A method for compressing three-dimensional models of human duct tissue
By homosimilar resampling, image segmentation and Gaussian smoothing of the three-dimensional model of human pipe-like tissues, combined with Marching cube and Garland and Heckbert algorithms, the problems of 3D model file size and rendering efficiency are solved, and the rendering effect is achieved without affecting the rendering effect. The model size and number of patches are significantly compressed in the case.
Patent Information
- Application Number
- CN202210476108.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-29
- Publication Date
- 2024-12-17
- Estimated Expiration
- 2042-04-29
AI Technical Summary
Due to the complexity of surface and high curvature of the three-dimensional model of human organs, especially blood vessels, bronchials and other pipe tissues, the three-dimensional model file size is large, the storage and transmission pressure is high, and the direct smoothing operation may lead to the reduction of the model and spike phenomenon, which affects the Rendering effect.
By acquiring preoperative CT images for homosexual resampling and normalizing CT values, the target tissue was extracted using image segmentation algorithm, and the connection noise cancellation and Gaussian smoothing were performed. The surface was reconstructed in combination with the Marching cube algorithm. Finally, the Garland and Heckbert surface film streamlining algorithm was used. Number of compressed dough sheets.
On the basis of maintaining the rendering smooth effect, the size of the three-dimensional model is greatly reduced, the storage and transmission pressure are reduced, and the rendering response performance is improved, solving the problem of model file size and transmission efficiency.
Smart Images
Figure CN114708415B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of medical technology, and in particular relates to a method for compressing a three-dimensional model of human duct tissue. Background Art
[0002] 3D reconstruction of medical images is a process of generating a surface patch fitting model of human organs through organ segmentation and 3D modeling based on human body tomography (CT or MRI); a 3D model of human organs is a digital model that uses numerous interconnected polygonal patches to fit the surface shape of human organs; it can be drawn on a computer screen through a 3D rendering method to present a 3D image of the human organs and achieve a virtual reality observation effect; patches are the basic units that make up the 3D model of organs, and the interconnection of numerous tiny patches enables the fitting of the 3D organ surface; patch smoothing is an algorithm for processing patches on the surface of a 3D reconstructed model, which can make the transition between adjacent patches smoother and achieve an overall 3D model rendering effect that is closer to the smooth surface of human organs; stl model is the abbreviation of stereo lithography, a file format for storing 3D surface models.
[0003] As medical technology develops towards digitalization and precision, three-dimensional reconstruction models of human organs are increasingly used. Through three-dimensional models, doctors can accurately and intuitively understand the spatial relationship between lesions and surrounding tissues, preoperative planning, and intraoperative surgical guidance, such as preoperative bronchoscopic approach planning in virtual bronchoscopy, preoperative resection plan planning for segmentectomy, and puncture path planning in percutaneous pulmonary nodule puncture operations.
[0004] However, human organs, especially blood vessels, bronchi and other tubular tissues, have complex structures and are distributed in a tree-like manner. Their three-dimensional models have a large average surface curvature, so more and finer patches are needed to achieve smooth surface fitting. This results in a large size of the three-dimensional model file. For example, an unprocessed three-dimensional model of the pulmonary artery often reaches several 10M bytes, which can reach hundreds of M bytes after being processed by the smoothing algorithm. Large-sized three-dimensional files put a lot of pressure on the storage of databases, the design of three-dimensional interactive algorithms, and the transmission of mobile medical applications. On the other hand, if the original three-dimensional reconstructed model is directly smoothed, the model may be reduced and spikes may be generated at the end of the tubular tissue, affecting the true effect of the three-dimensional rendering.
[0005] Surgery-assisted products based on 3D reconstruction of medical images, especially organs and systems with rich duct-like tissues, such as lungs and liver, require a model compression algorithm. To this end, we propose a 3D model compression method for human duct-like tissues. Summary of the invention
[0006] The purpose of the present invention is to provide a method for compressing a three-dimensional model of human duct tissue to solve the problems raised in the above background technology.
