Lumbar intervertebral disc scanning surface positioning method based on vertebral block segmentation
Through Unet segmentation network and maximum connected image denoising technology, the automation and accuracy of the lumbar disc scanning surface positioning line is realized, solving the time-consuming and labor-intensive problem of manually dividing positioning lines, improving diagnostic efficiency and safety, and supporting telemedicine images transmission.
Patent Information
- Application Number
- CN202411059021.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-07-25
- Filing Date
- 2024-08-02
- Publication Date
- 2024-11-15
AI Technical Summary
In the diagnosis of lumbar spondylosis in the prior art, manual division of positioning lines is time-consuming and labor-intensive, causing the spinal cord sequence to deviate from the target position, which may lead to misjudgment of the disease, and a great test for radiologists' physical strength, lacking automated positioning methods.
The lumbar disc scanning surface positioning method based on vertebral segmentation is adopted, and five sets of positioning lines are automatically positioned through the Unet segmentation network, combining the maximum connectivity map denoising and objective function optimization to ensure the accuracy and automation of the positioning lines and reduce the work burden of doctors.
It realizes high-precision, low noise, and rapid lumbar positioning line division, reduces the work burden of doctors, and reduces unnecessary radiation exposure in MRI imaging, supporting image transmission and diagnosis of telemedicine.
Smart Images

Figure CN118967814A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a lumbar intervertebral disc scanning surface positioning method based on vertebral block segmentation, and belongs to the field of computer image processing. Background Art
[0002] According to relevant data released by the National Health Commission, the number of patients with lumbar spondylosis in China has exceeded 200 million. Common lumbar spondylosis includes lumbar disc herniation, lumbar muscle strain, lumbar spinal stenosis, etc. Patients may experience symptoms such as low back pain, limited lumbar movement, intermittent claudication, and even paraplegia.
[0003] Magnetic resonance imaging is an important examination method in the diagnosis and treatment of lumbar spondylosis. During the MRI scan, the doctor needs to first scan the sagittal plane of the lumbar spine as a positioning image, then draw a positioning line along the direction of the intervertebral disc, and then scan the sequence of the spinal cord. The positioning lines are located between the first lumbar vertebra (L1), the second lumbar vertebra (L2), the third lumbar vertebra (L3), the fourth lumbar vertebra (L4), the fifth lumbar vertebra (L5) and the first coccygeal vertebra (S1), with a total of five groups. Figure 1 As shown, each group has 3 positioning lines, the middle positioning line is located at the midline of the longest direction of the intervertebral disc, and the upper and lower positioning lines are tangent to the adjacent vertebral blocks of the intervertebral disc.
[0004] Accurate localization line division is the guarantee for accurate diagnosis of lumbar spine diseases. Because incorrect localization line division will cause the spinal cord sequence to deviate from the target position, causing doctors to misjudge the disease. However, since the direction, position, and scanning range of each set of localization lines are different, dividing the lumbar localization lines is a time-consuming and laborious task, which is a great test for the physical strength of radiologists. Therefore, it is necessary to explore an automated lumbar localization line division method to reduce the burden on doctors. Summary of the invention
[0005] In view of the time-consuming and labor-intensive manual positioning method, the present invention proposes a lumbar disc scanning surface positioning method based on vertebral block segmentation. The entire process of the method is as follows: Figure 2 As shown, it can automatically locate the positions and directions of five groups of positioning lines according to the positioning plane (lumbar sagittal plane) to reduce the workload of doctors.
[0006] In order to achieve the above-mentioned purpose, the technical solution of the present invention is as follows: The present invention provides a method for locating the scanning surface of a lumbar intervertebral disc based on vertebral block segmentation, and the specific steps are as follows:
[0007] Step 1: Scan the sagittal plane of the lumbar spine as a positioning image for lumbar spine scanning.
[0008] Step 2: Input the positioning image in step 1 into the Unet segmentation network and output the segmentation images of L1, L2, L3, L4, L5, and S1 vertebral blocks. Figure 3As shown in the figure, the structure of Unet includes a downsampling path and an upsampling path, and the jump connection in the middle is used to transmit the detailed information of the image. At the same time, we set the number of channels of feature maps of different resolutions from large to small to 16-32-64-128-150, so as to occupy the GPU video memory and speed up the operation. The essence of the segmentation network is to classify each pixel. In this task, we divide the target into two categories: background and vertebral block. Therefore, the network outputs the probability that the pixel belongs to the vertebral block. During processing, pixels with values greater than 0.5 are considered to be vertebral blocks, and pixels with values less than or equal to 0.5 are considered to be background. Figure 4 The training process of the network is shown: After marking the vertebral block positions on the training data (T1-weighted, T2-weighted, and fat-suppressed modalities) provided by Xingaoyi, we use pictures and labels to train the network. During the training process, we first use online data augmentation to expand the data, including random translation, Gaussian blur, gamma correction, etc., to prevent network overfitting and enhance the generalization ability of the network. Then the image is z-score normalized and put into the Unet network to obtain the predicted segmentation map, and the loss function is calculated together with the label. The loss function here is set to the sum of the binary cross entropy and the Dice coefficient. The gradient is then calculated using the loss value, and the network parameters are updated using the stochastic gradient descent optimization method (SGD).
