Lumbar vertebra positioning method and device, electronic equipment and readable medium

Abstract

The invention relates to a lumbar vertebra positioning method and device, electronic equipment and a readable medium, and the method comprises the steps: obtaining a magnetic resonance lumbar vertebra image of a target object, and segmenting a lumbar vertebra segmentation image from the magnetic resonance lumbar vertebra image; determining voxel data corresponding to the lumbar vertebra according to the magnetic resonance lumbar vertebra image and the lumbar vertebra segmentation image; the voxel data are input into an intervertebral disc trailing edge point recognition feature extraction network to obtain a prediction probability graph and a prediction error deviation value output from the intervertebral disc trailing edge point recognition feature extraction network, and the prediction error deviation value is used for correcting predicted intervertebral disc trailing edge point position deviation; performing multi-loss function joint training based on the prediction probability graph and the prediction error deviation value to determine the position of a trailing edge point of the intervertebral disc; and determining the centroid position of each intervertebral disc according to the lumbar vertebra segmentation image, and determining the position of each cone by using the centroid position and the position of the trailing edge point. The problem that the lumbar vertebra position in magnetic resonance imaging cannot be accurately positioned is solved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method, device, electronic device, and readable medium for lumbar spine localization. Background Technology

[0002] In the field of clinical diagnosis, with the continuous advancement of medical imaging technology, MRI (Magnetic Resonance Imaging) has become a key means of disease diagnosis due to its advantages of high resolution and three-dimensional imaging, and can provide anatomical information for clinical practice.

[0003] However, traditional MRI lumbar spine localization relies on manual operation by medical technicians, which has significant drawbacks. For example, manual marking and adjustment of body parts is extremely inefficient and difficult to adapt to the large volume of clinical imaging diagnostic needs. If manual marking is not used, lumbar spine localization and identification must be performed using traditional image processing, mainly employing basic algorithms such as threshold segmentation and edge detection, which are actually insufficient to handle the complexity of MRI images. Therefore, none of the above methods can accurately locate the lumbar spine in MRI.

[0004] There is currently no effective solution to the problem of accurately locating the lumbar spine in magnetic resonance imaging. Summary of the Invention

[0005] This application provides a lumbar spine localization method, device, electronic device, and readable medium to solve the aforementioned technical problem of "inability to accurately locate the lumbar spine in magnetic resonance imaging".

[0006] According to one aspect of the embodiments of this application, this application provides a lumbar spine localization method, including: acquiring a magnetic resonance lumbar spine image of a target object, and segmenting a lumbar spine segmentation image from the magnetic resonance lumbar spine image; determining voxel data corresponding to the lumbar spine based on the magnetic resonance lumbar spine image and the lumbar spine segmentation image; inputting the voxel data into a feature extraction network for identifying the posterior edge of the intervertebral disc to obtain a prediction probability map and a prediction error offset value output by the feature extraction network for identifying the posterior edge of the intervertebral disc, wherein the prediction probability map is used to predict the position of the posterior edge of the intervertebral disc, and the prediction error offset value is used to correct the deviation of the predicted position of the posterior edge of the intervertebral disc; performing joint training of multiple loss functions based on the prediction probability map and the prediction error offset value to determine the position of the posterior edge of the intervertebral disc; determining the centroid position of each intervertebral disc based on the lumbar spine segmentation image, and determining the position of each vertebra using the centroid position and the posterior edge position, so as to complete the lumbar spine localization of the target object.

[0007] Optionally, segmenting a lumbar spine image from a magnetic resonance lumbar spine image includes: preprocessing the magnetic resonance lumbar spine image to obtain a first image; inputting the first image into a segmentation network for segmentation to obtain segmentation results for each vertebra and intervertebral disc; filtering segmentation impurities in the segmentation results to optimize the segmentation boundaries and obtain a lumbar spine segmentation image.

[0008] Optionally, the voxel data is input into the intervertebral disc posterior margin feature extraction network to obtain the prediction probability map and prediction error offset value output from the intervertebral disc posterior margin recognition feature extraction network. This includes: downsampling the voxel data through the intervertebral disc posterior margin recognition feature extraction network to obtain multiple first feature map groups; extracting features from each first feature map group based on the Split-Attention operation to obtain fused features; outputting the prediction probability map through a convolutional layer using the fused features; and outputting the prediction error offset value through a fully connected layer using the fused features.

[0009] Optionally, feature extraction is performed on each first feature map group based on the Split-Attention operation to obtain fused features, including: performing convolution operations on each first feature map group to obtain two second feature map groups; performing Split-Attention operations on the two second feature map groups respectively to obtain two third feature map groups; merging the two third feature map groups into a fourth feature map group; performing global average pooling on the fourth feature map group and compressing the spatial dimension to obtain a global feature vector; multiplying the global feature vector element-wise with the two third feature map groups respectively, and adding the two multiplication results to obtain the fused features.

