A spondylolisthesis detection method and device based on lumbar X-ray film

Abstract

The application discloses a spondylolisthesis detection method and device based on a lumbar vertebra X-ray film, and the method comprises the following steps: acquiring a lumbar vertebra X-ray film digital image; pre-processing the lumbar vertebra X-ray film digital image to obtain a pre-processed lumbar vertebra anteroposterior and lateral X-ray film digital image; and detecting the pre-processed lumbar vertebra X-ray film digital image to obtain a spondylolisthesis detection result of the lumbar vertebra X-ray film. The application completes full-process automatic processing from a pre-processed image to accurate positioning of key points, and greatly improves the detection efficiency compared with a traditional manual key point marking or a semi-automatic detection method.

Description

Technical Field

[0001] This invention relates to the field of image processing and application technology, and in particular to a method and device for detecting vertebral slippage based on lumbar spine X-ray films. Background Technology

[0002] Lumbar spondylolisthesis is a common orthopedic condition. Lumbar spine X-rays, as a routine diagnostic tool, have certain limitations. Mild lumbar spondylolisthesis is often not clearly visible on X-rays, leading to diagnostic difficulties. Furthermore, the projection angle and image quality of X-rays can further blur the image, increasing diagnostic uncertainty. Therefore, mild lumbar spondylolisthesis is easily missed on lumbar spine X-rays. While CT scans provide clear images, they are generally taken in a supine position, significantly affecting the results, and are also costly and involve high radiation exposure, making them unsuitable for screening. In recent years, the continuous development of deep learning technology has provided a new approach to lumbar spondylolisthesis detection. Training and recognizing lumbar spine X-ray images using deep learning models can achieve faster and more accurate detection, providing reliable support for the imaging diagnosis of vertebral spondylolisthesis. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a method and device for detecting vertebral slippage based on lumbar spine X-ray films. To address the aforementioned technical problems, the first aspect of this invention discloses a method for detecting vertebral slippage based on lumbar spine X-ray films, the method comprising: S1, acquire digital images of lumbar spine X-ray films; S2, preprocess the digital image of the lumbar spine X-ray to obtain preprocessed digital images of the lumbar spine anteroposterior and lateral X-rays; S3, the preprocessed lumbar spine X-ray digital image is detected to obtain the vertebral slippage detection result of the lumbar spine X-ray.

[0004] As an optional implementation, in the first aspect of the present invention, the preprocessing of the lumbar spine X-ray digital image to obtain a preprocessed lumbar spine X-ray digital image includes: S21, perform grayscale normalization on the digital image of the lumbar spine X-ray to obtain a normalized digital image of the lumbar spine X-ray; S22, Denoise the normalized lumbar spine X-ray digital image to obtain a denoised lumbar spine X-ray digital image; S23, the denoised lumbar spine X-ray digital image is augmented to obtain an augmented lumbar spine X-ray digital image; S24, the region of interest is extracted from the expanded lumbar spine X-ray digital image to obtain a preprocessed lumbar spine X-ray digital image.

[0005] As an optional implementation, in the first aspect of the present invention, the step of denoising the normalized lumbar spine X-ray digital image to obtain a denoised lumbar spine X-ray digital image includes: S221, Preset maximum number of iterations Scale factor Filtering direction ; S222, perform multi-scale decomposition on the normalized lumbar spine X-ray digital image to obtain the low-frequency contour component and the high-frequency texture noise component of each layer. S223, Process the high-frequency texture noise component of each layer to obtain the bone texture features of each layer; S224, fuse the bone texture features of each layer to obtain global high-frequency texture information; S225, regarding the global high-frequency texture information and the... The low-frequency contour components of the layers are fused to obtain fused image information; S226, the fused image information is compensated to obtain a denoised lumbar spine X-ray digital image.

[0006] As an optional implementation, in the first aspect of the present invention, the step of detecting the preprocessed lumbar spine X-ray digital image to obtain the lumbar spine X-ray vertebral slippage detection result includes: S31, perform key point detection on the preprocessed lumbar spine X-ray digital image to obtain key point information; the key point information includes the midpoint of the superior endplate of each vertebra, the midpoint of the inferior endplate of each vertebra, the anterior superior angle of each vertebra, the anterior inferior angle of each vertebra, the posterior superior angle of each vertebra, and the posterior inferior angle of each vertebra. S32, Process the key point information to obtain the vertebral endplate linear fitting information; S33, The linear fitting information of the vertebral endplate is processed to obtain the vertebral slippage detection result of the lumbar spine X-ray.

[0007] As an optional implementation, in the first aspect of the present invention, the key point detection of the preprocessed lumbar spine X-ray digital image to obtain key point information includes: S311, Construct a multi-scale feature fusion keypoint detection network model; The multi-scale feature fusion key point detection network model includes a backbone feature extraction layer, a multi-scale feature fusion layer, and a key point heatmap prediction layer. S312, using the skeletal feature extraction layer, the preprocessed lumbar spine X-ray digital image is processed to obtain first feature information, second feature information and third feature information; S313, using a multi-scale feature fusion layer, the first feature information, the second feature information, and the third feature information are fused to obtain fused feature information; S314, using the key point heatmap prediction layer, the fused feature information is processed to obtain key point information.

[0008] As an optional implementation, in the first aspect of the present invention, processing the key point information to obtain vertebral endplate linear fitting information includes: S321, The key point information is processed to obtain the endpoints of the superior endplate of each vertebra and the endpoints of the inferior endplate of each vertebra. S322, Process the endpoints of the superior endplate of each vertebral body to obtain a straight line of the superior endplate of each vertebral body; S323, Process the endpoints of the inferior endplate of each vertebral body to obtain a straight line of the inferior endplate of each vertebral body; S324, integrate the upper endplate line of each vertebra and the upper endplate line of each vertebra to obtain vertebral endplate line fitting information.

