Method, system, device and medium for segmenting lung lobe region in lung image

Abstract

The present invention relates to the field of image processing technology, and specifically provides a method, system, device and medium for segmenting the lobe region in a lung image, aiming to solve the problem of how to accurately segment the lobe in a lung image. To this end, the present invention uses a point cloud neural network model to correct the three-dimensional coordinates of the fissure region in the lung image, and performs lobe segmentation on the lung image according to the corrected fissure region to obtain the lobe region. Through the above-mentioned configuration, the present invention can correct the three-dimensional coordinates of the fissure region in the lung image for lung images with unclear fissures, accessory fissures, and pulmonary fibrosis, so that the prediction of the fissure region is more complete and accurate, and further the lobe region obtained by lobe segmentation according to the corrected fissure region is also more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and specifically provides a method, system, device and medium for segmenting a lung lobe region in a lung image. Background Art

[0002] The lung is one of the most important organs in the human body and also an organ with a high incidence rate. Common lung diseases include emphysema, lung cancer, bullae, pneumonia, etc. In clinical practice, surgeons or radiologists generally diagnose lung diseases and propose treatment plans by interpreting lung images such as CT (Computed Tomography) images of the lungs. The human lung can be divided into five independent functional units, called lobes. Among the technical methods currently used for lobe segmentation, the lobe segmentation technology based on the neural network method has relatively excellent performance, especially the two-step method, that is, the method of first segmenting the fissures and then segmenting the lobes, which has more precise segmentation results for the lobes near the fissures. This is because the fissures are the physical boundaries between the lobes and are the basis for the division of lung regions, the location of lung diseases, and quantitative evaluation. However, the fissures are not always obvious and easy to obtain. For some patients, their fissures are unclear due to factors such as lesions, and some patients have accessory fissures or pulmonary fibrosis, which will produce noise similar to the fissure signals and are easily misjudged as fissures by the neural network. These incorrect fissures will lead to subsequent lobe segmentation errors.

[0003] Accordingly, the art needs a new segmentation solution for the lung lobe region in lung images to solve the above problems. Summary of the invention

[0004] In order to overcome the above-mentioned defects, the present invention is proposed to provide a solution or at least partially solve the problem of how to accurately segment the lung lobes in the lung image when the lung fissures are not clear.

[0005] In a first aspect, the present invention provides a method for segmenting a lung lobe region in a lung image, the method comprising:

[0006] Acquire a lung region in a lung image and acquire a lung fissure region in the lung region;

[0007] Using a point cloud neural network model to correct the three-dimensional coordinates of the pulmonary fissure region in the lung image to remove noise in the pulmonary fissure region and complete the missing pulmonary fissure region;

[0008] The lung image is segmented into lobes according to the lung region and the corrected pulmonary fissure region to obtain the pulmonary lobe region.

[0009] In a technical solution of the above-mentioned method for segmenting the lung lobe region in the lung image, the step of "correcting the three-dimensional coordinates of the lung fissure region in the lung image using a point cloud neural network model to remove noise from the lung fissure region and complete the missing lung fissure region" specifically includes:

[0010] The point cloud neural network model is trained using training samples and sample labels, wherein the training samples include the coordinate values ​​of the x, y and z coordinates in the three-dimensional coordinates of the first type of lung fissure points without missing coordinate data, and the coordinate values ​​of the x and y coordinates in the three-dimensional coordinates of the second type of lung fissure points, and the sample labels are the coordinate values ​​of the z coordinates in the three-dimensional coordinates of the second type of lung fissure points; the second lung fissure point acquisition method is: randomly sampling the first type of lung fissure points to obtain a first type of lung fissure point as a first target lung fissure point; according to the first target lung fissure point, a plurality of first type of lung fissure points whose coordinate values ​​are continuous with the first target lung fissure point in three-dimensional space are acquired from other first type of lung fissure points as second target lung fissure points; the first target lung fissure point and the second target lung fissure point form the second type of lung fissure point;

[0011] Using the trained point cloud neural network model, respectively obtain the coordinate prediction value of the z coordinate in the three-dimensional coordinates of each pulmonary fissure point in the pulmonary fissure area;

[0012] For each lung fissure point, if the coordinate value of the z coordinate of the current lung fissure point is missing, the coordinate prediction value is set to the coordinate value of the z coordinate of the current lung fissure point;

[0013] If the current lung fissure point does not lack the coordinate value of the z coordinate and the actual coordinate value of the z coordinate is inconsistent with the predicted coordinate value, the actual coordinate value is replaced with the predicted coordinate value.