[0010] Optionally, multiple loss functions are jointly trained based on the predicted probability map and the prediction error offset value to determine the location of the posterior edge of the intervertebral disc, including: calculating the first loss of the predicted probability map using a first loss function, calculating the second loss of the prediction error offset value using a second loss function, and calculating the third loss between the predicted posterior edge point and the actual posterior edge point using a third loss function; weighting the first loss, the second loss, and the third loss to obtain the total loss; and determining the location of the posterior edge of the intervertebral disc using the total loss.

[0011] Optionally, the position of each cone is determined using the centroid position and the trailing edge position, including: determining the anteroposterior direction of the cone using the three-dimensional coordinates of the centroid position and the trailing edge position; determining the vertical direction of the cone based on the lumbar spine segmentation image; calculating the left-right direction of the cone using the anteroposterior and vertical directions; and determining the spatial position of the cone based on the anteroposterior, vertical, and left-right directions.

[0012] Optionally, determining the vertical direction of the vertebrae based on the lumbar spine segmentation image includes: obtaining a target plane by fitting each point in the lumbar spine segmentation image; and calculating the normal vector of the target plane to determine the vertical direction of the vertebrae.

[0013] According to another aspect of the embodiments of this application, this application provides a lumbar spine positioning device, including: an acquisition module, used to acquire a magnetic resonance lumbar spine image of a target object and segment a lumbar spine segmentation image from the magnetic resonance lumbar spine image; a first determination module, used to determine voxel data corresponding to the lumbar spine based on the magnetic resonance lumbar spine image and the lumbar spine segmentation image; an extraction module, used to input the voxel data into a feature extraction network for identifying the posterior edge of the intervertebral disc, to obtain a prediction probability map and a prediction error offset value output from the feature extraction network for identifying the posterior edge of the intervertebral disc, wherein the prediction probability map is used to predict the position of the posterior edge of the intervertebral disc, and the prediction error offset value is used to correct the deviation of the predicted position of the posterior edge of the intervertebral disc; a training module, used to perform joint training of multiple loss functions based on the prediction probability map and the prediction error offset value, to determine the position of the posterior edge of the intervertebral disc; and a second determination module, used to determine the centroid position of each intervertebral disc based on the lumbar spine segmentation image, and to determine the position of each vertebra using the centroid position and the posterior edge position, so as to complete the lumbar spine positioning of the target object.

[0014] According to another aspect of the embodiments of this application, this application provides an electronic device, including a memory, a processor, a communication interface and a communication bus. The memory stores a computer program that can run on the processor. The memory and the processor communicate with each other through the communication bus and the communication interface. When the processor executes the computer program, it implements the steps of the above method.

[0015] According to another aspect of the embodiments of this application, this application also provides a computer-readable medium having processor-executable non-volatile program code that causes the processor to perform the above-described method.

[0016] Compared with related technologies, the technical solutions provided in this application have the following advantages:

[0017] This application provides a method for lumbar spine localization, comprising: acquiring a magnetic resonance lumbar spine image of a target object, and segmenting a lumbar spine segmentation image from the magnetic resonance lumbar spine image; determining voxel data corresponding to the lumbar spine based on the magnetic resonance lumbar spine image and the lumbar spine segmentation image; inputting the voxel data into a feature extraction network for intervertebral disc posterior margin point recognition to obtain a prediction probability map and a prediction error offset value output by the feature extraction network for intervertebral disc posterior margin point recognition, wherein the prediction probability map is used to predict the position of the posterior margin point of the intervertebral disc, and the prediction error offset value is used to correct the deviation of the predicted position of the posterior margin point of the intervertebral disc; performing joint training of multiple loss functions based on the prediction probability map and the prediction error offset value to determine the position of the posterior margin point of the intervertebral disc; determining the centroid position of each intervertebral disc based on the lumbar spine segmentation image, and determining the position of each vertebral body using the centroid position and the posterior margin point position, thereby completing the lumbar spine localization of the target object. By employing joint training with multiple loss functions to identify the posterior edge of the intervertebral disc, and incorporating the loss due to deviation in the location of the posterior edge, the accuracy of identifying the posterior edge of the intervertebral disc can be further improved, thereby enhancing the accuracy of lumbar spine localization and solving the problem of inaccurate localization of the lumbar spine in magnetic resonance imaging. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of an optional lumbar spine positioning method provided according to an embodiment of this application;

[0021] Figure 2 A detailed flowchart of an optional lumbar spine positioning method provided according to an embodiment of this application;

[0022] Figure 3 This is a schematic diagram of segmentation of an optional lumbar spine image according to an embodiment of this application;

[0023] Figure 4 This is a schematic diagram of an optional intervertebral disc posterior margin point recognition feature extraction network provided according to an embodiment of this application;

[0024] Figure 5 This is a schematic diagram illustrating the identification of an optional posterior edge point of an intervertebral disc according to an embodiment of this application.

