Bone density measurement method, system, storage medium and electronic device
By optimizing the object detection algorithm and regression bone density detection model, using CT images to detect spinal areas and calculate bone density values, the complexity and inefficiency of bone density measurement in the existing technology is solved, and more efficient and accurate bones are achieved Density detection.
Patent Information
- Application Number
- CN202111404215.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-24
- Publication Date
- 2024-12-13
- Estimated Expiration
- 2041-11-24
AI Technical Summary
Existing bone density measurement methods are complex and inefficient, with low accuracy, making it difficult to effectively detect spinal areas and calculate bone density values.
The CT chest plain scanning sequence image was detected by an optimized target detection algorithm, combined with the spinal segmentation model and the regression bone density detection model, and the cancellous bone region was trained using the Densenet121 network structure to calculate the bone density value.
It improves the accuracy of spinal area detection and the accuracy of bone density calculation, improves the efficiency of bone density measurement, and achieves faster and higher measurement results.
Smart Images
Figure CN114240848B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of bone density measurement, and in particular to a bone density measurement method, system, storage medium and electronic equipment. Background Art
[0002] Bone density is an important indicator of bone quality. The degree of osteoporosis is an important basis for predicting the risk of fractures. Bone density tests can help detect calcium deficiency, so that children can be given calcium in a timely manner by adjusting their diet to ensure normal bone development. By evaluating bone density and determining the absorption and intake of osteogenic materials during growth and development, we can analyze internal and external factors that may affect their growth and development, so as to formulate targeted nutrition and exercise plans in a timely manner. The existing bone density measurement scheme is too complicated, and the actual operation process is inefficient, and the measurement accuracy is relatively low. Summary of the invention
[0003] The technical problem to be solved by the present invention is to provide a bone density measurement method, system, storage medium and electronic equipment.
[0004] The technical solution of the present invention to solve the above technical problem is as follows: a bone density measurement method, comprising:
[0005] Step 1, using the optimized target detection algorithm to detect the spine region of the CT chest plain scan sequence image to obtain a spine region image;
[0006] Step 2, inputting the spinal region image into a spinal segmentation model for segmentation to obtain a segmentation result;
[0007] Step 3, processing the segmentation result to obtain a cancellous bone region;
[0008] Step 4: training a regression bone density detection model based on the cancellous bone region, inputting the image to be measured into the regression bone density detection model, and obtaining a bone density value.
[0009] The beneficial effects of the present invention are as follows: the detection accuracy of the spinal area can be improved by optimizing the target detection algorithm, and the accuracy of the regression bone density detection model can be improved by obtaining the cancellous bone area. In addition, the calculation efficiency of bone density can be effectively improved by constructing a regression bone density detection model, which is more efficient and more accurate than the bone density calculation method in the prior art.
[0010] Based on the above technical solution, the present invention can also be improved as follows.
[0011] Furthermore, the process of obtaining the optimized target detection algorithm includes:
[0012] The size of the fixed reference frame in the target detection algorithm is set to a preset size, the output of the multi-scale feature layer in the target detection algorithm is removed, and the feature extraction layer in the target detection algorithm is replaced with a residual network structure to obtain the optimized target detection algorithm.
[0013] Furthermore, before step 2, the method further includes:
[0014] The spine region image is cropped according to a predetermined rule, and the cropped spine region image is enhanced to obtain an enhanced spine region image, and the enhanced spine region image is used as the spine region image in step 2 for processing.
[0015] The beneficial effect of adopting the above further solution is that the influence of useless information can be reduced and the segmentation speed and accuracy can be improved by preprocessing the spinal region image.
[0016] Furthermore, the segmentation result includes:
[0017] Thoracic vertebrae, first lumbar vertebrae, and second lumbar vertebrae.
[0018] Further, the step 3 is specifically as follows:
[0019] Step 301, respectively calculating the inclination angle of each part in the segmentation result;
[0020] Step 302, performing vertebral rotation on the corresponding part based on the inclination angle of each part, and cutting the image after the vertebral rotation to obtain a first image;
[0021] Step 303, removing cortical bone interference from the first image to obtain a second image;
[0022] Step 304: Calculate the vertebral center coordinates of the second image, and extract the cancellous bone region based on the vertebral center coordinates.
