WO2022063198A1

WO2022063198A1 - Lung image processing method, apparatus and device

Info

Publication number: WO2022063198A1
Application number: PCT/CN2021/120099
Authority: WO
Inventors: 黄钢; 聂生东; 段辉宏
Original assignee: 上海健康医学院; 上海理工大学
Priority date: 2020-09-24
Filing date: 2021-09-24
Publication date: 2022-03-31
Also published as: CN112184659A; CN112184659B

Abstract

The present invention relates to a lung image processing method, apparatus and device, the method comprising: acquiring a lung CT sequence diagram to be processed; segmenting to acquire a binary lung parenchyma mask; carrying out dot product processing on the binary lung parenchyma mask and the lung CT sequence diagram, so as to acquire a lung parenchyma region image, and enhancing a potential vascular region therein to obtain a vascular region of the lung; segmenting a candidate pulmonary nodule region from the lung CT sequence diagram; and determining whether the lung vascular region and the candidate lung nodule region have an intersection, and if so, then segmenting the intersection part of the lung vascular region and the candidate lung nodule region, performing three-dimensional reconstruction display on the segmented lung vascular region and candidate lung nodule region, so as to render the lung vascular region and the candidate lung nodule region by using different colors. Compared with the existing technology, the present invention may solve the problems in which determining the relationship between a lung nodule and surrounding blood vessels in a two-dimensional CT image is highly difficult, misjudgments are likely to occur, and so on.

Description

Lung image processing method, device and equipment

technical field

The present invention relates to the field of medical image processing, in particular to a lung image processing method, device and equipment.

Background technique

Lung cancer is the most common malignant tumor and the most lethal tumor. According to GLOBOCAN (Global Cancer Observatory) statistics, the number of new lung cancer cases worldwide in 2018 was 2.1 million, accounting for 11.6%; the number of deaths was 1.8 million, accounting for 18.4% of all tumor deaths. The proportion of new cases and fatal cases ranks first in the world. According to the latest statistics from the China Cancer Center in 2018, the incidence of lung cancer ranks first in my country, with an annual incidence of about 781,000, accounting for 20.55% of all tumors; about 626,000 lung cancer deaths each year, with a mortality rate of 48.5%. About 70% of lung cancer cases are diagnosed when the tumor metastasizes or develops into an advanced stage, which makes the five-year survival rate of lung cancer in my country as low as 16.1%, ranking the third from the bottom of all malignant tumors. The imaging manifestations of early lung cancer are pulmonary nodules, but because the imaging characteristics of early pulmonary nodules are not obvious, it is often easy to cause missed diagnosis or misdiagnosis.

For traditional pulmonary nodule screening, based on clinical experience, radiologists abstract the characteristics of the lesions from medical images by manually reviewing CT images and observing with the naked eye, and analyze, diagnose, and classify the nodules as benign and malignant. Whether there is vascular invasion and the positional relationship between peripheral blood vessels and pulmonary nodules are important references for radiologists to judge whether pulmonary nodules are benign or malignant. The pulmonary vascular tree has a complex structure with numerous branches. At the same time, in CT images used for routine clinical detection and diagnosis of pulmonary diseases, the gray values of blood vessels and pulmonary nodules are similar, and even the two-dimensional slice structure of some blood vessels is similar to that of nodules. Difficulty in diagnosis. At the same time, the image presented by the CT image is a two-dimensional slice image, even though the lesions can be observed from the transverse, sagittal and coronal planes. However, in most cases, the blood vessels and nodules will present the illusion of adhesion in the two-dimensional cutting, while in the actual three-dimensional space, the above-mentioned nodules and blood vessels do not have adhesions, and the blood vessels just approach or pass around the nodules. Due to the limitation of the imaging principle of CT images, the above-mentioned blood vessels appear to be adhered to the nodules or pass through the blood vessels, resulting in misjudgment by the radiologist.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a lung image processing method, device and equipment in order to overcome the above-mentioned defects of the prior art, which can provide a reliable reference for radiologists, improve the accuracy of diagnosis, and reduce the limitation of two-dimensional image information. resulting in misdiagnosis.

