An unsupervised liver tumor CT image segmentation method based on handcrafted features

Abstract

The application provides an unsupervised liver tumor CT image segmentation method based on manual features, relates to the technical field of deep learning, and first collects normal liver CT images as a data set, and selects a tumor position in combination with clinical knowledge. A texture similar to real imaging is generated through three-dimensional simple noise binary mask, and histological features are used for morphological modeling. Then, the tumor texture is superimposed with the liver CT image at the selected position to synthesize a new liver tumor CT image. A segmentation model is trained by using the CT image with annotations, and the synthesized image is used for medical image segmentation. Finally, the segmentation results are evaluated by using the Dice coefficient, the Hausdorff distance, the standardized surface distance and the surface distance index. The application is helpful for artificially synthesizing tumor lesion images and generating voxel-level annotations, and provides a new perspective for solving the medical image annotation challenge and promoting liver tumor diagnosis and treatment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of deep learning, and particularly relates to a handcrafted feature based unsupervised liver tumor CT image segmentation method. BACKGROUND

[0002] Three-dimensional medical imaging technology can accurately visualize and quantitatively analyze internal structures, thus completely changing clinical practice. Accurate segmentation of three-dimensional images is crucial for early detection and effective diagnosis of liver tumors. Tumor segmentation has become a basic task of medical image analysis and has important value for accurate diagnosis and treatment decision of lesions, and has been the focus of researchers and clinicians.

[0003] However, the existing tumor CT image segmentation method has the following problems: first, the volume characteristics of three-dimensional imaging make manual segmentation costly and subjective; second, a large amount of training data is usually required in conventional medical image segmentation methods, but the scale of well-labeled tumor data is limited; third, the complexity and morphological difference of tumors make tumor image segmentation more difficult.

[0004] Therefore, it is urgent to provide a scheme to improve the above problems. SUMMARY

[0005] The purpose of the present application is to provide a handcrafted feature based unsupervised liver tumor CT image segmentation method to improve the problems in the background art.

[0006] The handcrafted feature based unsupervised liver tumor CT image segmentation method provided by the present application adopts the following technical scheme:

[0007] Collect CT scan images of normal liver as a data set, and verify the volume and gradient of the potential tumor position of the liver to determine the tumor position of the liver;

[0008] Obtain the final tumor texture based on the tumor position of the liver, and model the morphology of the liver tumor based on the final tumor texture to synthesize new liver tumor CT images;

[0009] Use annotated CT scan images to train a segmentation model, perform medical image segmentation on the synthesized tumor images as a data set, and verify the segmentation performance based on evaluation indexes.

[0010] The handcrafted feature-based unsupervised liver tumor CT image segmentation method has the beneficial effects that: first, normal liver CT images are collected as a data set, and the tumor position is selected in combination with clinical knowledge; second, a three-dimensional simple noise binary mask is generated to generate a texture similar to real imaging, and histological features are used for morphological modeling; third, the tumor texture is superimposed on the selected position of the liver CT image to synthesize a new liver tumor CT image; fourth, a segmentation model is trained using the CT image with annotations, and the synthesized image is used for medical image segmentation; and fifth, the segmentation results are evaluated by using the Dice coefficient, Hausdorff distance, normalized surface distance and surface distance indicators. The method helps to artificially synthesize tumor lesion images and generate voxel-level annotations, and provides a new perspective for solving medical image annotation challenges and promoting liver tumor diagnosis and treatment.

[0011] Optionally, the process of volume and gradient verification on the potential tumor position of the liver to determine the tumor position of the liver comprises:

[0012] Generating spherical coordinates in the liver Converting the spherical coordinates into Cartesian coordinates x=(x, y, z), and converting the Cartesian coordinates and the spherical coordinates based on the elliptical axis and the half value;

[0013] Iteratively verifying each potential tumor position x to determine whether the point is located in the liver until the tumor position x is determined, and stopping the iteration, wherein the tumor position is located inside the liver and has sufficient distance from the liver boundary. t =(x t ,y t ,z t )

[0014] Optionally, the process of generating the tumor position of the liver comprises:

[0015] Obtaining the number of liver masks in the neighborhood volume around the predicted position based on the neighborhood volume check, obtaining the first tumor position based on the number of liver masks and the first threshold value;

[0016] Calculating the gradient value of the scanning amount at the predicted position based on the gradient amplitude evaluation, obtaining the second tumor position based on the gradient value and the second threshold value, and generating the tumor position of the liver based on the first tumor position and the second tumor position.

