Spine bone structure and bone mineral density automatic measuring method based on CT image

Abstract

The invention relates to a spine bone structure and bone mineral density automatic measurement method based on a CT image. The method comprises the following steps: performing CT scanning on a phantom with known bone mineral density to obtain a real bone mineral density calculation formula; all CT images in the DICOM format are read, DICOM analysis is carried out, and three-dimensional CT value data and other main information of all the images are obtained; performing outer contour segmentation on the whole spine in the CT image by using a deep learning network to obtain an outer contour of each vertebral body in the image; the method comprises the following steps: carrying out preliminary segmentation on cortical bones and cancellous bones by adopting a dynamic threshold method aiming at each vertebral body, selecting a preliminarily segmented cancellous bone region, using a maximum connected domain algorithm in a three-dimensional direction, retaining a maximum connected region, and further measuring the bone mineral density, namely calculating overall topology parameters, the bone mineral density and the bone mineral content of the current vertebral body; calculating the rubbing parameters, the bone mineral density and the bone mineral content of the current vertebral cortical bone and cancellous bone; and measuring the content of the cross section of the cone.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical technology, in particular to a method for automatically measuring spinal bone structure and bone density based on CT images. BACKGROUND

[0002] When osteopenia occurs, the shape of the bone mass does not change significantly, but the cortical bone becomes thinner, the bone marrow cavity gradually expands, and the trabeculae become thinner, shorter and fewer. Therefore, measuring the cancellous bone and cortical bone separately can reflect the osteoporosis state early. The current quantitative CT (also known as QCT) evaluates the cortical bone density and cancellous bone density through clinical CT scanning technology, which can indicate the early osteopenia of cancellous bone and cortical bone, and to some extent, can be used as an early indicator of bone strength changes. When using quantitative CT to detect bone density, some methods require manual delineation of the areas of cancellous bone and cortical bone on the CT image. However, due to different judgment standards of different staff for the delineated areas, the consistency of the results is poor, which leads to errors in the final calculated bone density results, and such methods cannot achieve fully automated measurement; some other methods use fixed empirical values, although they can achieve fully automated measurement, but since the human body grows and the skeleton changes automatically, and there are greater differences between individuals, the use of fixed empirical values is not universal, and the calculated results are not objective and cannot be quantitatively evaluated; in addition, some methods can recognize bone, but cannot distinguish between cancellous bone and cortical bone, and cannot perform specific data calculation and quantitative evaluation for the two different bone structures (cancellous bone and cortical bone) in the same area.

[0003] Chinese patent application CN202211541888.0 discloses a bone structure recognition method and system, which includes performing gray image conversion on the obtained bone image, and using a neural network algorithm of machine learning to remove the side small bones or interfering small bones and various fixed distance empirical value parameters to obtain the target bone region and the cortical bone and cancellous bone regions of the target bone. The technical problem solved by the above-mentioned disclosed patent is to obtain the target bone region and the cortical bone and cancellous bone regions of the target bone by relying on various fixed empirical value parameters, i.e., to bring multiple subjective empirical values into the method and system to complete automatic recognition calculation, which is not universal, and the calculated results are not objective and cannot be quantitatively evaluated; in addition, the bone density calculation method and formula provided by the method and system are completely wrong (formula: cancellous bone density = cancellous bone area divided by inner contour area. First of all, the calculation result is the ratio of areas; secondly, the two areas are not of the same material density, and the bone density is not uniform, so the calculation result has nothing to do with the density, not to mention bone density). SUMMARY

[0004] The present application aims at the deficiencies of the prior art, and provides a CT image-based automatic measurement method for spinal bone structure.

[0005] To solve the above technical problems, the present application adopts the following technical solutions: The CT image-based automatic measurement method for spinal bone structure comprises the following steps: Step one: CT scanning is performed on a body phantom with known bone density, and a DICOM format CT image is obtained, the body phantom ROI is circled, and the body phantom CT value is obtained, the linear correspondence between the body phantom CT value and the body phantom bone density is established to obtain the correspondence between the CT value and the real bone density (i.e. the real bone density calculation formula is obtained); Step two: all DICOM format CT images are read, and DICOM analysis is performed to obtain three-dimensional CT value data of all images, actual length between X-axis, Y-axis and Z-axis pixel points, image direction, position information of the human body relative to the CT device, gantry tilt angle and bed tilt angle, device name, KVP, FOV and other main information; Step three: a deep learning network is used to perform outer contour segmentation on the whole spine in the CT image to obtain the outer contour of each vertebral body in the image; Step four: a dynamic threshold method is used for preliminary segmentation of cortical bone and cancellous bone for each vertebral body, and then the largest connected region algorithm is used in the three-dimensional direction to select the preliminary segmented cancellous bone region to retain the largest connected region, and the outer contour of the region is the boundary between the cortical bone and the cancellous bone, and the cortical bone and the cancellous bone are segmented.