The object of the present invention can be realized through the following technical solutions:

A lung image processing method, the method comprises the following steps:

Obtain the CT sequence map of the lung to be processed;

segmenting the lung CT sequence map to obtain a binary lung parenchyma mask;

Perform dot product processing on the binary lung parenchyma mask and the lung CT sequence map to obtain an image of the lung parenchyma area, and perform enhancement processing on the potential blood vessel area therein to obtain the lung blood vessel area;

Using the first 3D U-net network to segment candidate lung nodule regions from the lung CT sequence map;

Determine whether there is an intersection between the pulmonary blood vessel area and the candidate pulmonary nodule area, and if so, use the second 3D U-net network to segment the intersection of the pulmonary blood vessel area and the candidate pulmonary nodule area. 3D reconstruction and display of the pulmonary blood vessel area and candidate pulmonary nodule area, if not, directly perform 3D reconstruction and display of the pulmonary blood vessel area and candidate pulmonary nodule area. The pulmonary blood vessel area and the candidate lung nodule area are rendered, so as to clearly display the positional relationship between the pulmonary blood vessel and the nodule.

Further, performing enhancement processing on the blood vessel region specifically includes:

Gaussian filtering is performed on the image of the lung parenchyma area with Gaussian filtering kernels of different scales to obtain a multi-scale image set;

The images of each scale in the multi-scale image set are respectively enhanced and filtered by using the Krissian function;

Taking the pixel as the calculation unit, obtain the maximum value of the Krissian filtering result of each pixel in the lung parenchyma image at all scales;

Perform skeletonization processing on the maximum value of the Krissian filtering result to obtain a preliminary three-dimensional skeleton of the vascular tree;

Binarization is performed on the preliminary three-dimensional skeleton, thereby obtaining an enhanced pulmonary blood vessel region.

Further, the formula of the Krissian function is:

Among them, R(x, σ, θ) represents the Krissian response value of the pixel point x at the angle θ in the multi-scale image fσ(x, y) with the scale σ; (x+θσυ _a ) represents the pixel point x’s Krissian response value; An edge point, θ represents the angle value of the currently detected edge point, and θ=θ+da, υ _a represents the rotation vector: υ _a =cosav ₃ +cosav ₂ , where v ₃ and v ₂ represent the Hessian matrix feature of the pixel x value.

Further, performing binarization processing on the preliminary three-dimensional skeleton specifically includes:

Take the minimum skeleton unit as (x _i-1 , x _i , x _i+1 ), where x _i-1 , x _i , x _i+1 are the three adjacent voxel points on the preliminary three-dimensional skeleton, take respectively The set of edge points whose three points obtain the maximum response value under the Krissian function

Obtain the connection line between each skeleton point in each minimum skeleton unit and each edge point in the corresponding edge point set, obtain the average gray value on the connection line, and use the average gray value as the local optimum of the corresponding skeleton point threshold;

Thresholding is performed on the minimum unit skeleton and the points on its connecting line based on the local optimal threshold, so as to complete the binarization process.

Further, segmenting the candidate lung nodules from the lung CT sequence diagram specifically includes:

Obtain the collection of pulmonary nodules center points to be observed;

Perform threshold segmentation on the area centered on the center point of each lung nodule in turn to obtain the preliminary lung nodule area;

trimming the preliminary lung nodule region to obtain a set of nodule image blocks;

A three-dimensional U-Net network is used to segment each nodule sub-image block in the nodule image block set, so as to obtain a refined candidate lung nodule region.

