Chest image three-dimensional reconstruction method for early-stage lung cancer patient

Abstract

The invention relates to the technical field of image data processing, in particular to a chest image three-dimensional reconstruction method for an early-stage lung cancer patient, which comprises the following steps: acquiring a lung area in each lung slice image in a lung slice image sequence of the early-stage lung cancer patient, segmenting a plurality of low-density tissue connected domains from the lung area, according to the area difference of the low-density tissue connected domains between the continuous adjacent lung slice images, the lesion possibility of the low-density tissue connected domains is determined, so that the low-density tissue connected domains are divided into lesion tissue connected domains and airway tissue connected domains, and different marks are given to the lesion tissue connected domains and the airway tissue connected domains; and converting the marked lung slice image sequence into a three-dimensional model. According to the method, the lesion tissue and the airway tissue are accurately distinguished, so that the capability of expressing the early lesion tissue by the three-dimensional reconstruction model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and particularly relates to a three-dimensional reconstruction method for chest images of early-stage lung cancer patients. Background Art

[0002] Three-dimensional reconstruction of chest CT images of lung cancer patients is an important medical imaging technology. By processing and analyzing two-dimensional slice images obtained from CT scans to generate three-dimensional stereoscopic images, it can effectively help doctors more comprehensively evaluate the patient's lung condition, including the size and location of tumors and the relationship with surrounding tissues, providing a key basis for subsequent surgical treatment plans.

[0003] Existing problems: Three-dimensional reconstruction based on CT images requires accurate threshold segmentation of two-dimensional discrete slice images. After accurately segmenting airway tissues and other diseased tissues, they are converted into pixels in three-dimensional coordinates for three-dimensional reconstruction. However, the diseased tissues of early-stage lung cancer patients are irregular and small in volume, and they show similar characteristics to other normal airway tissues under CT images. Therefore, it is difficult to effectively distinguish such diseased tissues, which may lead to a low accuracy of the final three-dimensional reconstruction model in expressing diseased tissues. Summary of the Invention

[0004] The present invention provides a three-dimensional reconstruction method for chest images of early-stage lung cancer patients to solve the existing problems.

[0005] A three-dimensional reconstruction method for chest images of early-stage lung cancer patients according to the present invention adopts the following technical solutions: An embodiment of the present invention provides a three-dimensional reconstruction method for chest images of early-stage lung cancer patients, and the method includes the following steps: Obtain a sequence of lung slice images of early-stage lung cancer patients; in the sequence of lung slice images, use a segmentation neural network to segment the lung regions in each lung slice image; According to the gray value differences between pixel points in the lung regions of each lung slice image, segment several low-density tissue connected domains from the lung regions in each lung slice image; According to the area differences of low-density tissue connected domains between continuously adjacent lung slice images in the sequence of lung slice images, determine the lesion possibility of each low-density tissue connected domain in each lung slice image; According to the lesion possibility of each low-density tissue connected domain in each lung slice image, classify the low-density tissue connected domains into lesion tissue connected domains and airway tissue connected domains; assign different labels to the lesion tissue connected domains and airway tissue connected domains in each lung slice image in the sequence of lung slice images to obtain a labeled sequence of lung slice images; convert the labeled sequence of lung slice images into a three-dimensional model.

[0006] Further, the steps of segmenting a plurality of low-density tissue connected domains from the lung regions in each lung slice image are as follows: In the lung slice image sequence, the lung region in any one lung slice image is denoted as the target lung region; Segment a plurality of initial airway tissue connected domains from the target lung region; Denote the boundary pixel points on each initial airway tissue connected domain as seed points; Preset a plurality of attenuation directions. According to the gray value difference of the pixel points along the th preset attenuation direction in the th initial airway tissue connected domain, determine the attenuation degree of the th initial airway tissue connected domain in the th preset attenuation direction; According to the attenuation degrees of the th initial airway tissue connected domain in all preset attenuation directions, determine the similarity threshold of each seed point on the th initial airway tissue connected domain;

[0007] Further, the steps of segmenting a plurality of initial airway tissue connected domains from the target lung region are as follows: Use the Otsu algorithm to obtain the optimal segmentation threshold of the target lung region. In the target lung region, denote the pixel points with gray values greater than the optimal segmentation threshold as target pixel points, and denote the connected domain formed by adjacent target pixel points as the initial airway tissue connected domain.