[0023] Furthermore, the cancellous bone region training regression bone density detection model specifically includes:
[0024] The cancellous bone area is used as input data, the bone density value of the cancellous bone area is used as a label, and the regression bone density detection model is trained through the Densenet121 network structure.
[0025] Another technical solution of the present invention to solve the above technical problem is as follows: a bone density measurement system, comprising:
[0026] A detection module, used to detect the spine region on the CT chest plain scan sequence images by using an optimized target detection algorithm to obtain a spine region image;
[0027] A segmentation module, used for inputting the spinal region image into a spinal segmentation model for segmentation to obtain a segmentation result;
[0028] A processing module, used for processing the segmentation result to obtain a cancellous bone region;
[0029] The measurement module is used to train a regression bone density detection model based on the cancellous bone area, input the image to be measured into the regression bone density detection model, and obtain the bone density value.
[0030] The beneficial effects of the present invention are as follows: the detection accuracy of the spinal area can be improved by optimizing the target detection algorithm, and the accuracy of the regression bone density detection model can be improved by obtaining the cancellous bone area. In addition, the calculation efficiency of bone density can be effectively improved by constructing a regression bone density detection model, which is more efficient and more accurate than the bone density calculation method in the prior art.
[0031] Furthermore, the process of obtaining the optimized target detection algorithm includes:
[0032] The size of the fixed reference frame in the target detection algorithm is set to a preset size, the output of the multi-scale feature layer in the target detection algorithm is removed, and the feature extraction layer in the target detection algorithm is replaced with a residual network structure to obtain the optimized target detection algorithm.
[0033] Furthermore, it also includes:
[0034] The optimization module is used to perform cropping on the spinal region image according to a predetermined rule, perform enhancement processing on the cropped spinal region image to obtain an enhanced spinal region image, and process the enhanced spinal region image as the spinal region image in the segmentation module.
[0035] The beneficial effect of adopting the above further solution is that the influence of useless information can be reduced and the segmentation speed and accuracy can be improved by preprocessing the spinal region image.
[0036] Furthermore, the segmentation result includes:
[0037] Thoracic vertebrae, first lumbar vertebrae, and second lumbar vertebrae.
[0038] Furthermore, the processing module is specifically used for:
[0039] The inclination angle of each part in the segmentation result is calculated respectively; based on the inclination angle of each part, the vertebrae of the corresponding part are rotated, and the image after the vertebral rotation is cropped to obtain a first image; the first image is processed to remove cortical bone interference to obtain a second image; the vertebral center coordinates of the second image are calculated, and the cancellous bone area is extracted based on the vertebral center coordinates.
[0040] Further, the training of the regression bone density detection model based on the cancellous bone region specifically includes:
[0041] The cancellous bone area is used as input data, the bone density value of the cancellous bone area is used as a label, and the regression bone density detection model is trained through the Densenet121 network structure.
[0042] Another technical solution of the present invention to solve the above technical problem is as follows: a storage medium, in which instructions are stored, and when a computer reads the instructions, the computer executes a bone density measurement method as described in any one of the above items.
[0043] The beneficial effects of the present invention are as follows: the detection accuracy of the spinal area can be improved by optimizing the target detection algorithm, and the accuracy of the regression bone density detection model can be improved by obtaining the cancellous bone area. In addition, the calculation efficiency of bone density can be effectively improved by constructing a regression bone density detection model, which is more efficient and more accurate than the bone density calculation method in the prior art.
[0044] Another technical solution of the present invention to solve the above technical problem is as follows: an electronic device, comprising a memory, a processor, and a program stored in the memory and running on the processor, wherein when the processor executes the program, a bone density measurement method as described in any one of the above items is implemented.
