A method and system for automatic tissue positioning based on CT images

Abstract

This invention provides a method and system for automatic tissue localization calculation based on CT images. It uses a deep learning network to segment the spinal bones in CT images to obtain a spinal mask; it then projects the spinal mask onto a coronal plane and smooths the projected image using a filtering kernel; by locating the center point of the spine, it performs spinal segmentation on the smoothed image to locate the spinal bone gray matter; and it performs a triangular search on both sides of the center point of the spinal bone gray matter to locate and obtain muscle and fat, wherein the search directions for muscle and fat are opposite. This invention directly uses local tomographic images and leverages the unique and stable relationship between cancellous bone, muscle, and fat in the human body to automatically identify and locate these components; the localization process uses a method that extrapolates from a cylindrical shape to a free form, resulting in more accurate measurements.

Description

Technical Field

[0001] This invention relates to the field of hospital image processing technology, and in particular to a method and system for automatic tissue localization calculation based on CT images. Background Technology

[0002] Today, the use of computed tomography (CT) scan data for predicting and assisting in the diagnosis of clinical diseases is becoming increasingly relied upon in clinical practice. Currently, the most common measurement is bone mineral density (BMD), which is one of the main bases for diagnosing osteoporosis. These measurements, based on CT values, require the measurement of regions of interest in muscle, fat, and spongy bone. The measurement process requires professionals who can identify muscle, fat, and bone regions in the scanned images, and the results are obtained only after complex manipulation by medical personnel.

[0003] Currently, the methods for diagnosing osteoporosis are DXA and QCT. DXA (using dual-x-ray absorptioometry) directly detects X-ray absorption, but its equipment requirements are high and its adoption rate is not high. QCT relies on phantom data for auxiliary measurements, and the specialized phantoms and software also cause inconvenience in the examination. There are also methods that directly segment cortical bone, muscle, and fat from the tomographic scan data. However, this method involves manually segmenting muscle and fat, which is prone to human error, tedious, and inefficient. Summary of the Invention

[0004] In order to overcome the above-mentioned technical defects, the purpose of this invention is to provide a method and system for automatic tissue localization calculation based on CT images that can automatically identify and locate bone ash, muscle, and fat.

[0005] This invention discloses an automatic tissue localization calculation method based on CT images, comprising: segmenting the spinal bones in the CT image using a deep learning network to obtain a spinal mask; projecting the spinal mask onto the coronal plane and smoothing the projected image using a filtering kernel; segmenting the smoothed image by locating the center point of the spine to locate the spinal bone gray matter; and performing a triangular search on both sides of the center point of the spinal bone gray matter to locate and obtain muscle and fat; wherein the search directions for muscle and fat are opposite.

[0006] Preferably, the step of projecting the spinal mask onto the coronal plane and smoothing the projected image using a filter kernel includes: the filter kernel size is...

[0007] Preferably, the step of segmenting the smoothed image by locating the center point of the spine to locate the vertebral urinary fossa includes: the vertebrae segmented in the smoothed image are near-cubic prisms; the coordinates (x1, y1) and (x2, y2) of the two opposite angles of the near-cubic prism in the Cartesian coordinate system are obtained; the center points (x1, x2) / 2 and (y1, y2) / 2 of the near-cubic prism are obtained; and the vertebral urinary fossa is determined based on the center points.

[0008] Preferably, the step of performing a triangular search on both sides of the center point of the spinal bone ash to locate and obtain muscle and fat includes: dividing a central axis to distinguish between muscle and fat regions, with muscle and fat regions on both sides of the central axis; and defining 120-degree search ranges in the muscle and fat regions on both sides of the central axis for searching.

[0009] Preferably, the step of performing a triangular search on both sides of the center point of the spinal bone ash to locate and obtain muscle and fat further includes: using a cylindrical area in the search area to determine whether muscle or fat tissue exists, thereby delineating the muscle or fat area.

[0010] Preferably, the step of determining the presence of muscle or fat tissue in the search area using a columnar region to delineate the muscle or fat region includes: calculating the Hu value of the muscle or fat region, and delineating the initial muscle region and initial fat region in the muscle or fat region based on the Hu value range of muscle tissue and the Hu value range of fat tissue.

[0011] Preferably, after defining the initial muscle region and initial fat region in the muscle or fat region based on the Hu value range of muscle tissue and the Hu value range of fat tissue, the method further includes: eroding and expanding the cylindrical region to obtain the precise muscle region and fat region again.

