A three-dimensional reconstruction method for focal areas before gastrointestinal tumor resection

Abstract

This invention relates to the field of image processing technology and proposes a three-dimensional reconstruction method for focal regions before gastrointestinal tumor resection. The method includes: obtaining a preliminary tumor region based on the brightness values ​​of pixels within the tissue region; obtaining the tumor boundary characterization based on the brightness values ​​of pixels along the outward extension line, thereby obtaining the true tumor region; acquiring a set of tumor regions; determining the fluctuation degree based on the area of ​​the true tumor regions within the set, thereby obtaining the region to be optimized and its overall region to be optimized; determining the optimization direction based on the tumor boundary characterization of pixels on the boundary of the region to be optimized; determining the optimization degree based on pixels on the boundary of the region to be optimized; and combining the overall region to be optimized to obtain the true boundary, thereby obtaining a three-dimensional reconstruction model. This invention assists gastrointestinal tumor resection surgery with a three-dimensional reconstruction model, improving the accuracy of gastrointestinal tumor resection.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to a method for three-dimensional reconstruction of focal areas before gastrointestinal tumor resection. Background Technology

[0002] Early diagnosis and accurate localization of gastrointestinal tumors are crucial for surgical treatment. Imaging techniques such as CT, MRI, and ultrasound have become important tools for the assessment of gastrointestinal tumors. Traditional two-dimensional images cannot fully reflect the spatial relationship between the tumor and surrounding organs and tissues, especially in terms of localized lesions and invasive expansion of the tumor, where the limitations of two-dimensional images are more pronounced. Three-dimensional reconstruction technology can integrate multiple planar image data into a three-dimensional model, intuitively showing the relationship between the tumor and surrounding structures. This helps doctors more accurately assess the location and size of the tumor and its contact with important blood vessels and tissues, thus providing more precise information for surgical planning.

[0003] Before gastrointestinal tumor resection surgery, the determination of tumor margins is crucial during the three-dimensional reconstruction of the focal area. More precise margins result in a lower recurrence rate. Currently, determining tumor margins faces challenges because some larger tumors may contain necrotic or cystic areas. These areas appear as low-density or signal-free regions on imaging, and their boundaries with surrounding normal tissue are often blurred, making them difficult to distinguish from inflammation or adipose tissue. Furthermore, in the transverse abdominal images acquired during scanning, the tumor may be obscured by other normal tissues, further complicating the identification of its actual boundaries. This leads to unclear tumor boundaries during three-dimensional reconstruction, reducing the accuracy of gastrointestinal tumor resection. Summary of the Invention

[0004] This invention provides a three-dimensional reconstruction method for focal areas before gastrointestinal tumor resection, to solve the problem that the accuracy of gastrointestinal tumor resection is reduced due to the difficulty in identifying the actual boundaries of existing tumors. The specific technical solution adopted is as follows:

[0005] This invention proposes a three-dimensional reconstruction method for focal areas before gastrointestinal tumor resection, the method comprising the following steps:

[0006] Obtain several tissue regions from each abdominal cross-sectional image;

[0007] The initial probability of a tumor in each tissue region is obtained based on the brightness value of the pixels within the tissue region. Several preliminary tumor regions are obtained by filtering based on the initial probability of the tumor. Several outward extension lines are obtained for each preliminary tumor region. The tumor boundary characterization degree of each pixel on each outward extension line is obtained based on the change of the brightness value of the pixels on the outward extension lines. Several real tumor regions are obtained based on the tumor boundary characterization degree.

[0008] Based on the positional similarity of real tumor regions between abdominal cross-sectional images, several sets of tumor regions are obtained; based on the area differences of real tumor regions within the sets of tumor regions, the fluctuation degree of each real tumor region is obtained; based on the fluctuation degree, several regions to be optimized and the total region to be optimized for each region to be optimized are obtained; based on the tumor boundary characterization of pixels on the boundary of the regions to be optimized, the optimization direction of each region to be optimized is obtained.

[0009] Based on the number of pixels on the boundary of the region to be optimized and the matching relationship of pixels on the boundary between regions to be optimized, the optimization degree of each pixel on the boundary of each region to be optimized is obtained; based on the optimization direction and the optimization degree, combined with the position of the pixels on the boundary of the entire region to be optimized, the true boundary of each region to be optimized is obtained.

[0010] A 3D reconstruction model is obtained based on the true boundaries of all regions to be optimized.

[0011] Furthermore, the method for obtaining the preliminary tumor probability of each tissue region based on the brightness value of pixels within the tissue region, and then filtering out several preliminary tumor regions based on the preliminary tumor probability, includes the following specific methods:

[0012] For any given tissue region, based on the discreteness of the brightness of the pixels within that tissue region and the overall low brightness of the pixels within that tissue region, the preliminary probability of a tumor in that tissue region can be obtained.

[0013] Tissue regions with a preliminary tumor probability greater than the preset preliminary screening threshold are categorized as preliminary tumor regions.

[0014] Furthermore, the specific method for obtaining several outward extension lines for each preliminary tumor region, and obtaining the tumor boundary characterization degree of each pixel on each outward extension line based on the change in the brightness value of the pixels on the outward extension lines, includes:

[0015] For any initial tumor region, obtain the centroid of the initial tumor region, and denote the line connecting the centroid to any edge pixel of the initial tumor region as the outward extension line of the initial tumor region.

