CN113689442A

CN113689442A - Method for realizing lung organ segmentation based on three-dimensional image

Info

Publication number: CN113689442A
Application number: CN202111095898.1A
Authority: CN
Inventors: 赵军; 武影影; 刘靖; 王传付; 关敏杰
Original assignee: Jingwei Beijing Technology Co ltd
Current assignee: Jingwei Beijing Technology Co ltd
Priority date: 2021-09-17
Filing date: 2021-09-17
Publication date: 2021-11-23

Abstract

The invention discloses a method for realizing lung organ segmentation based on a three-dimensional image, which relates to four modules; 1) a pulmonary parenchyma coarse segmentation module; 2) a trachea segmentation module; 3) a lung internal blood vessel rough segmentation module; 4) lung abnormality segmentation is achieved; the invention combines the traditional computer vision image algorithm and the U-net neural network to realize the automatic 3D model reconstruction of lung organs, effectively reduces the workload of doctors for manufacturing the 3D model and precious time, completes the 3D model reconstruction of each part subdivided into lung parenchyma, trachea, blood vessels inside the lung and abnormal parts inside the lung, has complete segmentation, and is more beneficial to 3D display, thereby being beneficial to improving doctor-patient communication and improving communication efficiency.

Description

Method for realizing lung organ segmentation based on three-dimensional image

Technical Field

The invention relates to the technical field of medical image processing, in particular to a method for realizing lung organ segmentation by combining a three-dimensional image with a traditional computer vision image algorithm and a U-net neural network.

Background

With social development and technological progress, more channels are available for people to acquire medical information, so that doctor-patient relationship becomes delicate and complicated. Medical technical problems are complex and difficult for a patient to understand, but obvious and easy for a physician to understand. When doctors and patients communicate with each other before surgery, the diagnosis of diseases, abnormal size, position and surrounding tissue organ relationship can be objectively and accurately known, the position and spatial relationship of tumors can be known as detailed as possible before surgery, the establishment of the optimal surgical scheme and more effective doctor-patient communication can be facilitated, the surgical risk can be reduced, and the success rate of surgery can be improved. In the prior art, doctors generate 3D model files by mainly using a standard model of human organs or manually and interactively segmenting by using software such as Mimics, 3D slicers and the like which runs on a PC desktop. Meanwhile, the deep learning provides an achievable means for extracting characteristic parts in the field of medical image processing, and the lung abnormity diagnosis based on the deep learning artificial intelligence is applied to the realization of segmentation of abnormal parts of the lung, so that the abnormal diagnosis problem is basically solved, but the spatial position relation between the abnormal parts and the surrounding trachea and blood vessels cannot be visually displayed.

In summary, some existing doctor-patient communication methods rely on standard human organ models, and some existing doctor-patient communication methods generate 3D model files by manual interactive segmentation based on a PC desktop, and the realization time is different depending on the proficiency of operators; some achieve only segmentation of the tumor part or the lung contour, and lack segmentation of the trachea and blood vessels inside the lung.

Disclosure of Invention

The invention aims to provide a lung organ segmentation method, which is used for solving the problems that the lung needs to be segmented semi-automatically through manual intervention, the segmentation process is complex, and the segmentation is incomplete in the prior art.

In order to achieve the purpose, the invention provides the following technical scheme: a method for segmenting lung organs comprises the steps of roughly extracting the lung organs on a three-dimensional image, wherein the lung organs comprise lung parenchyma, trachea and blood vessels inside the lung; 1): inputting 2D chest CT sequence images in a DICOM format on a computer; 2) coarse segmentation of lung parenchyma by module 1 module: and (3) processing each axial slice according to the axial size of the sequence in the input three-dimensional image: gaussian smoothing of traditional parameters is carried out, and noise interference is reduced; setting preferred determination threshold values T1 and T2 for simple binary segmentation, wherein a region which is larger than T1 and smaller than T2 is a region of interest ROI; performing morphological characteristic analysis on the ROI, wherein the region with the largest area is the obtained slice region C of the initial chest part; performing a binarization FillHoles cavity filling algorithm on the region of the initial chest cavity C to obtain a region H;