[0025] Figure 6This is an optional lumbar spine positioning diagram provided according to an embodiment of this application;

[0026] Figure 7 This is a block diagram of an optional lumbar spine positioning device provided according to an embodiment of this application;

[0027] Figure 8 This is a schematic diagram of an optional electronic device structure provided in an embodiment of this application. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustration and has no specific meaning in itself. Therefore, "module" and "part" may be used interchangeably.

[0030] In the field of clinical diagnosis, with the continuous advancement of medical imaging technology, MRI (Magnetic Resonance Imaging) has become a key means of disease diagnosis due to its advantages of high resolution and three-dimensional imaging, and can provide anatomical information for clinical practice.

[0031] However, traditional MRI lumbar spine localization relies on manual operation by medical technicians, which has significant drawbacks. For example, manual marking and adjustment of body parts is extremely inefficient and difficult to adapt to the large volume of clinical imaging diagnostic needs. If manual marking is not used, lumbar spine localization and identification must be performed using traditional image processing, mainly employing basic algorithms such as threshold segmentation and edge detection, which are actually insufficient to handle the complexity of MRI images. Therefore, none of the above methods can accurately locate the lumbar spine in MRI.

[0032] To address the problems mentioned in the background art, according to one aspect of the embodiments of this application, an embodiment of a lumbar spine positioning method is provided, such as... Figure 1 As shown, it includes:

[0033] Step 101: Obtain the magnetic resonance lumbar spine image of the target object, and segment the lumbar spine image from the magnetic resonance lumbar spine image;

[0034] Step 102: Determine the voxel data corresponding to the lumbar spine based on the magnetic resonance lumbar spine image and the lumbar spine segmentation image;

[0035] Step 103: Input the voxel data into the intervertebral disc posterior margin feature extraction network to obtain the prediction probability map and prediction error offset value output from the intervertebral disc posterior margin feature extraction network. The prediction probability map is used to predict the position of the intervertebral disc posterior margin point, and the prediction error offset value is used to correct the deviation of the predicted intervertebral disc posterior margin point position.

[0036] Step 104: Perform joint training of multiple loss functions based on the predicted probability map and the predicted error offset value to determine the location of the posterior edge of the intervertebral disc.

[0037] Step 105: Determine the centroid position of each intervertebral disc based on the segmented lumbar spine image, and use the centroid position and the posterior edge point position to determine the position of each vertebra, so as to complete the lumbar spine localization of the target object.

[0038] This application is divided into three stages: model training (vertebral body and intervertebral disc segmentation, lumbar posterior edge point recognition), model application, and 3D lumbar spine localization.

[0039] The training phase includes:

[0040] Data preparation: In the data preparation phase, a certain amount of MRI lumbar spine training data needs to be collected in advance. Subsequently, professional data personnel need to annotate the locations of various parts of the lumbar spine to complete the data annotation work.

[0041] Data preprocessing: The MRI lumbar spine data were preprocessed by normalizing and augmenting the data. The pixel values ​​of the data were normalized to between 0 and 1. Data augmentation included randomly horizontally flipping the image and adding noise.

[0042] Model training and updates: Vertebral body and intervertebral disc segmentation uses 64*256*96 pixel image data fed into the 3D NNUNNET V2 model for training. The training parameters are: batch size 8, SGD (stochastic gradient descent) optimizer, learning rate 0.001, training epochs 1000, and Dice loss function. Posterior disc point recognition uses 48*196*64 pixel image data fed into the posterior disc point recognition feature extraction network. The training parameters are: batch size 6, Adam (Adaptive Moment Estimation) optimizer, learning rate 0.001, training epochs 1000, and multiple loss functions used in combination.

[0043] The application phase of the model includes data preparation, data preprocessing, model inference and recognition, and data postprocessing. Specifically, it includes: Data preparation: acquiring lumbar MRI data; Data preprocessing: normalization processing; Model inference and recognition: cropping the acquired raw lumbar MRI data into 64*256*96 images and inputting them into the vertebral and intervertebral disc segmentation model. The segmentation result is the region of each vertebra and intervertebral disc. Based on the segmentation result, voxel data is extracted from the lumbar MRI data, and the data is cropped to 48*196*64 and input into the intervertebral disc posterior edge point recognition network to identify the location of the intervertebral disc posterior edge point; Data postprocessing: filtering impurities in the segmentation result using a connected component algorithm.