[0009] Step 3: Draw the positioning line of the lumbar intervertebral disc according to the vertebral block segmentation map in step 2. Due to the deviation of network prediction, the background may be considered as the segmentation target, resulting in the appearance of noise. Because our segmentation target is 6 vertebral blocks, we use the method of taking 6 maximum connected graphs to denoise the image. Then, we use the idea of maximizing the objective function to find the coordinates of the 4 vertebrae of each vertebral block. Figure 5 As shown in (a), we first establish a plane rectangular coordinate system with the center of gravity of the vertebral block as the origin, and search for the coordinates of the four vertices in all pixels belonging to the vertebral block category so that they meet the following conditions respectively:
[0010] (x1,y1)=argmin(a1x1-b1y1)
[0011] (x2,y2)=argmin(a2x2+b2y2)
[0012] (x3,y3)=argmax(a3x3-b3y3)
[0013] (x4,y4)=argmax(a4x4+b4y4)
[0014] Where (x i ,y i ) represents the coordinates of a point; (a i ,b i) represents a set of hyperparameters greater than 0, which can be adjusted according to the prior information of the vertebral block deflection; argmin(·) and argmax(·) represent the values of the variable when the objective function in the brackets takes the maximum and minimum values. After obtaining these four vertices, press Figure 5 (b) Method to divide the positioning line: connect the upper and lower vertices, connect the centers of the connecting lines, and get the positioning line in the middle. The directions of the upper and lower positioning lines are the same as the middle one, but they pass through the vertex on the right side of the vertebral block. According to the doctor's requirements, the positioning lines cannot intersect in the human body, so we need to adjust the direction of the positioning lines. Figure 5 As shown in (c), find the intersection of the nearest positioning lines between adjacent groups, take a 100×100 image block with this intersection as the center, and if there are pixels in the right half of this patch that are larger than the image average (indicating that the intersection is inside the human body), open the two groups of positioning lines outward by 1 degree, and repeat the above process until the intersection moves out of the human body.
[0015] Step 4: Calculate the parameters of the MRI scanning surface based on the positioning line obtained in step 3. It is known that the center point of the positioning line obtained in step 3 is (x0, y0) and the direction vector is (m, n). The coordinates of the upper left corner of the MRI target scanning surface are LT, the side length of the scanning surface is fov, and the horizontal axis direction vector is Vertical axis direction vector The coordinate of the upper left corner of the current positioning image is LT0, and the horizontal axis direction vector is Vertical axis direction vector The pixel spacing is s. The target scanning surface should be perpendicular to the surface of the positioning image (world coordinate system) and pass through the current positioning line. The calculation process of the target scanning surface parameters is as follows:
[0016]
[0017]
[0018] Where normalize(·) represents vector normalization and × represents cross product.
[0019] Step 5, using the scanning parameters obtained in step 4, input into a magnetic resonance scanning machine to obtain an intervertebral disc scanning surface.
[0020] Compared with the prior art, the advantages of the present invention are as follows: (1) high segmentation accuracy. The present invention collects multimodal data and uses data enhancement methods, and can accurately segment vertebral blocks on T1-weighted, T2-weighted and fat-suppressed magnetic resonance images; (2) low noise interference. The present invention uses the method of taking the maximum connected graph to remove noise in the vertebral block segmentation graph, eliminating the interference of noise on subsequent division positioning lines; (3) good generalization. The training data of the present invention are real clinical data of the three modalities of T1-weighted, T2-weighted and fat-suppressed provided by Xingaoyi Company, so the present invention is compatible with data of these three modalities. (4) High positioning accuracy, ensuring that the divided positioning lines do not intersect in the human body, avoiding positioning errors, and meeting the actual positioning needs of doctors; (5) Fast running speed, because the segmentation algorithm runs on the GPU, and the algorithm for dividing the positioning lines is simple and has low computational requirements, so in actual testing, it only takes 500ms to divide the positioning lines for a single positioning image; (6) Reduce radiation exposure. In MRI imaging, although there is no direct ionizing radiation like X-ray or CT scanning, reducing unnecessary repeated scanning or erroneous scanning still helps to reduce the overall radiation exposure risk of patients and medical staff; (7) High degree of automation. The present invention is an automated, end-to-end method that can be integrated into the scanning software of the magnetic resonance imaging machine. The doctor only needs to click a button after scanning the positioning image to obtain the information of the intervertebral disc positioning line. (8) Support telemedicine: With the development of telemedicine technology, the precise images and positioning information generated by this method can be transmitted to remote experts through the network for consultation, thereby expanding the coverage of medical resources and improving the accessibility of medical services. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 This is a schematic diagram of the lumbar block numbering and positioning line positions.