[0012] Preferably, the division is a central axis used to distinguish between muscle and fat regions, with muscle and fat regions on either side of the central axis; the search is performed by defining 120-degree search ranges in the muscle and fat regions on either side of the central axis, respectively: p1 is a point in the muscle region, p2 is a point in the fat region, and p0 is a point on the central axis; assuming p0(x0, y0, z) i p1(x1,y1,z) i p1(x2,y2,z) i p1 satisfies or

[0013]

[0014] The starting search position for p2 is:

[0015] or

[0016] And the search scope is:

[0017] or or or

[0018] Where H is the height of the i-th layer image and W is the image width.

[0019] Preferably, the step of segmenting the spinal bones in the CT image using a deep learning network to obtain a spinal mask includes: the deep learning network being a segmentation network, which includes ResUnet, nnUnet, Unet++, and Nestnet networks.

[0020] This invention also discloses an automatic tissue localization calculation system based on CT images, including a segmentation module and a localization module. The segmentation module includes a first segmentation unit and a second segmentation unit based on a deep learning network. The first segmentation unit segments the spinal bones in the CT image using the deep learning network to obtain a spinal mask. The second segmentation unit projects the spinal mask onto a coronal plane, and the projected image is smoothed using a filtering kernel. The smoothed image is segmented by locating the center point of the spine to locate the spinal bone gray matter. The localization module performs a triangular search on both sides of the center point of the spinal bone gray matter to locate and obtain muscle and fat. The search directions for muscle and fat are opposite.

[0021] Compared with existing technologies, the above technical solution has the following advantages:

[0022] 1. This invention directly uses local tomographic images and takes advantage of the special stable relationship between cancellous bone, muscle, and fat in the human body to automatically identify and locate cancellous bone, muscle, and fat; the localization process uses a method of extrapolating from a cylindrical shape to a free shape, making the measurement values ​​more accurate. Attached Figure Description

[0023] Figure 1 A flowchart of the tissue automatic localization calculation method based on CT images provided by the present invention;

[0024] Figure 2 This is a schematic diagram of the search area for the muscle and fat search and localization method provided by the present invention. Detailed Implementation

[0025] The advantages of the present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments.

[0026] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0027] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0028] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0029] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0030] In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "linking" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two components. They can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0031] In the following description, suffixes such as "module," "part," or "unit" used to denote elements are used only for the convenience of the description of the invention and have no specific meaning in themselves. Therefore, "module" and "part" can be used interchangeably.

[0032] Lumbar spine localization refers to the process of determining the location of the lumbar vertebrae or identifying specific lumbar vertebrae. The lumbar vertebrae are a group of vertebrae located at the lower part of the spine, typically numbered L1 to L5. Lumbar spine localization is commonly used in fields such as medical imaging, spinal surgery, and diagnostics. Computed tomography (CT) scans can provide more detailed images of the lumbar spine and allow for localization in three-dimensional space. Physicians can use CT scan images to accurately locate the lumbar vertebrae and determine the location and number of specific lumbar vertebrae.

[0033] See appendix Figure 1 This invention discloses an automatic tissue localization calculation method based on CT images, comprising:

[0034] S100. The spinal bones in the CT image are segmented using a deep learning network to obtain a spinal mask;

[0035] S200. Project the spinal mask onto the coronal plane and smooth the projected image using a filter kernel. Then, segment the smoothed image by locating the center point of the spine to locate the spinal bone ash.

[0036] S300: Perform a triangular search on both sides of the center point of the spinal bone ash to locate and obtain muscle and fat; wherein the search directions of muscle and fat are opposite.

[0037] The above three steps allow for the localization of bone ash, muscle, and fat, thus enabling the segmentation of individual vertebrae. Specifically,

[0038] In step S100, a number of data points (e.g., 50) are used to segment the spinal bones, producing a coarse spinal mask. This facilitates further image recognition technology for vertebral segmentation in the next step. The deep learning network used is a segmentation network, which includes, but is not limited to, ResUnet, nnUnet, Unet++, and Nestnet networks.

[0039] Step S200 involves using image processing methods to accurately locate the bone ash in the spine. First, based on the characteristics of the spine, coronal projection is performed, and a filter kernel is designed to smooth the bone tissue in the CT image. The smoothed image is then segmented into vertebrae within the spine, while maintaining the original center point of each vertebra.