[0044] Figure 2 A detailed flowchart of a lumbar spine localization method provided in this application is shown in the figure. The localization method includes: acquiring lumbar MRI data; preprocessing the lumbar MRI data; inputting the preprocessed data into a 3D nnunetV2 network for segmentation to obtain regional images of each vertebra and intervertebral disc; multiplying the regional images with the lumbar MRI data to extract voxel data of the lumbar spine; preprocessing the voxel data (i.e., normalization) and inputting the processed voxel data into a posterior disc recognition network (corresponding to the posterior disc recognition feature network of this application) to obtain the coordinates of the posterior disc recognition point; calculating the vertebral orientation based on the coordinates of the posterior disc recognition point and the regional images of the vertebrae and intervertebral disc, thereby achieving lumbar spine localization.

[0045] As an optional embodiment, segmenting a lumbar spine image from a magnetic resonance lumbar spine image includes: preprocessing the magnetic resonance lumbar spine image to obtain a first image; inputting the first image into a segmentation network for segmentation to obtain segmentation results for each vertebra and intervertebral disc; filtering segmentation impurities in the segmentation results to optimize the segmentation boundaries and obtain a lumbar spine segmentation image.

[0046] Preprocessing of lumbar MRI images includes data normalization, data cropping, and data augmentation.

[0047] The segmentation network provided in this application is a 3D nnunetV2 network architecture, through which the segmentation results of each cone and intervertebral disc are obtained.

[0048] To filter out segmentation impurities in the segmentation results and optimize the segmentation boundaries, the following methods are employed: using a connected component algorithm to filter out segmentation impurities and smooth segmentation edges.

[0049] Figure 3This diagram illustrates a segmentation method for lumbar spine images provided in this application. As shown, MRI lumbar spine data is acquired, and data preprocessing includes data normalization, data cropping, and data augmentation. The segmentation network employs a 3D nnunetV2 architecture, which segments the lumbar vertebrae L1 to L5 and the sacral vertebra S1, a total of six vertebrae, along with the corresponding intervertebral discs between each pair of vertebrae. A connected component algorithm is used to filter out segmentation impurities and smooth segmentation edges before outputting the final result.

[0050] As an optional embodiment, voxel data is input into a posterior disc feature extraction network to obtain a predicted probability map and a prediction error offset value output from the posterior disc feature extraction network. This includes: downsampling the voxel data through the posterior disc feature extraction network to obtain multiple first feature map groups; extracting features from each first feature map group based on a Split-Attention operation to obtain fused features; outputting a predicted probability map through a convolutional layer using the fused features; and outputting a prediction error offset value through a fully connected layer using the fused features.

[0051] The output of the intervertebral disc posterior edge point recognition feature extraction network provided in this application is in the form of multi-task output. The first task output 1 represents the output of the predicted probability map (which can be denoted as Probabilitymap_pre) through two 3x3x3 convolutions, which is used to calculate the position of each key point. The second task output 2 represents the output of the prediction error offset (which can be denoted as diff_pre) through two liner linear layers, which is used to calculate the error offset value of each key point (i.e., posterior edge point).

[0052] A probability map is generated by applying a Gaussian distribution function to the location of each keypoint, creating a Gaussian probability map centered on that keypoint. The Gaussian distribution function centered on the keypoint will reach its maximum value at that point and gradually decrease as the distance from the center increases.

[0053] Probability maps essentially transform discrete targets into continuous targets, providing directional guidance for network training. The closer a network is to the target point, the higher its probability value, allowing the network to quickly and purposefully reach the target.

[0054] We use a 3D Probability Map to represent the location of each key point:

[0055]

[0056] Where ln(H(x, y, z)) represents the natural logarithm of the probability density at the point (x, y, z), (a xa y a z ) represents the spatial coordinates of the key point, and σ is the standard deviation of the distribution. Setting σ = 2, the original coordinate points are transformed into a spatial sphere with a radius of 4. The probability value at the center of the sphere is 1, and the probability value gradually decreases until it becomes 0 as it gets closer to the edge.

[0057] The feature extraction network for identifying the posterior edge of the intervertebral disc in this application adopts a multi-task output format. The first task outputs a prediction probability map to calculate the position of each key point; the second task outputs a prediction error offset to calculate the error offset value of each key point, with the aim of reducing the error generated during image processing.

[0058] As an optional implementation, feature extraction is performed on each first feature map group based on the Split-Attention operation to obtain fused features, including: performing convolution operations on each first feature map group to obtain two second feature map groups; performing Split-Attention operations on the two second feature map groups respectively to obtain two third feature map groups; merging the two third feature map groups into a fourth feature map group; performing global average pooling on the fourth feature map group and compressing the spatial dimension to obtain a global feature vector; multiplying the global feature vector element-wise with the two third feature map groups respectively, and adding the two multiplication results to obtain the fused features.