[0009] As an optional implementation, in the first aspect of the present invention, processing the linear fitting information of the vertebral endplate to obtain the vertebral slippage detection result of the lumbar X-ray film includes: S331, Process the fitting information of the vertebral endplate straight line to obtain the parallelism deviation between the lower endplate straight line and the upper endplate straight line; S332, Based on the parallelism deviation, the linear fitting information of the vertebral endplate is corrected to obtain the corrected linear information of the vertebral endplate; S333, The straightness information of the corrected vertebral endplate is processed to obtain the horizontal displacement distance value; S334, The horizontal displacement distance value is processed to obtain the vertebral slippage detection result of the lumbar spine X-ray.

[0010] A second aspect of this invention discloses a vertebral slippage detection device based on lumbar spine X-ray films, the device comprising: The data acquisition module is used to acquire digital images of lumbar spine X-ray films; The preprocessing module is used to preprocess the digital images of the lumbar spine X-ray film to obtain preprocessed digital images of the lumbar spine anteroposterior and lateral X-ray films. The detection module is used to detect the preprocessed digital images of the lumbar spine X-ray to obtain the vertebral slippage detection results of the lumbar spine X-ray.

[0011] As an optional implementation, in the second aspect of the present invention, the preprocessing of the lumbar spine X-ray digital image to obtain a preprocessed lumbar spine X-ray digital image includes: S21, perform grayscale normalization on the digital image of the lumbar spine X-ray to obtain a normalized digital image of the lumbar spine X-ray; S22, Denoise the normalized lumbar spine X-ray digital image to obtain a denoised lumbar spine X-ray digital image; S23, the denoised lumbar spine X-ray digital image is augmented to obtain an augmented lumbar spine X-ray digital image; S24, the region of interest is extracted from the expanded lumbar spine X-ray digital image to obtain a preprocessed lumbar spine X-ray digital image.

[0012] As an optional implementation, in the second aspect of the present invention, the step of denoising the normalized lumbar spine X-ray digital image to obtain a denoised lumbar spine X-ray digital image includes: S221, Preset maximum number of iterations Scale factor Filtering direction ; S222, perform multi-scale decomposition on the normalized lumbar spine X-ray digital image to obtain the low-frequency contour component and the high-frequency texture noise component of each layer. S223, Process the high-frequency texture noise component of each layer to obtain the bone texture features of each layer; S224, fuse the bone texture features of each layer to obtain global high-frequency texture information; S225, regarding the global high-frequency texture information and the... The low-frequency contour components of the layers are fused to obtain fused image information; S226, the fused image information is compensated to obtain a denoised lumbar spine X-ray digital image.

[0013] As an optional implementation, in the second aspect of the present invention, the step of detecting the preprocessed lumbar spine X-ray digital image to obtain the lumbar spine X-ray vertebral slippage detection result includes: S31, perform key point detection on the preprocessed lumbar spine X-ray digital image to obtain key point information; the key point information includes the midpoint of the superior endplate of each vertebra, the midpoint of the inferior endplate of each vertebra, the anterior superior angle of each vertebra, the anterior inferior angle of each vertebra, the posterior superior angle of each vertebra, and the posterior inferior angle of each vertebra. S32, Process the key point information to obtain the vertebral endplate linear fitting information; S33, The linear fitting information of the vertebral endplate is processed to obtain the vertebral slippage detection result of the lumbar spine X-ray.

[0014] As an optional implementation, in the second aspect of the present invention, the key point detection of the preprocessed lumbar spine X-ray digital image to obtain key point information includes: S311, Construct a multi-scale feature fusion keypoint detection network model; The multi-scale feature fusion key point detection network model includes a backbone feature extraction layer, a multi-scale feature fusion layer, and a key point heatmap prediction layer. S312, using the skeletal feature extraction layer, the preprocessed lumbar spine X-ray digital image is processed to obtain first feature information, second feature information and third feature information; S313, using a multi-scale feature fusion layer, the first feature information, the second feature information, and the third feature information are fused to obtain fused feature information; S314, using the key point heatmap prediction layer, the fused feature information is processed to obtain key point information.

[0015] As an optional implementation, in the second aspect of the present invention, processing the key point information to obtain vertebral endplate linear fitting information includes: S321, The key point information is processed to obtain the endpoints of the superior endplate of each vertebra and the endpoints of the inferior endplate of each vertebra. S322, Process the endpoints of the superior endplate of each vertebral body to obtain a straight line of the superior endplate of each vertebral body; S323, Process the endpoints of the inferior endplate of each vertebral body to obtain a straight line of the inferior endplate of each vertebral body; S324, integrate the upper endplate line of each vertebra and the upper endplate line of each vertebra to obtain vertebral endplate line fitting information.

[0016] As an optional implementation, in the second aspect of the present invention, processing the linear fitting information of the vertebral endplate to obtain the vertebral slippage detection result of the lumbar X-ray film includes: S331, Process the fitting information of the vertebral endplate straight line to obtain the parallelism deviation between the lower endplate straight line and the upper endplate straight line; S332, Based on the parallelism deviation, the linear fitting information of the vertebral endplate is corrected to obtain the corrected linear information of the vertebral endplate; S333, The straightness information of the corrected vertebral endplate is processed to obtain the horizontal displacement distance value; S334, The horizontal displacement distance value is processed to obtain the vertebral slippage detection result of the lumbar spine X-ray.