[0006] Preferably, the body phantom with bone density in step one is a composite material of hydroxyapatite and polyethylene and other materials which is close to the performance of bone mineral in absorbing and attenuating X-rays, and the density of hydroxyapatite is known, the scanning parameters of the body phantom and the CT shooting parameters of the real human body are consistent, i.e. KVP, FOV and other parameters are used when the human body is shot, and the body phantom is scanned with the same scanning parameters and the body phantom calibration is performed on these images one by one, during the calibration, the correspondence between the CT value and the real bone density is calculated according to the linear relationship between the CT value of the body phantom ROI and the bone density of the body phantom, and after the calibration, the correspondence between the CT value and the real bone density is saved according to the device name, KVP and other main parameters as the key words.

[0007] Preferably, the deep learning network segmentation whole spine method in step three is to use nnUNet network as the backbone network, the same number of encoder and decoder network layers are used, the encoder uses multi-layer convolution and down-sampling, the decoder uses multi-layer deconvolution and up-sampling, the spine is labeled by multi-classification labeling according to the vertebral body part, and the training is also carried out by multi-classification, and the outer contour of each vertebral body in the three-dimensional direction can be obtained after prediction, and the outer contour of each vertebral body is the combined area of the cortical bone and cancellous bone of the vertebral body, wherein the vertebral foramen is segmented out of the area (at this time, there is a hollow vertebral foramen area in the shape of the outer contour of each vertebral body). For each vertebral body, find the hollow vertebral foramen area, find the vertebral foramen center coordinates of each layer of the transverse section of the vertebral body, remove the vertebral arch plate, transverse process and spinous process area by the image direction obtained in step 2 and the position information of the human body relative to the CT device, and obtain the outer contour of the vertebral body and part of the vertebral pedicle.

[0008] Preferably, the dynamic threshold method in step four: first, automatic threshold analysis is performed, a gray level histogram is made for all discrete points of the image f(x, y, z), and then a Levenberg-Marquardt method is used to perform nonlinear least squares optimization to obtain an optimal fitting function, so as to obtain the coefficients of the two Gaussian distributions, that is, , and further the intersection point of the two distributions can be obtained, and the CT value corresponding to the intersection point is the threshold value of the cortical bone and cancellous bone boundary to be found.

[0009] Another object of the present application is to provide a CT image-based automatic measurement method for spinal bone bone density, The technical scheme adopted is as follows: The CT image-based automatic measurement method for spinal bone bone density comprises the following steps: Calculate the overall topological parameters, bone density and bone mineral content of the current vertebral body; Calculate the topological parameters, bone density and bone mineral content of the cortical bone and cancellous bone of the current vertebral body, respectively; Measure the topological parameters, bone density and bone mineral content of each layer of the transverse section of the vertebral body (the measurement is only performed on the thoracic vertebrae and lumbar vertebrae).

[0010] Preferably, according to the actual length between the X-axis, Y-axis and Z-axis pixel points, the obtained gantry tilt angle and bed tilt angle and the obtained vertebral body outer contour area, the topological parameters of the current vertebral body, including volume, maximum height, maximum width of X-axis and Y-axis and centroid principal axis, are calculated. According to the corresponding relationship between the CT value and the true bone density, the obtained vertebral body outer contour area and the CT value of each pixel in the area, the average bone density value and the bone mineral content of the whole vertebral body are calculated.

[0011] Preferably, according to the actual length between the acquired X-axis Y-axis Z-axis pixel points, the acquired gantry tilt angle and the bed tilt angle, and the obtained vertebral cortical bone region and cancellous bone region, the current vertebral cortical bone volume is calculated, and the current vertebral cancellous bone volume is calculated. According to the correspondence between the CT value and the true bone density, and the obtained vertebral cortical bone region, cancellous bone region and CT value of each pixel, the current vertebral cortical bone average bone density value and bone mineral content are calculated, and the current vertebral cancellous bone average bone density value and bone mineral content are calculated.

[0012] Preferably, according to the actual length between the acquired X-axis Y-axis Z-axis pixel points, the acquired gantry tilt angle and the bed tilt angle, and the obtained vertebral cortical bone region and cancellous bone region, the current cross-sectional cortical bone circumference, cortical bone volume, cortical bone area, cortical bone maximum width, cortical bone average width, cancellous bone circumference, cancellous bone volume, cancellous bone area, cancellous bone maximum width, cancellous bone average width, curvature, curvature radius, flexion ratio, and eccentricity are calculated. According to the correspondence between the CT value and the true bone density, and the obtained calculation of the current cross-sectional cortical bone average bone density value and bone mineral content, and the calculation of the current cross-sectional cancellous bone average bone density value and bone mineral content.