[0008] Further, the steps of determining the attenuation degree of the th initial airway tissue connected domain in the th preset attenuation direction are as follows: Denote the perpendicular direction of the th preset attenuation direction as the reference direction; Take the straight line in the th preset attenuation direction and translate it within the th initial airway tissue connected domain along the reference direction to traverse the entire One initial airway tissue connected domain, and obtain a number of non-overlapping straight lines as target straight lines; In the th initial airway tissue connected domain, obtain the sequence formed by all pixel points on each target straight line with the th preset attenuation direction as the positive direction, as the pixel sequence of each target straight line; In the pixel sequence of each target straight line, calculate the difference between the gray value of the th pixel point and the gray value of the th pixel point, and take the mean value of the differences between the gray values of all adjacent pixel points as the attenuation degree of each target straight line; Take the mean value of the attenuation degrees of all target straight lines as the attenuation degree of the th initial airway tissue connected domain in the th preset attenuation direction.

[0009] Further, the specific steps for determining the similarity threshold of each seed point on the th initial airway tissue connected domain are as follows: Take the maximum attenuation degree among the attenuation degrees of the th initial airway tissue connected domain in all preset attenuation directions as the similarity threshold of each seed point on the th initial airway tissue connected domain.

[0010] Further, the specific steps for determining the lesion possibility of each low-density tissue connected domain in each lung slice image are as follows: In the lung slice image sequence, according to several lung slice images adjacent before and after the th lung slice image, obtain the reference lung slice image sequence segment of the th lung slice image; Denote any low-density tissue connected domain in the th lung slice image as the target low-density tissue connected domain; According to the area of the low-density tissue connected domain matched by the target low-density tissue connected domain in the reference lung slice image sequence segment, obtain the area sequence segment; According to the differences between the areas in the area sequence segment, determine the lesion possibility of the target low-density tissue connected domain in the th lung slice image.

[0011] Further, the specific steps for obtaining the reference lung slice image sequence segment of the th lung slice image are as follows: Preset quantity threshold , in the sequence of lung slice images, a sequence segment composed of the th lung slice image to the th lung slice image is used as the reference lung slice image sequence segment of the th lung slice image.

[0012] Furthermore, the steps for obtaining the area sequence segment are as follows: In the reference lung slice image sequence segment of the th lung slice image, the optical flow method is used to obtain the matching target low-density tissue connected domain of the target low-density tissue connected domain in each reference lung slice image, and a sequence segment of matching target low-density tissue connected domains is obtained; In the sequence segment of matching target low-density tissue connected domains, the number of pixel points in each matching target low-density tissue connected domain is counted as the area of each matching target low-density tissue connected domain, and an area sequence segment is obtained.

[0013] Furthermore, the steps for determining the lesion possibility of the target low-density tissue connected domain in the th lung slice image are as follows: In the area sequence segment, the variance of all areas is calculated, and then the absolute value of the difference between adjacent areas is calculated. The normalized value of the product of the mean of the absolute values of all adjacent area differences and the variance is denoted as the lesion possibility of the target low-density tissue connected domain in the th lung slice image.

[0014] Furthermore, the steps for classifying the low-density tissue connected domain into a diseased tissue connected domain and an airway tissue connected domain are as follows: The low-density tissue connected domain with a lesion possibility greater than a preset judgment threshold is denoted as a diseased tissue connected domain; The low-density tissue connected domain with a lesion possibility less than or equal to the preset judgment threshold is denoted as an airway tissue connected domain.