The present invention also provides a lung image processing device, comprising:

The CT image acquisition module is used to acquire the CT sequence map of the lung to be processed;

a first segmentation module, configured to segment the lung CT sequence map to obtain a binary lung parenchyma mask;

an enhancement module, configured to perform dot product processing on the binary lung parenchyma mask and the lung CT sequence map to obtain an image of the lung parenchyma area, and perform enhancement processing on the potential blood vessel area therein to obtain the lung blood vessel area;

The second segmentation module is used to segment the candidate lung nodule region from the lung CT sequence map by using the first 3D U-net network;

The third segmentation module is used for judging whether there is an intersection between the pulmonary blood vessel area and the candidate pulmonary nodule area, and when the judgment result is yes, use the second 3D U-net network to analyze the pulmonary blood vessel area and the candidate pulmonary nodule area. The intersection part of the section area is divided;

The display module performs three-dimensional reconstruction and display on the obtained pulmonary blood vessel area and the candidate lung nodule area, and renders the pulmonary blood vessel area and the candidate lung nodule area in different colors.

Further, the enhancement module includes:

a Gaussian filtering unit, configured to perform Gaussian filtering on the image of the lung parenchyma region with Gaussian filtering kernels of different scales to obtain a multi-scale image set;

an enhancement filtering unit, configured to perform enhancement filtering on the images of each scale in the multi-scale image set respectively by using the Krissian function;

The maximum filtering result calculation unit is used to obtain the maximum value of the Krissian filtering result of each pixel in the image of the lung parenchyma area under all scales by taking the pixel as the calculation unit;

a skeletonization processing unit, configured to perform skeletonization processing on the maximum value of the Krissian filtering result to obtain a preliminary three-dimensional skeleton of the blood vessel tree;

The binarization processing unit performs binarization processing on the preliminary three-dimensional skeleton, so as to obtain the enhanced pulmonary blood vessel area.

Further, the second segmentation module includes:

The preliminary threshold segmentation unit is used to obtain the set of pulmonary nodules center points to be observed, and sequentially performs threshold segmentation on the area centered on the center point of each pulmonary nodule to obtain the preliminary pulmonary nodule area;

a trimming unit, configured to trim the preliminary lung nodule region to obtain a set of nodule image blocks;

The fine segmentation unit is used for segmenting each nodule sub-image block in the nodule image block set by using a three-dimensional U-Net network, so as to obtain a fine candidate lung nodule region.

The present invention also provides a computer device, comprising:

processor;

a memory that stores processor-executable instructions;

Wherein, the processor is coupled to the memory for reading program instructions stored in the memory, and in response, executing the steps in the method as described above.

Compared with the prior art, the present invention has the following beneficial effects:

The invention separately performs segmentation processing on the pulmonary blood vessel area and pulmonary nodule area in the lung CT image, and performs fine segmentation on the intersection part, so that the three-dimensional display of the positional relationship between the pulmonary blood vessel and the nodule can be intuitively and accurately performed, and is suitable for the imaging department. Doctors can provide reliable reference, improve the accuracy of diagnosis, reduce misdiagnosis caused by the limitation of 2D image information, and solve the problem of difficulty in judging the relationship between pulmonary nodules and surrounding blood vessels on 2D CT images and easy to cause misjudgment.

Description of drawings

Fig. 1 is the schematic flow chart of the present invention;

2 is a schematic diagram of a 3D U-net network structure used in an embodiment of the present invention;

FIG. 3 is a schematic diagram of the segmentation result obtained by the present invention.

detailed description

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and provides a detailed implementation manner and a specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

Example 1

As shown in FIG. 1, this embodiment provides a lung image processing method, the method includes:

Step 1, obtain the lung CT sequence diagram I to be processed. The lung CT sequence map is in DICOM format.

Step 2: Segment the lung CT sequence map to obtain a binary lung parenchyma mask mask_lung.

In this embodiment, the method in the document "Automatic Lung Segmentation for Accurate Quantitation of Volumetric X-Ray CT Images" is used to perform the segmentation of this step. This method uses the optimal threshold method to automatically select the segmentation threshold that characterizes the grayscale characteristics of each subject's lung parenchyma image. specificity; and excellent performance in lung volume changes.