[0017] A third aspect of this invention discloses another vertebral slippage detection device based on lumbar spine X-ray films, the device comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute some or all of the steps in the lumbar spine X-ray-based vertebral slippage detection method disclosed in the first aspect of the present invention.

[0018] The fourth aspect of the present invention discloses a computer-storable medium storing computer instructions, which, when invoked, are used to execute some or all of the steps in the vertebral slippage detection method based on lumbar X-ray films disclosed in the first aspect of the present invention.

[0019] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: This invention employs an end-to-end multi-scale feature fusion network architecture, enabling fully automated processing from pre-processed images to precise keypoint localization without manual intervention. Compared to traditional manual keypoint annotation or semi-automated detection methods, this significantly improves detection efficiency. Simultaneously, through model pruning and quantization optimization, the model's inference speed is enhanced while maintaining detection accuracy. It can quickly output the precise coordinates (including pixel and world coordinates) of six key points for each vertebra, providing efficient data support for subsequent steps such as endplate linear fitting, slippage rate calculation, and grade determination. This advantage effectively reduces the workload of clinicians in interpreting images and making measurements, avoids subjective errors in manual measurements, helps doctors quickly obtain reliable image diagnostic evidence, and improves clinical diagnostic efficiency. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic flowchart of a method for detecting vertebral slippage based on lumbar spine X-rays disclosed in an embodiment of the present invention; Figure 2 This is a schematic flowchart of another method for detecting vertebral slippage based on lumbar spine X-rays disclosed in an embodiment of the present invention; Figure 3This is a structural diagram of the vertebral slippage detection model based on lumbar spine X-ray disclosed in an embodiment of the present invention; Figure 4 This is a structural diagram of the LSK module disclosed in an embodiment of the present invention; Figure 5 This is a comparison chart of the experimental results of the method of the present invention and the comparative method; Figure 6 This is a schematic diagram of the structure of a vertebral slippage detection device based on lumbar spine X-ray disclosed in an embodiment of the present invention; Figure 7 This is a schematic diagram of another vertebral slippage detection device based on lumbar spine X-ray disclosed in an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0024] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0025] In all embodiments of the present invention, the variables involved in all computational expressions or mathematical functions are dimensionless before calculation. The values ​​of the independent variables in all computational expressions or mathematical functions in these embodiments conform to the reasonable requirements of the input range of the computational expression or mathematical function, ensuring that the computational expression or mathematical function can be calculated smoothly without violating physical laws or mathematical rules.

[0026] This invention discloses a method and apparatus for detecting vertebral slippage based on lumbar spine X-ray films. The method includes acquiring digital images of lumbar spine X-ray films; preprocessing the digital images of lumbar spine X-ray films to obtain preprocessed digital images of anteroposterior and lateral lumbar spine X-ray films; and detecting the preprocessed digital images of lumbar spine X-ray films to obtain the vertebral slippage detection result of the lumbar spine X-ray films. This invention completes the fully automated processing from preprocessed images to precise localization of key points, significantly improving detection efficiency compared to traditional manual annotation of key points or semi-automated detection methods. Detailed descriptions follow.

[0027] Example 1 Please see Figure 1 , Figure 1 This is a schematic flowchart of a method for detecting vertebral slippage based on lumbar spine X-ray films, as disclosed in an embodiment of the present invention. Figure 1 The described method for detecting vertebral slippage based on lumbar spine X-rays is applied in the field of image processing and application technology, and the embodiments of this invention are not limited thereto. Figure 1 As shown, this method for detecting vertebral slippage based on lumbar spine X-rays may include the following operations: S1, acquire digital images of lumbar spine X-ray films; Digital images of lumbar spine X-rays are a dataset of X-ray images of lumbar isthmus fracture lesions. S2, preprocess the digital image of the lumbar spine X-ray to obtain preprocessed digital images of the lumbar spine anteroposterior and lateral X-rays; S3, the preprocessed lumbar spine X-ray digital image is detected to obtain the vertebral slippage detection result of the lumbar spine X-ray.

[0028] Optionally, the preprocessing of the lumbar spine X-ray digital image to obtain a preprocessed lumbar spine X-ray digital image includes: S21, perform grayscale normalization on the digital image of the lumbar spine X-ray to obtain a normalized digital image of the lumbar spine X-ray; In the formula, Pixels of a digital image of a lumbar spine X-ray grayscale value, , These represent the maximum and minimum grayscale values ​​of the digital image of the lumbar spine X-ray. To normalize the grayscale values ​​of pixels in a digital image of a lumbar spine X-ray film; S22, Denoise the normalized lumbar spine X-ray digital image to obtain a denoised lumbar spine X-ray digital image; S23, the denoised lumbar spine X-ray digital image is augmented to obtain an augmented lumbar spine X-ray digital image; in, For denoising digital images of lumbar spine X-rays The flattened pixel sequence, For the first Step spatial neighborhood sampling window, For two-dimensional image pixel flattening operator, for The height dimension, for The width dimension; For the first The frequency domain sequence of the steps, where F represents the two-dimensional discrete Fourier transform; in, The spatial domain time-invariant hidden state transition matrix characterizes the spatial correlation features of the cone pixels. Only the matrix positions corresponding to adjacent pixels are assigned minimum non-zero values, while the positions corresponding to distant pixels are set to 0. At the same time, a diagonal exponential decay constraint is applied. , For vector dimensions, The input-mapped matrix maps pixel inputs to the latent feature space. The spatial output mapping matrix is ​​used to restore the latent features to spatially enhanced image features. The spatial residual pass-through matrix preserves the basic anatomical information of the original denoised image. for The time-space hidden state vector; for Texture enhancement features output by the temporal-space branch. , , Orthogonal normalized random initialization is used. This represents the Hadamard product operation.