[0040] A filter kernel, also known as a convolution kernel or convolution matrix, is a small matrix or core used for filtering operations in image and signal processing. The filter kernel defines the weights and response characteristics of the filtering operation, used to perform convolution operations on the input image or signal to achieve different filtering effects. A filter kernel is typically a square matrix, where each element represents the weight of a pixel or signal at a corresponding position during the filtering process, used to calculate the value of the output pixel or signal. The specific value of the filter kernel determines the filtering effect; different values ​​can achieve different filtering operations, such as smoothing, sharpening, and edge detection. Here, we are implementing smoothing. The filter kernel size is...

[0041] The spine segmented from the smoothed image is a near-cubic prism. The coordinates (x1, y1) and (x2, y2) of the two opposite angles of the near-cubic prism in the Cartesian coordinate system are obtained. The center points (x1, x2) / 2 and (y1, y2) / 2 of the near-cubic prism are obtained based on the two opposite angles. Finally, the vertebral ash is determined based on the center points.

[0042] In step S300, due to the unique relationship of human tissues—bones are surrounded by muscles, and muscles are surrounded by fat—this physiological structural relationship is still preserved in computed tomography (CT) images. Therefore, this invention can summarize the relatively inherent and stable relationship between bones, muscles, and fat. To prevent errors in the calculation results caused by local tissue damage, a triangular search strategy radiating to both sides of the cancellous bone center point is used to locate muscles and fat. Furthermore, the search directions for muscles and fat are relative. A cylindrical region is used as a rough criterion for determining whether the corresponding tissue has been found in the search area; that is, a cylindrical region is used in the search area to determine whether muscle or fat tissue is present, thereby delineating the muscle or fat region.

[0043] The search is performed by searching 120 degrees to the left and right of the central axis towards the center. That is, a central axis is drawn to distinguish between muscle and fat areas, with muscle and fat areas on either side of the central axis; a 120-degree search area is defined within each of the muscle and fat areas on either side of the central axis.

[0044] For a given X-ray computed tomography (CT) image, different tissues have different Hu values. For lumbar spine data, the CT image information includes information on bones (lumbar spine), muscles, and fat, allowing the Hu values ​​to be calculated. The range of Hu values ​​is used to roughly determine the muscle and fat regions; that is, the Hu values ​​of the muscle or fat regions are calculated, and the initial muscle and fat regions are delineated based on the Hu value ranges of muscle and fat tissues.

[0045] Finally, after defining the initial muscle region and initial fat region within the muscle or fat region based on the Hu value range of muscle tissue and fat tissue, the columnar region is further eroded and expanded to obtain the precise muscle region and fat region again.

[0046] Triangle search is a spatial search method based on triangles, commonly used in computer graphics and computer vision for applications such as 3D model rendering, collision detection, and shape matching. It divides 3D space into a series of triangles and leverages the properties of triangles to accelerate querying and computation during the search process. Triangle search typically involves two main steps:

[0047] 1. Triangle Construction: First, divide the 3D model into a set of triangles. This can be done using a triangulation algorithm (such as Delaunay triangulation) or by manual modeling. Each triangle is defined by three vertices and corresponding attributes such as normals and texture coordinates.

[0048] 2. Triangle Search: This method utilizes the properties of triangles to perform searches or calculations. It includes the following aspects:

[0049] 1) Triangle Intersection Detection: Used to detect the intersection relationship between a ray and a triangle. For example, the ray-triangle intersection test can be used for picking operations or ray tracing.

[0050] 2) Bounding Box: Each triangle can have a bounding box defined to quickly exclude triangles that do not intersect with rays or other geometry. First, the minimum bounding box of the triangle is calculated, and then an intersection test is performed with the queried ray or bounding box to determine if intersection is possible.

[0051] 3) Spatial partitioning structures: Triangle-based spatial partitioning structures (such as grids, BVH, etc.) can further accelerate the search process. These structures group triangles into hierarchical structures, allowing queries to be performed only on specific spatial subsets.

[0052] By optimizing and accelerating triangle search, various query and calculation operations can be performed efficiently in 3D scenes, improving the performance and efficiency of applications such as rendering and collision detection.

