CN113870098B

CN113870098B - Automatic Cobb angle measurement method based on spine layered reconstruction

Info

Publication number: CN113870098B
Application number: CN202111053719.8A
Authority: CN
Inventors: 李立; 肖凯; 刘涛; 李余隆; 董瑜
Original assignee: Wuhan University WHU
Current assignee: Wuhan University WHU
Priority date: 2021-09-09
Filing date: 2021-09-09
Publication date: 2024-09-24
Anticipated expiration: 2041-09-09
Also published as: CN113870098A

Abstract

The invention discloses a Cobb angle automatic measurement method based on spine layered reconstruction, which comprises the following steps: (1) Inputting a transverse CT sequence image, preprocessing DICOM original data, reconstructing a superimposed sagittal image in a layering manner, selecting a sagittal range of a spine by adopting a template matching method, and reconstructing a bone coronal image in a middle third position interval; (2) Training a bone coronal graph-based vertebra segmentation model based on a deep learning network, detecting and segmenting each vertebra of the spine by adopting the model, and optimizing an intervertebral space by threshold processing; (3) Extracting the central point of each vertebra, and performing curve fitting on the central point set by using a sixth order polynomial to obtain a spine curve; (4) And (3) calculating the curvature among all the characteristic points in the spine curve, and iteratively solving the Cobb angle.

Description

Automatic Cobb angle measurement method based on spine layered reconstruction

Technical Field

The invention relates to the field of medical artificial intelligence, and particularly provides a Cobb angle automatic measurement method based on spine layered reconstruction.

Background

In recent years, medical imaging technology and artificial intelligence technology are continuously and rapidly developed, computed tomography is widely applied to clinical diagnosis, and research on spine lesions based on CT scanning combining deep learning and intelligent medical treatment has extremely important practical significance and research value, and the processing and analysis of medical image data by utilizing the artificial intelligence technology can provide powerful auxiliary effect for modern medical diagnosis.

When scoliosis is clinically judged, the Cobb angle is a common research object, and the magnitude of the Cobb angle can accurately reflect the severity of the scoliosis of a human body. At present, the measuring mode of the Cobb angle is mainly manual measurement, firstly, upper vertebrae and lower vertebrae with the most obvious concave side of lateral bending are searched in a spine image to serve as upper end plates and lower end plates, then, straight lines passing through the upper end plates and the lower end plates are drawn manually, finally, an angle meter is adopted to measure the angle between the two straight lines to obtain the Cobb angle, the manual measurement speed is too slow, a great deal of manpower and time are required to be consumed, and the accuracy rate depends on the expertise and the proficiency of doctors. The method has the advantages that the position of each vertebral segment is positioned with high accuracy from the transverse CT sequence images, the spinal bending characteristics are automatically judged, and the Cobb angle is calculated, so that the method has great significance for medical researches such as scoliosis and the like.

Disclosure of Invention

The invention aims to solve the technical problem of providing a Cobb angle automatic measurement method based on spine layering reconstruction, which specifically comprises the following steps:

(1) Inputting a transverse CT sequence image, preprocessing DICOM original data, reconstructing a superimposed sagittal image in a layering manner, selecting a sagittal range of a spine by adopting a template matching method, and reconstructing a bone coronal image at a part of middle position;

(2) Training a bone coronal graph-based vertebra segmentation model based on a deep learning network, detecting and segmenting each vertebra of the spine by adopting the model, and optimizing an intervertebral space by threshold processing;

(3) Extracting the central point of each vertebra, and performing curve fitting on the central point set by using a polynomial to obtain a spine curve;

(4) And (3) calculating the curvature among all the characteristic points in the spine curve, and iteratively solving the Cobb angle.

Further, the specific implementation manner of the step (1) is as follows;

(11) The CT sequence image refers to a transverse CT sequence image obtained by CT examination of chest and abdomen, at least comprises chest vertebrae and most lumbar vertebrae, and contains a complete spine shape;

(12) Sequencing the sequence pictures according to the time sequence of CT scanning according to the header file label field INSTANCENUMBER of the CT sequence images;

(13) According to DICOM image data, gray value statistics is carried out to determine a gray value threshold value of a bone region and a gray value threshold value of a metal object, and possible metal object influence is eliminated through threshold value limitation;

(14) Determining an effective DICOM image data range, setting a traversal range, and eliminating the influence generated by a CT machine bed plate;

(15) Based on the steps, each DICOM image is traversed and processed according to a CT scanning sequence, effective data of a bone part are extracted, a bone sagittal image is reconstructed in a three-dimensional mode, and the bone sagittal image is actually projection of three-dimensional data on a certain plane;

(16) Taking thoracic vertebrae and lumbar vertebrae areas of the spine in any bone sagittal image as characteristic templates, and binarizing the generated bone sagittal image and the characteristic templates;

(17) Performing multi-scale template matching on the binarized characteristic template and the bone coronal graph, and drawing a rectangular frame on a matching area;

(18) Returning to the abscissa x0 of the rectangular central point to obtain the sagittal range of the spine;

(19) And selecting a position interval of the middle third of the sagittal range of the spine according to x0 to reconstruct a bone coronal map.