[0007] To achieve the above object, the present invention provides the following technical solution: a method for compressing a three-dimensional model of human duct tissue, comprising the following specific steps:
[0008] Step 1: Obtain the patient's preoperative CT image and perform homogeneous resampling and CT value normalization preprocessing;
[0009] Step 2: Input the CT sequence into the pre-trained image segmentation algorithm to extract the target tissue; segment it into a target tissue mask image sequence, that is, the voxels belonging to the target tissue in the original medical image are marked as white (8-bit encoding, grayscale value = 255);
[0010] Step 3: Connectivity denoising is performed. According to the voxel connectivity of 6-connectivity or 8-connectivity, the first N largest connected domains are taken, and other smaller connected domains are removed as segmentation noise; N is set to 2-4 according to different tissues and different image quality characteristics;
[0011] Step 4: Gaussian smoothing is performed on the segmentation mask to obtain grayscale mask data with a transition from 255 to 0. The grayscale value of each voxel is encoded by floating-point data. Gaussian smoothing is based on the Gaussian distribution curve. For each voxel, the weighted average is obtained according to the voxels within a certain radius of its neighborhood. The closer the voxel is to the center, the higher the weight, and vice versa.
[0012] Step 5: Based on the segmentation mask after Gaussian smoothing, set the iso-threshold, and use the Marching cube algorithm to reconstruct the surface of the 3D model; the iso-threshold is pre-set to a value between 0 and 255 according to the characteristics of different tissues and organs;
[0013] Step 6: Use the Garland and Heckbert mesh simplification algorithm to simplify the 3D surface model; set the simplification parameter percentage, and the algorithm can compress the size of the original model to the specified ratio to achieve the effect of mesh compression.
[0014] Furthermore, in step 4, the standard deviation σ and the radius of action of the Gaussian curve are set in advance according to the characteristics of different tissues. The weight of the voxels within the action radius is taken according to the Gaussian distribution, and the weight of the voxels outside the action radius is 0, and does not participate in the weighted average of the current voxel.
[0015] Furthermore, the central voxel is Its neighborhood radius The voxel set of the range is denoted as Voxel With the central voxel The Euclidean distance is denoted as Voxel The Gaussian smoothing parameter is denoted as
[0016]
[0017] The value of the central voxel after Gaussian smoothing is but:
[0018] in is the number of neighborhood voxel points.
[0019] Furthermore, the higher the isosurface threshold in step 5, the more the generated 3D model conforms to the boundary of the original segmentation result, but the surface defects are more and need to be adjusted according to different reconstruction objectives and image quality.
[0020] Furthermore, the lower the isosurface threshold in step five, the higher the saturation and smoothness of the model, and the stronger the ability to maintain the smoothness of the model in the subsequent patch reduction algorithm, which needs to be adjusted according to different reconstruction objectives and image quality.
[0021] Furthermore, in the step 2, the voxels at other locations outside the target tissue are marked with a black (8-bit encoding, grayscale value = 0) background.
[0022] Compared with the prior art, the present invention has the following beneficial effects:
[0023] 1. This application can significantly reduce the number of surface patches, reduce the size of the three-dimensional model, and improve the efficiency of use without seriously reducing the smoothness of the three-dimensional model rendering. It can effectively solve the problem that the polygonal patches of the human body three-dimensional reconstruction model have complex surface topological structures and large curvatures, and require smaller sizes and larger numbers, which causes great pressure on file storage, algorithm processing and network transmission.
[0024] 2. While maintaining the smooth rendering effect of the model surface, this application compresses the model size to a great extent, which can better assist personnel in surgical planning for three-dimensional reconstruction of medical images. It can reduce the size of the three-dimensional model without affecting the three-dimensional rendering effect, improve the model rendering response performance, network transmission performance and reduce data storage pressure. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 Schematic diagram of lung CT scan image and pulmonary artery segmentation mask.
[0026] Figure 2 Schematic diagram of pulmonary artery segmentation mask.
[0027] Figure 3 This is a schematic diagram of the model rendering effect generated by the Marching Cube 3D reconstruction algorithm directly using the original segmentation mask with 41.9MB of file memory.
[0028] Figure 4 This is a schematic diagram of the rendering effect of the original segmentation mask with 228.3MB of file memory after Gaussian smoothing and then Marching Cube 3D reconstruction to generate the model.
