Vessel structure enhancement method and apparatus incorporating distribution and region constraints

Abstract

The method and device for introducing distribution and region constraints and enhancing blood vessel structure can obtain real information of liver blood vessels simply and quickly, assist doctors in preoperative planning, are favorable for expanding the use range of liver blood vessel data, and increase the flexibility of use. The method comprises the following steps: (1) obtaining three-dimensional image data, manually sketching the liver boundary to obtain a corresponding liver mask; (2) pre-processing the image based on liver labeling; (3) estimating the tissue distribution in the liver by using a Gaussian mixture model; (4) introducing distribution and region constraints, and calculating the liver blood vessel enhancement value under single-scale Gaussian filtering; and (5) calculating all scales, using maximum density projection on the image of each scale to obtain a liver blood vessel enhancement image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, in particular to a blood vessel structure enhancement method with distribution and region constraints and a blood vessel structure enhancement device with distribution and region constraints. BACKGROUND

[0002] Liver cancer is the fourth most common cancer and one of the highest mortality diseases in the world. The internal blood vessel system of the liver is complex, and in clinical practice, doctors need to interpret the relative position relationship between tumors and liver blood vessels from images based on experience. Therefore, it is very important to automatically and quickly obtain the true information of liver blood vessels. For liver data, the following points cannot be ignored:

[0003] 1. Due to the differences in actual situations such as acquisition equipment, modalities and patients, the data obtained are quite different, and the imaging quality is easily disturbed by noise.

[0004] 2. The liver is an irregular three-dimensional region and the gradient at the liver capsule is large.

[0005] 3. The liver blood vessel system is not completely wrapped by the liver, and it enters and exits the liver.

[0006] 4. Compared with CT imaging, MR is better for soft tissue imaging, and the intensity in the liver blood vessel is uneven.

[0007] 5. The contrast agent will diffuse into the liver tissue, and part of the liver tissue and tumor will also be enhanced.

[0008] However, the existing three-dimensional blood vessel enhancement method cannot cope with the above problems brought by liver data. SUMMARY

[0009] To overcome the defects of the prior art, the technical problem to be solved by the present application is to provide a blood vessel structure enhancement method with distribution and region constraints, which can quickly and simply obtain the true information of liver blood vessels to assist doctors in preoperative planning, and is beneficial to expand the use range of liver blood vessel data and increase the flexibility of use.

[0010] The technical scheme of the present application is that the blood vessel structure enhancement method with distribution and region constraints comprises the following steps:

[0011] (1) obtaining three-dimensional image data, manually outlining the liver boundary to obtain the corresponding liver mask;

[0012] (2) pre-processing the image based on liver annotation;

[0013] (3) estimating the tissue distribution in the liver using a Gaussian mixture model;

[0014] (4) using the estimated tissue distribution to guide the blood vessel structure enhancement;

[0015] (4) Introducing distribution and region constraints, calculating liver blood vessel enhancement under single scale Gaussian filtering

[0016] value;

[0017] (5) Calculating all scales, using maximum density projection of each scale image to obtain a liver blood vessel enhancement image.

[0018] The present application automatically estimates tissue distribution in the liver by a Gaussian mixture model, obtains large blood vessel tissue information, introduces distribution and region constraints, calculates liver blood vessel enhancement value under single scale Gaussian filtering, significantly enhances liver blood vessels while suppressing enhancement of non-blood vessel information such as liver capsule position and contrast agent diffusion liver parenchyma part, calculates all scales, uses maximum density projection of enhancement images under all scales to obtain a liver blood vessel enhancement image, and can ensure that both large blood vessels and small blood vessels can be enhanced. The present application takes into account both calculation efficiency and memory cost, can simply and quickly obtain real information of liver blood vessels to assist doctors in preoperative planning, is beneficial to expand the use range of liver blood vessel data, and increases flexibility of use.

[0019] Also provided is a blood vessel structure enhancement device introducing distribution and region constraints, which comprises:

[0020] An acquisition module acquires three-dimensional image data, and obtains a liver mask by manually outlining a liver boundary;

[0021] A preprocessing module pre-processes an image based on liver annotation;

[0022] An estimation module estimates tissue distribution in the liver by using a Gaussian mixture model;

[0023] A single scale calculation module introduces distribution and region constraints, and calculates liver blood vessel enhancement value under single scale Gaussian filtering;

[0024] An all scale calculation module calculates all scales, uses maximum density projection of each scale image to obtain a liver blood vessel enhancement image.