preferably, the region C is subtracted from the region H mathematically to obtain an initial lung region L, and the characteristic information region area LA of the region L and the area CA of the thorax region are calculated;

calculating the section with the largest area LA in all the sections, and simultaneously satisfying the condition that the area CA is larger than 0.125 of the largest area so as to cover the situation that a part of the lung of the patient is removed by the operation;

after the axial slice with the maximum lung area is obtained, pixel intensity values are searched in the area L within the range of [ -1200, -700], and the pixel intensity values of all pixels within 1 adjacent pixel range in the three-dimensional space are within the range of [ -1200, -700], so that the pixel coordinate position of the initial seed point of the lung area is automatically obtained through calculation;

setting a proper initial upper threshold T3 and a proper initial lower threshold T4, and performing a three-dimensional region growing method to obtain a complete initial lung region; the initial lung region encompasses a tracheal portion with a pixel value close to that of the initial lung region;

3): the complete initial lung region obtained by the module 1 is expanded through the trachea segmentation module of the module 2; and then calculating the volume change according to a region growing method and a gradient change threshold value to obtain a segmentation result of the trachea. Referring to fig. 3, the specific method is as follows:

(1) after binarization expansion is carried out in the lung area, seed points are automatically screened on the top layer in the axial direction;

(2) intercepting an ROI (region of interest), and setting an upper threshold and a lower threshold;

(3) obtaining an initial trachea position by a region growing algorithm;

(4) finding the position of the first branch of the main air pipe by judging the center of mass position of the left branch and the right branch near the branch of the air pipe;

(5) according to the coordinate position of the first fork, the whole three-dimensional image area is divided into a left sub-area and a right sub-area by taking the X coordinate of the fork center as a boundary line;

(6) the sub-region gradually increases the upper threshold of the region growing algorithm according to the gradient value T until the difference value of the sizes of the volumes of the air pipes of the adjacent 2 segmentation results is obviously changed, and the obtained threshold is a coarse threshold;

(7) on the basis of the coarse threshold, increasing the upper limit threshold by step 1 to obtain a reasonable threshold and reasonable main bronchus and bronchus regions (LR, RR) on the left side and the right side;

(8) and combining the regions LR and RR, and performing smoothing treatment to obtain a whole trachea segmentation result.

3): through a lung internal blood vessel rough segmentation module of the module 3, after a trachea region is subtracted from a lung initial region, binary image closure processing of a radius R-12 is carried out, the binary image closure processing is limited to be an input region of blood vessel segmentation, blood vessels are segmented in the region based on a multi-scale hessian matrix measurement mode, and rough precision segmentation of the lung internal blood vessels is achieved. The specific flow chart is shown in fig. 4.

4) Segmentation of lung abnormalities is achieved by block 4, as shown in the flow chart of fig. 5; carrying out binarization image closure processing of a radius R-12 in an initial lung region after subtracting a trachea region, limiting the binary image closure processing to be an input region of abnormal lung segmentation, grouping the binary image closure processing according to axial slices, inputting the input region into a trained 2D UNet neural network for feature extraction, and outputting a probability map;

preferably, setting an initial decomposition value of 0.3 in the probability map to obtain an initial anomaly;

taking the central point of each anomaly as a center, selecting a block ROI with a capture radius of 8, carrying out 3D canny edge detection, and taking the edge as a segmentation basis to realize further anomaly segmentation;

and performing characteristic judgment on each abnormal area, and finally determining whether the abnormal area is abnormal.

In conclusion, the segmentation result obtained by each module is binary Mask point cloud data, and a printable 3D model file is generated after the binary Mask point cloud data is converted into polygonal model data through a conversion method MarchingCubes.

Compared with the related technology, the method for realizing the segmentation of the lung organ based on the three-dimensional image has the following beneficial effects: the invention combines the traditional computer vision image algorithm and the U-net neural network to realize the automatic 3D model reconstruction of lung organs, effectively reduces the workload of doctors for manufacturing the 3D model and precious time, completes the 3D model reconstruction of each part subdivided into lung parenchyma, trachea, blood vessels inside the lung and abnormal parts inside the lung, has complete segmentation, and is more beneficial to 3D display, thereby being beneficial to improving doctor-patient communication and improving communication efficiency.