[0016] No. The first preliminary tumor region The first outward extension line The method for calculating the tumor boundary characterization of each pixel is as follows:

[0017] ;

[0018] In the formula, For the first The first preliminary tumor region The first outward extension line Tumor boundary characterization at each pixel; For the first The first preliminary tumor region The first outward extension line The variance of the brightness values ​​of all pixels preceding a given pixel; For the first The first preliminary tumor region The first outward extension line The variance of the brightness values ​​of all pixels after a given pixel; For the first The first preliminary tumor region The first outward extension line The brightness value of each pixel; For the first The first preliminary tumor region The first outward extension line The range of brightness values ​​of all pixels preceding a given pixel; It is a linear normalization function.

[0019] Furthermore, the specific method for obtaining several real tumor regions based on the tumor boundary characterization degree includes:

[0020] For any outward extension line of any initial tumor region, among all pixels on the outward extension line whose tumor boundary characterization is greater than a preset boundary threshold, the pixel with the largest tumor boundary characterization is recorded as the tumor boundary pixel.

[0021] For any initial tumor region, the Bezier curve technique is used to smoothly connect all the tumor boundary pixels in the initial tumor region to form a real tumor region.

[0022] Furthermore, the specific method for obtaining a set of tumor regions based on the positional similarity between actual tumor regions in abdominal cross-sectional images includes:

[0023] Any abdominal cross-sectional image is recorded as the target abdominal cross-sectional image, and any real tumor region in the target abdominal cross-sectional image is recorded as the target real tumor region; the abdominal cross-sectional images adjacent to the target abdominal cross-sectional image are recorded as the adjacent cross-sectional images of the target abdominal cross-sectional image.

[0024] Obtain the position of the centroid of the target real tumor region; in the adjacent cross-sectional images of the target abdominal cross-sectional image, obtain the position of the centroid of each real tumor region, obtain the relative distance between the centroid of the target real tumor region and the centroid of each real tumor region in the adjacent cross-sectional images of the target abdominal cross-sectional image, and record the real tumor regions whose relative distance between their centroids and the centroid of the target real tumor region is less than a preset distance threshold as the corresponding tumor regions of the target real tumor region in the adjacent cross-sectional images;

[0025] Obtain the corresponding tumor region for each real tumor region in adjacent cross-sectional images, and denote the set of real tumor regions that are corresponding to each other as a tumor region set.

[0026] Furthermore, the specific method for obtaining the fluctuation degree of each real tumor region based on the area differences of the real tumor regions within the tumor region set includes:

[0027] For any real tumor region, obtain the variance of the area of ​​all real tumor regions in the set of tumor regions to which the real tumor region belongs, obtain the mean of the area of ​​all real tumor regions in the set of tumor regions to which the real tumor region belongs, and record the ratio of the variance to the mean as the degree of fluctuation of the real tumor region.

[0028] Furthermore, the specific method for obtaining several regions to be optimized and the full region to be optimized for each region based on the degree of fluctuation includes:

[0029] For any real tumor region, if the fluctuation of the real tumor region is greater than the preset fluctuation threshold, the real tumor region is recorded as a region to be optimized.

[0030] The region with the most edge pixels in the tumor region set containing the region to be optimized is denoted as the total region to be optimized.

[0031] Furthermore, the specific method for obtaining the optimization direction of each region to be optimized based on the tumor boundary characterization of pixels on the boundary of the region to be optimized is as follows:

[0032] For any pixel on the boundary of any region to be optimized, the gradient direction of the tumor boundary characterization of the pixel is obtained by using the gradient descent method, and the unit vector in the gradient direction of the tumor boundary characterization of the pixel is denoted as the characterization vector of the pixel.

[0033] The sum of the representation vectors of all pixels on the boundary of the region to be optimized is denoted as the optimization vector of the region to be optimized; the direction of the optimization vector of the region to be optimized is denoted as the optimization direction of the region to be optimized.

[0034] Furthermore, the specific method for obtaining the optimization degree of each pixel on the boundary of each region to be optimized based on the number of pixels on the boundary of the region to be optimized and the matching relationship of pixels on the boundary between regions to be optimized is as follows:

[0035] Any region to be optimized is denoted as the target region to be optimized. Any region to be optimized in the set of tumor regions where the target region to be optimized is located is obtained. The target region to be optimized and the target region to be optimized are matched with the target region to be optimized using the SURF feature point matching algorithm to obtain several feature points in the target region to be optimized and the matching point of each feature point in the target region to be optimized.

[0036] For any feature point in the target region to be optimized, the number of matching points of that feature point in all regions to be optimized within the tumor region set containing the target region to be optimized is recorded as the matching number of that feature point; for any pixel on the boundary of the target region to be optimized, the feature point with the closest Euclidean distance to that pixel is recorded as the reference point of that pixel; the matching number of the reference point of that pixel is recorded as the neighboring matching number of that pixel;

[0037] The average number of pixels on the boundary of all corresponding tumor regions of the target region to be optimized is denoted as the neighbor boundary length of the target region to be optimized.