[0029] In the formula, , , Orthogonal normalized random initialization is used. Frequency domain output characteristics; in, To enhance the image, The adaptive fusion weight coefficients range from 0.4 to 0.6. To ensure fidelity, a compensation factor is used, ranging from 0.1 to 0.3. express Gradient characteristics; This invention utilizes the steady-state output characteristics of spatial branching. With frequency domain branch steady-state output characteristics Weighted fusion is performed to evenly integrate vertebral spatial contours, anatomical structures, trabecular microtexture, and high-frequency detail information, achieving complementary fusion of spatial and frequency domain features. A gradient compensation term is introduced, using coefficients... Adjusting the compensation level will preserve the gradient information of the denoised image. The image is superimposed onto the fusion result to accurately compensate for the edge details lost during the fusion process, while also constraining the image's anatomical structure to prevent distortion.

[0030] Total loss: Hidden alignment loss: In the formula, , For spatial and frequency domain learnable weights, =1, ; Enhanced fidelity loss: Where the subscript F denotes the Frobenius norm of the matrix, express The gradient characteristics are obtained; through multi-loss joint constraint iteration, the optimal fixed parameters and steady-state time series output characteristics are gradually converged to complete the model shaping.

[0031] Mean extraction: in, This indicates the calculation of the expected value. Joint Gaussian distribution: in, , , , for and covariance; in, To The new gradient characteristics obtained after displacement perturbation. To The new gradient characteristics obtained after phase perturbation and As weight, + =1; Mahalanobis distance filtering: in As a preset threshold, during the training phase, the Mahalanobis distance of all real lumbar spondylolisthesis images relative to the normal baseline image distribution is calculated in batches. The maximum Mahalanobis distance of all real pathological samples is counted, and a reasonable pathological fluctuation margin is reserved based on this as a benchmark to finally determine the steady-state compliance threshold.

[0032] For those filtered by Mahalanobis distance Processing: in, To expand the digital images of lumbar spine X-rays.

[0033] S24, the region of interest is extracted from the expanded lumbar spine X-ray digital image to obtain a preprocessed lumbar spine X-ray digital image.

[0034] Based on the grayscale distribution characteristics of the lumbar vertebral bodies, an adaptive local threshold segmentation algorithm is employed to accurately distinguish high-grayscale skeletal vertebral body regions from low-grayscale background and soft tissue regions, initially identifying the core target region of the lumbar spine. Building upon this, morphological opening and closing operations are introduced as joint constraints to remove isolated noise points and small pseudo-edges in the image, fill in tiny voids within the vertebral body regions, and optimize the contour integrity of the target region. Finally, combined with prior anatomical size constraints of the lumbar spine, trunk debris and invalid background regions are eliminated, accurately selecting the core region of interest for the lumbar vertebral bodies and spondylolisthesis lesions. This method can adaptively adapt to expanded samples with different degrees of slippage and different vertebral body morphologies, effectively filtering out invalid background interference, preserving complete lesion features and skeletal texture details, and providing high-quality standardized preprocessed images for subsequent model training and lesion recognition.

[0035] Optionally, the step of denoising the normalized lumbar spine X-ray digital image to obtain a denoised lumbar spine X-ray digital image includes: S221, Preset maximum number of iterations Scale factor Filtering direction ; Preset Scale factor Filtering direction ; S222, perform multi-scale decomposition on the normalized lumbar spine X-ray digital image to obtain the low-frequency contour component and the high-frequency texture noise component of each layer. In the formula, For the first Low-frequency profile components of the layer, For the first High-frequency texture noise components of the layer. , These are the low-pass and high-pass filter coefficients, which can be constructed based on DB4 wavelets; initial iteration state. , For the first The k-th local pixel value in the low-frequency contour component of the layer that participates in the convolution operation, the k-th local pixel value. A local window of pixels in a two-dimensional image is unfolded into a one-dimensional sequence, where k represents the pixel position index in this one-dimensional sequence. The wavelet interval downsampling offset, initial conditions ; S223, Process the high-frequency texture noise component of each layer to obtain the bone texture features of each layer; In the formula, ; For the first Layer, corresponding discrete angle skeletal texture features Using an anisotropic Gaussian kernel function, targeted texture enhancement is performed on four typical anatomical angles: In the formula, These are the local two-dimensional spatial coordinates centered on the current pixel. The Gaussian scaling smoothing factor; S224, fuse the bone texture features of each layer to obtain global high-frequency texture information; in, For global high-frequency texture information, , This is the weighting adjustment coefficient; S225, regarding the global high-frequency texture information and the... The low-frequency contour components of the layers are fused to obtain fused image information; in, To fuse image information, For a preset local neighborhood window, the present invention fixes the local neighborhood window size to 3×3. It is a minimal smoothing constant; S226, the fused image information is compensated to obtain a denoised lumbar spine X-ray digital image.

[0036] In the formula To denoise digital images of lumbar spine X-rays, This is the gray-level gradient operator (differential operator). This is a vertebral body region indicator function that performs detailed compensation only on the core regions of the vertebral body and endplate, leaving the background region unprocessed. The compensation coefficients are... .

[0037] To address the edge details lost during the denoising process in the vertebral region, and to avoid endplate blurring and keypoint offset issues caused by over-smoothing.