Drawings

Fig. 1 is a block diagram illustrating an overall implementation of the present invention.

Fig. 2 is a flow chart of an embodiment of the module for roughly segmenting lung parenchyma according to the present invention.

Fig. 3 is a flow chart of the implementation of the trachea rough segmentation module according to the present invention.

Fig. 4 is a flowchart illustrating an implementation of the module for roughly segmenting the blood vessels inside the lung according to the present invention.

Fig. 5 is a flowchart illustrating an implementation of the intra-lung abnormal rough segmentation module according to the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Example 1: referring to fig. 1-5, a DICOM format 2D breast CT sequence image is input on a computer; step 1, according to the sequence axial size in the input three-dimensional image, each axial slice is processed as follows:

gaussian smoothing of traditional parameters is carried out, and noise interference is reduced;

setting preferred determination threshold values T1 and T2 for simple binary segmentation, wherein a region which is larger than T1 and smaller than T2 is a region of interest ROI;

performing morphological characteristic analysis on the ROI, wherein the region with the largest area is the obtained slice region C of the initial chest part;

performing a binarization FillHoles cavity filling algorithm on the region of the initial chest cavity C to obtain a region H;

step 2, after obtaining the axial slice with the maximum lung area, searching pixel intensity values in the area L within the range of [ -1200, -700], and simultaneously satisfying that the pixel intensity values of all pixels in the adjacent 1 pixel range in the three-dimensional space are within the range of [ -1200, -700], and automatically obtaining the pixel coordinate position of the initial seed point of the lung area through calculation;

step 3, setting a proper initial upper threshold T3 and a proper initial lower threshold T4, and performing a three-dimensional region growing method to obtain a complete initial lung region; this region of course encompasses the portion of the trachea where the pixel values are close;

step 4, expanding the lung region obtained in the step 3; then, on the basis, the volume change is calculated mainly according to a region growing method and a gradient change threshold value to obtain a segmentation result of the trachea, in detail, see fig. 3, the method is subdivided into the following steps

(3) obtaining an initial trachea position by a region growing algorithm;

(4) preferably, the position of the first branch of the main air pipe is found by judging the center of mass position of the left branch and the right branch near the branch of the air pipe;

(7) preferably, fine to increase the upper threshold by step 1 on a coarse threshold basis, resulting in reasonable thresholds, and reasonable left and right main and bronchial regions (LR, RR);

And 5, subtracting the trachea area from the lung initial area, performing binary image closing processing with the radius R being 12, limiting the binary image closing processing to be an input area of blood vessel segmentation, segmenting blood vessels in the area based on a multi-scale hessian matrix measurement mode, and realizing coarse precision segmentation of the blood vessels in the lung. The specific flow chart is shown in fig. 4.

And 6, realizing the abnormal segmentation of the lung by the module, as shown in a flow chart 5. The method comprises the steps of (1) carrying out binarization image closing processing on a lung initial region with radius R being 12 after subtracting a trachea region, limiting the lung initial region as an input region of abnormal segmentation of the lung, carrying out grouping according to axial slices, inputting a trained 2D UNet neural network for feature extraction, and outputting a probability map;

The segmentation result obtained by each module is binary Mask point cloud data, and a printable 3D model file is generated after the binary Mask point cloud data is converted into polygonal model data through a conversion method MarchingCubes.

It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.

Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims

1. A method for segmenting lung organs is characterized in that a three-dimensional image is input into a computer, and the lung organs are extracted based on the rough extraction of the three-dimensional image, wherein the lung organs comprise lung parenchyma, trachea and blood vessels in the lung; and it is divided by four modules: module 1: a pulmonary parenchyma coarse segmentation module; and (3) module 2: a trachea segmentation module; and a module 3: a lung internal blood vessel rough segmentation module; and (4) module: lung abnormality segmentation is achieved; the method comprises the following specific steps: step 1): inputting 2D chest CT sequence images in a DICOM format on a computer; step 2) roughly segmenting the lung parenchyma through a module 1: step 3) expanding the complete initial lung region obtained by the module 1 through the trachea segmentation module of the module 2; then, calculating volume change according to a region growing method and a gradient change threshold value to obtain a segmentation result of the trachea; step 4), a module for roughly dividing the blood vessels in the lung through the module 3; step 5), the abnormal segmentation of the lung is realized through the module 4; and step 6), the segmentation results obtained by the modules 1-4 are binary Mask point cloud data, and printable 3D model files are generated after converting the binary Mask point cloud data into polygonal model data by a conversion method Marching Cubes.

