Scoliosis Cobb angle determination method and device based on Mama, equipment and storage medium

Abstract

The embodiment of the invention provides a scoliosis Cobb angle determination method and device based on Mama, equipment and a storage medium. The method is applied to the scoliosis prediction field. The method comprises the following steps: acquiring image information, and preprocessing the acquired image information to obtain an original image; performing multi-level local feature extraction on the original image through an encoder to obtain a first feature map; performing global feature capture on the first feature map through a Mama module to obtain a second feature map; decoding the second feature map through a decoder to obtain a target feature map; generating, by a prediction module, a central heat map, an offset map and a centripetal vector map based on the target feature map; according to the central heat map, the offset map and the centripetal vector map, extracting central point coordinates of the centripetal body, and according to the central point coordinates of the centripetal body, determining four angular point coordinates of the centripetal body; the Cobb angle is determined through a geometric algorithm according to the coordinates of the four angular points of the cone, and the measurement precision of the Cobb angle is improved.

Description

Technical Field

[0001] The present application relates to the field of scoliosis prediction, and in particular to a Mamba-based scoliosis Cobb angle determination method, apparatus, device and storage medium. Background Art

[0002] Accurate assessment of scoliosis is crucial for clinical diagnosis and treatment decisions. Traditional methods rely on doctors manually measuring the Cobb angle on X-ray images. However, manual measurement is inefficient, time-consuming, and labor-intensive. The results are also easily influenced by the operator's experience and subjective judgment, resulting in poor repeatability and consistency, making it difficult to meet the needs of precision medicine.

[0003] To overcome the limitations of manual measurement methods, automated measurement technologies based on deep learning have emerged. For example, the Cobb angle measurement method based on convolutional neural networks can automatically extract local features such as vertebral texture and edges from X-ray images, which can be used to locate key points or segment spinal areas, and realize automatic calculation of the Cobb angle, thereby improving the efficiency and objectivity of the evaluation.

[0004] However, existing automated measurement methods based on convolutional neural networks or other single models are insufficient in capturing the overall structural information and long-distance dependencies of the spine, making it difficult for the model to accurately understand the overall bending pattern and limiting the accuracy of Cobb angle measurement. Summary of the Invention

[0005] The present application provides a Mamba-based scoliosis Cobb angle determination method, apparatus, device and storage medium to address the problem that the existing technology is insufficient in capturing the overall structural information of the spine and long-distance dependencies, resulting in limited Cobb angle measurement accuracy.

[0006] In a first aspect, the present application provides a method for determining the Cobb angle of scoliosis based on Mamba, the method comprising:

[0007] Acquiring image information, and preprocessing the acquired image information to obtain an original image;

[0008] Performing multi-level local feature extraction on the original image through an encoder to obtain a first feature map, wherein the encoder includes: a plurality of residual modules;

[0009] The global features of the first feature map are captured by the Mamba module to obtain the second feature map;

[0010] Decoding the second feature map through a decoder to obtain a target feature map, wherein the decoder includes: a main decoding branch and an auxiliary decoding branch;

[0011] Generate center heat map, offset map and centroid vector map based on target feature map through prediction module;

[0012] Extract the center point coordinates of the vertebra according to the center heat map, offset map and centripetal vector map, and determine the coordinates of the four corner points of the vertebra according to the center point coordinates of the vertebra;

[0013] The Cobb angle is determined by a geometric algorithm based on the coordinates of the four corner points of the vertebra.

[0014] Optionally, performing global feature capture on the first feature map by using a Mamba module to obtain a second feature map includes:

[0015] Performing feature serialization processing on the first feature map to obtain a first feature vector sequence;

[0016] The first eigenvector sequence is analyzed and processed by the structured state space model of the Mamba module to obtain a second eigenvector sequence;

[0017] The second feature vector sequence is mapped back to the two-dimensional feature map to obtain a second feature map.

[0018] Optionally, decoding the second feature map by a decoder to obtain a target feature map includes:

[0019] Upsampling and convolution operations are performed on the second feature map through the main decoding branch to obtain a third feature map;

[0020] The features of the middle layer of the encoder are received through the auxiliary decoding branch, and back-propagation optimization is performed through the auxiliary loss function to obtain the fourth feature map;

[0021] The third feature map and the fourth feature map are fused to obtain the target feature map.