[0017] Optionally, the handcrafted feature-based unsupervised liver tumor CT image segmentation method comprises:

[0018] Obtaining a three-dimensional simple noise based on a pre-defined parameter, using nonlinear diffusion and wavelet transform to respectively perform artifact filtering and high-frequency detail preservation on the noise to obtain a texture to be processed;

[0019] normalizing the to-be-processed texture to obtain a normalized texture, and smoothing the normalized texture based on a first Gaussian filter to obtain a smoothed texture;

[0020] scaling and moving the smoothed texture to obtain a first texture;

[0021] obtaining a first binary mask based on random sampling, smoothing the first binary mask based on a second Gaussian filter to obtain a second binary mask, and generating a final tumor texture by scaling and cropping the first texture based on the second binary mask.

[0022] Optionally, the process of morphological modeling of liver tumors comprises:

[0023] dividing the synthetic tumor into four size categories: micro, small, medium, and large, and extracting the number of synthetic samples from a discrete uniform distribution;

[0024] simulating the tumor shape based on an ellipsoid defined by parameters, the semi-principal axes of the ellipsoid on the x, y, and z coordinates being taken from a uniform distribution U(0.75r, 1.25r) and the ellipsoid being centered at the coordinates x t =(x t ,y t ,z t ) within the scan volume domain f(x), and the initial tumor mask being defined within this range;

[0025] nonlinearly transforming the initial tumor mask based on nonlinear elastic deformation to obtain a deformed tumor mask:

[0026] convolving the deformed tumor mask based on a third Gaussian filter to obtain a synthetic mask.

[0027] Optionally, the process of synthesizing a new liver tumor CT image comprises:

[0028] quantifying the density of the liver: dividing the liver into three density intervals, and normalizing the intensity values within each interval;

[0029] adjusting the tumor density to reduce the density of high-density tissues;

[0030] integrating the tumor into the scan volume domain f(x) and the corresponding liver mask l(x).

[0031] Optionally, the process of quantifying the density of the liver comprises:

[0032] obtaining the lower limit L and the upper limit U of the liver region P, and calculating the density scale Δ:

[0033] Δ = (U-L) / 3;

[0034] For each three-dimensional coordinate point P in the liver region ijk The following operations are performed:

[0035] If P ijk <L+Δ, let P ijk =L; if L+Δ≤P ijk <L+2Δ, let P ijk =L+Δ; if L+Δ≤P ijk <U, let P ijk =L+2Δ; if P ijk ≥U, let P ijk =U;

[0036] According to the above operations, a density distribution map D is obtained, for each D ijk on the density distribution map D, the following processing is performed: if D ijk =U and P ijk ≠U, let D ijk =L+2Δ.

[0037] Optionally, the process of adjusting the tumor density includes:

[0038] Obtaining the lower limit L and the upper limit U of the liver region P, and calculating the density scale Δ:

[0039] Δ=(U-L) / 3;

[0040] For each coordinate point F ijk in the scanning volume domain F, the following processing is performed:

[0041] When the coordinate point F ijk coincides with the generated tumor position, if D ijk =U, let if D ijk =L+2Δ, let otherwise, F ijk is retained;

[0042] wherein D ijk represents a coordinate point on the density distribution map D, U represents the upper limit of the liver region, and L represents the lower limit of the liver region.

[0043] Optionally, the process of integrating the tumor into the scanning volume domain f(x) and the corresponding liver mask l(x) includes:

[0044] Integrating the tumor with the scanning volume:

[0045] f'(x)=[1-t"(x)]⊙f(x)+t"(x)⊙T"(x);

[0046] Wherein, ⊙ represents element multiplication, t''(x) represents a tumor mask array, T''(x) represents a tumor texture array, and a new scanning volume f'(x) can be obtained through the above transformation, and a new liver mask containing a tumor can be synthesized in the manner of l'(x) = l(x) + t''(x);

[0047] Determine the boundary region of the tumor:

[0048]

[0049] Wherein λ l represents a lower intensity threshold for defining the boundary region of the tumor, λ u represents an upper intensity threshold for defining the boundary region of the tumor;

[0050] A Gaussian filter G(x; σ3) is used to convolve b(x) to generate a smoothed boundary image b'(x):

[0051] b'(x) = b(x) * G(x; σ3);

[0052] Wherein σ3 determines the standard deviation of the Gaussian kernel, and controls the smoothness of the boundary conversion;

[0053] The boundary image b'(x) enhances the voxel intensity of the tumor periphery by adding a fixed intensity offset δ, which can simulate the increase of Hounsfield Unit (HU) of the cyst:

[0054] f'''(x) = f''(x) + δ·b'(x);

[0055] Wherein δ represents the HU intensity difference between the tumor core and the fibrous capsule, and f'''(x) represents the final volumetric scanning result.