[0013] Beneficial effects:

[0014] The present application uses a three-dimensional segmentation and measurement method based on the original CT image, which can objectively and automatically measure and calculate the cortical bone and cancellous bone topological parameters, bone mechanical parameters, bone density and bone mineral content of each cone of the spine, and the measurement and calculation is accurate and efficient. BRIEF DESCRIPTION OF DRAWINGS

[0015] The present application will be further described below in conjunction with the drawings: Figure 1 The flowchart of the automatic measurement method of the spinal bone structure and bone density based on the CT image is shown. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings.

[0017] Embodiment 1

[0018] The detailed operation steps of the automatic measurement method of the spinal bone structure based on the CT image are as follows: Pre-calibration before bone density measurement: by CT scanning the phantom with known bone density and obtaining the DICOM format CT image, the ROI of the phantom is circled and the CT value of the phantom is obtained, and the corresponding relationship between the CT value and the true bone density is obtained by establishing the linear correspondence between the CT value and the bone density of the phantom (i.e. obtaining the true bone density calculation formula); The phantom with known bone mineral density in the method is a composite material of hydroxyapatite and polyethylene, which is close to the performance of bone mineral in absorbing and attenuating X-rays, and the density of hydroxyapatite is known. The scanning parameters of the phantom and the CT shooting parameters of the real human body are consistent, that is, the same KVP, FOV and other parameters are used when shooting the human body, and the same scanning parameters are used to scan the phantom and the images are calibrated one by one. During calibration, the corresponding relationship between the CT value and the true bone density is calculated according to the linear relationship between the CT value of the phantom ROI and the bone density of the phantom. After calibration, the corresponding relationship between the CT value and the true bone density is saved according to the device name, KVP and other main parameters as the key.

[0019] Select the measurement type (i.e. spine) and the CT image to be measured, and execute the image loading; Read all CT images in DICOM format from the disk, and perform DICOM analysis to obtain three-dimensional CT value data, actual length between X-axis, Y-axis and Z-axis pixel points, image direction and human body relative to CT device position information, gantry tilt angle and bed tilt angle, device name, KVP, FOV and other main information.

[0020] The deep learning network is used to segment the whole spine in the CT image to obtain the outer contour of each vertebral body in the image. The deep learning network segmentation method of the whole spine in the application uses nnUNet network as the backbone network, and the number of layers of the encoder and the decoder network is the same, wherein the encoder uses multi-layer convolution and down-sampling, and the decoder uses multi-layer deconvolution and up-sampling. When labeling the spine, multi-classification labeling is performed according to the vertebral body part, and during training, multi-classification training is also performed. After prediction, the outer contour of each vertebral body in the three-dimensional direction can be obtained. The outer contour of each vertebral body is the combined area of the cortical bone and cancellous bone of the vertebral body, and the vertebral foramen is segmented out of this area (at this time, there is a hollow vertebral foramen area in the shape of the outer contour of each vertebral body).

[0021] For each vertebral body, find the hollow vertebral foramen area, find the vertebral foramen center coordinates of each layer of the transverse section of the vertebral body, and remove the vertebral arch plate, transverse process and spinous process area by using the image direction and human body relative to CT device position information obtained in step 2. In this way, the outer contour of the vertebral body and part of the vertebral pedicle is obtained.

[0022] The cortical bone and cancellous bone are automatically segmented for each vertebral body respectively; The dynamic threshold method is used for the preliminary segmentation of the cortical bone and cancellous bone for each vertebral body, and then the largest connected region is reserved by using the maximum connected region algorithm in the three-dimensional direction of the selected cancellous bone region which is preliminarily segmented, and the outer contour of the region is the boundary between the cortical bone and the cancellous bone, so that the cortical bone and the cancellous bone are accurately segmented.

[0023] The dynamic threshold method: first, automatic threshold analysis is performed, and a gray histogram is made for all discrete points of the image f(x, y, z), and then the Levenberg-Marquardt method is used for non-linear least square optimization to obtain the optimal fitting function, so as to obtain the coefficients of the two Gaussian distributions, i.e. , and further, the intersection point of the two distributions can be obtained, and the CT value corresponding to the intersection point is the threshold value of the boundary between the cortical bone and the cancellous bone.