[0038] The first on the boundary of the target region to be optimized The optimization level of each pixel is calculated as follows:

[0039] ;

[0040] In the formula, The first on the boundary of the target region to be optimized The degree of optimization per pixel; The length of the nearest boundary of the target region to be optimized; The mean of the neighboring boundary lengths of all regions to be optimized within the set of tumor regions containing the target region to be optimized; The range of the neighboring boundary lengths of all regions to be optimized within the set of tumor regions containing the target region to be optimized. The first on the boundary of the target region to be optimized The ratio of the number of neighbor matches for each pixel to the number of regions to be optimized in the tumor region set containing the target region to be optimized; It is a linear normalization function.

[0041] Furthermore, the specific method for obtaining the true boundary of each region to be optimized based on the optimization direction and the degree of optimization, combined with the position of the pixels on the boundary of the entire region to be optimized, includes:

[0042] For any pixel on the boundary of any region to be optimized, obtain the Euclidean distance between the position of the reference point of the pixel and the position of the matching point of the reference point in the entire region to be optimized in the region to be optimized. The product of the Euclidean distance and the degree of optimization of the pixel is recorded as the optimization distance of the pixel.

[0043] The optimized position of a pixel is defined as the position after moving the pixel by the optimized distance in the optimization direction of the region to be optimized.

[0044] Obtain the optimized position of each pixel on the boundary of the region to be optimized; use Bezier curve technology to smoothly connect the optimized positions of all pixels on the boundary of the region to be optimized to obtain the true boundary of the region to be optimized.

[0045] The beneficial effects of this invention are as follows: When performing three-dimensional reconstruction of focal areas before gastrointestinal tumor resection, some larger tumors may contain necrotic or cystic areas. Therefore, it is necessary to distinguish between necrotic or cystic areas and normal tissue areas while differentiating between tumors and normal tissue. This invention obtains a preliminary tumor area and, based on the changes in brightness values ​​of pixels along outward extension lines, obtains the tumor boundary characterization of each pixel on each outward extension line. Based on the tumor boundary characterization, several real tumor areas are obtained, thus distinguishing between necrotic or cystic areas and normal tissue areas. In multiple transverse abdominal images obtained through scanning, the tumor may be obscured by other normal tissues. The more the tumor is obscured by normal tissue, the more necrotic the tumor... The smaller the tumor, the higher the precision of its resection. This invention obtains the region to be optimized and its entire target region. Based on the tumor boundary characterization of pixels on the boundary of the region to be optimized, the optimization direction of each region to be optimized is obtained, and the optimization direction of pixels on the boundary of the region to be optimized is determined. To accurately obtain the distance the pixels on the boundary of the region to be optimized move in the optimization direction, this invention obtains the optimization degree of each pixel on the boundary of each region to be optimized by the number of pixels on the boundary of the region to be optimized and the matching relationship of pixels on the boundary between regions to be optimized. Based on the optimization direction and the optimization degree, combined with the position of pixels on the boundary of the entire target region, the true boundary of each region to be optimized is obtained. Thus, this invention obtains a three-dimensional reconstruction model from the true boundary of the region to be optimized. The three-dimensional reconstruction model is used to assist in gastrointestinal tumor resection surgery, improving the accuracy of gastrointestinal tumor resection. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of a three-dimensional reconstruction method for focal areas before gastrointestinal tumor resection, provided in one embodiment of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Please see Figure 1 The diagram illustrates a three-dimensional reconstruction method for focal areas before gastrointestinal tumor resection according to an embodiment of the present invention. The method includes the following steps:

[0050] Step S001: Obtain several tissue regions on each abdominal cross-sectional image.

[0051] It should be noted that when performing three-dimensional reconstruction of focal areas before gastrointestinal tumor resection, it is first necessary to use CT computed tomography (CT) technology to obtain corresponding transverse abdominal images through high-resolution thin-slice scanning, and then perform three-dimensional reconstruction based on the transverse abdominal images. Therefore, the transverse abdominal images are obtained first.

[0052] Specifically, the patient is first prepared, including fasting for 4 hours to reduce interference from contents, oral administration of water or low-concentration iodine contrast agent to fill the gastrointestinal tract, and intravenous injection of iodine contrast agent to enhance the imaging of blood vessels and tumors; then thin-slice high-resolution scanning is used, with a tube voltage of 120kV to acquire several consecutive transverse abdominal images.

[0053] It should be noted that in abdominal cross-sectional images, the main difference between tumors and body tissues lies in the difference in brightness. When obtaining the location of the tumor, the abdominal cross-sectional image first needs to be divided into regions based on the brightness value.

[0054] Specifically, firstly, seed points are generated at equal intervals on the cross-sectional image based on a preset grid spacing; wherein, the preset grid spacing is... This embodiment will be described using this as an example; taking each seed point as the center, a region growing algorithm is used to obtain several tissue segmentation regions; a closing operation is performed on each tissue segmentation region to obtain several tissue regions.

[0055] Step S002: Based on the brightness value of pixels within the tissue region, obtain the preliminary probability of tumors in each tissue region; based on the preliminary probability of tumors, select several preliminary tumor regions; obtain several outward extension lines for each preliminary tumor region; based on the change in brightness value of pixels on the outward extension lines, obtain the tumor boundary characterization degree of each pixel on each outward extension line; based on the tumor boundary characterization degree, obtain several real tumor regions.