[0056] Optionally, when the synthesized tumor image is used as a data set for medical image segmentation and the segmentation performance is verified based on evaluation indexes, the image segmentation is trained by using 118 CT scans with liver and liver tumor annotations to train a segmentation model, the data set includes 281 three-dimensional abdominal CTs, the resolution of each data set is 512*512 pixels, the spiral data is reconstructed in the axial plane, the slice thickness is 1.25 mm, and the interval is 1.0 mm, and the data set can be divided into a liver tumor segmentation data set and a normal liver CT data set, wherein,

[0057] The liver tumor segmentation data set contains detailed per-pixel annotations of liver tumors, and is part of the LiTS challenge;

[0058] The normal liver CT dataset is a normal liver dataset consisting of 116 CT scans compiled from healthy subjects in the CHAOS, Pancreas-CT, BTCV and LiTS datasets, which is the baseline for generating synthetic tumor images. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 is a flowchart of a manual feature-based unsupervised liver tumor CT image segmentation method provided by the present application. DETAILED DESCRIPTION

[0060] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of protection of the present application. Unless otherwise defined, the technical terms or scientific terms used herein should have the meanings commonly understood by a person of ordinary skill in the art to which the present application belongs. The terms such as “comprise” and the like used herein mean that the elements or objects before the terms encompass the elements or objects listed after the terms and their equivalents, without excluding other elements or objects.

[0061] The embodiments of the present application provide a manual feature-based unsupervised liver tumor CT image segmentation method, which comprises:

[0062] S1, collecting CT scan images of a normal liver as a dataset, and performing volume and gradient verification on potential tumor positions of the liver to determine tumor positions of the liver;

[0063] S2, obtaining final tumor textures based on the tumor positions of the liver, and performing morphological modeling on liver tumors based on the final tumor textures to synthesize new liver tumor CT images;

[0064] S3, training a segmentation model using annotated CT scan images, performing medical image segmentation on the synthesized tumor images as a dataset, and verifying segmentation performance based on evaluation indexes.

[0065] In some embodiments, in the process of performing step S1 to determine the tumor positions of the liver by volume and gradient verification on the potential tumor positions of the liver, specifically comprising:

[0066] S1-1, generating spherical coordinates in the liver Converting the spherical coordinates into Cartesian coordinates x=(x, y, z), and converting the Cartesian coordinates and the spherical coordinates based on the elliptical axis and the half value;

[0067] S1-2, iteratively verify each potential tumor position x to determine whether the point is located within the liver until the tumor position x is determined t = (x t ,y t ,z t ) is stopped, and the tumor position of the liver is generated, wherein the tumor position is located inside the liver and is at a sufficient distance from the liver boundary.

[0068] Specifically, when performing step S1-1, spherical coordinates The spherical coordinates are converted into Cartesian coordinates x = (x, y, z), which will be further converted according to the axis and radius of the ellipsoid to ensure that the tumor position is uniformly distributed within the defined ellipsoid.

[0069] Further, in the process of generating the tumor position of the liver in step S1-2, the process includes:

[0070] S1-2-1, based on the number of liver masks in the neighborhood volume around the predicted position, the first tumor position is obtained based on the number of liver masks and the first threshold value;

[0071] S1-2-2, calculate the gradient value of the scan volume at the predicted position based on the gradient magnitude evaluation, and obtain the second tumor position based on the gradient value and the second threshold value, and generate the tumor position of the liver based on the first tumor position and the second tumor position.

[0072] Specifically, when performing step S1-2-1, the mathematical expression of the neighborhood volume is:

[0073]

[0074] where τ is the first threshold value, l(x i ) is the liver mask value, V(x) represents the ellipsoid-shaped field volume around the predicted position x, x i represents each point or pixel in the field volume V(x).

[0075] Further, when performing step S1-2-2, the mathematical expression of the gradient value is:

[0076]

[0077] where represents the gradient value, x, y, z represents the three-dimensional coordinate point.