[0045] The beneficial effects of the present invention are as follows: the detection accuracy of the spinal area can be improved by optimizing the target detection algorithm, and the accuracy of the regression bone density detection model can be improved by obtaining the cancellous bone area. In addition, the calculation efficiency of bone density can be effectively improved by constructing a regression bone density detection model, which is more efficient and more accurate than the bone density calculation method in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 A schematic diagram of a flow chart of a bone density measurement method embodiment of the present invention;
[0047] Figure 2 A system framework diagram provided for a bone density measurement system embodiment of the present invention. DETAILED DESCRIPTION
[0048] The principles and features of the present invention are described below. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.
[0049] like Figure 1 A method for measuring bone density, comprising:
[0050] Step 1, using the optimized target detection algorithm to detect the spine region of the CT chest plain scan sequence image to obtain a spine region image;
[0051] Step 2, inputting the spinal region image into a spinal segmentation model for segmentation to obtain a segmentation result;
[0052] Step 3, processing the segmentation result to obtain a cancellous bone region;
[0053] Step 4: training a regression bone density detection model based on the cancellous bone region, inputting the image to be measured into the regression bone density detection model, and obtaining a bone density value.
[0054] In some possible implementations, the detection accuracy of the spinal region can be improved by optimizing the target detection algorithm, and the accuracy of the regression bone density detection model can be improved by obtaining the cancellous bone region. In addition, the calculation efficiency of bone density can be effectively improved by constructing a regression bone density detection model, which is more efficient and more accurate than the bone density calculation method in the prior art.
[0055] It should be noted that the optimized target detection algorithm is as follows: first, considering that the shape and size of the spine are relatively fixed, the size of the anchor (anchor: fixed reference frame) is set to (1, 1, 10) in length, width and height. In addition, the feature output layer removes the output of the multi-scale feature layer, and the fixed output layer is downsampled 32 times. The feature extraction layer is modified to use the residual network structure. At the same time, in order to expand the receptive field, the first three layers are downsampled and added with 3*3 void convolutions. Step 2 obtains the segmentation results of all supported thoracic vertebrae, the first lumbar vertebrae and the second lumbar vertebrae; Step 3 extracts the cancellous bone area together.
[0056] Processing the segmentation results includes calculating the spinal inclination angle and removing the interference of cortical bones. The complete process of this solution can be understood by referring to Example 1.
[0057] Example 1, data processing: CT chest plain scan sequence images, convert the dicom format to nii format. Spinal region target detection: Use the image data processed in the previous step to detect the spinal region. In order to reduce the impact of useless information and improve the segmentation speed and accuracy, the coordinate position of the spinal region detected in the second step is expanded by 10 pixels in each direction for cropping, and the cropped data is used as the input data for the next step of segmentation. The cropped data can completely contain the spinal region and a small part of the surrounding information. Spinal region segmentation: Using the relatively regular data cropped in the previous step, the data is enhanced (image normalization, flipping, etc.), and then input into the 3D Unet segmentation model for segmentation. The cropped data is relatively small, and the downsampling size is set to (16, 16, 64). The feature extraction network is a modified residual network structure, and the first two layers are added with hollow convolutions. The downsampled feature map is upsampled using bilinear interpolation. The segmentation result is finally output, and the thoracic 12, lumbar 1, and lumbar 2 vertebral regions in the segmentation result are used for bone density regression detection. The angle of the three vertebral areas segmented in the previous step is calculated. By traversing the angles and calculating the Iou of the two vertebrae after rotation, the angle of spinal inclination is calculated. The vertebrae are rotated according to the angle, and then cropped based on the segmentation position and the original image position.
[0058]
[0059] A and B are the images before and after rotation respectively. The ROI area (cancellous bone area, removing the interference of cortical bone) is extracted from the cropped vertebral body. The process is as follows:
[0060]
[0061] The pixel values of the original image and the values of the Mask image are used to pre-remove the influence of some cortical bones, where Mask is the segmentation result, src is the original image, thresh is the threshold for removing cortical bones, and (x, y, z) correspond to the coordinate positions on the image. Calculate the coordinates of the vertebral center, grow outward from the center coordinates, and extract the cancellous bone area; crop the cancellous bone area. Use the cancellous bone area of the spine cropped in the previous step, where each vertebra corresponds to a QCT bone density value. Use these two sets of information to train the regression bone density detection model, use the extracted ROI area as input data, and QCT floating point data as label training model. The bone density detection algorithm model is a modified Densenet121 network structure. Considering that the shallow semantic information of the vertebral texture is more important, the pooling layer before the Dense Block is removed during feature extraction, and the AvgPooling layer is added to the output layer. The output layer is modified to reduce the number of parameters and prevent overfitting. The output layer is 1 unit for training. At the same time, the output value during testing is the bone density value of the corresponding vertebra.