[0059] Figure 4 This diagram illustrates a feature extraction network for identifying posterior intervertebral disc points, as provided in this application. As shown, the input data undergoes two downsampling processes, and then features are extracted from the downsampled data. The main contribution of this network is the design of a Split-Attention module, which enables cross-channel attention. This design uses two SlipAttention structures, performing 3x3x3 (convolutional layer 2) and 1x1x1 (convolutional layer 1) convolutions on each feature map group to obtain two feature map groups, which are then subjected to the Split-Attention operation. The two feature maps are then concatenated element-wise to form a new feature map group. Global average pooling is then applied to this new group to obtain a feature vector, which is then multiplied by the original feature map and the corresponding element-wise summed to obtain the output of this base array.

[0060] As an optional embodiment, multiple loss functions are jointly trained based on the predicted probability map and the prediction error offset value to determine the location of the posterior edge of the intervertebral disc, including: calculating a first loss of the predicted probability map using a first loss function, calculating a second loss of the prediction error offset value using a second loss function, and calculating a third loss between the predicted posterior edge point and the actual posterior edge point using a third loss function; weighting the first loss, the second loss, and the third loss to obtain the total loss; and determining the location of the posterior edge of the intervertebral disc using the total loss.

[0061] The network regression provided in this application adopts a joint training method with multiple loss functions. By adjusting the weights between different loss functions, the network focuses on training a certain part. The first loss function (loss1) is used to calculate the probability map loss; the second loss function (loss2) is used to calculate the error offset loss of the trailing edge point; and the third loss function (loss3) is used to calculate the position loss of the true trailing edge point.

[0062] Network regression employs a joint training approach using multiple loss functions. The first loss function (loss1) uses the BCECoCal Loss function to calculate the probabilistic graphical loss, as shown in the following formula:

[0063] loss1 = -[-β(1-V)] γ ylog(V)+(1-β)V γ [(1-y)log(1-V)],

[0064] Where V is the model's predicted probability for the target class, β is the balancing factor used to adjust the influence between positive and negative samples, γ is the focus factor used to adjust the weights of easy and difficult samples, and y represents the positive sample.

[0065] Due to the large size of the network input data, two downsampling operations are required. After downsampling, the keypoint coordinates are rounded, resulting in deviations when reconstructing the original coordinates. Therefore, this application provides a second loss function (loss2), adding a center point offset loss to calculate the error offset loss of the keypoints, as shown in the following formula:

[0066]

[0067] Where K represents the number of key points, This represents the center point offset value, where pi represents the center point coordinates before scaling, and R represents the scaling factor. This indicates the coordinates of the predicted key points.

[0068] The third loss function (loss3) is used to calculate the positional loss between the predicted keypoints and the actual keypoints. The formula is as follows:

[0069]

[0070] Where M is the number of samples, and pi is the true value of the i-th sample. It is the predicted value of the i-th sample.

[0071] The final loss function, Loss, is expressed as:

[0072] loss=a1*loss1+a2*loss2+a3*loss3

[0073] Where a1+a2+a3=1, here a1=0.4, a2=0.3, a3=0.3. In practical applications, the values ​​of a1, a2 and a3 can be adjusted according to the requirements.

[0074] Figure 5 This is a schematic diagram for identifying the posterior edge point of the intervertebral disc provided in this application. MR represents the lumbar MRI image, and MASK represents the segmentation mask corresponding to the segmented lumbar spine image. Voxel data is obtained through MR and MASK. The voxel data is input into the feature extraction network to obtain two outputs, namely the prediction probability map (Probabilitymap_pre) and the prediction error offset (diff_pre). These two outputs are jointly trained using multiple loss functions (the specific steps have been described above), and finally, the three-dimensional keypoint coordinates (i.e., the posterior edge point coordinates) are output.

[0075] This application employs a multi-loss function for joint training to identify the posterior edge point of the intervertebral disc, and adds a deviation loss for the key point position, which can further improve the accuracy of the posterior edge point identification and increase the calculation speed.

[0076] As an optional embodiment, the position of each cone is determined using the centroid position and the trailing edge position, including: determining the anteroposterior direction of the cone using the three-dimensional coordinates of the centroid position and the trailing edge position; determining the vertical direction of the cone based on the lumbar spine segmentation image; calculating the left-right direction of the cone using the anteroposterior and vertical directions; and determining the spatial position of the cone based on the anteroposterior, vertical, and left-right directions.

