A method for segmenting cone point clouds from three-dimensional images

Abstract

This invention discloses a method for segmenting vertebral body point clouds in three-dimensional images, comprising: converting a three-dimensional image containing the spine into a three-dimensional point cloud, and filtering to obtain point clouds of vertebral body regions; projecting the point clouds of vertebral body regions along the left-right direction of the human body, sampling a set distance from the vertebral body direction to the spinous process direction on the projected image to obtain a sampling image range, filtering the three-dimensional point cloud with the sampling image range and generating a projected image, thereby obtaining the minimum bounding rectangle of the vertebral body region of each vertebral segment; and extending and filtering the three-dimensional point cloud according to the minimum bounding rectangle to obtain the point clouds of each vertebral body. This invention can quickly complete the segmentation of multiple vertebral body three-dimensional regions in a three-dimensional image, which is more convenient for surgical path planning and surgical registration and positioning operations, and has a wide range of applications in the field of orthopedic surgical assistance planning.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method for segmenting cone point clouds in three-dimensional images. Background Technology

[0002] With the development of biomedical engineering and computer technology, medical imaging provides multimodal medical images for clinical diagnosis. These medical images are playing an increasingly important role in today's clinical field, not only in medical diagnosis but also in surgical planning, radiotherapy planning, tracking pathological changes, and evaluating treatment effectiveness. Medical image segmentation technology is a crucial component of medical imaging, essentially involving the processing of medical images to extract features of areas of interest, thereby enabling more precise diagnosis and surgery.

[0003] Three-dimensional imaging data is the primary data used by doctors for pathological diagnosis. It is characterized by complete imaging data, high imaging accuracy, and rich data details. It is currently the commonly used imaging data for orthopedic assisted navigation robots. However, it generally contains details of multiple parts. For example, three-dimensional images taken before spinal surgery may contain one or more parts of the cervical, thoracic, and lumbar vertebrae. The number of vertebral segments is also generally multiple. Therefore, through vertebral segmentation technology, each vertebral segment region can be approximately segmented, which makes it easier to plan the surgical path and perform surgical registration and positioning. Summary of the Invention

[0004] Purpose of the invention: To address the above-mentioned shortcomings, this invention provides a three-dimensional image vertebral point cloud segmentation method based on image processing, which can quickly complete the segmentation of multiple vertebral three-dimensional regions in an image.

[0005] Technical solution:

[0006] A method for segmenting cone point clouds from three-dimensional images, comprising:

[0007] The three-dimensional image containing the spine is converted into a three-dimensional point cloud, and the point cloud of the approximate vertebral body region is obtained by filtering.

[0008] Project the point cloud of the approximate vertebral body region along the left and right direction of the human body, and sample a set distance from the vertebral body direction to the spinous process direction on the projected image to obtain the sampling image range;

[0009] The 3D point cloud is filtered by the sampling image range and a projection image is generated, thereby obtaining the minimum bounding rectangle of the vertebral body region of each vertebral segment;

[0010] The point clouds of each cone are obtained by extending and filtering the three-dimensional point cloud based on the minimum bounding rectangle.

[0011] The specific steps for obtaining the approximate vertebral body region point cloud through the screening process are as follows:

[0012] The point cloud of the spine region in the 3D point cloud is obtained, the point cloud of the spine region is projected along the height direction of the human body, and the projected image is binarized to obtain the image of the spine region.

[0013] In the spinal region image, the minimum bounding rectangle of the vertebral body region image is determined, and the 3D point cloud is filtered accordingly to obtain an approximate vertebral body region point cloud.

[0014] The specific steps for determining the minimum bounding rectangle of the vertebral body region image in the spinal region image are as follows:

[0015] In the spinal region image, the vertebral body region is determined. The minimum bounding rectangle of the vertebral body region image is determined by using the maximum coordinate difference between the two coordinate axes of each pixel point in the vertebral body region in the two-dimensional coordinate system of the spinal region image as the length and width of the minimum bounding rectangle.

[0016] Determining the vertebral body region in the spinal region image includes:

[0017] Morphological erosion and dilation operations are performed on the spinal region image to obtain an image containing the vertebral body, vertebral foramen region and its surrounding area.

[0018] The predetermined distance is determined on the projected image at the location of a characteristic height abrupt change in the image from the vertebral body to the pedicle, which serves as the termination point.