Further, in the step (13), the pixel gray value threshold value for distinguishing the bone region and the metal object is taken

4. The automatic Cobb angle measurement method based on the layered reconstruction of the spine as claimed in claim 2, wherein: in the step (14), the image size is 512×512, and when the portion with the y-axis direction larger than 450 contains the bed board and has almost no effective pixel information, the traversing range of the y-axis is set to be 0-450.

Further, the specific implementation manner of training the vertebra segmentation model based on the bone coronal image in the step (2) is as follows;

(21) Acquiring a plurality of bone coronal images based on layered reconstruction as training set images;

(22) Labeling the training set picture by taking each vertebra of the bone coronal image as the ROI to generate a training label picture;

(23) Training based on a deep learning U-Net network, and adjusting training parameters to obtain a vertebra segmentation model based on a skeleton coronary chart;

(24) Performing vertebrae segmentation by using the trained model and storing segmentation results;

(25) And optimizing the intervertebral space by using threshold processing to obtain a binary vertebra segmentation image, and storing a result.

Further, the specific implementation manner of extracting the center point for curve fitting in the step (3) is as follows;

(31) Extracting the outline of the binary vertebrae segmentation image;

(32) Extracting the central point of each contour, namely the central point of each vertebral segment by using the first moment;

(33) And performing curve fitting on the central points of the spines by using polynomials to obtain a fitting curve of the spines.

Further, in the step (4), the specific implementation process of calculating the Cobb angle according to the curvature is as follows;

(41) Calculating first-order derivatives of the fitted spine curves at the central points to obtain curvatures of the different central points;

(42) Iterative calculation of included angles among tangents of different center points according to curvatures of the different center points, namely slopes of the tangents;

(43) Judging the size of the included angle, wherein the largest included angle is the Cobb angle.

The beneficial effects of the invention are as follows: according to the automatic Cobb angle measuring method based on the layered spine reconstruction, each section of vertebral segment is automatically and accurately found without manual assistance, the spine bending characteristics are judged, the Cobb angle is calculated, and medical research and analysis on scoliosis are facilitated.

Drawings

FIG. 1 is a flow chart of the automatic Cobb angle measurement of the present invention.

FIG. 2 is a template of a multi-scale template matching.

Fig. 3 is a result of multi-scale template matching.

Fig. 4 is a reconstructed bone coronal view.

FIG. 5 is a schematic view of the thoracic and lumbar regions;

FIG. 6 is a schematic view of the traversal range in step 4 of the embodiment;

FIG. 7 is a multi-planar sample of a medical image;

FIG. 8 is a schematic cross-sectional view;

fig. 9 is a sagittal view schematic.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings.

In the prior art, cobb angle automatic measurement is performed on an X-ray image, and the X-ray image is planar imaging, the invention adopts a double-helix CT image which is cross section sequence imaging, has three-dimensional data of CT double helix, can perform data processing according to the characteristics of vertebrae, generates an effective coronal image, and removes interference, which is the problem that cannot be solved by X-ray imaging, so that the two images are essentially different in imaging principle, data latitude and accuracy. The invention focuses on layered reconstruction, firstly, a sagittal image is reconstructed according to a CT image, as shown in fig. 3 (side), and then a three-dimensional coronary image of the spine is reconstructed according to the sagittal image, as shown in fig. 4 (front), wherein the coronary image is an overlapped coronary image of the CT image and is formed by overlapping a plurality of coronary images (the problem that the superposition of x-ray imaging sternum and ribs can not effectively identify the spine can be completely overcome), so that the method not only comprises clear vertebral segments and is beneficial to segmentation, but also removes the interference of factors such as bone organs. And finally, carrying out unet segmentation on the coronal image to obtain each spinal column block, thereby obtaining a center point and obtaining a cobb angle by curve fitting.

As shown in fig. 1, the Cobb angle automatic measurement method based on spine layering reconstruction comprises the following steps:

(1) Inputting a transverse CT sequence image, preprocessing DICOM original data, reconstructing a superimposed sagittal image in a layering manner, selecting a sagittal range of a spine by adopting a template matching method, and reconstructing a bone coronal image at a middle third position;

The method for reconstructing the skeleton coronary image based on the transverse CT sequence image layering in the step (1) comprises the following steps:

(11) The CT sequence image refers to a transverse CT sequence image obtained by CT examination of chest and abdomen, at least comprises chest vertebrae and most lumbar vertebrae, and contains a complete spine shape.