[0022] Optionally, extracting the center point coordinates of the vertebra according to the central heat map, the offset map, and the centripetal vector map, and determining the coordinates of the four corner points of the vertebra according to the center point coordinates of the vertebra, includes:

[0023] Apply the non-maximum suppression algorithm to the center heat map, and select a preset number of peak points with the highest confidence scores in the processed center heat map. For each selected peak point, record its integer coordinates and confidence score on the target feature map;

[0024] Extract the corresponding offset vector from the offset map according to the integer coordinates of the peak point, and determine the coordinates of the center point of the vertebra according to the integer coordinates and the offset vector;

[0025] The eight corresponding vector components are extracted from the centripetal vector diagram according to the coordinates of the center point of the vertebral body, and the coordinates of the four corner points of the vertebral body are determined according to the coordinates of the center point of the vertebral body and the extracted vector components.

[0026] Optionally, determining the Cobb angle using a geometric algorithm based on the coordinates of the four corner points of the vertebral body includes:

[0027] Determine the direction line of each vertebra based on the coordinates of the four corner points of the vertebra;

[0028] Determine the angles between all pairs of vertebrae based on the direction lines of the vertebrae to obtain an angle matrix;

[0029] Determine the vertebral range corresponding to the main curve, and determine the Cobb angle of the main curve based on the angle matrix;

[0030] The vertebral range corresponding to the secondary curve is determined, and the Cobb angle of the secondary curve is determined by a geometric algorithm.

[0031] Optionally, performing back propagation optimization through an auxiliary loss function to obtain a fourth feature map includes:

[0032] Define the first loss function for the center heat map, which is used to measure the difference between the predicted center heat map and the true center heat map;

[0033] A second loss function is defined for the offset map, where the second loss function is used to measure the difference between the predicted offset map and the true offset map;

[0034] A third loss function is defined for the centripetal vector map, where the third loss function is used to measure the difference between the predicted centripetal vector map and the true centripetal vector map;

[0035] Perform weighted summation on the first loss function, the second loss function, and the third loss function to obtain the total loss;

[0036] Back propagation optimization is performed based on the total loss to obtain the fourth feature map.

[0037] In a second aspect, the present application provides a device for determining the Cobb angle of scoliosis based on Mamba, the device comprising:

[0038] An acquisition module, used for acquiring image information;

[0039] A first processing module, configured to preprocess the acquired image information to obtain an original image;

[0040] A second processing module is configured to perform multi-level local feature extraction on the original image through an encoder to obtain a first feature map, wherein the encoder includes: a plurality of residual modules;

[0041] The second processing module is further configured to perform global feature capture on the first feature map through a Mamba module to obtain a second feature map;

[0042] The second processing module is further configured to decode the second feature map through a decoder to obtain a target feature map, wherein the decoder includes: a main decoding branch and an auxiliary decoding branch;

[0043] The second processing module is further configured to generate a center heat map, an offset map, and a centripetal vector map based on the target feature map through the prediction module;

[0044] The second processing module is further configured to extract the coordinates of the center point of the vertebra according to the center heat map, the offset map and the centripetal vector map, and determine the coordinates of the four corner points of the vertebra according to the coordinates of the center point of the vertebra;

[0045] The second processing module is further configured to determine the Cobb angle using a geometric algorithm according to the coordinates of the four corner points of the vertebral body.

[0046] In a third aspect, the present application provides a Mamba-based scoliosis Cobb angle determination device, comprising:

[0047] Memory;

[0048] processor;

[0049] wherein the memory stores computer-executable instructions;

[0050] The processor executes the computer-executable instructions stored in the memory to implement the Mamba-based scoliosis Cobb angle determination method as described in the first aspect and various possible implementations of the first aspect.

[0051] In a fourth aspect, the present application provides a computer storage medium having a computer program stored thereon, which is executed by a processor to implement the Mamba-based scoliosis Cobb angle determination method as described in the first aspect and various possible implementations of the first aspect.

[0052] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the Mamba-based scoliosis Cobb angle determination method as described in the first aspect and various possible implementations of the first aspect.