[0019] The set distance is 3 / 4 of the distance from the vertebral body direction to the spinous process direction in each vertebral body region on the projected image.

[0020] The minimum bounding rectangle of the vertebral body region for each vertebral segment is specifically obtained as follows:

[0021] The distribution region of each vertebral segment is obtained by filtering the 3D point cloud with the sampling image range and generating a projection image using connected component extraction, and the minimum bounding rectangle of the vertebral body region of each vertebral segment is generated accordingly.

[0022] The point cloud of the spine region is obtained by filtering the three-dimensional point cloud based on the energy value of the bones.

[0023] The step of filtering the 3D point cloud based on the sampled image range and generating a projection image includes:

[0024] Filter the 3D point cloud based on the range of the sampled image;

[0025] The filtered 3D point cloud is projected along the left-right and front-back directions of the human body to generate corresponding projected images.

[0026] Beneficial effects: This invention can quickly segment multiple vertebral body 3D regions in a 3D image, making it easier to plan surgical paths and perform surgical registration and positioning, and has a wide range of applications in the field of orthopedic surgical planning. Attached Figure Description

[0027] Figure 1 This is a flowchart of the three-dimensional image cone point cloud segmentation steps of the present invention;

[0028] Figure 2 The image is a CT image to be segmented.

[0029] Figure 3 Point cloud image of the skeleton to be segmented;

[0030] Figure 4 A projection image of the skeletal point cloud along the spine height direction;

[0031] Figure 5 (a) and 5(b) are the projected images obtained by projecting along the left-right and front-back directions of the human body, respectively;

[0032] Figure 6 (a) and (b) are projection images obtained by projecting the three-dimensional point cloud along the left-right and front-back directions of the human body after filtering the point cloud distribution range of the vertebral body along the height of the spine.

[0033] Figure 7 (a) and 7(b) are schematic diagrams of the vertebral regions extracted from the projection maps projected along the left-right and front-back directions of the human body, respectively.

[0034] Figure 8 This is a schematic diagram of the segmented cone point cloud. Detailed Implementation

[0035] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0036] This invention relates to a three-dimensional image cone point cloud segmentation method based on two-dimensional image processing, as follows: Figure 1 As shown, the steps include:

[0037] (1) Read the three-dimensional image containing the spine, convert the three-dimensional image into a three-dimensional point cloud containing energy values ​​(CT values ​​or gray values ​​in this invention), and then filter the three-dimensional point cloud according to the energy values ​​of the bones to remove non-skeletal feature information in the three-dimensional image, such as tissues, organs, etc., and retain only the point cloud information of the spine. Since the obtained point cloud information of the spine generally contains noise, the noise can be removed by filtering algorithm to obtain a relatively accurate point cloud of the spine region.

[0038] In this invention, the three-dimensional image is a CT image containing the spine obtained through CT acquisition, such as... Figure 2 As shown, the energy value is the CT value. Therefore, the 3D image is converted into a 3D point cloud containing the CT value. Then, based on the CT value of the bone, the corresponding spinal region point cloud is obtained through the above processing, as shown... Figure 3 As shown.

[0039] (2) Project the point cloud of the spine region along the height direction of the human body in the three-dimensional image to obtain the projection image of the spine region.

[0040] The three-dimensional coordinate system is established using the length of the operating table supporting the human body in the three-dimensional image as the Z-axis and the horizontal direction perpendicular to the Z-axis as the X-axis, according to the right-hand rule. This is the aforementioned image coordinate system. Since the human body must lie in a normal position on the operating table to obtain a three-dimensional image containing the human spine, the height, left-right, and front-back directions of the human body in the three-dimensional image are parallel to the Z-axis, X-axis, and Y-axis of the image coordinate system, respectively. Furthermore, the human body may move during this process, which will cause changes in the actual height, left-right, and front-back directions. However, these three directions only have small angles with the Z-axis, X-axis, and Y-axis of the image coordinate system, respectively. Therefore, the corresponding height, left-right, and front-back directions of the human body can be clearly distinguished in the three-dimensional image. Thus, the point cloud of the spinal region can be projected along the height direction of the human body in the three-dimensional image, and the point cloud can be projected along the left-right and front-back directions of the human body in subsequent steps.

[0041] Of course, the present invention can also directly project the point cloud of the spine region along the Z-axis of the image coordinate system. In subsequent steps, the point cloud can also be projected along the X and Y axes of the image coordinate system. The present invention preferably projects the point cloud of the spine region along the height direction of the human body in the three-dimensional image.