[0015] The beneficial effects of the technical solution of the present invention are: In an embodiment of the present invention, the lung regions in each lung slice image of a sequence of lung slice images of an early lung cancer patient are obtained, and a number of low-density tissue connected domains are segmented from the lung regions. Thus, by accurately segmenting the low-density tissue connected domains containing diseased tissues and airway tissues, the accuracy of subsequent differentiation of diseased tissues is ensured. According to the area difference of the low-density tissue connected domains between continuously adjacent lung slice images, the lesion possibility of the low-density tissue connected domains is determined, so that the low-density tissue connected domains are classified into diseased tissue connected domains and airway tissue connected domains and different labels are assigned. Thus, according to the analysis between consecutive lung slice images, the accuracy of diseased tissue segmentation is further ensured. The labeled sequence of lung slice images is converted into a three-dimensional model. By accurately differentiating diseased tissues from airway tissues, the present invention can effectively help determine the positions of diseased tissues during the three-dimensional reconstruction process, thereby improving the ability of the three-dimensional reconstruction model to represent early diseased tissues. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to these drawings.

[0017] Figure 1 It is a flowchart of the steps of a method for three-dimensional reconstruction of chest images of an early lung cancer patient according to the present invention; Figure 2 It is a schematic diagram of chest CT shooting; Figure 3 It is a CT slice image of a lung nodule of early lung cancer; Figure 4 It is a slice image including only the lungs; Figure 5 It is an image after segmentation of the lung region by the Otsu algorithm; Figure 6 It is a schematic diagram of the division of a connected domain pixel sequence in the attenuation direction of the upper left; Figure 7 It is a three-dimensional schematic diagram of a CT slice of lung airway tissue. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following specifically describes, in conjunction with the accompanying drawings and preferred embodiments, a three-dimensional reconstruction method for chest images of early-stage lung cancer patients, including its specific implementation manner, structure, features, and effects. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0020] The following specifically describes the specific solution of a three-dimensional reconstruction method for chest images of early-stage lung cancer patients provided by the present invention in conjunction with the accompanying drawings.

[0021] Please refer to Figure 1 , which shows a flowchart of the steps of a three-dimensional reconstruction method for chest images of early-stage lung cancer patients provided by an embodiment of the present invention. The method includes the following steps: Step S001: Obtain a sequence of lung slice images of early-stage lung cancer patients; in the sequence of lung slice images, use a segmentation neural network to segment the lung regions in each lung slice image.

[0022] Use a 128-slice spiral CT device to collect chest images of early-stage lung cancer patients and obtain a sequence of lung slice images of early-stage lung cancer patients.

[0023] It should be noted that: the schematic diagram of chest CT shooting, as shown in Figure 2 , Figure 2 the 128 rows in it refer to a technical parameter of the CT scanner, indicating that the scanner can simultaneously obtain image data of 128 different levels in one rotation. In this embodiment, the layer thickness of the lung slice images is set to 1 mm, the number of lung slice images in the sequence of lung slice images is 200, and the lung slice images are grayscale images. This is used as an example for description.

[0024] In most cases, the clinical manifestations of early-stage lung cancer patients are lung nodules. The nodules may appear in various shapes. For example, ellipsoidal nodules are mostly benign tumors, and malignant tumors are mostly spiculated and lobulated. The cancerous tissue is generally small, and the density of the cancerous tissue is low. In the lung slice images, the normal airway tissue has a low density because it contains air. Therefore, in the lung slice images, the cancerous tissue and the normal airway tissue show similar grayscale, making it difficult to distinguish, and thus the cancerous tissue is easily overlooked during three-dimensional reconstruction.

[0025] However, since normal airway tissue has a tree-like bifurcated structure, it may appear discontinuous in a lung slice image, and multiple adjacent lung slice images can form continuous airway tissue. Since the diameter width gradually decreases from the main airway to the branched airway, and early cancerous lung nodules usually have irregular structures, although they may present various forms such as spiculated or lobulated, their continuity in the performance of continuous lung slice images is much lower than that of normal tissues, and the width is irregular. The CT slice images of early lung cancer lung nodules, such as Figure 3 shown.

[0026] Therefore, in this embodiment, first, the lung slice images are preliminarily segmented to determine the outer contour of the lungs and the tissues with lower density inside. Secondly, the possibility that the tissues with lower density are normal airway tissues is calculated. Finally, a segmentation threshold is set to distinguish normal airway tissues from cancerous tissues.