Step 3: Perform dot product processing on the binary lung parenchyma mask and the lung CT sequence map to obtain an image of the lung parenchyma area, and perform enhancement processing on the potential blood vessel area therein to obtain the lung blood vessel area.

In this embodiment, the enhancement method in the document "Model-Based Detection of Tubular Structures in 3D Images" is used to enhance the potential blood vessel area. This method takes the blood vessel scale as a reference and combines the advantages of two enhancement filter functions, which can simultaneously ensure good sensitivity and specificity for blood vessels of various sizes and contrasts.

Step 3-1: Do point multiplication of the binary lung parenchyma mask mask_lung obtained in step 2 with the original sequence image I, thereby obtaining the lung parenchyma region image lung_region.

Step 3-2: Set up the multi-scale analysis kernel σ set:

σ _i = {σ ₁ , σ ₂ ... σ _i }, with i = 1, 2 ... 10

Among them, i represents the multi-scale analysis kernel size, and σ ₁ represents the Gaussian filter kernel with a radius of 1 pixel.

The multi-scale analysis kernel σ can better preserve the high-frequency information of the image and reduce the blur of blood vessel edges in the subsequent multi-scale image analysis.

A Gaussian filter kernel with a filter kernel size of σ _i is respectively set, and Gaussian filtering is performed on the lung_region image of the lung parenchyma region, that is, multi-scale analysis, so as to obtain a multi-scale image set:

f _σ (x, y) = {f ₁ (x, y), f ₂ (x, y)...f ₁₀ (x, y)}

Wherein, f ₁ (x, y) indicates that the Gaussian filter kernel is σ ₁ , that is, the Gaussian filter kernel with a radius of 1 pixel, and the obtained filtering result.

Step 3-3: Use the Krissian function to perform enhancement filtering on the images of each scale in the multi-scale image set. The formula of the Krissian function is:

Among them, R(x, σ, θ) represents the Krissian response value of the pixel point x at the angle θ in the multi-scale image f _σ (x, y) with the scale σ; (x+θσυ _a ) represents the pixel point x An edge point (boundary point) of , which is obtained by adding the x-coordinate of the pixel to be detected and the detection radius θσυ _a , θ represents the angle value of the currently detected edge point, and θ=θ+da, υ _a represents the rotation vector: υ _a =cosav ₃ +cosav ₂ , where v ₃ and v ₂ represent the eigenvalues of the Hessian matrix of the pixel point x, and changing the value of a can obtain a series of edge point sets P _a . The above Krissian function utilizes the boundary information of the blood vessel, and further calculates the boundary degree of the voxel points on the basis of the Hessian matrix blood vessel enhancement filter, which can detect more small tubular structures.

The above Krissian function filters the images in the multi-scale image collection respectively, and the obtained results are:

K _σ (x, y) = {K ₁ (x, y), K ₂ (x, y)...K ₁₀ (x, y)}

Wherein, K ₁ (x, y) represents the Krissian function filtering result obtained from an image whose scale is 1.

After that, take the pixel as the calculation unit to obtain the maximum value of the Krissian filtering result of each pixel in the lung_region at all scales. The calculation formula is as follows:

K _max (x,y)=max( _σKσ (x,y)),withσ=1,2,...,10

Among them, K _max (x, y) represents the final response value of the pixel point x whose coordinate value is (x, y). K _max represents the final filtering result of the lung_region image, and the pixel value of the pixel in K _max represents the possibility that the point belongs to the blood vessel region. The larger the value, the greater the possibility.

Step 3-4: skeletonize the K _max result to obtain a preliminary three-dimensional skeleton of the vascular tree.

In this embodiment, the method in the document "Building Skeleton Models via 3-D Medial Surface Axis Thinning Algorithms" is used for skeletonization, and the method can fully guarantee the integrity of the segmented trachea.