[0042] Projected images such as Figure 4 As shown, the calculation method for each pixel value is as follows: First, calculate the sum of the energy values ​​of all points in the point cloud corresponding to each pixel and invert it, denoted as G. Then, calculate the natural constant e to the power of G. Finally, normalize the pixel values ​​of all points to the range of 0 to 255.

[0043] (3) Process the spinal region projection image obtained in step (2) to obtain the minimum bounding rectangle of the vertebral region image;

[0044] Specifically:

[0045] First, the non-vertebral regions are removed by binarization of the spinal region projection image obtained in step (2) to obtain the region of interest image, i.e., the spinal region image.

[0046] Secondly, morphological erosion and dilation operations are performed on the region of interest image to obtain a vertebral region image containing the vertebral body, vertebral foramen region and its surrounding areas. For example, some transverse process regions in the vertebral region image are filtered out by morphological erosion and dilation operations.

[0047] Then, the vertebral region image is further processed by Blob analysis to further filter out the noise around the vertebral body. Blob analysis is a further removal of the noise around the vertebral body.

[0048] During image processing, it can be determined whether morphological erosion, dilation, and Blob analysis are needed. After binarizing the spinal region image, morphological processing and Blob analysis are selected based on the noise around the vertebral body.

[0049] Finally, a two-dimensional coordinate system is determined based on the spinal region image. The minimum bounding rectangle of the vertebral region image is determined by using the maximum coordinate difference between the two coordinate axes of each pixel in the vertebral region image as the length and width of the minimum bounding rectangle. The two-dimensional coordinate system can be established using the X and Y axes of the three-dimensional coordinate system as coordinate axes, or it can be established using the left-right and front-back directions of the human body as coordinate axes. The former is preferred in this invention.

[0050] Then, the minimum bounding rectangle of each pixel in the region can be obtained from the minimum coordinate range of the cone region image in the XY direction of the two-dimensional coordinate system.

[0051] (4) Filter the 3D point cloud obtained in step (1) using the minimum bounding rectangle of the vertebral region image obtained in step (3) to obtain an approximate vertebral region point cloud. Then, project the filtered point cloud along the left-right and front-back directions of the human body to generate corresponding projection images, such as... Figure 5 As shown in (a) and 5(b), where, Figure 5 The projected image of (a) includes the vertebral body, the foramen, and part of the spinous process.

[0052] (5) Sample the projection image obtained in step (4) from the vertebral body direction to the spinous process direction at a set distance to obtain the sampling image range;

[0053] In this invention, such as Figure 5 As shown in (a), since the vertebral features are relatively clear in the projection image obtained by projecting along the left and right direction of the human body in step (4), the distribution range of the vertebral body along the spine height direction is approximately obtained by sampling from right to left along the projection image, which is also the sampling image range; of course, the projection image can also be projected from the left and right direction of the human body, then the sampling needs to be carried out from left to right from the corresponding projection image.

[0054] Specifically, since there is a sudden change in the height of the vertebral body to the pedicle in the projection image (that is, the characteristic height in the image), this sudden change position is set as the sampling termination area, and the sampled area is the approximate distribution range of the vertebral body along the height of the spine.

[0055] More specifically, in this invention, if sampling is performed from the vertebral body direction to the spinous process direction of the projected image, and the abrupt change position from the vertebral body to the pedicle is used as the sampling termination area, the sampling data will be inaccurate due to too much noise in the sampling data. Therefore, it is preferable to select 3 / 4 of each vertebral body region from right to left as the sampling termination data to ensure the accuracy of the sampling data.

[0056] (6) Filter the 3D point cloud obtained in step (1) based on the sampling image range obtained in step (5), and project the filtered 3D point cloud along the left-right and front-back directions of the human body to generate corresponding projection images, such as... Figure 6 As shown in (a) and 6(b).

[0057] (7) Perform connected component lookup on the projected image obtained in step (6) to obtain the distribution region of each vertebral segment, and generate the minimum bounding rectangle of the vertebral body region of each segment accordingly, such as... Figure 7 As shown in (a) and 7(b);

[0058] First, the projected image obtained in step (6) is processed by one or more of binarization, morphology, and Blob analysis to obtain an image containing part of the vertebral body;

[0059] Then, connected component extraction is used to obtain the distribution region of each vertebral segment, and the minimum bounding rectangle of the vertebral body region of each vertebral segment is generated accordingly, such as... Figure 7 As shown;

[0060] Specifically, the extension direction of each minimum enclosing rectangle is consistent with the width and height directions of that vertebral body, and the position of each vertebral body is consistent with... Figure 3 One-to-one correspondence.