[0014] In a technical solution of the above-mentioned method for segmenting the lung lobe region in the lung image, the step of "segmenting the lung lobe of the lung image according to the lung region and the corrected lung fissure region to obtain the lung lobe region" specifically includes:

[0015] Using a convolutional neural network and based on the lung area and the corrected lung fissure area, perform lobe segmentation on the lung image to obtain the lobe area of ​​each lobe in the lung image and determine the lobe label of each lobe area;

[0016] respectively obtaining the largest image connected area in each lung lobe region, and obtaining other image connected areas in the lung region except the largest image connected area;

[0017] For each of the other image connected areas, determine the largest image connected area that is closest to the current other image connected area, take the current other image connected area as a part of the lobe area to which the largest image connected area belongs, and set a corresponding lobe label for the current other image connected area.

[0018] In a technical solution of the above-mentioned method for segmenting the lung lobe area in the lung image, the point cloud neural network model is a model based on the PointNet network.

[0019] In a second aspect, the present invention provides a system for segmenting a lung lobe region in a lung image, the system comprising:

[0020] A lung fissure region acquisition module, configured to acquire a lung region in a lung image and acquire a lung fissure region in the lung region;

[0021] A lung fissure region correction module, which is configured to use a point cloud neural network model to correct the three-dimensional coordinates of the lung fissure region in the lung image to remove noise in the lung fissure region and complete the missing lung fissure region;

[0022] The lung lobe region segmentation module is configured to perform lung lobe segmentation on the lung image according to the lung region and the corrected lung fissure region to obtain the lung lobe region.

[0023] In a technical solution of the above-mentioned system for segmenting the lung lobe region in the lung image, the lung fissure region correction module includes:

[0024] A point cloud neural network training unit, configured to use training samples and sample labels to perform network training on a point cloud neural network model, wherein the training samples include coordinate values ​​of x, y and z coordinates in the three-dimensional coordinates of a first type of lung fissure point without missing coordinate data, and coordinate values ​​of x and y coordinates in the three-dimensional coordinates of a second type of lung fissure point, and the sample label is the coordinate value of the z coordinate in the three-dimensional coordinates of the second type of lung fissure point; the second lung fissure point acquisition method is: randomly sampling the first type of lung fissure point to obtain a first type of lung fissure point as a first target lung fissure point; according to the first target lung fissure point, obtaining multiple first type of lung fissure points with continuous coordinate values ​​with the first target lung fissure point in three-dimensional space from other first type of lung fissure points as second target lung fissure points; the first target lung fissure point and the second target lung fissure point form the second type of lung fissure point;

[0025] A z-coordinate prediction value acquisition unit is configured to respectively acquire a coordinate prediction value of a z-coordinate in the three-dimensional coordinates of each pulmonary fissure point in the pulmonary fissure region using a trained point cloud neural network model;

[0026] A z-coordinate setting unit is configured to set the coordinate prediction value to the coordinate value of the z-coordinate of the current lung fissure point for each lung fissure point if the current lung fissure point lacks the coordinate value of the z-coordinate;

[0027] A z-coordinate replacement unit is configured to replace the actual coordinate value with the predicted coordinate value if the current lung fissure point does not lack the coordinate value of the z-coordinate and the actual coordinate value of the z-coordinate is inconsistent with the predicted coordinate value.

[0028] In a technical solution of the above-mentioned system for segmenting the lung lobe region in the lung image, the lung lobe region segmentation module includes:

[0029] a lobe region and lobe label acquisition unit, configured to use a convolutional neural network and perform lobe segmentation on the lung image according to the lung region and the corrected fissure region, obtain the lobe region of each lobe in the lung image and determine the lobe label of each lobe region;

[0030] An image connected region acquisition unit, configured to respectively acquire the largest image connected region in each lung lobe region, and acquire other image connected regions in the lung region except the largest image connected region;

[0031] An image connected area processing unit is configured to determine, for each of the other image connected areas, the largest image connected area that is closest to the current other image connected area, take the current other image connected area as a part of the lobe area to which the largest image connected area belongs, and set a corresponding lobe label for the current other image connected area.

[0032] In a technical solution of the above-mentioned system for segmenting the lung lobe area in the lung image, the point cloud neural network model is a model based on the PointNet network.

[0033] In a third aspect, a control device is provided, which includes a processor and a storage device, wherein the storage device is suitable for storing multiple program codes, and the program codes are suitable for being loaded and run by the processor to execute the method for segmenting the lung lobe area in the lung image described in any one of the technical solutions of the method for segmenting the lung lobe area in the lung image.

[0034] In a fourth aspect, a computer-readable storage medium is provided, wherein a plurality of program codes are stored therein, wherein the program codes are suitable for being loaded and run by a processor to execute the method for segmenting the lung lobe area in the lung image as described in any one of the technical solutions of the method for segmenting the lung lobe area in the lung image.