[0053] The columnar search algorithm is a spatial search algorithm used to determine whether a given point lies inside a column of regions. A column typically consists of a base polygon and a top plane, and can be used to represent buildings, obstacles, or other objects with columnar structures. The following are the basic steps of a common columnar search algorithm:

[0054] 1. Define the cylinder: Determine the base polygon and top plane of the cylinder. The base polygon can be represented by a set of vertices and can be a rectangle, polygon, or other shapes. The top plane can be defined by the center point of the cylinder and the direction of the normal.

[0055] 2. Spatial Partition Structures: To accelerate the search process, spatial partition structures such as grids, Octrees, or BVH (Bounding Volume Hierarchy) can be constructed. These structures divide the space into hierarchical sub-regions, allowing the search to be performed only within specific sub-regions.

[0056] 3. Point Position Determination: For a given point, first traverse the spatial segmentation structure to determine the sub-region where the point is located. Then, for the columns within that sub-region, perform the following determination:

[0057] 1) Bottom polygon detection: Use a point-to-polygon interior testing algorithm, for example...

[0058] The intersection test between a ray and a polygon, or the test of whether a point is inside a polygon, can be used to determine whether a point is located inside the bottom polygon.

[0059] 2) Top plane determination: Determine whether the point is above the top plane by calculating the distance from the point to the top plane and comparing it with the height of the cylinder.

[0060] 3) Comprehensive judgment: Combine the results of the bottom polygon judgment and the top plane judgment to determine whether the point is located inside the cylinder.

[0061] By using a search region column judgment algorithm, it is possible to determine whether a given point is located inside a certain region column, which plays an important role in applications such as path planning, collision detection, and virtual environment modeling.

[0062] For further details, please see the appendix. Figure 2During the search process, a 120-degree search area is defined in the muscle and fat regions on both sides of the central axis. The search strategy formula is explained using one opposite corner as an example. Define: p1 is a point in the muscle region, p2 is a point in the fat region, and p0 is a point on the central axis; assume p0(x0, y0, z) i p1(x1,y1,z) i p1(x2,y2,z) i );

[0063] p1 satisfies or

[0064] The starting search position for p2 is:

[0065] or

[0066] And the search scope is:

[0067] or or or

[0068] Where H is the height of the i-th layer image and W is the image width.

[0069] At this point, after automatically segmenting each vertebra, since the scanned image includes the coccyx by default but may not include all lumbar vertebrae, based on the above image processing, using the last coccyx as the location, the image is sorted in reverse order to identify the target, resulting in formula L. i =5-i, to obtain the name of the lumbar vertebra. L i Including L1, L2, L3, L4, and L5.

[0070] This invention also discloses an automatic tissue localization calculation system based on CT images, including a segmentation module and a localization module. The segmentation module includes a first segmentation unit and a second segmentation unit based on a deep learning network. The first segmentation unit segments the vertebrae in the CT image using the deep learning network to obtain a spinal mask. The second segmentation unit projects the spinal mask onto a coronal plane, and the projected image is smoothed using a filtering kernel. The smoothed image is segmented by locating the center point of the spine to locate the spinal bone gray matter. The localization module performs a triangular search on both sides of the center point of the spinal bone gray matter to locate and obtain muscle and fat. The search directions for muscle and fat are opposite.

[0071] This invention directly uses local tomographic images and leverages the unique and stable relationship between cancellous bone, muscle, and fat in the human body to automatically identify and locate cancellous bone, muscle, and fat. During the localization process, a method of extrapolating from a cylindrical shape to a free form is used, making the measurement values ​​more accurate.