[0027] Since the CT images taken include all objects in the scanned area, including the bed board, clothes, other human tissues, etc., it is necessary to extract the lung area, and then preliminarily segment to obtain the tissues with lower density within the lung area. Since the low-density tissues (airway tissues or cancerous tissues) inside the lungs interfere with the clothes, bed board, etc. outside the lungs, the direct image segmentation effect is poor, while the lung area has obvious gray-scale boundaries and is relatively continuous, and can be easily recognized. Therefore, first, the lung area is recognized, and secondly, the low-density tissues inside the lungs are recognized.

[0028] In the sequence of lung slice images, taking any lung slice image as an example, the embodiment of the present invention uses a segmentation neural network to identify and segment the lung area in the lung slice image.

[0029] The relevant content of the segmentation neural network is as follows: The segmentation neural network used in this embodiment is the Mask R-CNN neural network; the dataset used is the lung slice image dataset. Among them, Mask R-CNN is a well-known technology, and the specific method will not be introduced here. The Chinese full name of Mask R-CNN is "Mask Region-based Convolutional Neural Network", and the English full name is "Mask Region-based Convolutional Neural Network".

[0030] The pixel points to be segmented are divided into 2 categories. That is, the label annotation process corresponding to the training set is: for the single-channel semantic label, the pixel points at the corresponding positions belonging to the background area are labeled as 0, and those belonging to the lung area are labeled as 1.

[0031] The task of the network is classification, so the loss function used is the cross-entropy loss function.

[0032] The lung region in each lung slice image is obtained by splitting the neural network. This process is a well-known technique, and the specific method will not be introduced here. Only the slice images including the lungs are as follows Figure 4 as shown.

[0033] Step S002: According to the gray value differences between the pixel points in the lung region of each lung slice image, several low-density tissue connected domains are segmented from the lung region of each lung slice image.

[0034] Because airway tissues or cancerous tissues diverge in all directions, for the tissues with divergence in the lung slice image, the diameter of these tissues narrows from the main end to the end, and the image gray value gradually decreases, which is the cross-section of the airway. At the same time, there are also tissues that penetrate the slice, which is the longitudinal section of the airway. These tissues may present isolated gray points in a slice plane, and there are even tissues combining the two characteristics. Therefore, the types of airway tissues in the lung slice image can be divided into divergent within the slice plane, penetrating the slice, and both penetrating the slice and diverging within the slice plane. When segmenting the lung slice image, it is easy to fail to completely segment the airway tissues and cancerous tissues that diverge within the slice plane, and only part of the tissues with similar gray values to the penetrating type can be segmented. Therefore, it is necessary to first determine the complete airway tissues.

[0035] Preferably, in an embodiment of the present invention, the method for obtaining the low-density tissue connected domain includes: In the lung slice image sequence, the lung region in any lung slice image is denoted as the target lung region.

[0036] Using the Otsu algorithm, the optimal segmentation threshold of the target lung region is obtained. In the target lung region, the pixel points with gray values greater than the optimal segmentation threshold are denoted as target pixel points, the connected domain formed by adjacent target pixel points is denoted as the initial airway tissue connected domain, and the boundary pixel points on each initial airway tissue connected domain are denoted as seed points.

[0037] It should be noted that: in this embodiment, the region growing algorithm is used to obtain the low-density tissue region in the lung region, and the low-density tissue region includes airway tissues and cancerous tissues. Both the region growing algorithm and the Otsu algorithm are well-known techniques, and the specific methods will not be introduced here. The image of the lung region after being segmented by the Otsu algorithm is as follows Figure 5 as shown. Figure 5The white connected region in the middle lung area is the area where the airway tissue is more obvious, that is, the initial airway tissue connected region. Therefore, the boundary pixel points on the initial airway tissue connected region are used as the seed points in the region growing process. Since the airway tissue or needle-shaped cancerous tissue on the section usually presents a linear shape, the gray scale attenuates from one side to the other side, and the diameter of the airway tissue can be considered to narrow uniformly within a small range, so the degree of gray scale attenuation is similar. Therefore, the attenuation degree in each direction in the initial airway tissue connected region can be analyzed, the maximum attenuation degree can be selected, and the similarity criterion in the region growing process can be determined.