[0024] Example 2

[0025] The CT image-based automatic measurement method of the spinal bone bone density automatically performs further topological, bone mechanical, bone density and bone mineral content measurement calculation for the cortical bone and cancellous bone regions of each vertebral body respectively. This step performs detailed measurement and calculation for each vertebral body of the spine, and the detailed measurement and calculation steps of each vertebral body are the same. The measurement and calculation steps of a vertebral body are described in detail as follows: a. Calculate the overall topological parameters, bone density and bone mineral content of the current vertebral body: According to the actual lengths between the X-axis, Y-axis and Z-axis pixel points obtained in step 2, the gantry inclination angle and the bed inclination angle, and the outer contour region of the vertebral body obtained in step 3, the topological parameters of the current vertebral body, such as volume, maximum height, maximum width of X-axis and Y-axis, and centroid principal axis, are calculated.

[0026] According to the corresponding relationship between the CT value and the true bone density in step 1 and the outer contour region of the vertebral body and the CT value of each pixel in the region obtained in step 3, the average bone density value and the bone mineral content of the entire vertebral body are calculated.

[0027] b. Calculate the topological parameters, bone density and bone mineral content of the cortical bone and cancellous bone of the current vertebral body respectively: According to the actual lengths between the X-axis, Y-axis and Z-axis pixel points obtained in step 2, the gantry inclination angle and the bed inclination angle, and the cortical bone region and the cancellous bone region of the vertebral body obtained in step 3.1, the volume of the cortical bone of the current vertebral body is calculated, and the volume of the cancellous bone of the current vertebral body is calculated.

[0028] According to the correspondence between CT value and true bone density in step 1 and the cortical bone region, cancellous bone region and CT value of each pixel obtained in step 3.1, the current vertebral cortical bone average bone density value and bone mineral content are calculated, and the current vertebral cancellous bone average bone density value and bone mineral content are calculated.

[0029] c. Topology, bone mechanics, bone density and bone mineral content measurement are performed on each layer cross section where the vertebral body is located (the measurement is only performed on thoracic vertebrae and lumbar vertebrae): According to the actual length between X-axis, Y-axis and Z-axis pixel points obtained in step 2, the obtained gantry tilt angle and bed tilt angle and the cortical bone region and cancellous bone region obtained in step 3.1, the current cross section cortical bone circumference, cortical bone volume, cortical bone area, cortical bone maximum width, cortical bone average width, cancellous bone circumference, cancellous bone volume, cancellous bone area, cancellous bone maximum width, cancellous bone average width, curvature, curvature radius, flexion ratio and centrifugal rate are calculated.

[0030] According to the correspondence between CT value and true bone density in step 1 and the cortical bone region, cancellous bone region and CT value of each pixel obtained in step 3.1, the current vertebral cortical bone average bone density value and bone mineral content are calculated, and the current vertebral cancellous bone average bone density value and bone mineral content are calculated.

[0031] The embodiments of the present application are described in detail above, but the present application is not limited to the described embodiments. For those skilled in the art, various changes, modifications, replacements and variations of the embodiments can be made without departing from the principles and spirits of the present application, and still fall within the protection scope of the present application.

Claims

1. A method for automatic measurement of spinal bone structures based on CT images, characterized by, The method comprises the following steps: Step 1: CT scanning of a body phantom with known bone density to obtain DICOM format CT images, circumscribing the ROI of the body phantom and obtaining the CT value of the body phantom, and establishing a linear correspondence between the CT value and the bone density of the body phantom to obtain the correspondence between the CT value and the real bone density; Step 2: reading all DICOM format CT images and performing DICOM analysis to obtain three-dimensional CT value data of all images, actual length between X-axis, Y-axis and Z-axis pixel points, image direction and position information of the human body relative to the CT device, gantry tilt angle and bed tilt angle, device name, KVP, FOV and other main information; Step 3: using a deep learning network to perform contour segmentation on the entire spine in the CT image to obtain the contour of each vertebral body in the image; Step 4: using a dynamic threshold method to preliminarily segment the cortical bone and cancellous bone for each vertebral body, and then selecting the preliminarily segmented cancellous bone region to use a maximum connected region algorithm in the three-dimensional direction to retain the largest connected region, and the contour of the region is the boundary between the cortical bone and the cancellous bone, thereby separating the cortical bone and the cancellous bone.