Step 3-5: Take the minimum skeleton unit as (x _i-1 , _xi , _xi+1 ), where _xi-1 , _xi , _xi+1 are three 26 adjacent voxel points on the three-dimensional skeleton . Take three points respectively and obtain the edge point set with the maximum response value under the Krissian function

Then connect the minimum skeleton units x _i-1 , x _i , x _i+1 with the corresponding edge point sets respectively

Thus, all the connecting lines between each minimum unit skeleton point and the corresponding edge set point are obtained, and the average gray value on the connecting line is obtained to obtain the local optimal value of the skeleton point. The calculation formula is as follows:

Among them, the L operation represents, respectively connecting _xi and the set of boundary points

And calculate the average value of the response value on the connection line and the gray value of the original image.

After obtaining the local optimal threshold Th _part , thresholding is performed on the minimum unit skeleton and the points on its connecting line to obtain a binary image.

Step 3-6: Perform the operations of Step 3-5 on all the smallest skeleton units in the preliminary three-dimensional skeleton, so as to obtain the final pulmonary blood vessel area.

Step 4. Use the first 3D U-net network to segment candidate lung nodule regions from the input original CT sequence image. Among them, the U-Net architecture shows very good performance in different medical image segmentation applications, and the 3D U-net network takes advantage of the three-dimensional connectivity of the nodule tissue to be segmented. Encoding is performed on the pulmonary nodule to ensure the continuity of changes between the interstitial lung nodule tissues, and to better select candidate nodule regions.

Step 4-1: In an interactive way, click to select the central area of the lung nodule to be observed, and obtain the combination of the central point of the lung nodule: n _i = {n ₁ , n ₂ ,..., n _n }, with i =1, 2, . . . , n, where _ni represents the three-dimensional space coordinate point of the ith nodule, ie, _ni =( _xi , _yi , _zi ).

Step 4-2: Perform threshold segmentation on the area centered on n _i in turn to obtain a preliminary lung nodule area.

Step 4-3: Trim the acquired lung nodule area, and trim an image block with a size of 50*50*Sz, where Sz represents the number of slices of the nodule in the direction of the Sz axis. After pruning, a node _i = {node ₁ , node ₂ , ..., node _i }, with i = 1, 2, ..., n, where node _i represents the i-th node child, is obtained. image block.

Step 4-4: Substitute the acquired nodule sub-image blocks into the first 3D U-net network (3D-UNET-1 network) for segmentation, so as to obtain fine nodule regions.

Step 5: Determine whether there is an intersection between the pulmonary blood vessel area and the candidate pulmonary nodule area, so as to avoid misdiagnosing the pulmonary blood vessel as a pulmonary nodule and increase the robustness. If yes, then use the second 3D U-net network (3D-UNET-2 network) to segment the intersection of the pulmonary blood vessel area and the candidate pulmonary nodule area, and perform step 6, if not, directly perform step 6 .

The above-mentioned 3D-UNET-1 and 3D-UNET-2 have the same network structure, but the data used to train the network are different. The 3D-UNET-2 network is more focused on the distinction between blood vessels and nodular areas, and the 3D-UNET-1 network is more focused on the distinction between pulmonary nodules and non-nodular areas.

Step 6. Perform 3D reconstruction and display on the segmented pulmonary blood vessel area and candidate pulmonary nodule area. When performing the 3D reconstruction and display, the pulmonary blood vessel area and the candidate pulmonary nodule area are rendered in different colors, so as to achieve a clear realization. Display of the relationship between the location of the pulmonary vessels and the nodule.

Example 2

This embodiment provides a lung image processing device, including: a CT image acquisition module, configured to acquire a CT sequence diagram of the lung to be processed; a first segmentation module, configured to segment the CT sequence diagram of the lung to acquire A binary lung parenchyma mask; an enhancement module is used to perform dot product processing on the binary lung parenchyma mask and the lung CT sequence map to obtain an image of the lung parenchyma area, and perform enhancement processing on the potential blood vessel area therein to obtain an image of the lung parenchyma area. Pulmonary blood vessel region; a second segmentation module for segmenting candidate lung nodule regions from the lung CT sequence map by using the first 3D U-net network; and a third segmentation module for judging the pulmonary blood vessels Whether there is an intersection between the area and the candidate lung nodule area, when the judgment result is yes, use the second 3D U-net network to segment the intersection part of the pulmonary blood vessel area and the candidate lung nodule area; display module, to obtain The pulmonary vascular region and candidate pulmonary nodule region are displayed by three-dimensional reconstruction, and the pulmonary vascular region and candidate pulmonary nodule region are rendered in different colors. The rest are the same as in Example 1.