[0038] Optionally, the step of detecting the preprocessed lumbar spine X-ray digital image to obtain the lumbar spine X-ray vertebral slippage detection result includes: S31, perform key point detection on the preprocessed lumbar spine X-ray digital image to obtain key point information; the key point information includes the midpoint of the superior endplate of each vertebra, the midpoint of the inferior endplate of each vertebra, the anterior superior angle of each vertebra, the anterior inferior angle of each vertebra, the posterior superior angle of each vertebra, and the posterior inferior angle of each vertebra. S32, Process the key point information to obtain the vertebral endplate linear fitting information; S33, The linear fitting information of the vertebral endplate is processed to obtain the vertebral slippage detection result of the lumbar spine X-ray.

[0039] Optionally, the step of performing key point detection on the preprocessed lumbar spine X-ray digital image to obtain key point information includes: S311, Construct a multi-scale feature fusion keypoint detection network model; The multi-scale feature fusion key point detection network model includes a backbone feature extraction layer, a multi-scale feature fusion layer, and a key point heatmap prediction layer. S312, using the skeletal feature extraction layer, the preprocessed lumbar spine X-ray digital image is processed to obtain first feature information, second feature information and third feature information; The first, second, and third feature information correspond to shallow details, mid-level transitions, and deep semantic features, respectively. The shallow features preserve the texture details of the vertebral body edges, while the deep features focus on the overall structural semantics of the vertebral body, providing rich and layered basic feature materials for subsequent multi-scale fusion.

[0040] S313, using a multi-scale feature fusion layer, the first feature information, the second feature information, and the third feature information are fused to obtain fused feature information; Cross-scale stitching and weighted fusion are performed to obtain fused feature information; S314, using the key point heatmap prediction layer, the fused feature information is processed to obtain key point information.

[0041] The key locations of the vertebral body are predicted pixel by pixel using a heatmap probability mapping method, and the information of six preset key points of the vertebral body is accurately output. Optionally, processing the key point information to obtain vertebral endplate linear fitting information includes: S321, The key point information is processed to obtain the endpoints of the superior endplate of each vertebra and the endpoints of the inferior endplate of each vertebra. Based on the coordinate information of various key points of the vertebral body output by the network, and combined with the prior rules of lumbar vertebral anatomy and geometry, the effective endpoint coordinates corresponding to the superior and inferior endplates of each vertebral body are screened and calculated. Abnormal interference points are eliminated by constraining the position of key points, and the positions of the two endpoints corresponding to the edge of the endplate are accurately matched. S322, Process the endpoints of the superior endplate of each vertebral body to obtain a straight line of the superior endplate of each vertebral body; For the coordinates of the two ends of the superior endplate of each vertebra, the least squares method is used to fit a straight line and construct a straight line equation of the superior endplate that fits the true edge morphology of the vertebra.

[0042] S323, Process the endpoints of the inferior endplate of each vertebral body to obtain a straight line of the inferior endplate of each vertebral body; The coordinates of the two endpoints of the inferior endplate of each vertebra are fitted with a least-squares straight line to obtain the corresponding straight line of the inferior endplate of each vertebra.

[0043] S324, integrate the upper endplate line of each vertebra and the upper endplate line of each vertebra to obtain vertebral endplate line fitting information.

[0044] By integrating and associating the superior endplate line and the inferior endplate line corresponding to a single vertebral body, a geometric correspondence between two straight lines for a single vertebral body is established. The endplate line parameters of all vertebral bodies are uniformly included to form a complete set of vertebral endplate line fitting information.

[0045] Optionally, the processing of the linear fitting information of the vertebral endplate to obtain the vertebral slippage detection result of the lumbar spine X-ray includes: S331, Process the fitting information of the vertebral endplate straight line to obtain the parallelism deviation between the lower endplate straight line and the upper endplate straight line; Based on the integrated vertebral endplate straight line fitting information, the slope parameters of the upper and lower endplate straight lines of each vertebra are solved separately, and the parallelism deviation of the two sets of straight lines is calculated by the slope difference.

[0046] S332, Based on the parallelism deviation, the linear fitting information of the vertebral endplate is corrected to obtain the corrected linear information of the vertebral endplate; Based on the calculated parallelism deviation value and combined with the prior knowledge of parallel anatomy of normal lumbar vertebral endplates, adaptive correction is performed on the endplate straight line with deviation to correct the straight line offset error caused by vertebral slippage and angular distortion, so as to obtain regular and standard corrected vertebral endplate straight line information, eliminate deformation interference, and ensure the accuracy of subsequent displacement calculation.

[0047] S333, The straightness information of the corrected vertebral endplate is processed to obtain the horizontal displacement distance value; Based on the corrected standard vertebral endplate linear information, a coordinate system for calculating vertebral horizontal displacement is constructed. By matching the horizontal projection offset of adjacent vertebral endplates, the horizontal slip distance of the diseased vertebra relative to the superior vertebra is accurately calculated, and a quantified horizontal displacement distance value is obtained, thus achieving a precise numerical characterization of lumbar spondylolisthesis displacement.

[0048] S334, The horizontal displacement distance value is processed to obtain the vertebral slippage detection results of the lumbar spine X-ray, including: Using a vertebral slippage detection model, the horizontal displacement distance value is processed to obtain the vertebral slippage detection results of the lumbar spine X-ray. Specifically, the quantitative vertebral horizontal displacement distance value is input into the trained vertebral slippage detection model. Combined with the clinical slippage grading criteria, the displacement data is subjected to feature analysis and grade determination. Finally, the vertebral slippage detection results of the lumbar spine X-ray are output, including the slippage displacement value, the degree of slippage distortion and the corresponding slippage grade, thus completing the detection.