[0056] It should be noted that three-dimensional reconstruction of the focal area before gastrointestinal tumor resection can help doctors accurately assess the size, location, and growth direction of the tumor, providing necessary spatial information for surgical resection. However, because some larger tumors may have necrotic or cystic areas, these areas often lack effective blood supply, resulting in low signal intensity in imaging images, or even almost no signal in some imaging modalities. Therefore, necrotic or cystic areas usually appear as low-density or non-signal areas on abdominal transverse images obtained by CT scans, making the boundary between the necrotic or cystic area and the surrounding tissues unclear.

[0057] It should be further noted that when gastrointestinal tumors exhibit necrosis, cystic degeneration, or hemorrhage in abdominal transverse images, they will appear as distinct low-density areas, appearing as darker regions on the image. The tumor region also exhibits uneven density, appearing as an area of ​​uneven brightness on the abdominal transverse image. Normal gastrointestinal tissue, adipose tissue, or surrounding normal structures typically show uniform density. On abdominal transverse images, normal gastrointestinal tract usually appears as a medium-density area, appearing as a relatively bright region. Therefore, the initial tumor location can be determined based on the brightness characteristics of the tumor.

[0058] Specifically, for any given tissue region, based on the dispersion of the brightness of pixels within that tissue region and the overall low brightness of pixels within that tissue region, the preliminary probability of a tumor in that tissue region is obtained.

[0059] As an example, the initial probability of a tumor can be expressed by the formula:

[0060] ;

[0061] In the formula, For the first Preliminary likelihood of a tumor in a specific tissue region; For the first The sum of the brightness values ​​of all pixels in the organization region; For the first The variance of the brightness values ​​of all pixels in a given tissue region; For the first The average brightness value of all pixels in a given tissue region; This is a hyperparameter to prevent the denominator from being 0; It is a linear normalization function, and the normalization object is all tissue regions. .

[0062] In the formula, The smaller the value, the better. The smaller the overall brightness value of the internal pixels of a tissue region, the more it matches the brightness characteristics of gastrointestinal tumors in abdominal cross-sectional images when there is necrosis, cystic degeneration, or hemorrhage. The larger The smaller the value, the better. The more uneven the brightness distribution in a tissue region, the more it matches the brightness characteristics of a tumor region.

[0063] Furthermore, tissue regions with a preliminary tumor probability greater than a preset preliminary screening threshold are denoted as preliminary tumor regions; wherein, the preset preliminary screening threshold is 0.5, and this embodiment will be described using this as an example.

[0064] It should be noted that the preliminary tumor area includes the actual tumor area, which consists of ordinary tumor areas as well as necrotic or cystic areas. To accurately assist in the resection of gastrointestinal tumors, it is necessary to more precisely determine the boundary between the actual tumor area and the normal tissue area.

[0065] It should be noted that the initial tumor area is divided into the tumor area, the necrotic or cystic area, and the normal tissue area from the inside out. Since the brightness characteristics of different areas are different, the brightness analysis of the initial tumor area is performed from the inside out.

[0066] Specifically, for any initial tumor region, the centroid of the initial tumor region is obtained, and the line connecting the centroid to any edge pixel of the initial tumor region is recorded as the outward extension line of the initial tumor region.

[0067] It should be noted that normal tissue has a relatively high density and usually appears as a brighter area with relatively uniform brightness in abdominal cross-sectional images; necrotic or cystic areas have low signal intensity in imaging images due to the lack of effective blood supply, and may even have almost no signal in some imaging modalities; tumor areas show uneven density and appear as areas with uneven brightness in abdominal cross-sectional images.

[0068] Specifically, the first The first preliminary tumor region The first outward extension line The method for calculating the tumor boundary characterization of each pixel is as follows:

[0069] ;

[0070] In the formula, For the first The first preliminary tumor region The first outward extension line Tumor boundary characterization at each pixel; For the first The first preliminary tumor region The first outward extension line The variance of the brightness values ​​of all pixels preceding a given pixel; For the first The first preliminary tumor region The first outward extension line The variance of the brightness values ​​of all pixels after a given pixel; For the first The first preliminary tumor region The first outward extension line The brightness value of each pixel; For the first The first preliminary tumor region The first outward extension line The range of brightness values ​​of all pixels preceding a given pixel; For linear normalization functions, the object being normalized is the th... The first preliminary tumor region All pixels on the outward extension line .

[0071] In the formula, The larger the value, the more likely it is to be the first. The first preliminary tumor region The first outward extension line The brightness of pixels after the first pixel is relatively uniform, exhibiting the brightness characteristics of normal tissue, while the brightness of earlier pixels varies significantly, exhibiting the brightness characteristics of tumor areas and necrotic or cystic areas. The first preliminary tumor region The first outward extension line The higher the number of pixels, the greater the likelihood that it represents the tumor boundary; since normal tissue appears as a brighter area, The larger the value, the more likely it is to be the first. The first preliminary tumor region The first outward extension line Each pixel is located near normal tissue; The larger the value, the more likely it is to be the first. The first preliminary tumor region The first outward extension line The more likely a pixel is to be located behind a tumor area or an area of ​​necrosis or cyst.

[0072] It should be noted that the first The first preliminary tumor region The tumor boundary characterization of the first and last pixels on the outward extension line is recorded as 0.