[0025] BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a flowchart of the blood vessel structure enhancement method introducing distribution and region constraints according to the present application.

[0027] Figure 2 is a whole flowchart when liver blood vessels are enhanced by the blood vessel structure enhancement method introducing distribution and region constraints according to the present application.

[0028] Figure 3 is a Gaussian mixture model structure for estimating tissue distribution in the liver according to the present application.

[0029] ​Figure 4 is a single-scale liver vessel enhancement method with distribution and region constraints according to the present application. DETAILED DESCRIPTION

[0030] As shown in the figure, the vessel structure enhancement method with distribution and region constraints comprises the following steps: Figure 1

[0031] (1) Obtain three-dimensional image data, manually draw the liver boundary to obtain the corresponding liver mask film;

[0032]

[0033] (2) Preprocess the image based on the liver annotation;

[0034] (3) Estimate the tissue distribution in the liver using a Gaussian mixture model;

[0035] (4) Introduce distribution and region constraints, calculate the liver vessel enhancement value under single-scale Gaussian filtering;

[0036]

[0037] (5) Calculate all scales, use maximum density projection on the scale images to obtain the liver vessel enhancement image.

[0038] The present application automatically estimates the tissue distribution in the liver using a Gaussian mixture model, obtains large vessel tissue information, introduces distribution and region constraints, calculates the liver vessel enhancement value under single-scale Gaussian filtering, significantly enhances the liver vessels while suppressing the enhancement of non-vessel information such as the liver capsule position and contrast agent diffusion in the liver parenchyma, calculates all scales, uses maximum density projection on the enhancement images under all scales to obtain the liver vessel enhancement image, and can ensure that both large vessels and small vessels can be enhanced. The present application takes into account the calculation efficiency and memory cost, can simply and quickly obtain the true information of the liver vessels to assist the doctor in preoperative planning, is beneficial to expand the use range of liver vessel data, and increases the flexibility of use.

[0039] Preferably, in the step (2), the image is intersected with the liver mask film to obtain a liver region image, and the liver region image is filled with a background outside the liver to obtain the preprocessed image.

[0040] Preferably, in the step (3), the number of classes c≤τ is adaptively selected based on the BIC criterion, τ is the maximum number of classes, the parameters of the Gaussian mixture model are estimated and initialized by k-means pre-classification, and the distribution information P of the liver region image F is obtained by iteration using the EM algorithm:

[0041]

[0042] where s represents a voxel in the image, δ j and μ​​​j Let w be the mean and variance. j Let be the probability that a voxel belongs to the j-th sub-Gaussian model.

[0043] Preferably, in step (4), the preprocessed medical image is convolved and filtered using a single-scale Gaussian function, the second-order derivative of the local space of the filtered image is calculated, and the Hessian matrix is ​​calculated:

[0044]

[0045] Where f represents the Gaussian function, and the eigenvalues ​​λ are obtained by decomposing the Hessian matrix. i,σ (s)(|λ 1,σ |≤|λ 2,σ |≤|λ 3,σ |,i=1,2,3), the hepatic vascular enhancement value under scale σ, which incorporates distribution information and region of interest constraints, is:

[0046]

[0047] in

[0048]

[0049] λ p,σ For λ 3,σ The corrected eigenvalue, γ is the curvature coefficient, MIP z This represents the maximum value of the intensity of all layer voxels calculated along the z-axis. This represents solving for the gradient, and Otsu represents the threshold obtained by the OTU automatic thresholding method.

[0050] Preferably, in step (5), the enhanced maps at all scales Δ={1,2,...,m} are calculated, where m is the maximum scale value. The maximum density projection is used to project the enhanced maps at each scale, and the formula is:

[0051] R(s) = max σ∈△ R σ (s) (4)

[0052] The final enhanced image of liver vessels is obtained by intersecting the output image with the liver mask.

[0053] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium. When executed, the program includes the steps of the methods of the above embodiments. The storage medium can be ROM / RAM, magnetic disk, optical disk, memory card, etc. Therefore, corresponding to the method of the present invention, the present invention also includes a vascular structure enhancement device that introduces distribution and regional constraints. This device is typically represented in the form of functional modules corresponding to the steps of the method. The device includes:

[0054] The acquisition module acquires 3D image data and manually delineates the liver boundary to obtain its corresponding liver mask.