[0029] Figure 5 This is a schematic diagram of the rendering effect of the 3D reconstructed model after Gaussian smoothing and 94% facet reduction with a file memory of 2.1MB. DETAILED DESCRIPTION
[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0031] Example 1
[0032] like Figure 1-5 As shown, a method for compressing a three-dimensional model of a human duct-like tissue comprises the following specific steps:
[0033] Step 1: Obtain the patient's preoperative CT image and perform homogeneous resampling and CT value normalization preprocessing;
[0034] Step 2: Input the CT sequence into the pre-trained image segmentation algorithm to extract the target tissue; segment it into a target tissue mask image sequence, that is, the voxels belonging to the target tissue in the original medical image are marked as white (8-bit encoding, grayscale value = 255);
[0035] Step 3: Connectivity denoising is performed. According to the voxel connectivity 6-connectivity, the first N largest connected domains are taken, and other smaller connected domains are removed as segmentation noise; N is set to 2 according to the characteristics of different tissues and different image quality;
[0036] Step 4: Gaussian smoothing is performed on the segmentation mask to obtain grayscale mask data with a transition from 255 to 0. The grayscale value of each voxel is encoded by floating-point data. Gaussian smoothing is based on the Gaussian distribution curve. For each voxel, the weighted average is obtained according to the voxels within a certain radius of its neighborhood. The closer the voxel is to the center, the higher the weight, and vice versa.
[0037] Step 5: Based on the segmentation mask after Gaussian smoothing, set the iso-threshold value, and use the Marching cube algorithm to reconstruct the surface of the three-dimensional model; the iso-threshold value is pre-set to 50 according to the characteristics of different tissues and organs;
[0038] Step 6: Use the Garland and Heckbert mesh simplification algorithm to simplify the 3D surface model; set the simplification parameter percentage, and the algorithm can compress the size of the original model to the specified ratio to achieve the effect of mesh compression.
[0039] Among them, in step 4, the standard deviation σ and the radius of action of the Gaussian curve are set in advance according to the characteristics of different tissues. The weight of the voxels within the action radius is taken according to the Gaussian distribution, and the weight of the voxels outside the action radius is 0, and does not participate in the weighted average of the current voxel.
[0040] Wherein, the central voxel is Its neighborhood radius The voxel set of the range is denoted as Voxel With the central voxel The Euclidean distance is denoted as Voxel The Gaussian smoothing parameter is denoted as
[0041]
[0042] The value of the central voxel after Gaussian smoothing is but:
[0043] in is the number of neighborhood voxel points.
[0044] Among them, the higher the isosurface threshold in step 5, the more the generated 3D model conforms to the boundary of the original segmentation result, but there are more surface defects and it needs to be adjusted according to different reconstruction goals and image quality.
[0045] Among them, the lower the isosurface threshold in step five, the higher the saturation and smoothness of the model, and the stronger the ability to maintain the smoothness of the model in the subsequent patch deletion algorithm, which needs to be adjusted according to different reconstruction goals and image quality.
[0046] In the step 2, the voxels at other locations outside the target tissue are marked with a black (8-bit encoding, grayscale value = 0) background.
[0047] Example 2
[0048] like Figure 1-5 As shown, a method for compressing a three-dimensional model of a human duct-like tissue comprises the following specific steps:
[0049] Step 1: Obtain the patient's preoperative CT image and perform homogeneous resampling and CT value normalization preprocessing;
[0050] Step 2: Input the CT sequence into the pre-trained image segmentation algorithm to extract the target tissue; segment it into a target tissue mask image sequence, that is, the voxels belonging to the target tissue in the original medical image are marked as white (8-bit encoding, grayscale value = 255);
[0051] Step 3: Connectivity denoising is performed. According to the voxel connectivity 8-connectivity, the first N largest connected domains are taken, and other smaller connected domains are removed as segmentation noise; N is set to 3 according to the characteristics of different tissues and different image quality;
[0052] Step 4: Gaussian smoothing is performed on the segmentation mask to obtain grayscale mask data with a transition from 255 to 0. The grayscale value of each voxel is encoded by floating-point data. Gaussian smoothing is based on the Gaussian distribution curve. For each voxel, the weighted average is obtained according to the voxels within a certain radius of its neighborhood. The closer the voxel is to the center, the higher the weight, and vice versa.