[0038] Preset several attenuation directions. In this embodiment, the eight-neighborhood directions of up, down, left, right, upper left, lower left, upper right, and lower right are used as the preset attenuation directions, and this is used as an example for description.

[0039] In the target lung area, taking the th initial airway tissue connected region as an example, and then taking the th preset attenuation direction as an example, the perpendicular direction of the th preset attenuation direction is used as the reference direction. Along the reference direction, the straight line in the th preset attenuation direction is translated within the th initial airway tissue connected region to traverse the entire th initial airway tissue connected region, and several non-overlapping straight lines are obtained as the target straight lines. In the th initial airway tissue connected region, for each target straight line, all the pixel points on it are used to form a sequence with the th preset attenuation direction as the positive direction, which is used as the pixel sequence of each target straight line.

[0040] It should be noted that: if the length of a certain pixel sequence is 1, it is discarded and not analyzed further. The schematic diagram of the pixel sequence division of the connected region with the upper left as the attenuation direction is as Figure 6 shown. Figure 6 The gray squares 1, 2, 3, 4, 5, 6, and 7 in the figure are the connected region pixels, that is, the gray area is the initial airway tissue connected region. With the upper left direction pointed by the dotted arrow as the attenuation direction, three pixel sequences {2, 1}, {5, 4, 3}, and {7, 6} can be obtained.

[0041] In the pixel sequence of each target straight line, calculate the difference between the gray value of the th pixel point and the gray value of the th pixel point. The mean value of the differences between the gray values of all adjacent pixel points is used as the attenuation degree of each target straight line, and the mean value of the attenuation degrees of all target straight lines is used as the attenuation degree of the th initial airway tissue connected region in the The attenuation degree in a preset attenuation direction.

[0042] In the above manner, obtain the attenuation degree of the th initial airway tissue connected domain in each preset attenuation direction.

[0043] Take the maximum attenuation degree among the attenuation degrees of the th initial airway tissue connected domain in all preset attenuation directions as the similarity threshold of each seed point on the th initial airway tissue connected domain.

[0044] In the above manner, obtain the similarity threshold of each seed point in the target lung region.

[0045] In the target lung region, according to each seed point and the similarity criterion of each seed point, use the region growing algorithm to start region growing from each seed point to obtain several connected domains as the low-density tissue connected domains. Among them, the similarity criterion of each seed point is: when the absolute value of the difference between the gray value of each seed point and each pixel point adjacent to each seed point is less than the similarity threshold of each seed point, merge each adjacent pixel point into the region where each seed point is located.

[0046] It should be noted that: The region growing algorithm is a well-known technology, and the selection of seed points and the definition of the similarity criterion of each seed point are the main operations of the region growing algorithm. When growing seed points, when the difference between the growing pixel and the seed pixel is less than the maximum attenuation degree, it is considered to be caused by the change of the airway tube or cancerous tissue, and continue to grow, otherwise it is considered to be the boundary between the airway tissue and other lung tissues, and stop growing.

[0047] In the above manner, obtain several low-density tissue connected domains in the lung region of each lung slice image in the lung slice image sequence.

[0048] So far, the airway or diseased low-density tissue in the lung region of each lung slice image is completely segmented.

[0049] Step S003: Determine the lesion possibility of each low-density tissue connected domain in each lung slice image according to the area difference of the low-density tissue connected domains between two consecutive adjacent lung slice images in the lung slice image sequence.

[0050] In the tissue with lower density, there are normal airway tissues and tissues suspected of canceration. However, the normal airway tissues present a structure of tree crown bifurcation in the three-dimensional structure. On the CT slices of adjacent intervals, the airway tissues at similar positions can form a continuous tissue, and the tissue width changes from small to large. The three-dimensional schematic diagram of the CT slice of the lung airway tissue is as Figure 7 shown.Figure 7 The continuous diamond boxes are consecutive lung CT slices. Both the solid line and the dashed line represent airway tissues. The solid line represents the distribution of the airway in the slice penetration direction, and the dashed line represents the airway tissue in the same slice direction.