[0073] Furthermore, for any outward extension line of any preliminary tumor region, among all pixels on the outward extension line whose tumor boundary characterization is greater than a preset boundary threshold, the pixel with the largest tumor boundary characterization is recorded as the tumor boundary pixel; where the preset boundary threshold is 0.9.

[0074] It should be noted that some outward extension lines may not have tumor boundary pixels, indicating that there is an unclear boundary between necrotic or cystic areas and normal tissue areas on these outward extension lines. It is necessary to obtain the actual tumor area based on the obtained tumor boundary pixels.

[0075] Specifically, for any initial tumor region, the Bezier curve technique is used to smoothly connect all the tumor boundary pixels in the initial tumor region to form a real tumor region; the Bezier curve technique is a well-known technique, and the specific method will not be described here.

[0076] Step S003: Based on the positional similarity of the real tumor regions between the abdominal cross-sectional images, obtain several sets of tumor regions; based on the area differences of the real tumor regions within the sets of tumor regions, obtain the fluctuation degree of each real tumor region; based on the fluctuation degree, obtain several regions to be optimized and the total region to be optimized for each region; based on the tumor boundary characterization of the pixels on the boundary of the region to be optimized, obtain the optimization direction of each region to be optimized.

[0077] It's important to note that during gastrointestinal tumor resection, if the tumor region's boundaries are not precise enough, the constructed 3D tumor region will also be inaccurate, affecting the surgical margin and causing residual tumor. This residual tumor may lead to postoperative local recurrence. Precise tumor region boundaries determine the optimal surgical margin, ensuring that while removing the tumor, micro-spread of the tumor is minimized, reducing the likelihood of postoperative recurrence. Furthermore, in multiple abdominal transverse images acquired through scanning, the tumor may be obscured by other normal tissues. The actual exposed tumor area will change from one slice of abdominal transverse image to another. The greater the variation in the area of ​​the exposed tumor area across different abdominal transverse images, the more the tumor is obscured by normal tissue, indicating a smaller tumor size and higher resection precision, thus requiring greater optimization of the initial boundaries.

[0078] It should be further explained that, since it is necessary to compare the changes of the same tumor in different abdominal transverse images, it is necessary to match the actual tumor regions in different abdominal transverse images. Because the abdominal transverse images are acquired using a high-resolution thin-slice scanning method, the actual tumor regions are relatively similar in location within adjacent abdominal transverse images, so it is necessary to match the actual tumor regions in different abdominal transverse images.

[0079] Specifically, any abdominal cross-sectional image is recorded as the target abdominal cross-sectional image, and any real tumor region in the target abdominal cross-sectional image is recorded as the target real tumor region; the abdominal cross-sectional images adjacent to the target abdominal cross-sectional image are recorded as the adjacent cross-sectional images of the target abdominal cross-sectional image.

[0080] The centroid of the target real tumor region is obtained. In adjacent cross-sectional images of the target abdominal cross-sectional image, the centroid of each real tumor region is obtained. The relative distance between the centroid of the target real tumor region and the centroid of each real tumor region in adjacent cross-sectional images of the target abdominal cross-sectional image is obtained. Real tumor regions whose relative distance to the centroid of the target real tumor region is less than a preset distance threshold are recorded as the corresponding tumor regions of the target real tumor region in the adjacent cross-sectional images. It should be noted that since the abdominal cross-sectional images are of the same size, the distance between the centroids of real tumor regions in different abdominal cross-sectional images is obtained as follows: the centroid of the target real tumor region is projected onto any abdominal cross-sectional image other than the target abdominal cross-sectional image to obtain the projection point of the target real tumor region in that abdominal cross-sectional image. For any real tumor region in that abdominal cross-sectional image, the Euclidean distance from the centroid of the real tumor region to the projection point is recorded as the relative distance between the centroids of the target real tumor region and the centroid of the real tumor region. The preset distance threshold is 5, and this embodiment uses this as an example for description.

[0081] Based on the above method, the corresponding tumor regions of each real tumor region in adjacent cross-sectional images are obtained. The set of real tumor regions that are corresponding to each other is denoted as a tumor region set. For example, real tumor region 1 in abdominal cross-sectional image 1 and real tumor region 2 in abdominal cross-sectional image 2 are corresponding to each other, and real tumor region 2 in abdominal cross-sectional image 2 and real tumor region 3 in abdominal cross-sectional image 3 are corresponding to each other. Then, real tumor region 1, real tumor region 2, and real tumor region 3 are considered as a tumor region set. It should be noted that each tumor region set represents the appearance of a tumor in different abdominal cross-sectional images.

[0082] It should be noted that in adjacent abdominal cross-sectional images, the greater the difference in area between the actual tumor area and its corresponding tumor area, the greater the fluctuation in the area of ​​the actual tumor area. This means that more precise cutting is required during surgery to avoid missing lesions. Therefore, further optimization and adjustment of the boundary of the actual tumor area is necessary.

[0083] Specifically, for any real tumor region, obtain the variance of the area of ​​all real tumor regions in the set of tumor regions to which the real tumor region belongs, obtain the mean of the area of ​​all real tumor regions in the set of tumor regions to which the real tumor region belongs, and record the ratio of the variance to the mean as the degree of fluctuation of the real tumor region.