[0072] It should be noted that the embodiments of the present invention have better implementability and are not intended to limit the present invention in any way. Any person skilled in the art may use the above-disclosed technical content to change or modify it into equivalent effective embodiments. However, any modifications or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A method for automatic tissue localization calculation based on CT images, characterized in that, include: The spinal bones in CT images are segmented using a deep learning network to obtain a spinal mask; The spinal mask is projected onto the coronal plane, and the projected image is smoothed using a filtering kernel. The smoothed image is then segmented by locating the center point of the spine to locate the spinal bone ash. A triangular search is performed on both sides of the center point of the spinal bone ash to locate and obtain muscle and fat; wherein the search directions of muscle and fat are opposite. The step of performing a triangular search on both sides of the center point of the spinal bone ash to locate and obtain muscle and fat includes: A central axis is used to distinguish between muscle and fat regions, with muscle and fat regions on either side of the central axis; A search range of 120 degrees is defined in the muscle and fat regions on both sides of the central axis, respectively; The step of performing a triangular search on both sides of the center point of the spinal bone ash to locate and obtain muscle and fat also includes: The search area uses bar-shaped regions to determine whether muscle or fat tissue is present, thereby defining the muscle or fat region. The step of using a cylindrical region in the search area to determine the presence of muscle or fat tissue, thereby defining the muscle or fat region, includes: Calculate the Hu value of the muscle or fat region, and delineate the initial muscle region and initial fat region in the muscle or fat region based on the Hu value range of muscle tissue and the Hu value range of fat tissue. After defining the initial muscle region and initial fat region within the muscle or fat region based on the Hu value range of muscle tissue and the Hu value range of adipose tissue, the method further includes: The cylindrical region is eroded and expanded to obtain precise muscle and fat regions again. The division is a central axis used to distinguish between muscle and fat regions, with muscle and fat regions on either side of the central axis; the search is performed by defining 120-degree search areas in the muscle and fat regions on either side of the central axis, including: Points within the muscle region, Points within the fat region, A point on the central axis; Assumption , , ; satisfy ,or ; but The starting search position is: ,or ; And the search scope is: ,or ,or ,or ; in, Let be the height value of the i-th layer image. This represents the image width.

2. The tissue automatic positioning calculation method according to claim 1, characterized in that, The step of projecting the spinal mask onto the coronal plane and smoothing the projected image using a filtering kernel includes: The size of the filter kernel is .

3. The tissue automatic positioning calculation method according to claim 1, characterized in that, The step of segmenting the smoothed image by locating the center point of the spine to locate the vertebral ash includes: The spine segmented from the smoothed image is a near-tetracubic prism. The coordinates of two opposite angles of this near-tetracubic prism in a Cartesian coordinate system are calculated. and ; Obtain the center point of the near-cubic prism , ; The spinal bone ash is determined based on the center point.

4. The tissue automatic positioning calculation method according to claim 1, characterized in that, The step of segmenting the vertebrae in CT images using a deep learning network to obtain a spinal mask includes: The deep learning network is a segmentation network, which includes ResUnet, nnUnet, Unet++, and Nestnet networks.

5. A tissue automatic localization calculation system based on CT images, characterized in that, It includes a segmentation module and a localization module. The segmentation module includes a first segmentation unit and a second segmentation unit based on a deep learning network. The deep learning network in the first segmentation unit is used to segment the spinal bones in the CT image to obtain a spinal mask; The second segmentation unit projects the spinal mask onto the coronal plane, and the projected image is smoothed using a filtering kernel. The smoothed image is then segmented by locating the center point of the spine to locate the spinal bone ash. The positioning module performs a triangular search on both sides of the center point of the spinal bone ash to locate and obtain muscle and fat; wherein the search directions of muscle and fat are opposite. The step of performing a triangular search on both sides of the center point of the spinal bone ash to locate and obtain muscle and fat includes: A central axis is used to distinguish between muscle and fat regions, with muscle and fat regions on either side of the central axis; A search range of 120 degrees is defined in the muscle and fat regions on both sides of the central axis, respectively; The step of performing a triangular search on both sides of the center point of the spinal bone ash to locate and obtain muscle and fat also includes: The search area uses bar-shaped regions to determine whether muscle or fat tissue is present, thereby defining the muscle or fat region. The step of using a cylindrical region in the search area to determine the presence of muscle or fat tissue, thereby defining the muscle or fat region, includes: Calculate the Hu value of the muscle or fat region, and delineate the initial muscle region and initial fat region in the muscle or fat region based on the Hu value range of muscle tissue and the Hu value range of fat tissue. After defining the initial muscle region and initial fat region within the muscle or fat region based on the Hu value range of muscle tissue and the Hu value range of adipose tissue, the method further includes: The cylindrical region is eroded and expanded to obtain precise muscle and fat regions again. The division is a central axis used to distinguish between muscle and fat regions, with muscle and fat regions on either side of the central axis; the search is performed by defining 120-degree search areas in the muscle and fat regions on either side of the central axis, including: Points within the muscle region, Points within the fat region, A point on the central axis; Assumption , , ; satisfy ,or ; but The starting search position is: ,or ; And the search scope is: ,or ,or ,or ; in, Let be the height value of the i-th layer image. This represents the image width.

Images