[0053] The present application provides a Mamba-based method, apparatus, device, and storage medium for determining the Cobb angle of scoliosis. The method acquires image information and preprocesses the acquired image information to obtain an original image; performs multi-level local feature extraction on the original image using an encoder to obtain a first feature map; performs global feature capture on the first feature map using a Mamba module to obtain a second feature map; decodes the second feature map using a decoder to obtain a target feature map; generates a center heat map, an offset map, and a centripetal vector map based on the target feature map using a prediction module; extracts the center point coordinates of the vertebra based on the center heat map, offset map, and centripetal vector map, and determines the coordinates of the four corner points of the vertebra based on the center point coordinates of the vertebra; and determines the Cobb angle using a geometric algorithm based on the coordinates of the four corner points of the vertebra, thereby improving the accuracy of Cobb angle measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0055] Figure 1 A schematic flow chart of a method for determining the Cobb angle of scoliosis based on Mamba provided in an embodiment of the present application;

[0056] Figure 2 A schematic diagram of the structure of a Mamba-based scoliosis Cobb angle determination device provided in an embodiment of the present application;

[0057] Figure 3 This is a schematic structural diagram of a Mamba-based scoliosis Cobb angle determination device provided in an embodiment of the present application.

[0058] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0059] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions in this application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0060] The terms "first," "second," "third," "fourth," and so forth (if any) in the description and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequential sequence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein.

[0061] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described in this application as "exemplary" or "for example" should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0062] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0063] Figure 1 A flowchart of a method for determining the Cobb angle of scoliosis based on Mamba is provided in an embodiment of the present application. Figure 1 As shown, the method for determining the Cobb angle of scoliosis based on Mamba provided in this embodiment includes:

[0064] S1: Acquire image information, and preprocess the acquired image information to obtain an original image.

[0065] The preprocessing of the acquired image information to obtain the original image includes:

[0066] Data cropping: remove the brain, neck, and areas below the pelvis to ensure recognition stability;

[0067] Resizing: All input images are resized to a fixed size (e.g. 1024×512) to ensure uniform data format.

[0068] Image enhancement: Enhance images through random cropping, expansion, contrast and brightness adjustment, etc., to increase data diversity and reduce the risk of overfitting;

[0069] Standardization: Normalize pixel values and adjust their value range to [0, 1] to facilitate model processing.

[0070] S2: Perform multi-level local feature extraction on the original image through the encoder to obtain the first feature map.

[0071] The encoder includes multiple residual modules.

[0072] As you can understand, the original image enters the encoder part, and the initial convolution layer of the convolutional neural network is used to extract multi-level local features in the original image. The extracted features include key detail information such as the texture, edges, corners, etc. of the vertebral body.

[0073] In an optional embodiment, ResNet34 is used as the backbone network, and the initial convolutional layers include: Cov1, Cov2, Cov3, and Cov4, which perform multi-level local feature extraction on the original image. Specifically, shallow convolution (Cov1, Cov2) captures low-level local features in the original image, such as the edge contour of the vertebra, bone texture, and other basic visual elements. The above features help to accurately identify the vertebral boundary; as the network deepens, the receptive field gradually increases, and deep convolution (Cov3, Cov4) can learn more complex and abstract combined features, such as the corner points of the vertebra, the specific morphology of the vertebral plate, and other local structural information.

[0074] It is also understandable that the residual connection mechanism of ResNet effectively alleviates the gradient vanishing problem in deep network training, enabling the network to learn deeper and more discriminative local features.

[0075] Furthermore, after being processed by the backbone network, the original input image Among them, H0 is the height of the original image, W0 is the width of the original image, 3 is the number of channels, and it is converted into a high-dimensional feature map F∈R (H×W×C) ,in, C=512, the output high-dimensional feature map contains rich, multi-scale local detail information of the vertebral body.

[0076] The purpose of this step is to utilize the local feature extraction capability of convolutional neural networks to provide high-quality and refined local anatomical structure representation for subsequent global information modeling.

[0077] S3: The global features of the first feature map are captured through the Mamba module to obtain the second feature map.