[0035] The above one or more technical solutions of the present invention have at least one or more of the following beneficial effects:

[0036] In the technical solution of the present invention, the lung area and the fissure area in the lung image can be obtained, and the three-dimensional coordinates of the fissure area in the lung image can be corrected by using a point cloud neural network model to remove the noise in the fissure area and complete the missing fissure area. The lung image is segmented into lobes according to the lung area and the corrected fissure area to obtain the lobe area. Through the above configuration, the present invention can correct the three-dimensional coordinates of the fissure area in the lung image for lung images with unclear fissures, accessory fissures, and pulmonary fibrosis, so as to remove the noise in the fissure area and complete the missing fissure area, so that the prediction of the fissure area is more complete and accurate, and further the lobe area obtained by lobe segmentation according to the corrected fissure area is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The disclosure of the present invention will become more easily understood with reference to the accompanying drawings. It is easy for those skilled in the art to understand that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. Among them:

[0038] Figure 1 is a schematic flow chart of main steps of a method for segmenting a lung lobe region in a lung image according to an embodiment of the present invention;

[0039] Figure 2 It is a schematic diagram of the result of obtaining the lung fissure area in the lung image with interference using only the convolutional neural network in the prior art;

[0040] Figure 3 is a schematic diagram of a network structure of a model based on a Point Net network according to an implementation of an embodiment of the present invention;

[0041] Figure 4 is a schematic diagram of two-dimensional comparison between a pulmonary fissure region and a pulmonary lobe region without and with three-dimensional coordinate correction of the pulmonary fissure region according to an implementation manner of an embodiment of the present invention;

[0042] Figure 5 is with Figure 4 A three-dimensional comparison schematic diagram corresponding to the two-dimensional comparison schematic diagram;

[0043] Figure 6 is a schematic flow chart of main steps of a method for segmenting a lung lobe region in a lung image according to an implementation of an embodiment of the present invention;

[0044] Figure 7 It is a main structural block diagram of a system for segmenting a lung lobe region in a lung image according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] Some embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0046] In the description of the present invention, "module" and "processor" may include hardware, software or a combination of the two. A module may include hardware circuits, various suitable sensors, communication ports, and memories, and may also include software parts, such as program codes, or a combination of software and hardware. The processor may be a central processing unit, a microprocessor, an image processor, a digital signal processor, or any other suitable processor. The processor has data and / or signal processing functions. The processor may be implemented in software, hardware, or a combination of the two. Non-temporary computer-readable storage media include any suitable medium that can store program codes, such as a magnetic disk, a hard disk, an optical disk, a flash memory, a read-only memory, a random access memory, and the like. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The term "at least one A or B" or "at least one of A and B" has a similar meaning to "A and / or B", and may include only A, only B, or A and B. The singular terms "one" and "the" may also include plural forms.

[0047] The lung is one of the important organs of the human body. The human lung includes the left lung and the right lung. The left lung is divided into the upper and lower lobes by the oblique fissure, and the right lung is divided into the upper, middle and lower lobes by the horizontal fissure and the oblique fissure. The oblique fissure of the left lung, the horizontal fissure and the oblique fissure of the right lung are collectively referred to as the fissure. Among the current technical methods for segmenting the lobe region in lung images, the lobe region segmentation method based on the convolutional neural network has better performance, especially the two-step method, that is, first segmenting the fissure region, and then segmenting the lobe region according to the fissure region. In this way, the segmentation results of the lobe region near the fissure region are more refined. This is because the fissure region is the physical boundary between the lobe regions and is the basis for the division of the lung region.

[0048] However, the fissure area is not always obvious and easy to obtain. The fissure area of ​​some patients' lungs is blurred due to lesions and other factors. Some patients have accessory fissures or pulmonary fibrosis, which causes the accessory fissures and pulmonary fibrosis to produce noise similar to the fissure area. These will cause the convolutional neural network to misjudge the fissure area, and the misjudgment of the fissure area will further lead to subsequent segmentation errors of the lung lobe area. Among them, accessory fissure refers to the additional pleural structure that segments a part of lung tissue. See the attached Figure 2 , Figure 2is a schematic diagram of the result of obtaining the lung fissure area in the lung image with interference using only the convolutional neural network in the prior art, where Figure 2 The upper left picture shows the right lung. Figure 2 The image on the right is a schematic diagram of the lung fissure results obtained in the right lung using only the convolutional neural network. Figure 2 The right image shows the right lung fissure results obtained using only the convolutional neural network. Figure 2 The white area in the right image can still be distinguished, but the horizontal fissure is almost absent. In this case, the missing lung fissure will cause errors in the subsequent segmentation of the lung lobe area, such as Figure 2 As shown in the lower left picture, the segmentation of the upper and lower lobes of the right lung is obviously unreasonable.