[0053] Step 5: Based on the segmentation mask after Gaussian smoothing, set the iso-threshold value, and use the Marching cube algorithm to reconstruct the surface of the three-dimensional model; the iso-threshold value is pre-set to 100 according to the characteristics of different tissues and organs;
[0054] Step 6: Use the Garland and Heckbert mesh simplification algorithm to simplify the 3D surface model; set the simplification parameter percentage, and the algorithm can compress the size of the original model to the specified ratio to achieve the effect of mesh compression.
[0055] Among them, in step 4, the standard deviation σ and the radius of action of the Gaussian curve are set in advance according to the characteristics of different tissues. The weight of the voxels within the action radius is taken according to the Gaussian distribution, and the weight of the voxels outside the action radius is 0, and does not participate in the weighted average of the current voxel.
[0056] Wherein, the central voxel is Its neighborhood radius The voxel set of the range is denoted as Voxel With the central voxel The Euclidean distance is denoted as Voxel The Gaussian smoothing parameter is denoted as
[0057]
[0058] The value of the central voxel after Gaussian smoothing is but:
[0059] in is the number of neighborhood voxel points.
[0060] Among them, the higher the isosurface threshold in step 5, the more the generated 3D model conforms to the boundary of the original segmentation result, but there are more surface defects and it needs to be adjusted according to different reconstruction goals and image quality.
[0061] Among them, the lower the isosurface threshold in step five, the higher the saturation and smoothness of the model, and the stronger the ability to maintain the smoothness of the model in the subsequent patch deletion algorithm, which needs to be adjusted according to different reconstruction goals and image quality.
[0062] In the step 2, the voxels at other locations outside the target tissue are marked with a black (8-bit encoding, grayscale value = 0) background.
[0063] Example 3
[0064] like Figure 1-5 As shown, a method for compressing a three-dimensional model of a human duct-like tissue comprises the following specific steps:
[0065] Step 1: Obtain the patient's preoperative CT image and perform homogeneous resampling and CT value normalization preprocessing;
[0066] Step 2: Input the CT sequence into the pre-trained image segmentation algorithm to extract the target tissue; segment it into a target tissue mask image sequence, that is, the voxels belonging to the target tissue in the original medical image are marked as white (8-bit encoding, grayscale value = 255);
[0067] Step 3: Connectivity denoising is performed. According to the voxel connectivity of 8-connectivity, the first N largest connected domains are taken, and other smaller connected domains are removed as segmentation noise; N is set to 4 according to the characteristics of different tissues and different image quality;
[0068] Step 4: Gaussian smoothing is performed on the segmentation mask to obtain grayscale mask data with a transition from 255 to 0. The grayscale value of each voxel is encoded by floating-point data. Gaussian smoothing is based on the Gaussian distribution curve. For each voxel, the weighted average is obtained according to the voxels within a certain radius of its neighborhood. The closer the voxel is to the center, the higher the weight, and vice versa.
[0069] Step 5: Based on the segmentation mask after Gaussian smoothing, set the iso-threshold value, and use the Marching cube algorithm to reconstruct the surface of the three-dimensional model; the iso-threshold value is pre-set to a value between 200 according to the characteristics of different tissues and organs;
[0070] Step 6: Use the Garland and Heckbert mesh simplification algorithm to simplify the 3D surface model; set the simplification parameter percentage, and the algorithm can compress the size of the original model to the specified ratio to achieve the effect of mesh compression.
[0071] Among them, in step 4, the standard deviation σ and the radius of action of the Gaussian curve are set in advance according to the characteristics of different tissues. The weight of the voxels within the action radius is taken according to the Gaussian distribution, and the weight of the voxels outside the action radius is 0, and does not participate in the weighted average of the current voxel.
[0072] Wherein, the central voxel is Its neighborhood radius The voxel set of the range is denoted as Voxel With the central voxel The Euclidean distance is denoted as Voxel The Gaussian smoothing parameter is denoted as
[0073]
[0074] The value of the central voxel after Gaussian smoothing is but:
[0075] in is the number of neighborhood voxel points.