[0078] Specifically, the first feature map is subjected to feature serialization processing to obtain a first feature vector sequence; the first feature vector sequence is analyzed and processed through the structured state space model of the Mamba module to obtain a second feature vector sequence; the second feature vector sequence is mapped back to the two-dimensional feature map to obtain a second feature map.

[0079] In an optional embodiment, feature serialization is performed on the first feature map to obtain a first feature vector sequence, including: flattening the H×W spatial dimensions into a sequence of length H×W, where each sequence element is a C-dimensional feature vector; and traversing the blocks on the feature map using a specific scanning order (such as raster scanning or Z-order scanning) to form a feature vector sequence. This method can retain the original spatial proximity.

[0080] In an optional embodiment, the first feature vector sequence is analyzed and processed by the structured state space model of the Mamba module to obtain the second feature vector sequence, including: the structured state space model of the Mamba module is processed by a hidden state h t To compress and transmit historical information, its basic dynamics can be expressed as:

[0081] x′ t =Bx t

[0082] h′ t =Ah t +x′ t

[0083] y t =Ch′ t

[0084] Among them, x t is the current input feature vector in the sequence, x′ t is the intermediate feature vector, h′ is the updated hidden state, y t is the current output feature vector in the sequence, and A, B, and C are the state space parameters.

[0085] Understandably, the state-space parameters can be dynamically adjusted based on the input. This means the model can selectively “remember” or “forget” information in the sequence based on what it currently sees, allowing it to more effectively focus on long-range dependencies relevant to the current task. Mamba further enhances the model’s ability to control information flow by incorporating a gating mechanism, enabling it to dynamically combine and filter information transmitted through the structured state-space model.

[0086] The Mamba module captures global features from the first feature map to generate the second feature map. This overcomes the shortcomings of traditional convolutional neural networks in understanding global context and capturing long-range dependencies. It effectively simulates the curvature of the entire spine, the relative positions and rotational relationships between vertebrae, and facilitates the accurate assessment of complex scoliosis patterns. The combination of convolutional neural networks and Mamba effectively integrates fine-grained local features with global structural information.

[0087] S4: Decode the second feature map through the decoder to obtain the target feature map.

[0088] The decoder includes a main decoding branch and an auxiliary decoding branch.

[0089] Specifically, the second feature map is upsampled and convolved through the main decoding branch to obtain the third feature map; the features of the middle layer of the encoder are received through the auxiliary decoding branch, and backpropagation optimization is performed through the auxiliary loss function to obtain the fourth feature map; the third feature map and the fourth feature map are fused to obtain the target feature map.

[0090] It can be understood that the main decoding branch gradually restores the spatial resolution through upsampling and convolution operations to generate the final feature map; the auxiliary decoding branch uses a self-supervised learning strategy to supervise the intermediate features and improve the feature expression ability.

[0091] In an optional embodiment, back propagation optimization is performed through an auxiliary loss function to obtain a fourth feature map, including: defining a first loss function for the center heat map, the first loss function is used to measure the difference between the predicted center heat map and the true center heat map, and the first loss function can be, for example, focal loss; defining a second loss function for the offset map, the second loss function is used to measure the difference between the predicted offset map and the true offset map, and the second loss function can be, for example, L1 loss; defining a third loss function for the centripetal vector map, the third loss function is used to measure the difference between the predicted centripetal vector map and the true centripetal vector map, and the third loss function can be, for example, L1 loss; performing weighted sum processing on the first loss function, the second loss function and the third loss function to obtain the total loss; performing back propagation optimization based on the total loss to obtain the fourth feature map.

[0092] S5: Generate center heat map, offset map and centroid vector map based on the target feature map through the prediction module.

[0093] In an optional embodiment, the target feature map is mapped to a single-channel heat map through a 1×1 convolution layer, and the probability value of each pixel being the target center is output to obtain a center heat map; the fused feature map is mapped to 2K channels (K is the number of key points) through a 1×1 convolution layer, and the x-axis and y-axis offsets of each key point relative to the center point are generated to obtain an offset map; the fused feature map is mapped to 2 channels through a 1×1 convolution layer, and the centroid vector of each pixel is generated, and the vector is normalized to obtain a centroid vector map.

[0094] S6: Extract the center point coordinates of the vertebra according to the center heat map, the offset map and the centripetal vector map, and determine the coordinates of the four corner points of the vertebra according to the center point coordinates of the vertebra.