[0049] In an embodiment of the present invention, a method for segmenting a lung lobe region in a lung image is provided to solve the above problem.

[0050] See attached Figure 1 , Figure 1 FIG. 1 is a flow chart showing the main steps of a method for segmenting a lung lobe region in a lung image according to an embodiment of the present invention. Figure 1 As shown, the method for segmenting the lung lobe region in the lung image in the embodiment of the present invention mainly includes the following steps S101 to S103.

[0051] Step S101: Acquire a lung region in a lung image and acquire a lung fissure region in the lung region.

[0052] In this embodiment, a lung region in a lung image may be acquired, and a lung fissure region in the lung region may be acquired based on the acquired lung region.

[0053] In one embodiment, the lung image may be a CT (Computed Tomography) image of the lung.

[0054] In one embodiment, a traditional image processing method can be used, such as the method disclosed in the authorized patent with the patent name: A method for automatic segmentation of lung parenchyma in CT images, application number: 201510426463.9, and publication number: CN104992445A, to obtain the lung area from the CT image of the lung.

[0055] In one embodiment, a deep learning method, such as the method disclosed in the document "Skourt BA, Hassani AE, Majda A. Lung CT Image Segmentation Using Deep Neural Networks [J]. Procedia Computer Science, 2018, 127: 109-113.", can be used to obtain the lung area from the lung CT image.

[0056] In one embodiment, a conventional image processing method may be used to first enhance the possible lung regions in the lung region, and then extract the lung fissure regions from the lung region by using a priori knowledge of the thin film 3D structure of the lung fissure region through methods such as energy functions, so as to obtain the lung fissure regions. The energy function refers to a function for classifying images according to the degree of difference between elements in the image, that is, the degree of difference between elements in the image is regarded as the energy between the elements in the image.

[0057] In one embodiment, a deep learning method can be used, such as the method disclosed in the document "Sarah E Gerard, Taylor JPatton, Gary E Christensen, John E Bayouth, Joseph M Reinhardt. FissureNet: A Deep Learning Approach For Pulmonary Fissure Detection in CT Images [J]. IEEE TransMed Imaging. 2019 Jan; 38(1): 156-166." First, the approximate area of ​​the pulmonary fissure in the lung area, that is, the ROI (Region Of Interest) area, is identified, and then the position of the pulmonary fissure area in the ROI area is accurately segmented.

[0058] Step S102: using a point cloud neural network model to correct the three-dimensional coordinates of the pulmonary fissure region in the lung image, so as to remove noise in the pulmonary fissure region and complete the missing pulmonary fissure region.

[0059] In this embodiment, a point cloud neural network model can be used to correct the three-dimensional coordinates of the pulmonary fissure region in the lung image to remove the noise in the pulmonary fissure region and complete the missing pulmonary fissure region. That is, by correcting the three-dimensional coordinates, the noise caused by accessory fissure, blood vessels and trachea, pulmonary fibrosis, etc. in the pulmonary fissure region is removed, and the missing pulmonary fissure region caused by lung lesions, etc. is completed.

[0060] In one implementation, the point cloud neural network model is a model based on the PointNet network.

[0061] In one embodiment, three point cloud neural network models can be trained for the horizontal fissure of the right lung, the oblique fissure of the right lung, and the oblique fissure of the left lung, respectively, and the three-dimensional coordinates of the fissure area in the lung image can be corrected by applying the corresponding point cloud neural network models respectively.

[0062] Step S103: performing lung lobe segmentation on the lung image according to the lung region and the corrected lung fissure region to obtain the lung lobe region.

[0063] In this embodiment, the lung image may be segmented into lobes according to the lung region obtained in step S101 and the corrected pulmonary fissure region obtained in step S102 to obtain the pulmonary lobe region.

[0064] In one embodiment, the lung lobe region can be obtained according to the method disclosed in the document “Pulmonary Lobe Segmentation Using A Sequence of Convolutional Neural Networks For Marginal Learning, 2019 IEEE16th International Symposium on Biomedical Imaging (ISBI 2019)”.

[0065] Based on the above steps S101-S103, the present invention can obtain the lung area and the fissure area in the lung image, and use the point cloud neural network model to correct the three-dimensional coordinates of the fissure area in the lung image to remove the noise in the fissure area and complete the missing fissure area, and perform lobe segmentation on the lung image according to the lung area and the corrected fissure area to obtain the lobe area. Through the above configuration, the present invention can correct the three-dimensional coordinates of the fissure area in the lung image for lung images with unclear fissures, accessory fissures, and pulmonary fibrosis, so as to remove the noise in the fissure area and complete the missing fissure area, so that the prediction of the fissure area is more complete and accurate, and further the lobe area obtained by lobe segmentation according to the corrected fissure area is more accurate.