[0077] The centroid A of each intervertebral disc is calculated using segmented images of the lumbar spine, and the posterior edge point B of each intervertebral disc is obtained through point recognition results. This indicates the anterior-posterior direction of the intervertebral disc. The disc is fitted to a plane using the least squares method, and the normal vector of this plane represents the superior-inferior direction of the vertebral body. Cross product of front-back and up-down directions That is, the left and right directions of the vertebrae. After obtaining the front-back, up-down, and left-right directions, the position of each vertebra in space can be calculated.

[0078] Optionally, in the centroid calculation formula, the numerator is: traversing all voxels, and only for voxels belonging to the lumbar region (mask=1), accumulating their indices in the d (depth), h (height), and w (width) dimensions; the denominator is: the total number of voxels in the lumbar region (i.e., the number of voxels with a value of 1 in the mask), and the two are divided.

[0079] As an optional embodiment, determining the vertical direction of the vertebrae based on lumbar spine segmentation images includes: obtaining a target plane by fitting each point in the lumbar spine segmentation image; and calculating the normal vector of the target plane to determine the vertical direction of the vertebrae.

[0080] The target screen is obtained by fitting a plane to the points representing the lumbar spine region in the segmented lumbar spine image using the least squares method. The normal vector of this plane represents the vertical direction of the vertebral body.

[0081] By calculating the spatial position of each vertebral segment and combining it with the posterior edge of the intervertebral disc, lumbar spine localization is achieved.

[0082] Figure 6 The lumbar spine positioning diagram provided in this application is shown in the figure, where T12 represents the 12th thoracic vertebra, L1 represents the first lumbar vertebra, L2 represents the second lumbar vertebra, L3 represents the third lumbar vertebra, L4 represents the fourth lumbar vertebra, L5 represents the fifth lumbar vertebra, and S1 represents the sacrum.

[0083] This application proposes a method for identifying the posterior edge of the intervertebral disc and automatically locating the lumbar spine based on joint training of multiple loss functions. This method is used to determine the position of the vertebral body, the intervertebral disc, and the orientation of the vertebral body, thereby achieving lumbar spine localization. Compared with traditional 2D lumbar spine localization, it is not affected by lesions such as scoliosis or fractures.

[0084] This application provides a method for lumbar spine localization, comprising: acquiring a magnetic resonance lumbar spine image of a target object, and segmenting a lumbar spine segmentation image from the magnetic resonance lumbar spine image; determining voxel data corresponding to the lumbar spine based on the magnetic resonance lumbar spine image and the lumbar spine segmentation image; inputting the voxel data into a feature extraction network for intervertebral disc posterior margin point recognition to obtain a prediction probability map and a prediction error offset value output by the feature extraction network for intervertebral disc posterior margin point recognition, wherein the prediction probability map is used to predict the position of the posterior margin point of the intervertebral disc, and the prediction error offset value is used to correct the deviation of the predicted position of the posterior margin point of the intervertebral disc; performing joint training of multiple loss functions based on the prediction probability map and the prediction error offset value to determine the position of the posterior margin point of the intervertebral disc; determining the centroid position of each intervertebral disc based on the lumbar spine segmentation image, and determining the position of each vertebral body using the centroid position and the posterior margin point position, thereby completing the lumbar spine localization of the target object. By employing joint training with multiple loss functions to identify the posterior edge of the intervertebral disc, and incorporating the loss due to deviation in the location of the posterior edge, the accuracy of identifying the posterior edge of the intervertebral disc can be further improved, thereby enhancing the accuracy of lumbar spine localization and solving the problem of inaccurate localization of the lumbar spine in magnetic resonance imaging.

[0085] According to another aspect of the embodiments of this application, this application provides a lumbar spine positioning device, such as... Figure 7 As shown, it includes:

[0086] The acquisition module 701 is used to acquire the magnetic resonance lumbar spine image of the target object and segment the lumbar spine segmentation image from the magnetic resonance lumbar spine image.

[0087] The first determining module 702 is used to determine the voxel data corresponding to the lumbar spine based on the magnetic resonance lumbar spine image and the lumbar spine segmentation image;

[0088] The extraction module 703 is used to input voxel data into the intervertebral disc posterior edge point recognition feature extraction network to obtain the prediction probability map and prediction error offset value output from the intervertebral disc posterior edge point recognition feature extraction network. The prediction probability map is used to predict the position of the intervertebral disc posterior edge point, and the prediction error offset value is used to correct the deviation of the predicted intervertebral disc posterior edge point position.

[0089] Training module 704 is used to perform joint training of multiple loss functions based on the prediction probability map and the prediction error offset value to determine the location of the posterior edge point of the intervertebral disc.