[0049] The vertebral slippage detection model includes: The entire architecture is built on YOLOv8, including the Stage4 module in Backbone. An LSKP dynamic kernel selection sub-block is added after the C2f layer. The LK Selection sub-block on the right can dynamically adjust the receptive field of the network as needed. It consists of a sequence of a fully connected layer, a GELU activation, an LSKP, and a second fully connected layer.

[0050] The LSKP module is built on the LSK structure [17,25,74] and has three main building blocks. Two of these modules use the LSK feature calculation method, while a simpler and more efficient feature extraction method is added to the smallest convolutional kernel.

[0051] First, global spatial information is compressed into a single channel descriptor. This is achieved by generating channel-level statistics using global average pooling. Formally, the statistics... Through its spatial dimensions shrink Generate, make The Each element is calculated as follows: Excitation: To utilize the information aggregated in the excitation operation, a second operation is performed to fully capture channel-level dependencies. To achieve this, the function must meet two criteria: first, it must be flexible (specifically, it must be able to learn interactions between non-linear channels); second, it must learn a non-mutually exclusive relationship, as we want to ensure that multiple channels are allowed to be emphasized (rather than performing a hot activation). To meet these criteria, the sigmoid activation function is chosen: in, For ReLU

[63] functions, and To limit model complexity and aid generalization, a bottleneck is formed around the nonlinearity with two fully connected (FC) layers: a dimensionality reduction layer with a dimensionality reduction ratio r, a ReLU layer, and then a dimensionality augmentation layer returning to the channel dimension of the transformed output U. The final output is obtained by rescaling U with activation functions s. in, and It is an indicator quantity and feature mapping Channel-level multiplication between them.

[0052] Example 2 Lumbar spondylolisthesis is a common orthopedic condition. Lumbar spine X-rays, as a routine diagnostic tool, have certain limitations. Mild lumbar spondylolisthesis is often not clearly visible on X-rays, leading to diagnostic difficulties. Furthermore, the projection angle and image quality of X-rays can further blur the image, increasing diagnostic uncertainty. Therefore, mild lumbar spondylolisthesis is easily missed on lumbar spine X-rays. While CT scans provide clear images, they are generally taken in a supine position, significantly affecting the results, and are also costly and involve high radiation exposure, making them unsuitable for screening. In recent years, the continuous development of deep learning technology has provided a new approach to lumbar spondylolisthesis detection. Training and recognizing lumbar spine X-ray images using deep learning models can achieve faster and more accurate detection, providing reliable support for the imaging diagnosis of vertebral spondylolisthesis. Please refer to [link to relevant documentation]. Figure 2 , Figure 2 This is a schematic flowchart of another method for detecting vertebral slippage based on lumbar spine X-rays disclosed in an embodiment of the present invention. Figure 2 The described method for detecting vertebral slippage based on lumbar spine X-rays is applied in the field of image processing and application technology, and the embodiments of this invention are not limited thereto. It includes: an image input module, a preprocessing module, a deep learning model, a post-processing module, and an output module.

[0053] The entire architecture is built on YOLOv8. As shown in the Stage4 module of Backbone, we added an LSKP dynamic kernel selection sub-block after the C2f layer. The LK Selection sub-block on the right can dynamically adjust the receptive field of the network as needed. It consists of a sequence of a fully connected layer, a GELU activation, an LSKP, and a second fully connected layer. Figure 3 This is a structural diagram of the vertebral slippage detection model based on lumbar spine X-ray disclosed in an embodiment of the present invention; The LSKP module is built on the recently popular LSK architecture [17,25,74] and has three main building blocks. Two of these modules use the LSK feature calculation method, while a simpler and more efficient feature extraction method [se] is added to the smallest convolutional kernel.

[0054] Table 1 compares the theoretical efficiency and mAP of a representative example of expanding a single large-depth kernel into a sequence with a given channel size of 512. This represents a 5.8% improvement in mAP for the 50-95 range with only a small increase in the number of parameters. k: kernel size; d: expansion.

[0055] Table 1. Comparison of theoretical efficiency and mAP Figure 4This is a structural diagram of the LSK module disclosed in an embodiment of the present invention; the LSKP module used is a complex structure composed of a series of kernel convolutions and a spatial kernel selection mechanism. As shown in the figure, the module we use consists of three parts, namely 3 3, 5 5 and 7 A convolution kernel of 7. It's worth noting that 3... The convolution kernel of 3 uses a feature calculation method different from the other two, and the ingenious application of this method is the key to our improvement in saliency performance.

[0056] Through a series of experiments and observations, we found that using convolutional kernels of the same size and similar feature extraction methods has certain limitations, failing to fully extract various features from images. Therefore, based on this observation, we first expanded the number of convolutional kernels from two to three, with these kernels having different sizes. This change produced some positive results in our experiments; convolutional kernels of different sizes can extract different features, and concatenating these features can improve accuracy.

[0057] Small convolution kernel: First, global spatial information is compressed into a single channel descriptor. This is achieved by generating channel-level statistics using global average pooling. Formally, the statistics... Through its spatial dimensions shrink Generate, make The Each element is calculated as follows: To utilize the information aggregated in the squash operation, a second operation is performed to fully capture channel-level dependencies. To achieve this, the function must meet two criteria: first, it must be flexible (specifically, it must be able to learn interactions between non-linear channels); second, it must learn a non-mutually exclusive relationship, as we want to ensure that multiple channels are allowed to be emphasized (rather than performing a hot activation). To meet these criteria, the sigmoid activation function is chosen: in, For ReLU

[63] functions, and To limit model complexity and aid generalization, a bottleneck is formed around the nonlinearity with two fully connected (FC) layers: a dimensionality reduction layer with a dimensionality reduction ratio r, a ReLU layer, and then a dimensionality augmentation layer returning to the channel dimension of the transformed output U. The final output is obtained by rescaling U with activation functions s. in, and It is an indicator quantity and feature mapping Channel-level multiplication between them.