[0066] Steps S101 to S103 are further described below.

[0067] In one embodiment, step S101 may include steps S1011 to S1013, and a UNet-like convolutional neural network may be used to obtain a lung region from a CT image of the lung by following the steps S1011 to S1013:

[0068] Step S1011: pre-processing the CT image of the lungs.

[0069] In this embodiment, the original lung CT image is input, the window width is set to 1000, and the window level is set to -500 to achieve preprocessing of the lung CT image. Among them, the window width and the window level are parameters in the key technology window technology in the field of medical imaging. The key technology window technology is a display technology used to observe normal tissues or lesions of different densities in CT examination.

[0070] Step S1012: segmenting the lung region of the preprocessed lung CT image.

[0071] In this embodiment, the two-dimensional UNet-like convolutional neural network model can be trained to obtain a trained two-dimensional UNet-like convolutional neural network model, which includes 4 downsampling structures and 4 upsampling structures. The trained two-dimensional UNet-like convolutional neural network model can be used to scan the two-dimensional transverse images of the CT images of the lungs one by one. The upsampling structure of the two-dimensional UNet-like convolutional neural network model uses a 1×1 convolution kernel. After the convolution operation, the convolution results of the two channels of the two-dimensional UNet-like convolutional neural network model are obtained, and the convolution results are classified by the Softmax function to obtain two output results of the two-dimensional UNet-like convolutional neural network model, and the two output results are background and lung areas. The lung areas in the output results of each two-dimensional transverse image are merged to obtain a three-dimensional image, which is the segmentation result of the lung area. Among them, downsampling is to reduce the image so that the image conforms to the size of the actual area, and generate a thumbnail of the corresponding image; upsampling is to enlarge the image so that it can be displayed on a display device with higher resolution. The two-dimensional cross-sectional image of the CT image of the lung refers to the two-dimensional image of each horizontal plane in the CT image of the lung. The CT image of the lung is obtained by merging all the two-dimensional cross-sectional images. The convolution kernel is when an input image is given, and the pixels in a small area of ​​the input image are weighted averaged to become each corresponding pixel in the output image, where the weight is defined by a function, which is called the convolution kernel. The Softmax function refers to a normalized exponential function, which is mostly used in classification problems. It can be understood by those skilled in the art that when training a two-dimensional UNet-like convolutional neural network model, a model training method commonly used in the field of neural networks can be used. For the sake of simplicity, the principle of the model training method will not be described in detail here.

[0072] Step S1013: Post-process the segmentation result of the lung region to obtain the final result of the lung region.

[0073] In this embodiment, for the segmentation result of the lung area obtained in step S1012, the image connected area of ​​the segmentation result is obtained, and the largest image connected area in the image connected area is further obtained, and the final result of the lung area is obtained according to the largest image connected area. The image connected area is an image area composed of foreground pixels with the same pixel value and adjacent positions in the image, and obtaining the image connected area means finding and marking each connected area from the image.

[0074] In one embodiment, step S101 may include steps S1014 to S1016, and a 3DUNet-like convolutional neural network may be used to obtain a pulmonary fissure region from a lung region by following the steps S1014 to S1016:

[0075] Step S1014: performing image preprocessing on the lung area.

[0076] In this embodiment, the minimum circumscribed rectangle of the acquired lung region can be taken, the pixels of the lung region can be adjusted to 256×256×256, the window width can be set to 1000, and the window level can be set to -500 to achieve preprocessing of the lung region.

[0077] Step S1015: segment the preprocessed lung area into lung fissure areas.

[0078] In this embodiment, a 3DUNet-like convolutional neural network model can be applied to segment the pulmonary fissure area. The 3DUNet-like convolutional neural network model can be trained to obtain a trained 3DUNet-like convolutional neural network model, which includes 4 downsampling results and 4 upsampling structures. The preprocessed lung area image is input into the trained 3DUNet-like convolutional neural network model. The upsampling structure uses a 1×1 convolution kernel, and the convolution results of 2 channels are obtained through the convolution operation. The convolution results are classified by the Softmax function to obtain two output results, and the two output results are the segmentation results of the background and the pulmonary fissure area, respectively.

[0079] Step S1016: Post-process the segmentation result of the lung fissure region to obtain the lung fissure region in the lung region.