[0078] Further, the gradient value is compared with the second threshold value k, when the gradient value exceeds the second threshold value, it indicates that the gradient value The corresponding point is close to the liver boundary.

[0079] Further, combining the neighborhood volume check and the gradient magnitude evaluation ensures that the selected tumor position is located within the liver tissue and far away from the boundary, expressed as:

[0080]

[0081] where ∧ denotes the logical AND operator, meaning that both conditions must be satisfied, x i represents each point or pixel in the field volume V(x).

[0082] In practice, is an iterative process that continues until the optimal tumor position x t = (x t , y t , z t ) is determined, which is located inside the liver and has sufficient distance from the liver boundary. t

[0083] In some embodiments, the process of obtaining the final tumor texture in step S2 includes:

[0084] S2-1, obtaining a three-dimensional simple noise based on predefined parameters, using nonlinear diffusion and wavelet transform to respectively perform artifact filtering and high-frequency detail preservation on the noise to obtain a to-be-processed texture;

[0085] S2-2, normalizing the to-be-processed texture to obtain a normalized texture, and performing smoothing processing on the normalized texture based on a first Gaussian filter to obtain a smoothed texture;

[0086] S2-3, scaling and moving the smoothed texture in proportion to obtain a first texture;

[0087] S2-4, obtaining a first binary mask based on random sampling, performing smoothing processing on the first binary mask based on a second Gaussian filter to obtain a second binary mask, and generating a final tumor texture after scaling and cropping the first texture based on the second binary mask.

[0088] Specifically, in the process of performing smoothing processing on the normalized texture in step S2-2 to obtain a smoothed texture, a Gaussian filter G(x; σ b ) with a standard deviation of σ b is used to perform smoothing processing on the normalized texture to obtain M'(x). The standard deviation of the first Gaussian filter G(x; σ1) is σ1.

[0089] ​Further, in the process of obtaining the first texture in step S2-3, the mathematical expression of the first texture is:

[0090] T(x) = a-(S"(x) * G(x; σ1)) + β;

[0091] where * represents convolution operation, a represents scaling ratio, β represents displacement, and T(x) represents the first texture. Further, in the process of generating the final tumor texture in step S2-4, the process includes:

[0092] T"(x) = T(x) * M'(x);

[0093] where T"(x) represents the final tumor texture.

[0094] Further, in the process of performing step S2-4, the standard deviation of the second Gaussian filter G(x; σ b ) is σ b .

[0095] In some embodiments, in the process of morphologically modeling the liver tumor in step S2, the process includes:

[0096] S2-5, dividing the synthetic tumor into four size categories: micro, small, medium, and large, and extracting the number of synthetic samples n from a discrete uniform distribution;

[0097] S2-6, simulating the tumor shape based on the parameter-defined ellipsoid, the semi-principal axes of the ellipsoid on the x, y, and z coordinates are taken from a uniform distribution U(0.75r, 1.25r) and centered at the coordinates x t = (x t , y t , z t ) within the scan volume domain f(x), and the initial tumor mask is defined within this range;

[0098] S2-7, performing a nonlinear elastic deformation on the initial tumor mask to obtain a deformed tumor mask:

[0099] S2-8, performing convolution processing on the deformed tumor mask based on a third Gaussian filter to obtain a synthetic mask.

[0100] Specifically, in the process of performing step S2-5, each tumor category has a specific radius r and a number of synthetic samples n extracted from a discrete uniform distribution U[a, b];

[0101] Specifically, in the process of performing step S2-7, the deformed tumor mask is closer to the anisotropic growth pattern observed in vivo, and its expression is:

[0102]

[0103] where t'(x) represents the deformed tumor mask, and t(x) represents the tumor mask before deformation.

[0104] Specifically, in the execution of step S2-8, the mathematical expression of the synthesized mask is:

[0105] t"(x) = t'(x) * G(x; σ2);

[0106] where t"(x) represents the synthesized mask, and G(x; σ2) represents a Gaussian filter with a deviation of σ2.

[0107] In some embodiments, in the execution of step S2, the process of synthesizing a new liver tumor CT image comprises:

[0108] S2-9, quantifying the density of the liver: dividing the liver into three density intervals, and standardizing the intensity values in each interval;

[0109] S2-10, adjusting the density of the tumor, reducing the density of high-density tissues;

[0110] S2-11, integrating the tumor into the scanning volume f(x) and the corresponding liver mask l(x).