[0084] It should be noted that the greater the fluctuation of the actual tumor area, the higher the dispersion of the tumor area in different abdominal cross-sectional images, and the more necessary it is to further optimize and adjust the boundary of the actual tumor area.

[0085] It should be noted that for the blurred boundary between normal tissue areas and necrotic or cystic areas, a relatively complete boundary needs to be found in the abdominal cross-sectional image in order to more accurately optimize the boundary and determine the direction of boundary optimization, that is, to determine the direction of the boundary where there is unclear boundary.

[0086] Specifically, for any real tumor region, if the fluctuation of the real tumor region is greater than a preset fluctuation threshold, the real tumor region is recorded as a region to be optimized; wherein, the preset fluctuation threshold is 0.5, and this embodiment is described using this as an example.

[0087] The region with the most edge pixels in the tumor region set containing the region to be optimized is denoted as the full region to be optimized.

[0088] For any pixel on the boundary of any region to be optimized, the gradient direction of the tumor boundary characterization of that pixel is obtained using the gradient descent method, and the unit vector in the gradient direction of the tumor boundary characterization of that pixel is denoted as the characterization vector of that pixel. The gradient descent method is a well-known technique, and the specific method will not be described here. It should be noted that the region to be optimized is a real tumor region, and the pixels on the boundary of the real tumor region are obtained from the tumor boundary pixels, so they all have tumor boundary characterization.

[0089] The sum of the representation vectors of all pixels on the boundary of the region to be optimized is denoted as the optimization vector of the region to be optimized; the direction of the optimization vector of the region to be optimized is denoted as the optimization direction of the region to be optimized.

[0090] Step S004: Based on the number of pixels on the boundary of the region to be optimized and the matching relationship of pixels on the boundary between regions to be optimized, obtain the optimization degree of each pixel on the boundary of each region to be optimized; based on the optimization direction and the optimization degree, combined with the position of pixels on the boundary of the entire region to be optimized, obtain the true boundary of each region to be optimized.

[0091] It should be noted that after obtaining the optimization direction of the region to be optimized, the positions of the pixels on the boundary of the region to be optimized need to be optimized along the optimization direction to obtain the specific boundary of the tumor.

[0092] It should be further explained that for any boundary of the region to be optimized, if the length of the boundary of the region to be optimized varies in different abdominal cross-sectional images and the degree of length variation is large, then the boundary of the region to be optimized has boundary fluctuations in the direction perpendicular to the scanning direction in the 3D image. Therefore, it needs to be optimized more accurately and given greater weight. At the same time, for the pixels on the boundary of the region to be optimized, if there are corresponding pixels in other regions to be optimized in the tumor region set where the region to be optimized is located, then the pixels are relatively accurate and do not need to be given greater weight for optimization.

[0093] Specifically, any region to be optimized is designated as the target region to be optimized. Any region to be optimized in the set of tumor regions where the target region to be optimized is located is obtained. The target region to be optimized and the target region to be optimized are matched using the SURF feature point matching algorithm to obtain several feature points in the target region to be optimized and the matching point of each feature point in the target region to be optimized. The SURF feature point matching algorithm is a well-known technology, and the specific method is not described here.

[0094] For any feature point in the target region to be optimized, the number of matching points of that feature point in all regions to be optimized within the tumor region set containing the target region to be optimized is recorded as the matching number of that feature point; for any pixel on the boundary of the target region to be optimized, the feature point with the closest Euclidean distance to that pixel is recorded as the reference point of that pixel; the matching number of the reference point of that pixel is recorded as the neighboring matching number of that pixel;

[0095] The average number of pixels on the boundary of all corresponding tumor regions of the target region to be optimized is denoted as the neighbor boundary length of the target region to be optimized.

[0096] The first on the boundary of the target region to be optimized The optimization level of each pixel is calculated as follows:

[0097] ;

[0098] In the formula, The first on the boundary of the target region to be optimized The degree of optimization per pixel; The length of the nearest boundary of the target region to be optimized; The mean of the neighboring boundary lengths of all regions to be optimized within the set of tumor regions containing the target region to be optimized; The range of the neighboring boundary lengths of all regions to be optimized within the set of tumor regions containing the target region to be optimized. The first on the boundary of the target region to be optimized The ratio of the number of neighbor matches for each pixel to the number of regions to be optimized in the tumor region set containing the target region to be optimized; This is a linear normalization function, which normalizes all pixels on the boundary of each region to be optimized. .

[0099] It should be noted that after obtaining the optimization direction of each region to be optimized and the optimization degree of each pixel on the boundary of each region to be optimized, it is necessary to optimize the pixels on the boundary of each region to be optimized.

[0100] Specifically, for any pixel on the boundary of any region to be optimized, the Euclidean distance between the position of the reference point of the pixel and the position of the matching point of the reference point in the entire region to be optimized is obtained, and the product of the Euclidean distance and the degree of optimization of the pixel is recorded as the optimization distance of the pixel.

[0101] The optimized position of a pixel is defined as the position after moving the pixel by the optimized distance in the optimization direction of the region to be optimized.