[0080] In this embodiment, the segmentation result of the lung fissure region obtained in step S1015 is subjected to image expansion, and then the image connected area of ​​the expanded segmentation result is calculated, and only the two largest image connected areas in the obtained image connected areas are retained, that is, the lung fissure areas in the lung area are obtained, wherein the two largest image connected areas correspond to the lung fissure areas of the right lung and the lung fissure areas of the left lung, respectively. Image expansion refers to adding pixel values ​​at the edge of the image so that the pixel values ​​of the entire image are expanded, and the boundary of the image is expanded after the image expansion.

[0081] In one implementation of the embodiment of the present invention, step S102 may include the following steps S1021 to S1024:

[0082] Step S1021: using training samples and sample labels to perform network training on the point cloud neural network model, wherein the training samples include the coordinate values ​​of the x, y and z coordinates in the three-dimensional coordinates of the first type of lung fissure points without missing coordinate data, and the coordinate values ​​of the x and y coordinates in the three-dimensional coordinates of the second type of lung fissure points, and the sample labels are the coordinate values ​​of the z coordinates in the three-dimensional coordinates of the second type of lung fissure points; the second lung fissure point acquisition method is: randomly sampling the first type of lung fissure points to obtain a first type of lung fissure point as a first target lung fissure point; according to the first target lung fissure point, obtaining multiple first type of lung fissure points whose coordinate values ​​are continuous with the first target lung fissure point in three-dimensional space from other first type of lung fissure points as second target lung fissure points; and the first target lung fissure point and the second target lung fissure point form a second type of lung fissure point.

[0083] In this embodiment, the point cloud neural network model can be iteratively trained by the following steps:

[0084] The training samples and sample labels are simultaneously input into the point cloud neural network model to obtain the coordinate prediction value of the z coordinate in the three-dimensional coordinates of the second type of pulmonary fissure point pre-output by the point cloud neural network model; the model loss of the point cloud neural network model is calculated according to the coordinate prediction value and the coordinate value of the z coordinate in the three-dimensional coordinates of the second type of pulmonary fissure point in the sample label, the gradient of the model parameters of the point cloud neural network model is calculated according to the model loss, and the model parameters of the point cloud neural network model are updated according to the gradient back propagation to complete an iterative training. The point cloud neural network model is iteratively trained for multiple times until the point cloud neural network model meets the convergence condition and stops training to obtain a trained point cloud neural network model. Among them, the convergence condition is that the number of iterative training reaches a preset number threshold or the model loss is less than the preset loss threshold.

[0085] In one embodiment, the first type of lung fissure points may be all the lung fissure points in the lung fissure region in the lung image. The first type of lung fissure points may be randomly sampled, that is, a lung fissure point may be randomly selected from the first type of lung fissure points as the first target lung fissure point. The method for obtaining the second target lung fissure point according to the first target lung fissure point may be, according to the first target lung fissure point, a three-dimensional closed surface containing the first target lung fissure point may be randomly determined, and all the first type of lung fissure points in the three-dimensional closed surface and other first type of lung fissure points on the three-dimensional closed surface excluding the first target lung fissure point may be obtained as the second target lung fissure point. The method for obtaining the second target lung fissure point according to the first target lung fissure point may also be, according to the first target lung fissure point, a three-dimensional closed surface containing the first target lung fissure point may be obtained, and all the first type of lung fissure points on the three-dimensional closed surface and all other first type of lung fissure points on the three-dimensional closed surface excluding the first target lung fissure point may be obtained as the second target lung fissure point.

[0086] Step S1022: Use the trained point cloud neural network model to obtain the coordinate prediction value of the z coordinate in the three-dimensional coordinates of each pulmonary fissure point in the pulmonary fissure area.

[0087] In this embodiment, the point cloud neural network model trained in step S1021 can be used to respectively obtain the coordinate prediction value of the z coordinate in the three-dimensional coordinates of each pulmonary fissure point in the pulmonary fissure area.

[0088] Specifically, the coordinate prediction value of the z coordinate in the three-dimensional coordinates of each pulmonary fissure point in the pulmonary fissure region is obtained by the following steps:

[0089] The coordinate values ​​of the x, y and z coordinates in the three-dimensional coordinates of a part of the pulmonary fissure points in the pulmonary fissure area (hereinafter referred to as the third type of pulmonary fissure points for simplicity of description) are obtained, and the coordinate values ​​of the x and y coordinates in the three-dimensional coordinates of a part of the pulmonary fissure points in the pulmonary fissure area (hereinafter referred to as the fourth type of pulmonary fissure points for simplicity of description) are obtained. The coordinate values ​​of the x, y and z coordinates in the three-dimensional coordinates of the third type of pulmonary fissure points and the coordinate values ​​of the x and y coordinates in the three-dimensional coordinates of the fourth type of pulmonary fissure points are simultaneously input into the trained point cloud neural network model, and the point cloud neural network model can output the coordinate prediction value of the z coordinate in the three-dimensional coordinates of the fourth type of pulmonary fissure points. By repeatedly executing the above steps, the coordinate prediction value of the z coordinate in the three-dimensional coordinates of each pulmonary fissure point in the pulmonary fissure area is obtained. Among them, the acquisition method of the third type of pulmonary fissure points can be random sampling with replacement, which means that after randomly sampling a part of the pulmonary fissure points in the pulmonary fissure area, the sampled pulmonary fissure points are placed back to the pulmonary fissure points in the original pulmonary fissure area, that is, after random sampling with replacement, the total number of pulmonary fissure points in the pulmonary fissure area remains unchanged. The fourth type of lung fissure points may be obtained by a random sampling method commonly used in the art, which will not be described here for simplicity.

[0090] Step S1023: For each pulmonary fissure point, if the coordinate value of the z coordinate of the current pulmonary fissure point is missing, the coordinate prediction value is set to the coordinate value of the z coordinate of the current pulmonary fissure point.

[0091] In this embodiment, it is determined whether the coordinate value of the z coordinate of the current pulmonary fissure point is missing. If so, the coordinate prediction value can be set as the coordinate value of the z coordinate of the current pulmonary fissure point.

[0092] Step S1024: If the current pulmonary fissure point is not missing the coordinate value of the z coordinate and the actual coordinate value of the z coordinate is inconsistent with the predicted coordinate value, the actual coordinate value is replaced with the predicted coordinate value.

[0093] In this embodiment, if the current pulmonary fissure point does not lack the coordinate value of the z coordinate, but the actual coordinate value of the z coordinate is inconsistent with the predicted coordinate value, the actual coordinate value of the z coordinate can be replaced with the predicted coordinate value.

[0094] In one implementation, step S102 may include steps S1025 to S1026, see the attached Figure 3 , Figure 3 is a schematic diagram of the network structure of a model based on a PointNet network according to an implementation of an embodiment of the present invention. Taking the horizontal fissure of the right lung as an example, the three-dimensional coordinates of the horizontal fissure of the right lung are corrected through the following steps S1025 to S1026:

[0095] Step S1025: Preprocess the lung fissure area.

[0096] In this embodiment, the corresponding area can be cut off from the lung fissure area of ​​the right lung according to the minimum circumscribed rectangle of the lung area of ​​the right lung. Then the pixels of the cut off area are adjusted to 160×160×128, and the three-dimensional coordinates of all predicted lung fissure points of the horizontal fissure of the right lung are obtained in the 160×160×128 area. There are a total of M lung fissure points, and the obtained three-dimensional coordinate matrix can be recorded as (M, 3). The three-dimensional coordinate matrix is ​​sampled with replacement, and 1024 lung fissure points are sampled, which is recorded as (N, 3), and N=1024.

[0097] Step S1026: Correct the three-dimensional coordinates of all lung fissure points.

[0098] In this embodiment, the network structure of the model based on the Point Net network is as follows: Figure 3 As shown in the figure, B is batch_size, the number of samples in each batch of model training. Its value is affected by the GPU (Graphics Processing Unit). If the GPU allows, it can be any non-zero natural number. The network structure of the model based on the PointNet network is as follows: The first type of lung split point matrix B×Nin ×D in and the second type of lung split point matrix B×N un ×D un , where B represents the number of samples, N in Indicates the number of first-class lung fissure points (e.g. 1024), D in represents the dimension of the first type of lung fissure point (3), N un represents the number of second-type pulmonary fissure points (1), D un The dimension (2) of the second type of lung fissure points is represented, that is, the input lung fissure points (Input points) are the first type of lung fissure point matrix is ​​B×1024×3, and the second type of lung fissure point matrix is ​​B×1×2. Three fully connected operations (mlp, Multi-layer Perceptron, i.e., an artificial neural network containing a multi-layer structure) are performed on the dimensions of the first type of lung fissure point matrix and the second type of lung fissure point matrix respectively. Each fully connected layer can include a connected BN normalization layer and a ReLu activation function layer. The number of channels of the three fully connected layers is 64, 128, and 1024 respectively. After three fully connected operations, the first type of lung fissure point matrix is ​​converted to B×1024×1024; the second type of lung fissure point matrix is ​​converted to B×1×1024. After three fully connected operations, the first type of lung split point matrix and the second type of lung split point matrix can be pooled on the intermediate dimensions of the feature matrix B×1024×1024 and B×1×1024 respectively through the maximum pooling layer (Max pooling), and two B×1×1024 feature matrices are obtained. The last dimensions of the two B×1×1024 feature matrices are concatenated to obtain a matrix of dimension B×2048; the B×2048 matrix is ​​operated through two layers of fully connected layers to obtain a B×1 matrix, where the number of channels of the two fully connected layers is 512 and 256 respectively, and each fully connected layer can include a connected BN normalization layer and a ReLu activation function layer. The B×1 matrix is ​​the coordinate prediction value of the z coordinate of the second type of lung split point with missing coordinate data.