[0076] Among them, the higher the isosurface threshold in step 5, the more the generated 3D model conforms to the boundary of the original segmentation result, but there are more surface defects and it needs to be adjusted according to different reconstruction goals and image quality.
[0077] Among them, the lower the isosurface threshold in step five, the higher the saturation and smoothness of the model, and the stronger the ability to maintain the smoothness of the model in the subsequent patch deletion algorithm, which needs to be adjusted according to different reconstruction goals and image quality.
[0078] In the step 2, the voxels at other locations outside the target tissue are marked with a black (8-bit encoding, grayscale value = 0) background.
[0079] The working principle and use process of the present invention are as follows: while not seriously reducing the smooth rendering effect of the 3D model, the number of surface patches can be greatly reduced, the size of the 3D model can be reduced, and the use efficiency can be improved. It can effectively solve the problem that the polygonal patches of the human body 3D reconstruction model have complex organ surface topological structures and large curvatures, and require smaller sizes and larger numbers, which causes greater pressure on file storage, algorithm processing and network transmission. While maintaining the smooth rendering effect of the model surface, the model size is compressed to a great extent, which can better assist personnel in surgical planning for 3D reconstruction of medical images.
[0080] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A method for compressing a three-dimensional model of human duct tissue, characterized in that: The specific steps include: Step 1: Obtain the patient's preoperative CT image and perform homogeneous resampling and CT value normalization preprocessing; Step 2: Input the CT sequence into the pre-trained image segmentation algorithm to extract the target tissue; Segmentation into a target tissue mask image sequence, that is, the voxels belonging to the target tissue in the original medical image are marked in white; Step 3: Connectivity denoising is performed. According to the voxel connectivity of 6-connectivity or 8-connectivity, the first N largest connected domains are taken, and other smaller connected domains are removed as segmentation noise; N is set to 2-4 according to different tissues and different image quality characteristics; Step 4: Gaussian smoothing is performed on the segmentation mask to obtain grayscale mask data with a transition from 255 to 0. The grayscale value of each voxel is encoded by floating-point data. Gaussian smoothing is based on the Gaussian distribution curve. For each voxel, the weighted average is obtained according to the voxels within a certain radius of its neighborhood. The closer the voxel is to the center, the higher the weight, and vice versa. Step 5: Based on the segmentation mask after Gaussian smoothing, set the iso-threshold, and use the Marching cube algorithm to reconstruct the surface of the 3D model; the iso-threshold is pre-set to a value between 0 and 255 according to the characteristics of different tissues and organs; Step 6: Use the Garland and Heckbert mesh simplification algorithm to simplify the 3D surface model; set the simplification parameter percentage, and the algorithm can compress the size of the original model to the specified ratio to achieve the effect of mesh compression.
2. A method for compressing a three-dimensional model of human duct tissue according to claim 1, characterized in that: In step 4, the standard deviation σ and the action radius r of the Gaussian curve are set in advance according to the characteristics of different tissues. The weight of the voxels within the action radius is taken according to the Gaussian distribution. The weight of the voxels outside the action radius is 0 and does not participate in the weighted average of the current voxel.
3. The method for compressing a three-dimensional model of human duct tissue according to claim 2, characterized in that: The central voxel is The voxel set within the neighborhood radius r is recorded as Voxel With the central voxel The Euclidean distance is denoted as Voxel The Gaussian smoothing parameter is denoted as The value of the central voxel after Gaussian smoothing is but: in is the number of neighborhood voxel points.
4. The method for compressing a three-dimensional model of human duct tissue according to claim 1, characterized in that: The higher the isosurface threshold in step 5, the more the generated 3D model conforms to the boundary of the original segmentation result, but the surface defects are more and need to be adjusted according to different reconstruction goals and image quality.
5. The method for compressing a three-dimensional model of human duct tissue according to claim 1, characterized in that: The lower the isosurface threshold in step 5, the higher the saturation and smoothness of the model, and the stronger the ability to maintain the smoothness of the model in the subsequent patch deletion algorithm. It needs to be adjusted according to different reconstruction goals and image quality.
6. The method for compressing a three-dimensional model of human duct tissue according to claim 1, characterized in that: In the step 2, the voxels at other locations outside the target tissue are marked with a black background.