2. The CT image-based automatic measurement method of a spinal bone structure according to claim 1, characterized by, The body phantom with known bone density in step 1 is a composite material of hydroxyapatite and polyethylene and other materials, which is close to the performance of bone mineral in absorbing and attenuating X-rays, and the density of hydroxyapatite is known. The scanning parameters of the body phantom are consistent with the CT shooting parameters of the real human body, that is, the same scanning parameters are used to scan the body phantom and the images are calibrated one by one. During calibration, the correspondence between the CT value and the real bone density is calculated according to the linear relationship between the CT value of the body phantom ROI and the bone density of the body phantom. After calibration, the correspondence between the CT value and the real bone density is saved according to the main parameters such as device name and KVP as the key.

3. The method of claim 1, wherein the method further comprises: The deep learning network segmentation method in step 3 uses nnUNet network as the backbone network, and the number of layers of the encoder and the decoder network is the same. The encoder uses multi-layer convolution and downsampling, and the decoder uses multi-layer deconvolution and upsampling. When labeling the spine, multi-classification labeling is performed according to the vertebral body part. During training, multi-classification training is also performed. After prediction, the contour of each vertebral body in the three-dimensional direction can be obtained. The contour of each vertebral body is the combined region of the cortical bone and the cancellous bone of the vertebral body, and the vertebral foramen is segmented out of this region (at this time, there is a hollow vertebral foramen region in the shape of the contour of each vertebral body); For each vertebral body, find the hollow vertebral foramen region, find the vertebral foramen center coordinates of each layer of the transverse section of the vertebral body, and remove the vertebral arch plate, transverse process and spinous process region by using the image direction and position information of the human body relative to the CT device obtained in step 2 to obtain the contour of the vertebral body and part of the vertebral arch.

4. The method of claim 1, wherein the CT image-based spinal bone structure and bone density automatic measurement method is characterized by, The dynamic threshold method described in step four: first, automatic threshold analysis is performed, and a gray level histogram is made for all discrete points of the image f(x, y, z), and then a Levenberg-Marquardt method is used to perform a non-linear least square optimization to obtain an optimal fitting function, so as to obtain the coefficients of two Gaussian distributions, i.e. , and further, the intersection point of the two distributions can be obtained, and the CT value corresponding to the intersection point is the boundary threshold value of the cortical bone and the cancellous bone that we want to find.

5. A method for further measuring the bone density of the spine based on the CT image-based automatic measurement method of the spine bone structure according to any one of claims 1-4, comprising the following steps: calculating the overall topological parameters, bone density and bone mineral content of the current vertebral body; Do not calculate the current vertebral cortical bone and cancellous bone topology parameters, bone density and bone mineral content; Topological, bone mechanics, bone density, and bone mineral content measurements are performed on each transverse section of the vertebral body (the measurements are only performed on thoracic and lumbar vertebrae).

6. The CT image-based automatic measurement method of spinal bone bone density according to claim 5, characterized in that, According to the actual length between the X-axis, Y-axis and Z-axis pixels, the obtained gantry tilt angle and bed tilt angle, and the obtained outer contour region of the vertebral body, the topology parameters of the current vertebral body, volume, maximum height, X-axis and Y-axis maximum width, and centroid principal axis are calculated. According to the correspondence between the CT value and the true bone density, and the obtained outer contour region of the vertebral body and the CT value of each pixel in the region, the average bone density value and bone mineral content of the entire vertebral body are calculated.

7. The CT image-based automatic measurement method of spinal bone bone density according to claim 5, characterized in that, According to the actual length between the X-axis, Y-axis and Z-axis pixels, the obtained gantry tilt angle and bed tilt angle, and the obtained cortical bone region and cancellous bone region of the vertebral body, the current vertebral cortical bone volume is calculated, and the current vertebral cancellous bone volume is calculated. According to the correspondence between the CT value and the true bone density, and the obtained cortical bone region and cancellous bone region of the vertebral body and the CT value of each pixel, the average bone density value and bone mineral content of the current vertebral cortical bone are calculated, and the average bone density value and bone mineral content of the current vertebral cancellous bone are calculated.

8. The CT image-based automatic measurement method of spinal bone bone density according to claim 5, characterized in that, According to the actual length between the X-axis, Y-axis and Z-axis pixels, the obtained gantry tilt angle and bed tilt angle, and the obtained cortical bone region and cancellous bone region of the vertebral body, the current transverse section cortical bone circumference, cortical bone volume, cortical bone area, cortical bone maximum width, cortical bone average width, cancellous bone circumference, cancellous bone volume, cancellous bone area, cancellous bone maximum width, cancellous bone average width, curvature, curvature radius, flexion ratio, and eccentricity are calculated. According to the correspondence between the CT value and the true bone density, and the obtained current transverse section cortical bone average bone density value and bone mineral content, and the obtained current transverse section cancellous bone average bone density value and bone mineral content are calculated.

Images