[0051] Correspondingly, in the consecutive slices, the change in continuity of the diseased tissue is not obvious, and the continuity of the width change is poor. Therefore, if the current low-density tissue has poor continuity in adjacent lung slice images and the width change is not continuous in a single direction, the possibility that it is diseased tissue is relatively high.

[0052] In the sequence of lung slice images, the size change of the lung region contour in the local continuous image is small. At the same time, the width of the tissue diameter is also reflected in the number of pixels in the tissue region. The more the number of pixels, the larger the width of the tissue diameter. In addition, in the direction where the low-density tissue penetrates the slice, most of the locally continuous slices represent the same tissue at the same position. Therefore, the tissue with a regular change in the number of pixels in the connected domain at the same position on the consecutive slices is more likely to be normal tissue and less likely to be diseased tissue.

[0053] Preferably, in an embodiment of the present invention, the method for obtaining the lesion possibility of each low-density tissue connected domain includes: Preset a quantity threshold Take 3 as an example for description.

[0054] In the sequence of lung slice images, taking the th lung slice image as an example, the sequence segment composed of the th lung slice image to the th lung slice image is used as the reference lung slice image sequence segment of the th lung slice image.

[0055] It should be noted that: when the th lung slice image is the first images or the last images in the sequence of lung slice images, select the existing lung slice images between the th lung slice image and the th lung slice image to form a reference lung slice image sequence segment.

[0056] Denote any low-density tissue connected domain in the th lung slice image as the target low-density tissue connected domain.

[0057] In the In the reference lung slice image sequence segment of reference lung slice images, the optical flow method is used to obtain the matching target low-density tissue connected domain of the target low-density tissue connected domain in each reference lung slice image, and the sequence segment of the matching target low-density tissue connected domain of the

[0058] In the sequence segment of the matching target low-density tissue connected domain of the reference lung slice images, the number of pixel points in each matching target low-density tissue connected domain is counted as the area of each matching target low-density tissue connected domain, and the sequence segment of the areas corresponding to the target low-density tissue connected domain in the reference lung slice images is obtained.

[0059] It should be noted that: The optical flow method is a well-known technology, and the specific method will not be introduced here. This algorithm can be used to track the changes of lung tissue in consecutive slice images. When there is no matching target low-density tissue connected domain of the target low-density tissue connected domain in a certain reference lung slice image, the number of pixel points in the matching target low-density tissue connected domain of this certain reference lung slice image is set to 0.

[0060] In the sequence segment of the areas corresponding to the target low-density tissue connected domain in the reference lung slice images, the variance of all areas is calculated, and then the absolute value of the difference between adjacent areas is calculated. The normalized value of the product of the mean of all absolute values of the differences between adjacent areas and the variance is denoted as the lesion possibility of the target low-density tissue connected domain in the reference lung slice images.

[0061] It should be noted that: Among them, for the normalized value of the product, in this embodiment, the linear normalization function is used to normalize the product to between 0 and 1. When the above variance is smaller and the absolute value of the difference between adjacent areas is smaller, it indicates that the area change of the target low-density tissue connected domain in the reference lung slice images is uniform in consecutive slice images and is less likely to be a lesion.

[0062] According to the above method, the lesion possibility of each low-density tissue connected domain in each lung slice image in the lung slice image sequence is obtained.

[0063] Step S004: According to the lesion possibility of each low-density tissue connected domain in each lung slice image, the low-density tissue connected domains are classified into lesion tissue connected domains and airway tissue connected domains; different labels are assigned to the lesion tissue connected domains and airway tissue connected domains in each lung slice image in the lung slice image sequence, and the labeled lung slice image sequence is obtained; the labeled lung slice image sequence is converted into a three-dimensional model.

[0064] Preferably, in an embodiment of the present invention, the method for obtaining a three-dimensional model includes: Taking the preset judgment threshold as 0.8 as an example for description.

[0065] The connected domain of low-density tissues with a lesion possibility greater than the preset judgment threshold is denoted as the connected domain of lesion tissues.

[0066] The connected domain of low-density tissues with a lesion possibility less than or equal to the preset judgment threshold is denoted as the connected domain of airway tissues.