2. The method of pulmonary organ segmentation of claim 1, wherein: the specific operation steps of the step 2) are as follows: and (3) according to the sequence axial size in the input three-dimensional image, performing the following processing on each axial slice: gaussian smoothing of traditional parameters is carried out, and noise interference is reduced; setting preferred determination threshold values T1 and T2 for simple binary segmentation, wherein a region which is larger than T1 and smaller than T2 is a region of interest ROI; performing morphological characteristic analysis on the ROI, wherein the region with the largest area is the obtained slice region C of the initial chest part; performing a binarization FillHoles cavity filling algorithm on the region of the initial chest cavity C to obtain a region H;

mathematically subtracting the region C from the region H to obtain an initial lung region L, and calculating the characteristic information region area LA of the region L and the area CA of the thoracic region;

after the axial slice with the maximum lung area is obtained, pixel intensity values are searched in the area L within the range of [ -1200, -700], and the pixel intensity values of all pixels within 1 adjacent pixel range in the three-dimensional space are within the range of [ -1200, -700], so that the pixel coordinate position of the initial seed point of the lung area is automatically obtained through calculation; setting a proper initial upper threshold T3 and a proper initial lower threshold T4, and performing a three-dimensional region growing method to obtain a complete initial lung region; the initial lung region encompasses the portion of the trachea where the pixel values are close.

3. The method of pulmonary organ segmentation of claim 2, wherein: the specific operation steps of the trachea segmentation module of the module 2 in the step 3) are as follows: (1) after binarization expansion is carried out in the lung area, seed points are automatically screened on the top layer in the axial direction; (2) intercepting an ROI (region of interest), and setting an upper threshold and a lower threshold; (3) obtaining an initial trachea position by a region growing algorithm; (4) finding the position of the first branch of the main air pipe by judging the center of mass position of the left branch and the right branch near the branch of the air pipe; (5) according to the coordinate position of the first fork, the whole three-dimensional image area is divided into a left sub-area and a right sub-area by taking the X coordinate of the fork center as a boundary line; (6) the sub-region gradually increases the upper threshold of the region growing algorithm according to the gradient value T until the difference value of the sizes of the volumes of the air pipes of the adjacent 2 segmentation results is obviously changed, and the obtained threshold is a coarse threshold; (7) on the basis of the coarse threshold, the upper limit threshold is increased by step length 1 to obtain a reasonable threshold, and reasonable left and right main bronchus LR and bronchial region RR; (8) and combining the regions LR and RR, and performing smoothing treatment to obtain a whole trachea segmentation result.

4. A method of pulmonary organ segmentation as set forth in claim 3, wherein: the operation steps of the module for roughly segmenting the blood vessels in the lung through the module 3 in the step 4) are as follows: and (3) subtracting the trachea area from the lung initial area, performing binarization image closing processing with the radius R being 12, limiting the binarization image closing processing to be an input area for blood vessel segmentation, and segmenting blood vessels in the input area based on a multi-scale hessian matrix measurement mode to realize coarse precision segmentation of blood vessels in the lung.

5. The antibody detection kit according to claim 4, wherein: the operation steps of segmenting the lung abnormality through the module 4 in the step 5) are as follows: the method comprises the steps of (1) carrying out binarization image closing processing on a lung initial region with radius R being 12 after subtracting a trachea region, limiting the lung initial region as an input region of abnormal segmentation of the lung, carrying out grouping according to axial slices, inputting a trained 2D UNet neural network for feature extraction, and outputting a probability map; setting an initial decomposition value of 0.3 in the probability map to obtain initial anomaly; taking the central point of each anomaly as a center, selecting a block ROI with a capture radius of 8, carrying out 3D canny edge detection, and taking the edge as a segmentation basis to realize further anomaly segmentation; and performing characteristic judgment on each abnormal area, and finally determining whether the abnormal area is abnormal.