[0062] Preferably, in any of the above embodiments, the process of acquiring the optimized target detection algorithm includes:
[0063] The size of the fixed reference frame in the target detection algorithm is set to a preset size, the output of the multi-scale feature layer in the target detection algorithm is removed, and the feature extraction layer in the target detection algorithm is replaced with a residual network structure to obtain the optimized target detection algorithm.
[0064] It should be noted that setting the size of the fixed reference frame to the preset size can be understood as: setting the length, width and height of the anchor (anchor: fixed reference frame) to (1, 1, 10). In addition, the optimization of the target detection algorithm also includes: the feature output layer removes the output of the multi-scale feature layer, the fixed output layer is downsampled 32 times, the feature extraction layer is modified to use the residual network structure, and in order to expand the receptive field, the first three layers are downsampled and added with 3*3 void convolution.
[0065] Preferably, in any of the above embodiments, before step 2, the step further includes:
[0066] The spine region image is cropped according to a predetermined rule, and the cropped spine region image is enhanced to obtain an enhanced spine region image, and the enhanced spine region image is used as the spine region image in step 2 for processing.
[0067] In some possible implementations, the influence of useless information can be reduced and the segmentation speed and accuracy can be improved by preprocessing the spinal region image.
[0068] It should be noted that the predetermined rule is to expand the detected spine region coordinate position by 10 pixels in each direction for cropping. In addition, the number of expansions can be adjusted according to actual conditions, and the present invention only takes 10 as an example.
[0069] Preferably, in any of the above embodiments, the segmentation result includes:
[0070] Thoracic vertebrae, first lumbar vertebrae, and second lumbar vertebrae.
[0071] It should be noted that the segmentation result includes but is not limited to the above three positions, and the thoracic vertebrae include 12 vertebrae (T1-T12), the first lumbar vertebrae and the second lumbar vertebrae (L1-L2).
[0072] Preferably, in any of the above embodiments, step 3 is specifically:
[0073] Step 301, respectively calculating the inclination angle of each part in the segmentation result;
[0074] Step 302, performing vertebral rotation on the corresponding part based on the inclination angle of each part, and cutting the image after the vertebral rotation to obtain a first image;
[0075] Step 303, removing cortical bone interference from the first image to obtain a second image;
[0076] Step 304: Calculate the vertebral center coordinates of the second image, and extract the cancellous bone region based on the vertebral center coordinates.
[0077] It should be noted that the angle of the segmented three vertebral areas is calculated, and the angle of spinal inclination is calculated by traversing the angle and calculating the Iou of the two vertebrae after rotation. The vertebrae are rotated according to the angle, and then cropped in combination with the segmentation position and the original image position.
[0078]
[0079] A and B are the images before and after rotation respectively. The ROI area (cancellous bone area, removing the interference of cortical bone) is extracted from the cropped vertebral body. The process is as follows:
[0080]
[0081] Use the pixel values of the original image and the values of the Mask image to pre-remove the influence of some cortical bones, where Mask is the segmentation result, src is the original image, thresh is the threshold for removing cortical bones, and (x, y, z) correspond to the coordinate positions on the image. Calculate the center coordinates of the vertebral body, grow outward from the center coordinates, extract the cancellous bone area, and crop the cancellous bone area.
[0082] Preferably, in any of the above embodiments, the cancellous bone region training regression bone density detection model specifically includes:
[0083] The cancellous bone area is used as input data, the bone density value of the cancellous bone area is used as a label, and the regression bone density detection model is trained through the Densenet121 network structure.