[0111] In some embodiments, the specific process of quantifying the density of the liver in the execution of step S2-9 is as follows:

[0112] S2-9-1, obtaining the lower limit L and the upper limit U of the liver region P, and calculating the density scale Δ:

[0113] Δ = (U - L) / 3;

[0114] S2-9-2, for each three-dimensional coordinate point P in the liver region ijk the following operations are performed:

[0115] If P ijk <L + Δ, let P ijk = L; if L + Δ ≤ P ijk <L + 2Δ, let P ijk = L + Δ; if L + Δ ≤ P ijk <U, let P ijk = L + 2Δ; if P ijk ≥ U, let P ijk = U.

[0116] S2-9-3, obtaining the density distribution graph D according to the above operations, for each D ijk on the density distribution graph D, the following processing is performed: if D ijk = U and Pijk ≠ U, then let D ijk = L + 2Δ.

[0117] Specifically, the process of adjusting the tumor density in step S2-10 includes:

[0118] S2-10-1, obtain the lower limit L and the upper limit U of the liver region P, and calculate the density scale Δ:

[0119] Δ = (U - L) / 3;

[0120] S2-10-2, for each coordinate point F ijk in the scanning volume domain F, the following processing is performed:

[0121] When the coordinate point F ijk coincides with the generated tumor position, if D ijk = U, then let D = L + 2Δ. If D ijk = L, then let D = L + Δ. In other cases, keep F ijk unchanged.

[0122] Wherein, D ijk represents the coordinate point on the density distribution map D, U represents the upper limit of the liver region, and L represents the lower limit of the liver region.

[0123] In some embodiments, the specific steps of integrating the tumor into the scanning volume domain f(x) and the corresponding liver mask l(x) in step S2-11 are as follows:

[0124] S2-11-1, integrate the tumor with the scanning volume:

[0125] f'(x) = [1 - t"(x)] O f(x) + t"(x) O T"(x);

[0126] Wherein, O represents element multiplication, t"(x) represents the tumor mask array, and T"(x) represents the tumor texture array. The new scanning volume f'(x) can be obtained by the above transformation, and the new liver mask containing the tumor can be synthesized in the form of l'(x) = l(x) + t"(x).

[0127] In fact, the formation of the capsule is due to the fibrosis reaction caused by the chronic compression of the surrounding liver tissue. The characteristic of this capsule is the high echo edge around the tumor, representing a layer of dense connective tissue. The formation of the capsule can be simulated by amplifying the intensity of the tumor boundary, and the boundary region b(x) of the tumor is usually identified as:

[0128]

[0129] Wherein, λ la lower intensity threshold, λ, representing a definition of a tumor boundary region u an upper intensity threshold, λu, representing a definition of a tumor boundary region;

[0130] S2-11-2, using a Gaussian filter G(x; σ3) to convolve with b(x) to generate a smoothed boundary image b'(x):

[0131] b'(x) = b(x) * G(x; σ3);

[0132] where σ3determines the standard deviation of the Gaussian kernel, controlling the smoothness of the boundary transformation;

[0133] In fact, the boundary image b'(x) enhances the voxel intensity of the tumor periphery by adding a fixed intensity offset δ, which can simulate cysts increasing Hounsfield Units (HU):

[0134] f'''(x) = f"(x) + δ b'(x);

[0135] where δ represents the HU intensity difference between the tumor core and fibrous capsule, and the final volume scan f'''(x) reflects both the mass effect and the encapsulation formation.

[0136] In some embodiments, when performing step S3 to perform medical image segmentation on the synthesized tumor image as a data set, and verify the segmentation performance based on evaluation indicators, the image segmentation is trained using 118 CT scans with liver and liver tumor annotations, and the data set includes 281 three-dimensional abdominal CTs, each data set has a resolution of 512x512 pixels, the spiral data is reconstructed in the axial plane, the slice thickness is 1.25 mm, and the interval is 1.0 mm.

[0137] wherein the type of the data set can be divided into a liver tumor segmentation data set and a normal liver CT data set, the liver tumor segmentation data set contains detailed per-pixel annotations of liver tumors, and is part of the LiTS challenge, and the normal liver CT data set is a normal liver data set composed of 116 CT scans compiled from healthy subjects in the CHAOS, Pancreas-CT, BTCV and LiTS data sets, which is the baseline for generating the synthesized tumor image.