[0067] In the sequence of lung slice images, different labels are assigned to the connected domain of lesion tissues and the connected domain of airway tissues in each lung slice image, and the labeled sequence of lung slice images is obtained. The three-dimensional reconstruction algorithm is used to convert the labeled sequence of lung slice images into a three-dimensional model.

[0068] It should be noted that: The three-dimensional reconstruction algorithm is a well-known technology, and the specific method will not be introduced here. In this embodiment, the connected domain of lesion tissues in each lung slice image is labeled red, and the connected domain of airway tissues is labeled yellow as an example for description.

[0069] So far, the present invention is completed.

[0070] In summary, in the embodiment of the present invention, the lung region in each lung slice image of the sequence of lung slice images of early lung cancer patients is obtained, several connected domains of low-density tissues are segmented from the lung region, and the lesion possibility of the connected domains of low-density tissues is determined according to the area difference of the connected domains of low-density tissues between consecutive adjacent lung slice images, so as to distinguish the connected domains of low-density tissues into the connected domain of lesion tissues and the connected domain of airway tissues, and different labels are assigned. The labeled sequence of lung slice images is converted into a three-dimensional model. The present invention improves the ability of the three-dimensional reconstruction model to express early lesion tissues by accurately distinguishing lesion tissues from airway tissues.

[0071] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for three-dimensional reconstruction of chest images of patients with early lung cancer, characterized in that: The method comprises the following steps: Acquire a sequence of lung slice images of a patient with early lung cancer; in the sequence of lung slice images, use a segmentation neural network to segment the lung region in each lung slice image; Segmenting a plurality of low-density tissue connected domains from the lung region in each lung slice image according to gray value differences between pixel points in the lung region in each lung slice image; Determining the possibility of a lesion in each low-density tissue connected domain in each lung slice image according to a difference in the area of ​​the low-density tissue connected domain between consecutive adjacent lung slice images in the lung slice image sequence; According to the lesion possibility of each low-density tissue connected domain in each lung slice image, the low-density tissue connected domain is divided into a lesion tissue connected domain and an airway tissue connected domain; different labels are assigned to the lesion tissue connected domain and the airway tissue connected domain in each lung slice image in the lung slice image sequence to obtain a labeled lung slice image sequence; and the labeled lung slice image sequence is converted into a three-dimensional model.

2. According to claim 1, a method for three-dimensional reconstruction of chest images of patients with early lung cancer, characterized in that: The specific steps of segmenting a plurality of low-density tissue connected domains from the lung region in each lung slice image are as follows: In the sequence of lung slice images, a lung region in any lung slice image is recorded as a target lung region; Segmenting a plurality of initial airway tissue connected domains from the target lung region; The boundary pixel points on each initial airway tissue connected domain are recorded as seed points; Preset several attenuation directions, according to The initial airway tissue connectivity domain is along the The gray value difference of the pixel points in the preset attenuation direction is used to determine the The initial airway tissue connectivity domain is The attenuation degree in a preset attenuation direction; According to The attenuation degree of the initial airway tissue connectivity domain in all preset attenuation directions is determined by The similarity threshold of each seed point on the initial airway tissue connectivity domain; In the target lung area, according to each seed point and the similarity criterion of each seed point, a region growing algorithm is used to start region growing from each seed point to obtain several connected domains as low-density tissue connected domains; the similarity criterion of each seed point is: when the absolute value of the difference between the grayscale values ​​of each seed point and each pixel point adjacent to each seed point is less than the similarity threshold of each seed point, each adjacent pixel point is merged into the region where each seed point is located.

3. The method for three-dimensional reconstruction of chest images of patients with early lung cancer according to claim 2, characterized in that: The step of segmenting a plurality of initial airway tissue connected domains from the target lung region includes the following specific steps: The Otsu algorithm is used to obtain the optimal segmentation threshold of the target lung area. In the target lung area, the pixels with grayscale values ​​greater than the optimal segmentation threshold are recorded as target pixels, and the connected domain formed by adjacent target pixels is recorded as the initial airway tissue connected domain.