[0084] It should be noted that the training of the regression bone density detection model specifically includes: using the cropped cancellous bone area of the spine, each of which corresponds to a QCT bone density value, using these two sets of information to train the regression bone density detection model, using the extracted ROI area as input data, and QCT floating-point data as labels to train the model. The bone density detection algorithm model is a modified Densenet121 network structure. Considering that the shallow semantic information of the vertebral texture is more important, the pooling layer before the Dense Block is removed during feature extraction, and the AvgPooling layer is added to the output layer. The output layer is modified to reduce the number of parameters and to prevent overfitting. The output layer is 1 unit for training. At the same time, the output value during testing is the bone density value of the corresponding spine.
[0085] Another technical solution of the present invention to solve the above technical problem is as follows: a bone density measurement system, comprising:
[0086] The detection module 100 is used to detect the spine region of the CT chest plain scan sequence images by using an optimized target detection algorithm to obtain a spine region image;
[0087] The segmentation module 200 is used to input the spinal region image into the spinal segmentation model for segmentation to obtain a segmentation result;
[0088] A processing module 300 is used to process the segmentation result to obtain a cancellous bone region;
[0089] The measurement module 400 is used to train a regression bone density detection model based on the cancellous bone area, input the image to be measured into the regression bone density detection model, and obtain a bone density value.
[0090] In some possible implementations, the detection accuracy of the spinal region can be improved by optimizing the target detection algorithm, and the accuracy of the regression bone density detection model can be improved by obtaining the cancellous bone region. In addition, the calculation efficiency of bone density can be effectively improved by constructing a regression bone density detection model, which is more efficient and more accurate than the bone density calculation method in the prior art.
[0091] Preferably, in any of the above embodiments, the process of acquiring the optimized target detection algorithm includes:
[0092] The size of the fixed reference frame in the target detection algorithm is set to a preset size, the output of the multi-scale feature layer in the target detection algorithm is removed, and the feature extraction layer in the target detection algorithm is replaced with a residual network structure to obtain the optimized target detection algorithm.
[0093] Preferably, in any of the above embodiments, it further comprises:
[0094] The optimization module is used to perform cropping on the spinal region image according to a predetermined rule, perform enhancement processing on the cropped spinal region image to obtain an enhanced spinal region image, and process the enhanced spinal region image as the spinal region image in the segmentation module 200.
[0095] In some possible implementations, the influence of useless information can be reduced and the segmentation speed and accuracy can be improved by preprocessing the spinal region image.
[0096] Preferably, in any of the above embodiments, the segmentation result includes:
[0097] Thoracic vertebrae, first lumbar vertebrae, and second lumbar vertebrae.
[0098] Preferably, in any of the above embodiments, the processing module 300 is specifically used for:
[0099] The inclination angle of each part in the segmentation result is calculated respectively; based on the inclination angle of each part, the vertebrae of the corresponding part are rotated, and the image after the vertebral rotation is cropped to obtain a first image; the first image is processed to remove cortical bone interference to obtain a second image; the vertebral center coordinates of the second image are calculated, and the cancellous bone area is extracted based on the vertebral center coordinates.
[0100] Preferably, in any of the above embodiments, the training of the regression bone density detection model based on the cancellous bone region specifically includes:
[0101] The cancellous bone area is used as input data, the bone density value of the cancellous bone area is used as a label, and the regression bone density detection model is trained through the Densenet121 network structure.
[0102] Another technical solution of the present invention to solve the above technical problem is as follows: a storage medium, in which instructions are stored, and when a computer reads the instructions, the computer executes a bone density measurement method as described in any one of the above items.
[0103] In some possible implementations, the detection accuracy of the spinal region can be improved by optimizing the target detection algorithm, and the accuracy of the regression bone density detection model can be improved by obtaining the cancellous bone region. In addition, the calculation efficiency of bone density can be effectively improved by constructing a regression bone density detection model, which is more efficient and more accurate than the bone density calculation method in the prior art.
[0104] Another technical solution of the present invention to solve the above technical problem is as follows: an electronic device, comprising a memory, a processor, and a program stored in the memory and running on the processor, wherein when the processor executes the program, a bone density measurement method as described in any one of the above items is implemented.