[0095] Specifically, the central heat map is processed by applying the non-maximum suppression algorithm, and a preset number of peak points with the highest confidence scores are selected from the processed central heat map. For each selected peak point, its integer coordinates and confidence score on the target feature map are recorded; the corresponding offset vector is extracted from the offset map according to the integer coordinates of the peak point, and the center point coordinates of the vertebra are determined according to the integer coordinates and the offset vector; the corresponding eight vector components are extracted from the centripetal vector map according to the center point coordinates of the vertebra, and the coordinates of the four corner points of the vertebra are determined according to the center point coordinates of the vertebra and the extracted vector components.

[0096] In an optional embodiment, the center heat map H heat The non-maximum suppression (NMS) algorithm is applied to eliminate redundant responses and retain local peak points. In the heat map processed by NMS, the K (for example, K = 17, corresponding to 12 thoracic vertebrae + 5 lumbar vertebrae, a total of 17) peak points p with the highest confidence scores are selected. peak , for each selected peak point i (i=1,...,K), record its integer coordinates on the feature map and confidence score s i .

[0097] Furthermore, using the offset map O to H heat The integer coordinates obtained in the compensation are used to obtain the center point coordinates with sub-pixel accuracy. Specifically, according to the integer coordinates of the peak point Extract the corresponding offset vector (Δx i ,Δy i ). The exact center point coordinates (x c,i ,y c,i )The calculation process satisfies the following formula:

[0098]

[0099] Furthermore, using the centripetal vector graph V and the refined center point coordinates (x c,i ,y c,i ) to calculate the coordinates of the four corner points of each vertebra (e.g., upper left TL, upper right TR, lower left BL, lower right BR). According to the integer coordinates of the peak point or the coordinates of the center point, the corresponding 8 vector components are extracted from the centripetal vector map V. The extracted components represent the vectors pointing from the four corner points to the center point, which are recorded as:

[0100] (v tl,x,i ,v tl,y,i ),(v tr,x,i ,v tr,y,i ),(v bl,x,i ,v bl,y,i ),(vbr,x,i ,v br,y,i ).

[0101] Since the predicted vector points from the corner point to the center point, the coordinates of each corner point are obtained by subtracting the corresponding vector component from the coordinates of the center point. For example, the coordinates of the upper left corner point TL (x tl,i ,y tl,i ) is calculated in the following way:

[0102] x tl,i =x c,i -v tl,x,i

[0103] x tl,i =x c,i -v tl,x,i

[0104] Similarly, the coordinates of the upper right TR, lower left BL, and lower right BR corner points can be calculated:

[0105] x tr,i =x c,i -v tr,x,i ,y tr,i =y c,i -v tr,y,i

[0106] x bl,i =x c,i -v bl,x,i ,y bl,i =y c,i -v bl,y,i

[0107] x br,i =x c,i -v br,x,i ,y br,i =y c,i -v br,y,i

[0108] For each detected vertebral center point i, the four corner coordinates (x tl,i ,y tl,i ),(x tr,i ,y tr,i ),(x bl,i ,y bl,i ),(x br,i ,y br,i ) are combined as the final detection result of the key point for subsequent Cobb angle geometry calculation.

[0109] S7: Determine the Cobb angle using a geometric algorithm based on the coordinates of the four corner points of the vertebra.

[0110] Specifically, the direction line of each vertebra is determined based on the coordinates of the four corner points of the vertebra; the angles between all vertebrae pairs are determined based on the direction lines of the vertebrae to obtain an angle matrix; the vertebrae range corresponding to the main curve is determined, and the Cobb angle of the main curve is determined based on the angle matrix; the vertebrae range corresponding to the secondary curve is determined, and the Cobb angle of the secondary curve is determined through a geometric algorithm.