4. The method for three-dimensional reconstruction of chest images of patients with early lung cancer according to claim 2, characterized in that: The determination The initial airway tissue connectivity domain is The attenuation degree of a preset attenuation direction includes the following specific steps: The first The vertical direction of the preset attenuation direction is used as the reference direction; First A straight line in the preset attenuation direction, The initial airway tissue connection domain is translated along the reference direction to traverse the entire Initial airway tissue connectivity domains are obtained, and several non-overlapping straight lines are obtained as target straight lines; In the In the initial airway tissue connected domain, all pixels on each target straight line are obtained. A sequence consisting of a preset attenuation direction as a positive direction is used as the pixel sequence of each target line; In the pixel sequence of each target straight line, calculate the The gray value of the pixel is subtracted from the The difference of the grayscale values ​​of pixels is taken as the mean of the differences of the grayscale values ​​of all adjacent pixels as the attenuation of each target line; The mean of the attenuation of all target lines is taken as the The initial airway tissue connectivity domain is The attenuation degree in the preset attenuation direction.

5. The method for three-dimensional reconstruction of chest images of patients with early lung cancer according to claim 2, characterized in that: The determination The similarity threshold of each seed point on the initial airway tissue connectivity domain includes the following specific steps: The first The maximum attenuation degree of the initial airway tissue connectivity domain in all preset attenuation directions is recorded as The similarity threshold of each seed point on the initial airway tissue connectivity domain.

6. The method for three-dimensional reconstruction of chest images of patients with early lung cancer according to claim 1, characterized in that: The specific steps of determining the lesion possibility of each low-density tissue connected domain in each lung slice image are as follows: In the lung slice image sequence, according to A plurality of lung slice images adjacent to and before the first lung slice image are obtained. a reference lung slice image sequence segment of a lung slice image; The first Any low-density tissue connected domain in the lung slice image is recorded as the target low-density tissue connected domain; Acquire an area sequence segment according to the area of ​​the low-density tissue connected domain matched by the target low-density tissue connected domain in the reference lung slice image sequence segment; According to the difference between the areas in the area sequence segments, the first The lesion probability of the target low-density tissue connected area in the lung slice image.

7. The method for three-dimensional reconstruction of chest images of patients with early lung cancer according to claim 6, characterized in that: The obtaining of the The specific steps of preparing a reference lung slice image sequence segment of a lung slice image are as follows: Preset quantity threshold In the lung slice image sequence, lung slice image to The sequence segment consisting of lung slice images is A reference lung slice image sequence segment of a lung slice image.

8. The method for three-dimensional reconstruction of chest images of patients with early lung cancer according to claim 6, characterized in that: The specific steps of obtaining the area sequence segment are as follows: In the said In the reference lung slice image sequence segments of the reference lung slice images, using the optical flow method to obtain the matching target low-density tissue connected domain of the target low-density tissue connected domain in each reference lung slice image, to obtain the matching target low-density tissue connected domain sequence segment; In the matching target low-density tissue connected domain sequence segment, the number of pixels in each matching target low-density tissue connected domain is counted as the area of ​​each matching target low-density tissue connected domain, and an area sequence segment is obtained.

9. The method for three-dimensional reconstruction of chest images of patients with early lung cancer according to claim 6, characterized in that: The determination The pathological possibility of the target low-density tissue connected area in the lung slice image is determined, and the specific steps are as follows: In the area sequence segment, the variance of all areas is calculated, and then the absolute value of the difference between adjacent areas is calculated. The normalized value of the product of the mean of the absolute value of the difference between all adjacent areas and the variance is recorded as The lesion probability of the target low-density tissue connected area in the lung slice image.

10. The method for three-dimensional reconstruction of chest images of patients with early lung cancer according to claim 1, characterized in that: The step of dividing the low-density tissue connected area into the lesion tissue connected area and the airway tissue connected area includes the following specific steps: The low-density tissue connected area with a lesion possibility greater than a preset judgment threshold is recorded as the lesion tissue connected area; The low-density tissue connectivity domain whose lesion possibility is less than or equal to the preset judgment threshold is recorded as the airway tissue connectivity domain.

Images