[0102] The optimized position of each pixel on the boundary of the region to be optimized is obtained according to the above method; the optimized positions of all pixels on the boundary of the region to be optimized are smoothly connected using the Bezier curve technique to obtain the true boundary of the region to be optimized; wherein, the Bezier curve technique is a known technique.

[0103] Step S005: Obtain the three-dimensional reconstruction model based on the true boundaries of all regions to be optimized.

[0104] It should be noted that after obtaining the true boundary, a 3D model reconstruction is required based on the true boundary.

[0105] For any real boundary, the pixels within the real boundary are recorded as 1, and the pixels outside the real boundary are recorded as 0, thus obtaining a binary mask of the region to be optimized. This binary mask is then superimposed onto the abdominal cross-sectional image where the region to be optimized is located to generate the region of interest.

[0106] Furthermore, by combining the three-dimensional model reconstruction with the region of interest, a three-dimensional reconstruction model is obtained; this three-dimensional reconstruction model is then used to assist doctors in tumor resection surgery; the method of three-dimensional model reconstruction is a well-known technique, and the specific method will not be described here.

[0107] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for three-dimensional reconstruction of focal areas before gastrointestinal tumor resection, characterized in that, The method includes the following steps: Obtain several tissue regions from each abdominal cross-sectional image; The initial probability of a tumor in each tissue region is obtained based on the brightness value of the pixels within the tissue region. Several preliminary tumor regions are obtained by filtering based on the initial probability of the tumor. Several outward extension lines are obtained for each preliminary tumor region. The tumor boundary characterization degree of each pixel on each outward extension line is obtained based on the change of the brightness value of the pixels on the outward extension lines. Several real tumor regions are obtained based on the tumor boundary characterization degree. Based on the positional similarity of real tumor regions between abdominal cross-sectional images, several sets of tumor regions are obtained; based on the area differences of real tumor regions within the sets of tumor regions, the fluctuation degree of each real tumor region is obtained; based on the fluctuation degree, several regions to be optimized and the total region to be optimized for each region to be optimized are obtained; based on the tumor boundary characterization of pixels on the boundary of the regions to be optimized, the optimization direction of each region to be optimized is obtained. Based on the number of pixels on the boundary of the region to be optimized and the matching relationship of pixels on the boundary between regions to be optimized, the optimization degree of each pixel on the boundary of each region to be optimized is obtained; based on the optimization direction and the optimization degree, combined with the position of the pixels on the boundary of the entire region to be optimized, the true boundary of each region to be optimized is obtained. A 3D reconstruction model is obtained based on the true boundaries of all regions to be optimized.

2. The method for three-dimensional reconstruction of focal areas before gastrointestinal tumor resection according to claim 1, characterized in that, The method for determining the preliminary tumor probability of each tissue region based on the brightness value of pixels within the tissue region, and then filtering out several preliminary tumor regions based on the preliminary tumor probability, includes the following specific methods: For any given tissue region, based on the discreteness of the brightness of the pixels within that tissue region and the overall low brightness of the pixels within that tissue region, the preliminary probability of a tumor in that tissue region can be obtained. Tissue regions with a preliminary tumor probability greater than the preset preliminary screening threshold are categorized as preliminary tumor regions.

3. The method for three-dimensional reconstruction of focal areas before gastrointestinal tumor resection according to claim 1, characterized in that, The specific method for obtaining several outward extension lines for each preliminary tumor region, and obtaining the tumor boundary characterization degree of each pixel on each outward extension line based on the change in the brightness value of the pixels on the outward extension lines, includes: For any initial tumor region, obtain the centroid of the initial tumor region, and denote the line connecting the centroid to any edge pixel of the initial tumor region as the outward extension line of the initial tumor region. No. The first preliminary tumor region The first outward extension line The method for calculating the tumor boundary characterization of each pixel is as follows: ； In the formula, For the first The first preliminary tumor region The first outward extension line Tumor boundary characterization at each pixel; For the first The first preliminary tumor region The first outward extension line The variance of the brightness values ​​of all pixels preceding a given pixel; For the first The first preliminary tumor region The first outward extension line The variance of the brightness values ​​of all pixels after a given pixel; For the first The first preliminary tumor region The first outward extension line The brightness value of each pixel; For the first The first preliminary tumor region The first outward extension line The range of brightness values ​​of all pixels preceding a given pixel; It is a linear normalization function.

4. The method for three-dimensional reconstruction of focal areas before gastrointestinal tumor resection according to claim 1, characterized in that, The specific method for obtaining several real tumor regions based on the tumor boundary characterization degree includes: For any outward extension line of any initial tumor region, among all pixels on the outward extension line whose tumor boundary characterization is greater than a preset boundary threshold, the pixel with the largest tumor boundary characterization is recorded as the tumor boundary pixel. For any initial tumor region, the Bezier curve technique is used to smoothly connect all the tumor boundary pixels in the initial tumor region to form a real tumor region.