[0111] In an optional embodiment, for each detected vertebra k (k=1,...,N v , where N v is the number of vertebrae), using the coordinates of its four corner points (x tl,k ,y tl,k ),(x tr,k ,y tr,k ),(x bl,k ,y bl,k ),(x br,k ,y br,k ) Calculate the midpoints of its left and right edges to satisfy the following formula:

[0112]

[0113] Among them, p left_mid,k represents the midpoint of the left edge, p right_mid,k represents the midpoint of the right edge, p tl,k Indicates the coordinates of the corner point (x tl,k ,y tl,k ) coordinate vector, p bl,k Indicates the coordinates of the corner point (x bl,k ,y bl,k ) coordinate vector, p tr,k Represents the corner point coordinates (x tr,k ,y tr,k ) coordinate vector, p br,k Represents the corner point coordinates (x br,k ,y br,k ) coordinate vector. The line segment connecting these two midpoints defines the direction line of the vertebra k (or called the medial axis horizontal line), and its direction vector The calculation process satisfies the following formula:

[0114]

[0115] Furthermore, calculate the relationship between any two vertebrae i and j (i, j = 1, ..., N v ) direction vector and The angle θ between ij This angle represents the Cobb angle if vertebrae i and j serve as endplate vertebrae. The calculation formula is:

[0116]

[0117] Where · represents the vector dot product, ‖·‖ represents the Euclidean norm of the vector. Calculate the angles between all pairs of vertebrae to obtain an angle matrix Θ = [θ ij ].

[0118] Furthermore, at all calculated angles θ ij The angle with the largest absolute value is the Cobb angle θ of the main curve (main thoracic vertebra / thoracolumbar segment). MT :

[0119] θ MT =max i,j {θ ij}

[0120] Determine the two vertebral indices i that reach this maximum value * and j * , corresponding to the upper and lower endplate vertebral bodies of the main curve, respectively.

[0121] Furthermore, by analyzing the overall shape of the alignment of all vertebral center points (e.g., by evaluating their deviation from a straight line connecting the center points of the first and last vertebrae), it is determined whether the spine has a distinct hyperbolic shape ("S" shape) or a predominantly single curve ("C" shape).

[0122] For the single curve case: Cobb angle θ of the proximal thoracic vertebra PT It is usually calculated as the direction line of the first (uppermost) vertebra and the vertebral body at the upper end of the main curve i * The angle θ between the direction lines 1,i* Cobb angle θ of the thoracolumbar / lumbar spine TL Usually calculated as the last (lowest) vertebra N v The direction line and the end plate vertebral body at the lower end of the main curve * Angles between direction lines .

[0123] For hyperbolic cases: usually at the end plate vertebral body at the upper end of the main curve * Find the endplate vertebra of the proximal thoracic curve in the vertebral body above, and then * ) Find the maximum angle θ again ij To determine the Cobb angle θ of the proximal thoracic spine PT and its corresponding endplate vertebral body. Similarly, the endplate vertebral body of the thoracolumbar / lumbar curve will be at the endplate vertebral body at the lower end of the main curve. * The vertebral body below is searched by looking in this area (for example, vertebral body j * to N v ) Find the maximum angle to determine the Cobb angle θ of the thoracolumbar / lumbar spineTL and its corresponding endplate vertebral body.

[0124] Finally, according to the position of the upper and lower end plates in the spine, the three angles calculated are determined in order from top to bottom as the Cobb angle θ of the proximal thoracic spine. PT 、Cobb angle θ of the main curve (main thoracic vertebra / thoracolumbar segment) MT , Cobb angle θ of the thoracolumbar segment / lumbar spine TL After obtaining the accurate Cobb angle, it can provide doctors with a quantitative reference to assist in the diagnosis and treatment decision-making of cervical spondylosis.

[0125] The embodiment of the present application provides a Mamba-based method for determining the Cobb angle of scoliosis. The method obtains image information and preprocesses the obtained image information to obtain an original image; performs multi-level local feature extraction on the original image through an encoder to obtain a first feature map; performs global feature capture on the first feature map through a Mamba module to obtain a second feature map; decodes the second feature map through a decoder to obtain a target feature map; generates a center heat map, an offset map, and a centripetal vector map based on the target feature map through a prediction module; extracts the center point coordinates of the vertebra according to the center heat map, the offset map, and the centripetal vector map, and determines the coordinates of the four corner points of the vertebra according to the center point coordinates of the vertebra; determines the Cobb angle through a geometric algorithm based on the coordinates of the four corner points of the vertebra, thereby improving the accuracy of Cobb angle measurement.