[0055] The preprocessing module preprocesses the image based on liver annotations;

[0056] The estimation module uses a Gaussian mixture model to estimate the tissue distribution within the liver.

[0057] The single-scale calculation module introduces distribution and region constraints to calculate the hepatic vascular enhancement value under single-scale Gaussian filtering.

[0058] All scale calculation modules calculate all scales, and use maximum density projection of images at each scale to obtain enhanced images of liver vessels.

[0059] Preferably, in the preprocessing module, the intersection of the image and the liver mask is taken to obtain the intrahepatic region image, and the background outside the liver is filled into the intrahepatic region image to obtain the preprocessed image.

[0060] Preferably, in the estimation module, the number of classes c≤τ is adaptively selected based on the BIC criterion, where τ is the maximum number of classes. The parameters of the Gaussian mixture model are estimated and initialized by k-means pre-classification, and the distribution information P of the intrahepatic region image F is obtained iteratively using the EM algorithm.

[0061]

[0062] Where s represents a voxel in the image, δ j and μ j Let w be the mean and variance. j Let be the probability that a voxel belongs to the j-th sub-Gaussian model.

[0063] Preferably, in the single-scale calculation module, the preprocessed medical image is convolved and filtered using a single-scale Gaussian function, the second-order derivative of the local space of the filtered image is calculated, and the Hessian matrix is ​​calculated.

[0064]

[0065] Where f represents the Gaussian function, and the eigenvalues ​​λ are obtained by decomposing the Hessian matrix. i,σ (s)(|λ 1,σ |≤|λ 2,σ |≤|λ 3,σ |,i=1,2,3), the hepatic vascular enhancement value under scale σ, which incorporates distribution information and region of interest constraints, is:

[0066]

[0067] in

[0068]

[0069] λ p,σ For λ 3,σ The corrected eigenvalue, γ is the curvature coefficient, MIP z This represents the maximum value of the intensity of all layer voxels calculated along the z-axis. This represents solving for the gradient, and Otsu represents the threshold obtained by the OTU automatic thresholding method.

[0070] Preferably, in all the scale calculation modules, the augmented maps at all scales Δ={1,2,...,m} are calculated, where m is the maximum scale value. The maximum density projection is used to project the augmented maps at each scale, and the formula is:

[0071] R(s) = max σ∈△ R σ (s) (4)

[0072] The final enhanced image of liver vessels is obtained by intersecting the output image with the liver mask.

[0073] Due to the adoption of the above technical solutions, the present invention has the following advantages and positive effects compared with the prior art: The present invention significantly enhances hepatic blood vessels while inhibiting the enhancement of non-vascular information such as the location of the liver capsule and the diffusion of contrast agent in the liver parenchyma by introducing distribution information and region of interest constraints. In addition, the present invention takes into account both computational efficiency and memory cost, and makes it applicable to a wider range of medical scenarios.

[0074] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method of vessel structure enhancement incorporating distribution and region constraints, characterized by, The method comprises the following steps: (1) obtaining three-dimensional image data, manually drawing the boundary of the liver to obtain a corresponding liver mask; (2) pre-processing the image based on liver labeling; (3) estimating the distribution of tissues in the liver using a Gaussian mixture model; (4) introducing distribution and region constraints to calculate the liver vessel enhancement value under single-scale Gaussian filtering; (5) calculating all scales, using maximum density projection of the images of all scales to obtain a liver vessel enhancement image; In step (3), the number of classes c is adaptively selected based on the BIC criterion, and the maximum number of classes is τ. The parameters of the Gaussian mixture model are estimated and initialized by k-means pre-classification, and the distribution information P of the intrahepatic region image F is obtained by iteration using the EM algorithm: where s represents a voxel in the image, δ j and μ j are the mean and variance, w j is the probability that the voxel belongs to the jth sub-Gaussian model, w j ≥ 0, In step (4), the pre-processed medical image is convoluted and filtered using a single-scale Gaussian function, the local spatial second derivative of the filtered image is calculated, and the Hessian matrix is calculated: where f represents a Gaussian function, and λ is the eigenvalue obtained by decomposing the Hessian matrix i,σ (s), |λ 1,σ |≤|λ 2,σ |≤|λ 3,σ |, i = 1, 2, 3, and the liver vessel enhancement value under the scale σ is introduced with distribution information and region of interest constraints: In step (2), the intersection of the image and the liver mask is obtained to obtain an intrahepatic region image, and the background outside the liver is filled in the intrahepatic region image to obtain a pre-processed image. λ p,σ 3,σ corrected eigenvalue, γ is the curvature factor, MIP z represents the maximum value of all layer voxel intensities calculated along the z-axis direction, represents the gradient, and Otsu represents the threshold value obtained by the OSTU automatic threshold segmentation method.​ 2. The method of claim 1, wherein: In step (5), the enhancement map under all scales Δ={1, 2, …, m} is calculated, m is the maximum scale value, and maximum density projection of the enhancement maps of all scales is used, and the formula is:

3. The method of claim 2, wherein: The output image and the liver mask are intersected to obtain a final liver vessel enhancement image. R(s) = max σ∈△ R σ (s) (4) The method comprises the following steps:

4. A blood vessel structure enhancement apparatus that introduces distribution and region constraints, characterized by, An acquisition module is configured to obtain three-dimensional image data, manually draw the boundary of the liver to obtain a corresponding liver mask; A preprocessing module is configured to pre-process the image based on liver labeling; An estimation module is configured to estimate the distribution of tissues in the liver using a Gaussian mixture model; A single-scale calculation module is configured to introduce distribution and region constraints to calculate the liver vessel enhancement value under single-scale Gaussian filtering; An all-scale calculation module is configured to calculate all scales, using maximum density projection of the images of all scales to obtain a liver vessel enhancement image; In the estimation module, the number of classes c is adaptively selected based on the BIC criterion, and the maximum number of classes is τ. The parameters of the Gaussian mixture model are estimated and initialized by k-means pre-classification, and the distribution information P of the intrahepatic region image F is obtained by iteration using the EM algorithm: In the preprocessing module, the intersection of the image and the liver mask is obtained to obtain an intrahepatic region image, and the background outside the liver is filled in the intrahepatic region image to obtain a pre-processed image. where s represents a voxel in the image, δ j and μ j is the mean variance, w j is the probability that the voxel belongs to the jth sub-Gaussian model, w j ≥ 0, In the single scale calculation module, the preprocessed medical image is convolved and filtered by a single scale Gaussian function, the local spatial second derivative of the filtered image is calculated, and the Hessian matrix is calculated: where f represents a Gaussian function, and λ is the eigenvalue obtained by decomposing the Hessian matrix i,σ (s), |λ 1,σ |≤|λ 2,σ |≤|λ 3,σ |, i = 1, 2, 3, and the liver vessel enhancement value under the scale σ is introduced with distribution information and region of interest constraints: In the all-scale calculation module, the enhancement map under all scales Δ={1, 2, …, m} is calculated, m is the maximum scale value, and maximum density projection of the enhancement maps of all scales is used, and the formula is: λ p,σ For λ 3,σ The corrected eigenvalue, γ is the curvature coefficient, MIP z This represents the maximum value of the intensity of all layer voxels calculated along the z-axis. The output image and the liver mask are intersected to obtain a final liver vessel enhancement image. The method comprises the following steps: characterized in that An acquisition module is configured to obtain three-dimensional image data, manually draw the boundary of the liver to obtain a corresponding liver mask; A preprocessing module is configured to pre-process the image based on liver labeling; characterized in that An estimation module is configured to estimate the distribution of tissues in the liver using a Gaussian mixture model; A single-scale calculation module is configured to introduce distribution and region constraints to calculate the liver vessel enhancement value under single-scale Gaussian filtering; R(s) = max σ∈△ R σ (s)(4) An all-scale calculation module is configured to calculate all scales, using maximum density projection of the images of all scales to obtain a liver vessel enhancement image; In the estimation module, the number of classes c is adaptively selected based on the BIC criterion, and the maximum number of classes is τ. The parameters of the Gaussian mixture model are estimated and initialized by k-means pre-classification, and the distribution information P of the intrahepatic region image F is obtained by iteration using the EM algorithm: In the preprocessing module, the intersection of the image and the liver mask is obtained to obtain an intrahepatic region image, and the background outside the liver is filled in the intrahepatic region image to obtain a pre-processed image. In the all-scale calculation module, the enhancement map under all scales Δ={1, 2, …, m} is calculated, m is the maximum scale value, and maximum density projection of the enhancement maps of all scales is used, and the formula is: The output image and the liver mask are intersected to obtain a final liver vessel enhancement image.

Images