5. The method for three-dimensional reconstruction of focal areas before gastrointestinal tumor resection according to claim 1, characterized in that, The method for obtaining a set of tumor regions based on the positional similarity between actual tumor regions in abdominal cross-sectional images includes the following specific methods: Any abdominal cross-sectional image is recorded as the target abdominal cross-sectional image, and any real tumor region in the target abdominal cross-sectional image is recorded as the target real tumor region; the abdominal cross-sectional images adjacent to the target abdominal cross-sectional image are recorded as the adjacent cross-sectional images of the target abdominal cross-sectional image. Obtain the position of the centroid of the target real tumor region; in the adjacent cross-sectional images of the target abdominal cross-sectional image, obtain the position of the centroid of each real tumor region, obtain the relative distance between the centroid of the target real tumor region and the centroid of each real tumor region in the adjacent cross-sectional images of the target abdominal cross-sectional image, and record the real tumor regions whose relative distance between their centroids and the centroid of the target real tumor region is less than a preset distance threshold as the corresponding tumor regions of the target real tumor region in the adjacent cross-sectional images; Obtain the corresponding tumor region for each real tumor region in adjacent cross-sectional images, and denote the set of real tumor regions that are corresponding to each other as a tumor region set.

6. The method for three-dimensional reconstruction of focal areas before gastrointestinal tumor resection according to claim 1, characterized in that, The method for determining the fluctuation level of each real tumor region based on the area differences of real tumor regions within the tumor region set includes the following specific methods: For any real tumor region, obtain the variance of the area of ​​all real tumor regions in the set of tumor regions to which the real tumor region belongs, obtain the mean of the area of ​​all real tumor regions in the set of tumor regions to which the real tumor region belongs, and record the ratio of the variance to the mean as the degree of fluctuation of the real tumor region.

7. The method for three-dimensional reconstruction of focal areas before gastrointestinal tumor resection according to claim 1, characterized in that, The specific method for obtaining several regions to be optimized and the full region to be optimized for each region based on the fluctuation level is as follows: For any real tumor region, if the fluctuation of the real tumor region is greater than the preset fluctuation threshold, the real tumor region is recorded as a region to be optimized. The region with the most edge pixels in the tumor region set containing the region to be optimized is denoted as the total region to be optimized.

8. The method for three-dimensional reconstruction of focal areas before gastrointestinal tumor resection according to claim 1, characterized in that, The method for obtaining the optimization direction of each region to be optimized based on the tumor boundary characterization of pixels on the boundary of the region to be optimized includes the following specific methods: For any pixel on the boundary of any region to be optimized, the gradient direction of the tumor boundary characterization of the pixel is obtained by using the gradient descent method, and the unit vector in the gradient direction of the tumor boundary characterization of the pixel is denoted as the characterization vector of the pixel. The sum of the representation vectors of all pixels on the boundary of the region to be optimized is denoted as the optimization vector of the region to be optimized; the direction of the optimization vector of the region to be optimized is denoted as the optimization direction of the region to be optimized.

9. A method for three-dimensional reconstruction of focal areas before gastrointestinal tumor resection according to claim 5, characterized in that, The method for determining the optimization level of each pixel on the boundary of each region to be optimized based on the number of pixels on the boundary of the region to be optimized and the matching relationship of pixels on the boundary between regions to be optimized includes the following specific methods: Any region to be optimized is denoted as the target region to be optimized. Any region to be optimized in the set of tumor regions where the target region to be optimized is located is obtained. The target region to be optimized and the target region to be optimized are matched with the target region to be optimized using the SURF feature point matching algorithm to obtain several feature points in the target region to be optimized and the matching point of each feature point in the target region to be optimized. For any feature point in the target region to be optimized, the number of matching points of that feature point in all regions to be optimized within the tumor region set where the target region to be optimized is recorded as the matching number of that feature point; for any pixel on the boundary of the target region to be optimized, the feature point with the closest Euclidean distance to that pixel is recorded as the reference point of that pixel; the matching number of the reference point of that pixel is recorded as the neighboring matching number of that pixel; The average number of pixels on the boundary of all corresponding tumor regions of the target region to be optimized is denoted as the neighbor boundary length of the target region to be optimized. The first on the boundary of the target region to be optimized The optimization level of each pixel is calculated as follows: ； In the formula, The first on the boundary of the target region to be optimized The degree of optimization per pixel; The length of the nearest boundary of the target region to be optimized; The mean of the neighboring boundary lengths of all regions to be optimized within the set of tumor regions containing the target region to be optimized; The range of the neighboring boundary lengths of all regions to be optimized within the set of tumor regions containing the target region to be optimized. The first on the boundary of the target region to be optimized The ratio of the number of neighbor matches for each pixel to the number of regions to be optimized in the tumor region set containing the target region to be optimized; It is a linear normalization function.

10. A method for three-dimensional reconstruction of focal areas before gastrointestinal tumor resection according to claim 9, characterized in that, The specific method for obtaining the true boundary of each region to be optimized based on the optimization direction and the degree of optimization, combined with the position of the pixels on the boundary of the entire region to be optimized, includes: For any pixel on the boundary of any region to be optimized, obtain the Euclidean distance between the position of the reference point of the pixel and the position of the matching point of the reference point in the entire region to be optimized in the region to be optimized. The product of the Euclidean distance and the degree of optimization of the pixel is recorded as the optimization distance of the pixel. The optimized position of a pixel is defined as the position after moving the pixel by the optimized distance in the optimization direction of the region to be optimized. Obtain the optimized position of each pixel on the boundary of the region to be optimized; use Bezier curve technology to smoothly connect the optimized positions of all pixels on the boundary of the region to be optimized to obtain the true boundary of the region to be optimized.

Images