[0126] Figure 2 This is a schematic diagram of the structure of the Mamba-based scoliosis Cobb angle determination device provided in the embodiment of the present application. Figure 2 As shown, the Mamba-based scoliosis Cobb angle determination device 200 provided in this embodiment includes:

[0127] Acquisition module 201, used to acquire image information;

[0128] A first processing module 202 is used to pre-process the acquired image information to obtain an original image;

[0129] The second processing module 203 is configured to perform multi-level local feature extraction on the original image through an encoder to obtain a first feature map, wherein the encoder includes: a plurality of residual modules;

[0130] The second processing module 203 is further configured to perform global feature capture on the first feature map through a Mamba module to obtain a second feature map;

[0131] The second processing module 203 is further configured to decode the second feature map through a decoder to obtain a target feature map, wherein the decoder includes: a main decoding branch and an auxiliary decoding branch;

[0132] The second processing module 203 is further configured to generate a center heat map, an offset map, and a centripetal vector map based on the target feature map through a prediction module;

[0133] The second processing module 203 is further configured to extract the coordinates of the center point of the vertebra according to the center heat map, the offset map and the centripetal vector map, and determine the coordinates of the four corner points of the vertebra according to the coordinates of the center point of the vertebra;

[0134] The second processing module 203 is further configured to determine the Cobb angle according to the coordinates of the four corner points of the vertebral body through a geometric algorithm.

[0135] The Mamba-based scoliosis Cobb angle determination device provided in this embodiment can execute the Mamba-based scoliosis Cobb angle determination method provided in the above method embodiment. Its implementation principle and technical effects are similar, and will not be described in detail in this embodiment.

[0136] Figure 3 This is a schematic diagram of the structure of the Mamba-based scoliosis Cobb angle determination device provided in the embodiment of the present application. Figure 3 As shown, the embodiment of the present application provides a Mamba-based scoliosis Cobb angle determination device 300, which includes: a receiver 301, a transmitter 302, a processor 303 and a memory 304.

[0137] Receiver 301, for receiving instructions and data;

[0138] Transmitter 302, used to send instructions and data;

[0139] Memory 304, for storing computer-executable instructions;

[0140] The processor 303 is configured to execute computer-executable instructions stored in the memory 304 to implement the steps of the Mamba-based scoliosis Cobb angle determination method in the above embodiment. For details, please refer to the relevant description in the above Mamba-based scoliosis Cobb angle determination method embodiment.

[0141] Optionally, the memory 304 may be independent or integrated with the processor 303 .

[0142] When the memory 304 is independently provided, the electronic device further includes a bus for connecting the memory 304 and the processor 303 .

[0143] An embodiment of the present application also provides a computer storage medium, which stores computer-executable instructions. When a processor executes the computer-executable instructions, it implements the Mamba-based scoliosis Cobb angle determination method performed by the above-mentioned Mamba-based scoliosis Cobb angle determination device.

[0144] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned Mamba-based scoliosis Cobb angle determination method.

[0145] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As is well known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable, and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media generally embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0146] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of the present application and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0147] It should be understood that the present application is not limited to the exact structure described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for determining the Cobb angle of scoliosis based on Mamba, characterized in that: The method comprises: Acquiring image information, and preprocessing the acquired image information to obtain an original image; Performing multi-level local feature extraction on the original image through an encoder to obtain a first feature map, wherein the encoder includes: a plurality of residual modules; The global features of the first feature map are captured by the Mamba module to obtain the second feature map; Decoding the second feature map through a decoder to obtain a target feature map, wherein the decoder includes: a main decoding branch and an auxiliary decoding branch; Generate center heat map, offset map and centroid vector map based on target feature map through prediction module; Extract the center point coordinates of the vertebra according to the center heat map, offset map and centripetal vector map, and determine the coordinates of the four corner points of the vertebra according to the center point coordinates of the vertebra; The Cobb angle is determined by a geometric algorithm based on the coordinates of the four corner points of the vertebra.

2. The method according to claim 1, characterized in that The method of capturing global features of the first feature map by the Mamba module to obtain the second feature map includes: Performing feature serialization processing on the first feature map to obtain a first feature vector sequence; The first eigenvector sequence is analyzed and processed by the structured state space model of the Mamba module to obtain a second eigenvector sequence; The second feature vector sequence is mapped back to the two-dimensional feature map to obtain a second feature map.