[0022] Figure 2 is a flow chart of the present invention,
[0023] Figure 3 This is a schematic diagram of the Unet network structure.
[0024] Figure 4 This is the Unet network training flow chart.
[0025] Figure 5 The geometric relationship diagram of the sub-steps divided for the positioning line,
[0026] Figure 6 This is the algorithm test result diagram. DETAILED DESCRIPTION
[0027] The present invention is further explained below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.
[0028] Example: Figure 2 As shown, the present invention provides a lumbar disc scanning surface positioning method based on vertebral block segmentation, and the final test results are shown in Figure 6 As shown, the specific steps are as follows:
[0029] Step 1: Scan the sagittal plane of the lumbar spine as a positioning image for lumbar spine scanning.
[0030] Step 2: Input the positioning image in step 1 into the Unet segmentation network and output the segmentation images of L1, L2, L3, L4, L5, and S1 vertebral blocks. The Unet segmentation network needs to be trained on a labeled dataset first.
[0031] Step 3: Draw the positioning line of the lumbar intervertebral disc according to the vertebral block segmentation map in step 2. First, take the 6 largest connected graphs in the vertebral block segmentation map to remove noise. Then use the following conditions to calculate the coordinates of the 4 vertebrae of each vertebral block:
[0032] (x1,y1)=argmin(a1x1-b1y1)
[0033] (x2,y2)=argmin(a2x2+b2y2)
[0034] (x3,y3)=argmax(a3x3-b3y3)
[0035] (x4,y4)=argmax(a4x4+b4y4)
[0036] Where (x i ,y i ) represents the coordinates of a point; (a i ,b i ) represents a set of hyperparameters greater than 0, which can be adjusted according to the prior information of the vertebral block deflection; argmin(·) and argmax(·) represent the values of the variable when the objective function in the brackets takes the maximum and minimum values. Then connect the upper and lower vertices, connect the centers of the connecting lines, and draw the middle positioning line. The directions of the upper and lower positioning lines are the same as the middle one, but they pass through the vertex on the right side of the vertebral block. Finally, adjust the positioning lines so that they do not intersect in the human body: find the intersection of the nearest positioning lines between adjacent groups, take a 100×100 image block with this intersection as the center, and if there are pixels in the right half of this patch that are greater than the image average (indicating that the intersection is inside the human body), open the two groups of positioning lines outward by 1 degree, and repeat the above process until the intersection moves out of the human body.
[0037] Step 4: Calculate the parameters of the MRI scanning surface based on the positioning line obtained in step 3. It is known that the center point of the positioning line obtained in step 3 is (x0, y0) and the direction vector is (m, n). The coordinates of the upper left corner of the MRI target scanning surface are LT, the side length of the scanning surface is fov, and the horizontal axis direction vector is Vertical axis direction vector The coordinate of the upper left corner of the current positioning image is LT0, and the horizontal axis direction vector is Vertical axis direction vector The pixel spacing is s. The target scanning surface should be perpendicular to the surface of the positioning image (world coordinate system) and pass through the current positioning line. The calculation process of the target scanning surface parameters is as follows:
[0038]
[0039]
[0040] Where normalize(·) represents vector normalization and × represents cross product.
[0041] Step 5, using the scanning parameters obtained in step 4, input into a magnetic resonance scanning machine to obtain an intervertebral disc scanning surface.
[0042] Effect evaluation:
[0043] The present invention discloses a method for locating the scanning surface of the lumbar intervertebral disc based on vertebral block segmentation. The test results of the method are as follows: Figure 6 As shown. The figure shows 3 groups of test images, including T1-weighted, T2-weighted and fat-suppression mode sagittal images of the lumbar spine. The test was tested in the integrated software. The test process is: import the positioning image into the software, then click the "Calculate positioning line" button to get the positioning line shown in the figure, and click "Save" to save the scanning parameters to the file. The average running time of the algorithm is about 500ms, the speed is fast, and the positioning line division is relatively accurate. It shows that this method can quickly and accurately locate the lumbar spine, greatly reducing the workload of doctors.