The main characteristic of the thoracic vertebrae is that the thoracic vertebrae are ribbed and have ribs connected with them. The lumbar vertebra is characterized in that the vertebral body is large, the spinous process is plate-shaped, but no rib is connected with the spinous process, two sides of the thoracic vertebra T12 are connected with the rib, and two sides of the lumbar vertebra bone part are only provided with transverse processes and no rib is connected with the transverse processes. Thus, the thoracic and lumbar vertebrae areas can be divided by using the thoracic bones T12 and the lumbar vertebrae L1 as features, and as shown in fig. 5, only the thoracic and lumbar vertebrae including most of them show the complete spine, thereby judging the degree of curvature of the spine.

The normal range threshold of each non-human tissue is generally >1800, the range of human skeleton and calcium is approximately 1200-1800, when the reconstructed pixel threshold is taken In this case, a bone sagittal image or a coronal image can be generated after the influence of the metallic foreign matter is eliminated.

setting the pixel range of the reconstruction region, the image size is 512 x 512, and traversing all pixel points of the image, but the part with the y-axis direction larger than 450 comprises a CT bed board and does not influence the generation of a bone coronal image or a sagittal image, so the y-axis traversing range is set to be 0-450, as shown in fig. 6.

(15) Based on the steps, each DICOM image is processed through traversing according to a CT scanning sequence, and effective data of a bone part is extracted to reconstruct a bone sagittal image in a three-dimensional mode (reconstruction of the sagittal image is actually projection of three-dimensional data on a certain plane);

As shown in FIG. 7, the sagittal plane refers to a longitudinal plane that divides the body into left and right parts, the plane being perpendicular to the ground plane, and the left and right aspects of the anatomy being visible; the coronal plane refers to a longitudinal section dividing the body into anterior and posterior parts, the section being visible in the anterior-posterior aspect of the anatomy, the section being perpendicular to the sagittal and horizontal planes. The cross section refers to a cross section dividing a human body into an upper part and a lower part, as shown in the following figure 8, because the acquired image data are CT cross section data and three-dimensional data, the three-dimensional gray data are directly projected and displayed on a two-dimensional plane according to different directions during multi-plane reconstruction, and a tomographic image with any angle can be generated, and the specific implementation method is as follows: the effective data is averaged to obtain pixel values, and the coronal plane is generated by projection in the y-axis direction and the sagittal plane is generated by projection in the x-axis direction. When generating a bone coronal or sagittal view, the pixel threshold described in (3) is used to take 1200-1800, and other useless pixel values are removed, and the resulting sagittal view is shown in fig. 9.

(17) Performing multi-scale template matching on the binarized characteristic template and the bone coronal graph, and drawing a red rectangular frame for a matching area;

The reconstructed crown is shown in fig. 4, with a selected third of the area corresponding to the inner small rectangular box in fig. 3 and the area in cross section corresponding to the lateral rectangular box in fig. 8.

The method for training the vertebrae segmentation model based on the bone coronal map and performing vertebrae segmentation in the step (2) comprises the following steps:

The method for extracting the center point to perform curve fitting in the step (3) comprises the following steps:

(31) Extracting the outline of the binary vertebrae segmentation image;

The method for calculating the Cobb angle according to the curvature in the step (4) comprises the following steps:

The foregoing description clearly illustrates the technical solution, flow and advantages of the present invention, and it is obvious to those skilled in the art that the present invention is not limited by the foregoing embodiments, but the above embodiments and descriptions are merely not representative of the technical solution and principles of the present invention, and the present invention makes improvements of the corresponding algorithm without departing from the spirit and content of the present invention, and all the improvements fall within the scope of the present invention claimed, and the experimental results of the present invention are realized in the specific form, and the scope of the present invention is defined by the appended claims and equivalents.

Claims

1. The automatic Cobb angle measurement method based on the spine layering reconstruction is characterized by comprising the following steps:

2. The automatic Cobb angle measurement method based on the layered reconstruction of the spine as claimed in claim 1, wherein: the specific implementation mode of the step (1) is as follows;

3. The automatic Cobb angle measurement method based on the layered reconstruction of the spine as claimed in claim 2, wherein: step (13) pixel gray value threshold value taking for distinguishing skeleton region and metal object

5. The automatic Cobb angle measurement method based on the layered reconstruction of the spine as claimed in claim 1, wherein: the specific implementation mode of training the vertebrae segmentation model based on the bone coronal diagram in the step (2) is as follows;

6. The automatic Cobb angle measurement method based on the layered reconstruction of the spine as claimed in claim 1, wherein: the specific implementation mode of extracting the center point for curve fitting in the step (3) is as follows;

(31) Extracting the outline of the binary vertebrae segmentation image;

7. The automatic Cobb angle measurement method based on the layered reconstruction of the spine as claimed in claim 1, wherein: in the step (4), the specific implementation process of calculating the Cobb angle according to the curvature is as follows;