[0105] In some possible implementations, the detection accuracy of the spinal region can be improved by optimizing the target detection algorithm, and the accuracy of the regression bone density detection model can be improved by obtaining the cancellous bone region. In addition, the calculation efficiency of bone density can be effectively improved by constructing a regression bone density detection model, which is more efficient and more accurate than the bone density calculation method in the prior art.
[0106] The reader should understand that in the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples without contradiction.
[0107] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the method embodiments described above are only illustrative, for example, the division of steps is only a logical function division, and there may be other division methods in actual implementation, such as multiple steps can be combined or integrated into another step, or some features can be ignored or not executed.
[0108] If the above method is implemented in the form of 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 present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.
[0109] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.
Claims
1. A method for measuring bone density, characterized in that: include: Step 1, using the optimized target detection algorithm to detect the spine region of the CT chest plain scan sequence image to obtain a spine region image; Step 2, inputting the spinal region image into a spinal segmentation model for segmentation to obtain a segmentation result; Step 3, processing the segmentation result to obtain a cancellous bone region; Step 4, training a regression bone density detection model based on the cancellous bone area, inputting the image to be measured into the regression bone density detection model, and obtaining a bone density value; The process of obtaining the optimized target detection algorithm includes: The size of the fixed reference frame in the target detection algorithm is set to a preset size, the output of the multi-scale feature layer in the target detection algorithm is removed, and the feature extraction layer in the target detection algorithm is replaced with a residual network structure to obtain the optimized target detection algorithm; The segmentation results include: Thoracic vertebrae, first lumbar vertebrae, and second lumbar vertebrae; The step 3 is specifically as follows: Step 301, respectively calculating the inclination angle of each part in the segmentation result; Step 302, performing vertebral rotation on the corresponding part based on the inclination angle of each part, and cutting the image after the vertebral rotation to obtain a first image; Step 303, removing cortical bone interference from the first image to obtain a second image; Step 304: Calculate the vertebral center coordinates of the second image, and extract the cancellous bone region based on the vertebral center coordinates.
2. A bone density measurement method according to claim 1, characterized in that: The step 2 also includes: The spine region image is cropped according to a predetermined rule, and the cropped spine region image is enhanced to obtain an enhanced spine region image, and the enhanced spine region image is used as the spine region image in step 2 for processing.
3. A bone density measurement method according to claim 1, characterized in that: The training of the regression bone density detection model based on the cancellous bone region specifically includes: The cancellous bone area is used as input data, the bone density value of the cancellous bone area is used as a label, and the regression bone density detection model is trained through the Densenet121 network structure.
4. A bone density measurement system, characterized in that: include: A detection module, used to detect the spine region on the CT chest plain scan sequence images by using an optimized target detection algorithm to obtain a spine region image; A segmentation module, used for inputting the spinal region image into a spinal segmentation model for segmentation to obtain a segmentation result; A processing module, used for processing the segmentation result to obtain a cancellous bone region; A measurement module, used for training a regression bone density detection model based on the cancellous bone area, inputting the image to be measured into the regression bone density detection model, and obtaining a bone density value; The process of obtaining the optimized target detection algorithm includes: The size of the fixed reference frame in the target detection algorithm is set to a preset size, the output of the multi-scale feature layer in the target detection algorithm is removed, and the feature extraction layer in the target detection algorithm is replaced with a residual network structure to obtain the optimized target detection algorithm; The segmentation results include: Thoracic vertebrae, first lumbar vertebrae, and second lumbar vertebrae; The processing module is specifically used to: calculate the inclination angle of each part in the segmentation result respectively; rotate the vertebrae of the corresponding part based on the inclination angle of each part, and crop the image after the vertebral rotation to obtain a first image; remove cortical bone interference from the first image to obtain a second image; calculate the vertebral center coordinates of the second image, and extract the cancellous bone area based on the vertebral center coordinates.
5. A storage medium, characterized in that: The storage medium stores instructions, and when a computer reads the instructions, the computer executes a bone density measurement method according to any one of claims 1 to 3.
6. An electronic device, characterized in that: The invention comprises the storage medium as claimed in claim 5 and a processor for executing instructions in the storage medium.