Example 3

This embodiment provides a computer device, including a processor and a memory that stores processor-executable instructions, wherein the processor is coupled to the memory, and is configured to read program instructions stored in the memory, and in response, Perform the steps in the method described in Example 1.

The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make many modifications and changes according to the concept of the present invention without creative efforts. Therefore, any technical solutions that can be obtained by those skilled in the art through logical analysis, reasoning or limited experiments on the basis of the prior art according to the concept of the present invention shall fall within the protection scope determined by the claims.

Claims

A lung image processing method, characterized in that the method comprises the following steps:

Obtain the CT sequence map of the lung to be processed;

segmenting the lung CT sequence map to obtain a binary lung parenchyma mask;

Perform dot product processing on the binary lung parenchyma mask and the lung CT sequence map to obtain an image of the lung parenchyma area, and perform enhancement processing on the potential blood vessel area therein to obtain the lung blood vessel area;

Using the first 3D U-net network to segment candidate lung nodule regions from the lung CT sequence map;

Determine whether there is an intersection between the pulmonary blood vessel area and the candidate pulmonary nodule area, and if so, use the second 3D U-net network to segment the intersection of the pulmonary blood vessel area and the candidate pulmonary nodule area. 3D reconstruction and display of the pulmonary blood vessel area and candidate pulmonary nodule area, if not, directly perform 3D reconstruction and display of the pulmonary blood vessel area and candidate pulmonary nodule area. The lung vessel regions and candidate lung nodule regions are rendered.
The lung image processing method according to claim 1, wherein the enhancing processing on the blood vessel region specifically comprises:

Gaussian filtering is performed on the image of the lung parenchyma area with Gaussian filtering kernels of different scales to obtain a multi-scale image set;

The images of each scale in the multi-scale image set are respectively enhanced and filtered by using the Krissian function;

Taking the pixel as the calculation unit, obtain the maximum value of the Krissian filtering result of each pixel in the lung parenchyma image at all scales;

Perform skeletonization processing on the maximum value of the Krissian filtering result to obtain a preliminary three-dimensional skeleton of the vascular tree;

Binarization is performed on the preliminary three-dimensional skeleton, thereby obtaining an enhanced pulmonary blood vessel region.
The lung image processing method according to claim 2, wherein the formula of the Krissian function is:

Among them, R(x, σ, θ) represents the Krissian response value of the pixel point x at the angle θ in the multi-scale image f σ (x, y) with the scale σ; (x+θσυ a ) represents the pixel point x An edge point of , θ represents the angle value of the currently detected edge point, and θ=θ+da, υ a represents the rotation vector: υ a =cosav 3 +cosav 2 , where v 3 and v 2 represent the Hessian matrix of the pixel point x Eigenvalues.
The lung image processing method according to claim 2, wherein the binarization processing on the preliminary three-dimensional skeleton specifically comprises:

Take the minimum skeleton unit as (x i-1 , x i , x i+1 ), where x i-1 , x i , x i+1 are the three adjacent voxel points on the preliminary three-dimensional skeleton, take respectively The set of edge points whose three points obtain the maximum response value under the Krissian function

Obtain the connection line between each skeleton point in each minimum skeleton unit and each edge point in the corresponding edge point set, obtain the average gray value on the connection line, and use the average gray value as the local optimum of the corresponding skeleton point threshold;

Thresholding is performed on the minimum unit skeleton and the points on its connecting line based on the local optimal threshold, so as to complete the binarization process.
The lung image processing method according to claim 1, wherein segmenting the candidate lung nodules from the lung CT sequence diagram specifically comprises:

Obtain the collection of pulmonary nodules center points to be observed;

Perform threshold segmentation on the area centered on the center point of each lung nodule in turn to obtain the preliminary lung nodule area;

trimming the preliminary lung nodule region to obtain a set of nodule image blocks;

A three-dimensional U-Net network is used to segment each nodule sub-image block in the nodule image block set, so as to obtain a refined candidate lung nodule region.
A lung image processing device, comprising:

The CT image acquisition module is used to acquire the CT sequence map of the lung to be processed;

a first segmentation module, configured to segment the lung CT sequence map to obtain a binary lung parenchyma mask;

an enhancement module, configured to perform dot product processing on the binary lung parenchyma mask and the lung CT sequence map to obtain an image of the lung parenchyma area, and perform enhancement processing on the potential blood vessel area therein to obtain the lung blood vessel area;

The second segmentation module is used to segment the candidate lung nodule region from the lung CT sequence map by using the first 3D U-net network;

The third segmentation module is used for judging whether there is an intersection between the pulmonary blood vessel area and the candidate pulmonary nodule area, and when the judgment result is yes, use the second 3D U-net network to analyze the pulmonary blood vessel area and the candidate pulmonary nodule area. The intersection part of the section area is divided;

The display module performs three-dimensional reconstruction and display on the obtained pulmonary blood vessel area and the candidate lung nodule area, and renders the pulmonary blood vessel area and the candidate lung nodule area in different colors.
The lung image processing apparatus according to claim 6, wherein the enhancement module comprises:

a Gaussian filtering unit, configured to perform Gaussian filtering on the image of the lung parenchyma region with Gaussian filtering kernels of different scales to obtain a multi-scale image set;

an enhancement filtering unit, configured to perform enhancement filtering on the images of each scale in the multi-scale image set respectively by using the Krissian function;

The maximum filtering result calculation unit is used to obtain the maximum value of the Krissian filtering result of each pixel in the image of the lung parenchyma area under all scales by taking the pixel as the calculation unit;

a skeletonization processing unit, configured to perform skeletonization processing on the maximum value of the Krissian filtering result to obtain a preliminary three-dimensional skeleton of the blood vessel tree;

The binarization processing unit performs binarization processing on the preliminary three-dimensional skeleton, so as to obtain the enhanced pulmonary blood vessel area.
The lung image processing device according to claim 7, wherein the binarization processing on the preliminary three-dimensional skeleton specifically comprises:

Take the minimum skeleton unit as (x i-1 , x i , x i+1 ), where x i-1 , x i , x i+1 are the three adjacent voxel points on the preliminary three-dimensional skeleton, take respectively The set of edge points whose three points obtain the maximum response value under the Krissian function

Obtain the connection line between each skeleton point in each minimum skeleton unit and each edge point in the corresponding edge point set, obtain the average gray value on the connection line, and use the average gray value as the local optimum of the corresponding skeleton point threshold;

Thresholding is performed on the minimum unit skeleton and the points on its connecting line based on the local optimal threshold, so as to complete the binarization process.
The lung image processing apparatus according to claim 6, wherein the second segmentation module comprises:

The preliminary threshold segmentation unit is used to obtain the set of pulmonary nodules center points to be observed, and sequentially performs threshold segmentation on the area centered on the center point of each pulmonary nodule to obtain the preliminary pulmonary nodule area;

a trimming unit, configured to trim the preliminary lung nodule region to obtain a set of nodule image blocks;

The fine segmentation unit is used for segmenting each nodule sub-image block in the nodule image block set by using a three-dimensional U-Net network, so as to obtain a fine candidate lung nodule region.
A computer equipment, characterized in that, comprising:

processor;

a memory that stores processor-executable instructions;

Wherein, the processor is coupled to the memory for reading program instructions stored in the memory, and in response, performing the steps in the method according to any one of claims 1-5.