[0061] (8) Since the sampling in step (5) may not include the complete vertebral body region, the minimum bounding rectangle of each vertebral body region extracted in step (7) is extended along the front-back direction of the human body on the three-dimensional point cloud obtained in step (1) to the skeletal point cloud (i.e., Figure 3 By defining the boundary, an approximate vertebral segment region encompassing the entire vertebral body can be obtained. Filtering the 3D point cloud obtained in step (1) based on the rectangular coordinates of each vertebral body region yields the point cloud for each vertebral body, such as... Figure 8 As shown, the segmentation ends here.

[0062] This invention segments the vertebral body region of each vertebra by analyzing the 3D point cloud and grayscale information in a 3D image. Then, based on the segmentation result, the corresponding 3D point cloud is extracted, ultimately obtaining the point cloud and grayscale information for each vertebra, thus completing the segmentation of the vertebral body in the 3D image. This invention can quickly segment multiple vertebral body 3D regions in a 3D image.

[0063] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations (such as quantity, shape, position, etc.) can be made to the technical solution of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for segmenting a three-dimensional image vertebral point cloud, characterized in that: The application relates to a method for extracting a vertebra region point cloud from a three-dimensional point cloud. The method comprises the following steps: converting a three-dimensional image containing a vertebra into a three-dimensional point cloud, obtaining a vertebra region point cloud in the three-dimensional point cloud, projecting the vertebra region point cloud along a human body height direction, and obtaining a vertebra region image through binarization processing of the projection image; determining a minimum enclosing rectangle of a vertebral body region image in the vertebra region image, and filtering the three-dimensional point cloud according to the minimum enclosing rectangle to obtain an approximate vertebral body region point cloud; projecting the approximate vertebral body region point cloud along a left-right direction of the human body, and obtaining a sampling image range by sampling a set distance from a vertebral body direction to a spinous process direction on the projection image; filtering the three-dimensional point cloud with the sampling image range and generating a projection image, and obtaining a minimum enclosing rectangle of a vertebral body region of each vertebra according to the projection image; extending and screening the three-dimensional point cloud according to the minimum enclosing rectangle to obtain a vertebral body point cloud of each vertebra.

2. The method of claim 1, wherein: The minimum enclosing rectangle of the vertebral body region image in the vertebra region image is determined as follows: a vertebral body region is determined in the vertebra region image, and the maximum coordinate difference of each pixel point in the vertebral body region in two coordinate axes of a two-dimensional coordinate system of the vertebra region image is taken as the length and width of the minimum enclosing rectangle to determine the minimum enclosing rectangle of the vertebral body region image.

3. The method of claim 2, wherein: The determination of the vertebral body region in the vertebra region image comprises the following steps: morphological erosion and expansion operations are performed on the vertebra region image to obtain an image containing a vertebral body, a vertebral foramen region and a surrounding partial region.

4. The method of claim 1, wherein: The set distance is determined by taking a feature height mutation position on an image from the vertebral body to the pedicle as a termination position on the projection image.

5. The method of claim 1, wherein: The set distance is 3 / 4 of the distance from the vertebral body direction to the spinous process direction of each vertebral body partial region on the projection image.

6. The method of claim 1, wherein: The minimum enclosing rectangle of the vertebral body region of each vertebra is obtained as follows: a distribution region of each vertebra is extracted by using a connected domain extraction on the projection image generated by filtering the three-dimensional point cloud with the sampling image range, and a minimum enclosing rectangle of the vertebral body region of each vertebra is generated according to the distribution region.

7. The method of claim 1, wherein: The three-dimensional point cloud is screened according to the energy value of the bone to obtain the vertebra region point cloud.

8. The method of claim 1, wherein: The filtering of the three-dimensional point cloud with the sampling image range and the generation of the projection image comprise the following steps: the three-dimensional point cloud is filtered with the sampling image range; the filtered three-dimensional point cloud is projected along a left-right direction and a front-rear direction of the human body to generate corresponding projection images.

Images