[0090] The second determining module 705 is used to determine the centroid position of each intervertebral disc based on the segmented lumbar spine image, and to determine the position of each vertebra using the centroid position and the posterior edge point position, so as to complete the lumbar spine localization of the target object.

[0091] It should be noted that the acquisition module 701 in this embodiment can be used to execute step 101 in this application embodiment, the first determination module 702 in this embodiment can be used to execute step 102 in this application embodiment, the extraction module 703 in this embodiment can be used to execute step 103 in this application embodiment, the training module 704 in this embodiment can be used to execute step 104 in this application embodiment, and the second determination module 705 in this embodiment can be used to execute step 105 in this application embodiment.

[0092] Optionally, the acquisition module 701 is specifically used to preprocess the magnetic resonance lumbar spine image to obtain a first image; input the first image into a segmentation network for segmentation to obtain segmentation results for each vertebra and intervertebral disc; filter segmentation impurities in the segmentation results to optimize the segmentation boundary and obtain a lumbar spine segmentation image.

[0093] Optionally, the extraction module 703 includes:

[0094] The processing submodule is used to downsample the voxel data through the intervertebral disc posterior margin point recognition feature extraction network to obtain multiple first feature map groups;

[0095] The extraction submodule is used to extract features from each first feature map group based on the Split-Attention operation to obtain fused features;

[0096] The output submodule is used to output a prediction probability map by passing the fused features through a convolutional layer, and to output prediction error offset values ​​by passing the fused features through a fully connected layer.

[0097] Optionally, the extraction submodule is further configured to perform convolution operations on each of the first feature map groups to obtain two sets of second feature map groups; perform Split-Attention operations on the two second feature map groups to obtain two sets of third feature map groups; merge the two sets of third feature map groups into a fourth feature map group; perform global average pooling on the fourth feature map group and compress the spatial dimension to obtain a global feature vector; multiply the global feature vector element-wise with the two sets of third feature map groups respectively, and add the two multiplication results to obtain the fused features.

[0098] Optionally, the training module 704 is specifically used to calculate the first loss of the predicted probability map using the first loss function, the second loss of the prediction error offset value using the second loss function, and the third loss of the predicted posterior edge point and the actual posterior edge point using the third loss function; the first loss, the second loss and the third loss are weighted and calculated to obtain the total loss; and the position of the posterior edge point of the intervertebral disc is determined by the total loss.

[0099] Optionally, the second determining module 705 includes:

[0100] The first determining submodule is used to determine the front-back direction of the cone using the three-dimensional coordinates of the centroid position and the trailing edge position;

[0101] The second determination submodule is used to determine the vertical direction of the vertebrae based on the segmented images of the lumbar spine.

[0102] The calculation submodule is used to calculate the left and right directions of the cone using the front-back and up-down directions;

[0103] The third determination submodule is used to determine the spatial position of the cone based on the front-back direction, the up-down direction, and the left-right direction.

[0104] Optionally, the second determining submodule is further configured to obtain a target plane by fitting each point in the segmented image of the lumbar spine; and to calculate the normal vector of the target plane to determine the vertical direction of the cone.

[0105] It should be noted that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments.

[0106] According to another aspect of the embodiments of this application, this application provides an electronic device, such as... Figure 8 As shown, it includes a memory 801, a processor 803, a communication interface 805, and a communication bus 807. The memory 801 stores a computer program that can run on the processor 803. The memory 801 and the processor 803 communicate through the communication interface 805 and the communication bus 807. When the processor 803 executes the computer program, it implements the steps of the above method.

[0107] The memory and processor in the aforementioned electronic devices communicate with each other via a communication bus and a communication interface. The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0108] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0109] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0110] According to another aspect of the embodiments of this application, a computer-readable medium having processor-executable non-volatile program code is also provided.

[0111] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0112] In specific implementation, the embodiments of this application can be referred to the above embodiments and have corresponding technical effects.

[0113] It is understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or combinations thereof.

[0114] For software implementation, the techniques described herein can be implemented by units that perform the functions described herein. The software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or external to the processor.

[0115] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0116] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0117] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0118] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0119] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0120] If the aforementioned function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks. It should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0121] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for lumbar spine localization, characterized in that, include: Acquire magnetic resonance lumbar spine images of the target object, and segment lumbar spine images from the magnetic resonance lumbar spine images; Based on the magnetic resonance lumbar spine image and the lumbar spine segmentation image, determine the voxel data corresponding to the lumbar spine; The voxel data is input into the intervertebral disc posterior margin point recognition feature extraction network to obtain the prediction probability map and prediction error offset value output from the intervertebral disc posterior margin point recognition feature extraction network. The prediction probability map is used to predict the position of the intervertebral disc posterior margin point, and the prediction error offset value is used to correct the deviation of the predicted intervertebral disc posterior margin point position. Based on the predicted probability map and the predicted error offset value, multiple loss functions are jointly trained to determine the location of the posterior edge point of the intervertebral disc; The centroid position of each intervertebral disc is determined based on the segmented lumbar spine image, and the position of each vertebra is determined using the centroid position and the posterior edge point position, so as to complete the lumbar spine localization of the target object.