[0058] Kernel convolution: Following the convolution method of LSK, the originally large kernel convolution is decomposed into a series of depthwise convolutions with gradually increasing kernel size and dilation rate. Specifically, for the i-th depthwise convolution in the series, the kernel size k, dilation rate d, and expansion of the receptive field RF are defined as shown in the formula: By increasing the kernel size and dilation rate, rapid expansion of the receptive field is ensured. At the same time, an upper limit is set on the dilation rate to ensure that dilated convolutions do not introduce gaps between feature maps.

[0059] This design has two main advantages. First, it explicitly generates multiple properties with various large receptive fields, making the subsequent kernel selection process more flexible. Second, sequential decomposition is significantly more effective than directly applying a single large kernel. As shown in Table 2, decomposition can greatly improve accuracy by adding only a small number of parameters while maintaining the same theoretical receptive field.

[0060] To capture rich contextual information from different ranges of the input X, a series of decomposed depthwise convolutions with different receptive fields are introduced. This approach performs exceptionally well in our experiments, providing the model with more comprehensive and richer feature representations, thus achieving better performance across various tasks.

[0061] in It is related to the core and expansion Depthwise convolution, assuming we have Each kernel is composed of a 1×1 convolutional layer. Further processing: , for i in Channel blending of each spatial feature vector is allowed. Based on this, a kernel selection mechanism that dynamically selects kernels for various targets based on the obtained multi-scale features is described below.

[0062] Spatial kernel selection: To enhance the network's ability to focus on the most relevant spatial context regions for target detection, a spatial selection mechanism is used to select feature maps from convolutional kernels of different scales. First, features obtained from different kernels with different receptive domains are concatenated: Then, by channel-based averaging and max pooling (denoted as...) and ) Applied to Effectively extract spatial relationships: , in, and These are the average and maximum merged spatial feature descriptors. To allow information exchange between different spatial descriptors, the spatial pooling features are concatenated and a convolutional layer is used. The aggregated features (with 2 channels) are converted into N spatial maps: For each spatial map, Applying the sigmoid activation function yields a single spatial selection mask for each decomposed large kernel: in, This represents the sigmoid activation function. Then, the features in the decomposed large kernel sequence are weighted using the corresponding spatial selection mask and then combined with the convolutional layer. By fusing the features, we obtain feature S: The final output of the LSKP module is the element-wise product of the input characteristics X and S, and similarly in [17,18,25]: Table 2. Experimental comparison between the present invention and other methods Figure 5This is a comparison of experimental results between the method of this invention and a comparative method. The significant effects of this invention include a substantial improvement in the accuracy, efficiency, and stability of lumbar spondylolisthesis detection. By introducing an improved LSKP module, our model achieved excellent results on evaluation metrics. Compared to the traditional YOLOv8m model, our LSKP module improved precision from 0.983 to 0.994, recall from 0.882 to 0.941, mAPval50 from 0.963 to 0.99, and mAPval50-95 from 0.799 to 0.857 at the same GPU speed. These data reflect that our model has higher accuracy and stability in detecting lumbar spondylolisthesis, better captures the target area, and reduces the risk of missed and misdiagnosed cases. Simultaneously, our model achieves a higher level of detection performance while maintaining a fast GPU processing speed, providing more reliable support for clinical diagnosis.

[0063] Example 3 Please see Figure 6 , Figure 6 This is a schematic diagram of a vertebral slippage detection device based on lumbar spine X-ray films disclosed in an embodiment of the present invention. Figure 6 The described lumbar spine X-ray-based vertebral slippage detection device is applied in the field of image processing and application technology, and the embodiments of this invention are not limited thereto. Figure 6 As shown, the vertebral slippage detection device based on lumbar spine X-rays may include the following operations: S301, Data Acquisition Module, used to acquire digital images of lumbar spine X-ray films; S302, preprocessing module, used to preprocess the digital image of the lumbar spine X-ray film to obtain preprocessed digital images of the lumbar spine anteroposterior and lateral X-ray films; S303, Detection module, used to detect the preprocessed lumbar spine X-ray digital image to obtain the vertebral slippage detection result of the lumbar spine X-ray.

[0064] Example 4 Please see Figure 7 , Figure 7 This is a schematic diagram of another vertebral slippage detection device based on lumbar spine X-ray films disclosed in an embodiment of the present invention. Figure 7 The described lumbar spine X-ray-based vertebral slippage detection device is applied in the field of image processing and application technology, and the embodiments of this invention are not limited thereto. Figure 7 As shown, the vertebral slippage detection device based on lumbar spine X-rays may include the following operations: Memory 401 storing executable program code; Processor 402 coupled to memory 401; The processor 402 calls the executable program code stored in the memory 401 to execute the steps in the lumbar spine X-ray-based vertebral slippage detection method described in Embodiments 1 and 2.

[0065] Example 5 This invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program enables a computer to perform the steps in the lumbar spine X-ray-based vertebral slippage detection method described in Embodiments 1 and 2.