[0044] The above description is only a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technology can understand and think of any changes or substitutions within the technical scope disclosed by the present invention, which should be included in the scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. A lumbar intervertebral disc scanning surface positioning method based on vertebral block segmentation, characterized in that: The specific steps are as follows: Step 1: Scan the sagittal plane of the lumbar spine as a positioning image for lumbar spine scanning. Step 2: Input the positioning image in step 1 into the Unet segmentation network and output the segmentation map of L1, L2, L3, L4, L5, and S1 vertebral blocks. The Unet segmentation network is trained on the labeled data. Step 3: Draw the positioning line of the lumbar intervertebral disc according to the vertebral block segmentation map in step 2, including 4 sub-steps: denoising, calculation of the four vertices of the vertebral block, positioning line division, and positioning line adjustment. Step 4: Calculate the parameters of the intervertebral disc scanning surface based on the positioning line obtained in step 3. Step 5, using the scanning parameters obtained in step 4, input into a magnetic resonance scanning machine to obtain an intervertebral disc scanning surface.
2. The method for lumbar intervertebral disc scanning surface positioning based on vertebral block segmentation according to claim 1, characterized in that: The positioning image data mentioned in step 1 is composed of multimodal data, including T1-weighted, T2-weighted and fat-suppressed data, and the positions of the L1, L2, L3, L4, L5, and S1 vertebral blocks are manually marked on each lumbar spine image.
3. The method for lumbar intervertebral disc scanning surface positioning based on vertebral block segmentation according to claim 1, characterized in that: In the training process of the Unet segmentation network in step 2, online data augmentation is used to expand the data, including random translation, Gaussian blur, and gamma correction, to prevent network overfitting and enhance the generalization ability of the network. Then the image is z-score normalized and put into the Unet network to obtain the predicted segmentation map. The loss function is calculated together with the label. The loss function is set to the sum of the binary cross entropy and the Dice coefficient. After that, the gradient is calculated by the loss value, and the network parameters are updated using the stochastic gradient descent optimization method (SGD).
4. The method for lumbar intervertebral disc scanning surface positioning based on vertebral block segmentation according to claim 1, characterized in that: In step 3, the vertebral block segmentation map is denoised as follows: first, all connected graphs are found in the segmentation map, the area of each connected graph is calculated, the first 6 connected graphs with the largest area are found, and the other connected graphs are set as background.
5. The method for lumbar intervertebral disc scanning surface positioning based on vertebral block segmentation according to claim 1, characterized in that: The four vertices of the vertebral block are calculated in the following steps. Specifically, a plane rectangular coordinate system is established with the center of gravity of the vertebral block as the origin, and the coordinates of the four vertices are searched in all foreground pixels so that they meet the following conditions respectively: (x1,y1)=argmin(a1x1-b1y1) (x2,y2)=argmin(a2x2+b2y2) (x3,y3)=argmax(a3x3-b3y3) (x4,y4)=argmax(a4x4+b4y4) Where (x i ,y i ) represents the coordinates of a point; (a i ,b i ) represents a set of hyperparameters greater than 0, which are adjusted according to the prior information of the vertebral block deflection; argmin(·) and argmax(·) represent the values of the variable when the objective function in the brackets takes the maximum and minimum values.
6. The method for lumbar intervertebral disc scanning surface positioning based on vertebral block segmentation according to claim 1, characterized in that: The specific method of dividing the positioning line mentioned in step 3 is to connect the upper and lower vertebrae of the two vertebral blocks, connect the centers of the connecting line, and obtain the positioning line in the middle. The directions of the upper and lower positioning lines are the same as the middle one, but pass through the vertex on the right side of the vertebral block.
7. The method for lumbar intervertebral disc scanning surface positioning based on vertebral block segmentation according to claim 1, characterized in that: The purpose of the positioning line adjustment in step 3 is to prevent the positioning lines from intersecting inside the human body. Specifically, find the intersection point of the nearest positioning lines between adjacent groups, take a 100×100 image block centered on this intersection point, and if there are pixels in the right half of this patch that are larger than the image average, it means that the intersection point is inside the human body. Then, open the two groups of positioning lines outward by 1 degree, and repeat the above process until the intersection point moves out of the human body.
8. The method for lumbar intervertebral disc scanning surface positioning based on vertebral block segmentation according to claim 1, characterized in that: In step 4, the parameters of the intervertebral disc scanning surface are calculated as follows. The center point of the positioning line obtained in step 3 is known to be (x0, y0), the direction vector is (m, n), the coordinates of the upper left corner of the magnetic resonance target scanning surface are LT, the side length of the scanning surface is fov, and the horizontal axis direction vector is Vertical axis direction vector The coordinate of the upper left corner of the current positioning image is LT0, and the horizontal axis direction vector is Vertical axis direction vector The pixel spacing is s, the target scanning surface should be perpendicular to the surface of the positioning image (world coordinate system) and pass through the current positioning line. The calculation process of the target scanning surface parameters is as follows: Where normalize(·) represents vector normalization and × represents cross product.
Citation Information
Cited By
Lumbar vertebra scanning processing method and device, storage medium and electronic equipment
CN121694727A