[0138] Specifically, the synthesized tumor image is crucial for creating diverse image label pairs. These synthetic tumors are systematically inserted into normal CT scans to increase the training data of the segmentation model.

[0139] Further, in the step S3, the evaluation index includes: Dice coefficient (DSC index), Hausdorff distance index (HD), normalized surface distance (NSD) and surface distance (SD) evaluation index.

[0140] Further, the DSC index is calculated, first, true positive (TP), true negative (TN), false positive (FP) and false negative (FN) four indexes of the model segmentation result should be calculated, the calculation method is as follows:

[0141]

[0142] Wherein, p i , q i ∈{0,1}, 0 represents negative voxel, 1 represents positive voxel, p i is the i-th voxel result predicted by the model, q i is the synthetic tumor label on the liver, n represents the number of all pixel points in the image, the calculation method of Dice is as follows:

[0143]

[0144] Further, the Hausdorff distance index (HD) calculation method is as follows:

[0145]

[0146] Wherein, h(A,B)=max a∈A {min b∈B ||a-b||}, h(B,A)=max b∈B {min a∈A ||b-a||}, a and b respectively represent the points in point set A and point set B, specifically, a is a point in point set A, b is a point in point set B.

[0147] Wherein, ||·|| is the distance norm between A and B point sets.

[0148] Further, the normalized surface distance (NSD) calculation method is as follows:

[0149]

[0150] Wherein, S ref represents the reference surface, S seg represents the surface point set of the segmentation result, d(x,S) represents the minimum distance from point x to point set S, |·| represents the number of points in the point set, τ is the set distance threshold, x represents the pixel point in S ref or S seg .

[0151] Further, the surface distance (SD) calculation method is as follows:

[0152]

[0153] wherein S ref represents the reference surface, S seg represents the surface point set of the segmentation result, d(x, S ref ) represents the minimum distance from the point x to the point set S ref .

[0154] Further, Table 1 is a comparison table of the segmentation results of the present application and other existing medical image segmentation methods.

[0155] Table 1 Comparison table of the segmentation results of the present application and other existing medical image segmentation methods

[0156]

[0157] The crossed cells in the table represent that the corresponding method does not use the evaluation index. It can be seen from the evaluation results that the method of the present application achieves better segmentation results in the four different evaluation indexes.

[0158] Although the embodiments of the present application have been described in detail above, it is obvious to those skilled in the art that various modifications and changes can be made to the embodiments. However, it should be understood that such modifications and changes all fall within the scope and spirit of the present application described in the claims. Moreover, the present application described herein can have other embodiments and can be implemented or realized in various ways.

Claims

1. An unsupervised liver tumor CT image segmentation method based on handcrafted features, characterized in that, include: We collected CT scan images of normal livers as a dataset and used volumetric and gradient validation to determine the location of potential liver tumors. The final tumor texture is obtained based on the location of the liver tumor, and morphological modeling of the liver tumor is performed based on the final tumor texture to synthesize a new liver tumor CT image, including: quantifying the density of the liver: dividing the liver into three density intervals and standardizing the intensity values ​​within each interval; adjusting the tumor density to reduce the density of high-density tissue; and integrating the tumor into the scanning volume domain. and the corresponding liver mask Including: Integrating tumor data with scan volume: ； in, This indicates element-wise multiplication. Indicates a tumor mask array. This represents a tumor texture array, and a new scan volume can be obtained through transformation. A new liver mask containing tumors can be pressed... Synthesis by method; Determine the boundary region of the tumor: ； in This represents the lower intensity threshold that defines the tumor boundary region. This represents the intensity threshold used to define the tumor boundary region; Use a Gaussian filter and Phase convolution generates a smooth boundary image. : ； in, The standard deviation of the Gaussian kernel is determined, which controls the smoothness of the boundary transition; Boundary map By increasing a fixed intensity offset To enhance the voxel intensity around the tumor, this offset can mimic the increase in Hounsfield units in a cyst: ； in This indicates the difference in Hounsfield unit intensity between the tumor core and the fibrous capsule. This indicates the final volumetric scan result; Annotated CT scan images were used to train the segmentation model, synthetic tumor images were used as a dataset for medical image segmentation, and the segmentation performance was validated based on evaluation metrics.

