar spine image, and the pooling layer is used to classify the feature data to obtain the lesion category of the user's lumbar spine.
4. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the program is executed by the processor, it implements the steps of the method described in claim 1 or 2.
5. An electronic device, comprising: include: A memory on which computer programs are stored; A processor for executing the computer program in the memory to implement the steps of the method of claim 1 or 2.