[0066] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0067] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0068] Finally, it should be noted that the vertebral slippage detection method and device based on lumbar spine X-ray disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting vertebral slippage based on lumbar spine X-ray films, characterized in that, The method includes: S1, acquire digital images of lumbar spine X-ray films; S2, preprocess the digital image of the lumbar spine X-ray to obtain preprocessed digital images of the lumbar spine anteroposterior and lateral X-rays; S3, the preprocessed lumbar spine X-ray digital image is detected to obtain the vertebral slippage detection result of the lumbar spine X-ray.

2. The method for detecting vertebral slippage based on lumbar spine X-rays according to claim 1, characterized in that, The preprocessing of the lumbar spine X-ray digital image to obtain a preprocessed lumbar spine X-ray digital image includes: S21, perform grayscale normalization on the digital image of the lumbar spine X-ray to obtain a normalized digital image of the lumbar spine X-ray; S22, Denoise the normalized lumbar spine X-ray digital image to obtain a denoised lumbar spine X-ray digital image; S23, the denoised lumbar spine X-ray digital image is augmented to obtain an augmented lumbar spine X-ray digital image; S24, the region of interest is extracted from the expanded lumbar spine X-ray digital image to obtain a preprocessed lumbar spine X-ray digital image.

3. The method for detecting vertebral slippage based on lumbar spine X-rays according to claim 2, characterized in that, The step of denoising the normalized lumbar spine X-ray digital image to obtain a denoised lumbar spine X-ray digital image includes: S221, Preset maximum number of iterations Scale factor Filtering direction ; S222, perform multi-scale decomposition on the normalized lumbar spine X-ray digital image to obtain the low-frequency contour component and the high-frequency texture noise component of each layer. S223, Process the high-frequency texture noise component of each layer to obtain the bone texture features of each layer; S224, fuse the bone texture features of each layer to obtain global high-frequency texture information; S225, regarding the global high-frequency texture information and the... The low-frequency contour components of the layers are fused to obtain fused image information; S226, the fused image information is compensated to obtain a denoised lumbar spine X-ray digital image.

4. The method for detecting vertebral slippage based on lumbar spine X-rays according to claim 1, characterized in that, The step of detecting the preprocessed lumbar spine X-ray digital image to obtain the lumbar spine X-ray vertebral slippage detection result includes: S31, perform key point detection on the preprocessed lumbar spine X-ray digital image to obtain key point information; the key point information includes the midpoint of the superior endplate of each vertebra, the midpoint of the inferior endplate of each vertebra, the anterior superior angle of each vertebra, the anterior inferior angle of each vertebra, the posterior superior angle of each vertebra, and the posterior inferior angle of each vertebra. S32, Process the key point information to obtain the vertebral endplate linear fitting information; S33, The linear fitting information of the vertebral endplate is processed to obtain the vertebral slippage detection result of the lumbar spine X-ray.

5. The method for detecting vertebral slippage based on lumbar spine X-rays according to claim 4, characterized in that, The key point detection of the preprocessed lumbar spine X-ray digital image to obtain key point information includes: S311, Construct a multi-scale feature fusion keypoint detection network model; The multi-scale feature fusion key point detection network model includes a backbone feature extraction layer, a multi-scale feature fusion layer, and a key point heatmap prediction layer. S312, using the skeletal feature extraction layer, the preprocessed lumbar spine X-ray digital image is processed to obtain first feature information, second feature information and third feature information; S313, using a multi-scale feature fusion layer, the first feature information, the second feature information, and the third feature information are fused to obtain fused feature information; S314, using the key point heatmap prediction layer, the fused feature information is processed to obtain key point information.

6. The method for detecting vertebral slippage based on lumbar spine X-ray films according to claim 4, characterized in that, The process of processing the key point information to obtain the linear fitting information of the vertebral endplate includes: S321, The key point information is processed to obtain the endpoints of the superior endplate of each vertebra and the endpoints of the inferior endplate of each vertebra. S322, Process the endpoints of the superior endplate of each vertebral body to obtain a straight line of the superior endplate of each vertebral body; S323, Process the endpoints of the inferior endplate of each vertebral body to obtain a straight line of the inferior endplate of each vertebral body; S324, integrate the upper endplate line of each vertebra and the upper endplate line of each vertebra to obtain vertebral endplate line fitting information.

7. The method for detecting vertebral slippage based on lumbar spine X-rays according to claim 4, characterized in that, The process of processing the linear fitting information of the vertebral endplate to obtain the vertebral slippage detection results of the lumbar spine X-ray includes: S331, Process the fitting information of the vertebral endplate straight line to obtain the parallelism deviation between the lower endplate straight line and the upper endplate straight line; S332, Based on the parallelism deviation, the linear fitting information of the vertebral endplate is corrected to obtain the corrected linear information of the vertebral endplate; S333, The straightness information of the corrected vertebral endplate is processed to obtain the horizontal displacement distance value; S334, The horizontal displacement distance value is processed to obtain the vertebral slippage detection result of the lumbar spine X-ray.

8. A device for detecting vertebral slippage based on lumbar spine X-ray films, characterized in that, The device includes: The data acquisition module is used to acquire digital images of lumbar spine X-ray films; The preprocessing module is used to preprocess the digital images of the lumbar spine X-ray film to obtain preprocessed digital images of the lumbar spine anteroposterior and lateral X-ray films. The detection module is used to detect the preprocessed digital images of the lumbar spine X-ray to obtain the vertebral slippage detection results of the lumbar spine X-ray.

9. A device for detecting vertebral slippage based on lumbar spine X-ray films, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the vertebral slippage detection method based on lumbar X-ray as described in any one of claims 1-7.

10. A computer-storable medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked, are used to execute the vertebral slippage detection method based on lumbar X-ray as described in any one of claims 1-7.

Images