2. The unsupervised liver tumor CT image segmentation method based on handcrafted features according to claim 1, characterized in that, The process of determining the location of a potential liver tumor through volumetric and gradient validation includes: Generate spherical coordinates within the liver. Converting spherical coordinates to Cartesian coordinates And based on the elliptical axis and half-value, transform Cartesian coordinates and spherical coordinates; Each potential tumor location is iteratively validated to determine whether it is located within the liver, until the tumor location is definitively identified. The iteration stops when the time is right, and the location of the liver tumor is generated, wherein the tumor location is located inside the liver and at a sufficient distance from the liver boundary.

3. The unsupervised liver tumor CT image segmentation method based on handcrafted features according to claim 2, characterized in that, The process by which a tumor site forms in the liver includes: The number of liver masks in the neighborhood volume surrounding the predicted location is obtained based on the neighborhood volume check, and the first tumor location is obtained based on the number of liver masks and a first threshold. The gradient value of the scan volume at the predicted location is calculated based on the gradient magnitude assessment, and the second tumor location is obtained based on the gradient value and a second threshold. The tumor location of the liver is generated based on the first tumor location and the second tumor location.

4. The unsupervised liver tumor CT image segmentation method based on handcrafted features according to claim 1, characterized in that, The process of obtaining the final tumor texture includes: Based on predefined parameters, three-dimensional simplified noise is obtained, and nonlinear diffusion and wavelet transform are used to perform artifact filtering and high-frequency detail preservation on the noise to obtain the texture to be processed. The texture to be processed is normalized to obtain a normalized texture, and the normalized texture is smoothed based on a first Gaussian filter to obtain a smoothed texture. The smoothed texture is scaled and moved proportionally to obtain the first texture; The first binary mask is obtained by random sampling, the first binary mask is smoothed by a second Gaussian filter to obtain a second binary mask, and the first texture is scaled and cropped based on the second binary mask to generate the final tumor texture.

5. The unsupervised liver tumor CT image segmentation method based on handcrafted features according to claim 1, characterized in that, The process of morphological modeling of liver tumors includes: Synthetic tumors were categorized into four size classes: micro, small, medium, and large, and the number of synthetic samples was drawn from a discrete uniform distribution. The ellipsoid simulates tumor shape based on parameter-defined ellipsoids. , and The semi-principal axis on the coordinate system is taken from a uniform distribution. And with scanning volume domain Tumor location inside Define an initial tumor mask within this area, centered on the tumor. A deformed tumor mask is obtained by performing a nonlinear transformation on the initial tumor mask based on nonlinear elastic deformation: A synthetic mask is obtained by convolving the deformed tumor mask using a third Gaussian filter.

6. The unsupervised liver tumor CT image segmentation method based on handcrafted features according to claim 1, characterized in that, The process of quantifying liver density includes: Obtain liver region offline and upper limit Calculate density scale : ； For each three-dimensional coordinate point within the liver region Perform the following operations: if ,make ;if Then let ;if Then let ;if Then let ; The density distribution map is obtained based on the above operations. For density distribution map Every one of them Process: If and Then let .

7. The unsupervised liver tumor CT image segmentation method based on handcrafted features according to claim 1, characterized in that, The process of adjusting tumor density includes: Obtain liver region offline and upper limit Calculate density scale : ； For the scanning volume domain Each coordinate point in Perform the following processing: When coordinate point When the location coincides with the location where the tumor was formed, if Then let ,if Then let In other cases, retain ; in, Density distribution map coordinates on Indicates the upper limit of the liver region. This indicates the lower limit of the liver region.

8. The unsupervised liver tumor CT image segmentation method based on handcrafted features according to claim 1, characterized in that, When using synthesized tumor images as a dataset for medical image segmentation and validating the segmentation performance based on evaluation metrics, the image segmentation was trained using 118 CT scans with annotations for the liver and liver tumors. The dataset included 281 three-dimensional abdominal CT scans, each with a resolution of 512×512 pixels. Helical data was reconstructed in the axial plane, with a slice thickness of 1.25 mm and an interval of 1.0 mm. The dataset was categorized into liver tumor segmentation datasets and normal liver CT datasets. The liver tumor segmentation dataset contains detailed per-pixel annotations of liver tumors and is part of the LiTS Challenge; The normal liver CT dataset is a normal liver dataset consisting of 116 CT scans compiled from healthy subjects in the CHAOS, Pancreas-CT, BTCV, and LiTS datasets. This dataset serves as the baseline for generating synthetic tumor images.

Images