3. The method according to claim 1, characterized in that The decoding the second feature map by the decoder to obtain the target feature map includes: Upsampling and convolution operations are performed on the second feature map through the main decoding branch to obtain a third feature map; The features of the middle layer of the encoder are received through the auxiliary decoding branch, and back-propagation optimization is performed through the auxiliary loss function to obtain the fourth feature map; The third feature map and the fourth feature map are fused to obtain the target feature map.

4. The method according to claim 1, wherein The method of extracting the center point coordinates of the vertebra according to the center heat map, the offset map, and the centripetal vector map, and determining the coordinates of the four corner points of the vertebra according to the center point coordinates of the vertebra, includes: Apply the non-maximum suppression algorithm to the center heat map, and select a preset number of peak points with the highest confidence scores in the processed center heat map. For each selected peak point, record its integer coordinates and confidence score on the target feature map; Extract the corresponding offset vector from the offset map according to the integer coordinates of the peak point, and determine the coordinates of the center point of the vertebra according to the integer coordinates and the offset vector; The eight corresponding vector components are extracted from the centripetal vector diagram according to the coordinates of the center point of the vertebral body, and the coordinates of the four corner points of the vertebral body are determined according to the coordinates of the center point of the vertebral body and the extracted vector components.

5. The method according to claim 1, wherein Determining the Cobb angle using a geometric algorithm based on the coordinates of the four corner points of the vertebral body includes: Determine the direction line of each vertebra based on the coordinates of the four corner points of the vertebra; Determine the angles between all pairs of vertebrae based on the direction lines of the vertebrae to obtain an angle matrix; Determine the vertebral range corresponding to the main curve, and determine the Cobb angle of the main curve based on the angle matrix; The vertebral range corresponding to the secondary curve is determined, and the Cobb angle of the secondary curve is determined by a geometric algorithm.

6. The method according to claim 3, characterized in that The back propagation optimization is performed through the auxiliary loss function to obtain the fourth feature map, including: Define the first loss function for the center heat map, which is used to measure the difference between the predicted center heat map and the true center heat map; A second loss function is defined for the offset map, where the second loss function is used to measure the difference between the predicted offset map and the true offset map; A third loss function is defined for the centripetal vector map, where the third loss function is used to measure the difference between the predicted centripetal vector map and the true centripetal vector map; Perform weighted summation on the first loss function, the second loss function, and the third loss function to obtain the total loss; Back propagation optimization is performed based on the total loss to obtain the fourth feature map.

7. A Mamba-based scoliosis Cobb angle determination device, characterized in that: The device comprises: An acquisition module, used for acquiring image information; A first processing module, configured to preprocess the acquired image information to obtain an original image; A second processing module is configured to perform multi-level local feature extraction on the original image through an encoder to obtain a first feature map, wherein the encoder includes: a plurality of residual modules; The second processing module is further configured to perform global feature capture on the first feature map through a Mamba module to obtain a second feature map; The second processing module is further configured to decode the second feature map through a decoder to obtain a target feature map, wherein the decoder includes: a main decoding branch and an auxiliary decoding branch; The second processing module is further configured to generate a center heat map, an offset map, and a centripetal vector map based on the target feature map through the prediction module; The second processing module is further configured to extract the coordinates of the center point of the vertebra according to the center heat map, the offset map and the centripetal vector map, and determine the coordinates of the four corner points of the vertebra according to the coordinates of the center point of the vertebra; The second processing module is further configured to determine the Cobb angle using a geometric algorithm according to the coordinates of the four corner points of the vertebral body.

8. A Mamba-based scoliosis Cobb angle determination device, characterized in that: The device comprises: Memory; processor; wherein the memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the Mamba-based scoliosis Cobb angle determination method according to any one of claims 1 to 6.

9. A computer storage medium, characterized in that The computer storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the Mamba-based scoliosis Cobb angle determination method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The invention comprises a computer program, which is used to implement the Mamba-based scoliosis Cobb angle determination method according to any one of claims 1 to 6 when the computer program is executed by a processor.

Images