2. The method according to claim 1, characterized in that, The step of segmenting the lumbar spine from the magnetic resonance lumbar spine image includes: The magnetic resonance lumbar spine image is preprocessed to obtain the first image; The first image is input into a segmentation network for segmentation to obtain the segmentation results of each vertebra and intervertebral disc; The segmentation impurities in the segmentation results are filtered out to optimize the segmentation boundaries, thereby obtaining the lumbar spine segmentation image.

3. The method according to claim 1, characterized in that, The step of inputting the voxel data into the intervertebral disc posterior margin feature extraction network to obtain the predicted probability map and prediction error offset value output from the intervertebral disc posterior margin feature extraction network includes: The voxel data is downsampled using the intervertebral disc posterior margin point recognition feature extraction network to obtain multiple first feature map groups; Based on the Split-Attention operation, feature extraction is performed on each of the first feature map groups to obtain fused features; The fused features are passed through a convolutional layer to output the predicted probability map, and the fused features are passed through a fully connected layer to output the predicted error offset value.

4. The method according to claim 3, characterized in that, The step of extracting features from each of the first feature map groups based on the Split-Attention operation to obtain fused features includes: Perform a convolution operation on each of the first feature map groups to obtain two groups of second feature maps. The Split-Attention operation is performed on the two second feature map groups respectively to obtain two third feature map groups; The two third feature map groups are merged into a fourth feature map group; Global average pooling is performed on the fourth feature map group, and the spatial dimension is compressed to obtain a global feature vector; The global feature vector is multiplied element-wise by the two third feature map groups respectively, and the two multiplication results are added together to obtain the fused feature.

5. The method according to claim 1, characterized in that, The step of performing joint training of multiple loss functions based on the predicted probability map and the predicted error offset value to determine the location of the posterior edge point of the intervertebral disc includes: The first loss of the predicted probability map is calculated using a first loss function, the second loss of the prediction error offset is calculated using a second loss function, and the third loss between the predicted trailing edge point and the actual trailing edge point is calculated using a third loss function. The first loss, the second loss, and the third loss are weighted and calculated to obtain the total loss. The location of the posterior edge of the intervertebral disc is determined by the total loss.

6. The method according to claim 1, characterized in that, Determining the position of each cone using the centroid position and the trailing edge position includes: The front-back direction of the cone is determined by the three-dimensional coordinates of the centroid position and the trailing edge position; The vertical direction of the vertebra is determined based on the segmented image of the lumbar spine. The left-right direction of the cone is calculated using the front-back direction and the up-down direction; The spatial position of the cone is determined based on the front-back direction, the up-down direction, and the left-right direction.

7. The method according to claim 6, characterized in that, Determining the vertical direction of the vertebrae based on the segmented lumbar spine image includes: The target plane is obtained by fitting each point in the segmented lumbar spine image; Calculate the normal vector of the target plane to determine the vertical direction of the cone.

8. A lumbar spine positioning device, characterized in that, include: The acquisition module is used to acquire magnetic resonance lumbar spine images of the target object and segment lumbar spine images from the magnetic resonance lumbar spine images. The first determining module is used to determine the voxel data corresponding to the lumbar spine based on the magnetic resonance lumbar spine image and the lumbar spine segmentation image; An extraction module is used to input the voxel data into the intervertebral disc posterior margin feature extraction network to obtain a prediction probability map and a prediction error offset value output from the intervertebral disc posterior margin feature extraction network. The prediction probability map is used to predict the position of the intervertebral disc posterior margin, and the prediction error offset value is used to correct the deviation of the predicted position of the intervertebral disc posterior margin. The training module is used to perform joint training of multiple loss functions based on the predicted probability map and the predicted error offset value to determine the location of the posterior edge point of the intervertebral disc. The second determining module is used to determine the centroid position of each intervertebral disc based on the lumbar spine segmentation image, and to determine the position of each vertebra using the centroid position and the posterior edge point position, so as to complete the lumbar spine localization of the target object.

9. An electronic device comprising a memory, a processor, a communication interface, and a communication bus, wherein the memory stores a computer program executable on the processor, and the memory and the processor communicate via the communication bus and the communication interface, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable medium having processor-executable non-volatile program code, characterized in that, The program code causes the processor to execute the method of any one of claims 1 to 7.

Images