[0099] When training a model based on the PointNet network, the second type of lung fissure points can be obtained by randomly obtaining a number of continuous lung fissure points in the 160×160×128 lung fissure area as training samples, and the remaining lung fissure points are used as the first type of lung fissure points. The loss function of the training process uses L2Loss (mean square error MSE, Mean Square Error), uses the Adam optimizer, the learning rate is 1e-4, the number of samples is 4, and the iteration is 100 epochs.

[0100] When testing the model based on the PointNet network, for a 160×160×128 lung fissure region sample, all xy coordinates of the lung fissure points in the lung fissure region sample can be predicted, that is, the number of prediction points required is 160x160=25600. Many of these points are not lung fissure points, so it is necessary to obtain the prediction points in the lung region among the 25600 prediction points. These prediction points in the lung region are the corrected lung fissure points. The corrected lung fissure region is obtained based on the corrected lung fissure points, and the corrected lung fissure region is adjusted to the original size of the CT image of the lung, which is the final corrected lung fissure region.

[0101] In one implementation of the embodiment of the present invention, step S103 may further include the following steps S1031 to S1033:

[0102] Step S1031: Use a convolutional neural network to perform lobe segmentation on the lung image according to the lung area and the corrected lung fissure area, obtain the lobe area of ​​each lobe in the lung image and determine the lobe label of each lobe area.

[0103] In this embodiment, the lung area and the corrected lung fissure area can be used as input, and a convolutional neural network can be applied to perform lobe segmentation of the lung image to obtain the lobe areas of the five lobes in the lung image and the lobe labels corresponding to the five lobe areas.

[0104] Step S1032: respectively obtain the largest image connected area in each lung lobe region, and obtain other image connected areas in the lung region except the largest image connected area.

[0105] In this embodiment, the largest image connected area and other image connected areas in the lung area except the largest image connected area can be obtained for each lung lobe area.

[0106] Step S1033: For each other image connected area, determine the largest image connected area closest to the current other image connected area, take the current other image connected area as a part of the lobe area to which the largest image connected area belongs, and set a corresponding lobe label for the current other image connected area.

[0107] In this embodiment, the largest image connected area closest to each other image connected area obtained in step S1032 can be determined, and according to the proximity principle, the same lung lobe label as the largest image connected area closest to it is set for each other image connected area.

[0108] In this embodiment, the image connected region can be obtained using an image connected region obtaining method commonly used in the art. For simplicity of description, the method for obtaining the image connected region will not be described in detail herein.

[0109] In one embodiment, a 3D convolutional neural network may be used to segment the lung lobes in the lung image. Step S103 may include the following steps S1033 to S1035, and the lung lobes are segmented on the lung image through steps S1033 to S1035:

[0110] S1033: Perform image preprocessing on the lung region image and the corrected lung fissure region image.

[0111] In this embodiment, the minimum circumscribed rectangle of the lung area can be taken from the original lung CT image, and then the pixels of the minimum circumscribed rectangle are adjusted to 256×256×256, and the window width is set to 1000 and the window level is set to -500; the minimum circumscribed rectangle of the lung area image and the corrected lung fissure area is also taken, and then the pixels of the minimum circumscribed rectangle are adjusted to 256×256×256 to achieve image preprocessing, and the image preprocessed lung CT image, lung area and corrected lung fissure area are used as input of the 3D convolutional neural network, that is, the input matrix size of the 3D convolutional neural network is 256×256×256×3.

[0112] S1034: training a 3D convolutional neural network, and applying the trained 3D convolutional neural network to perform lung lobe segmentation.

[0113] In this embodiment, the network model based on the 3D convolutional neural network can be trained, wherein the network model based on the 3D convolutional neural network includes 4 downsampling structures and 4 upsampling structures. The upsampling structure of the trained network model based on the 3D convolutional neural network uses a 1×1 convolution kernel to perform a convolution operation to obtain the convolution results of 6 channels, and is classified by the Softmax function to obtain 6 segmentation results, namely the background and 5 lung lobes.

[0114] Step S1035: post-process the lung lobe segmentation results.

[0115] In this embodiment, the image connected areas of the five lobes can be respectively acquired to post-process the pulmonary lobe segmentation results, so as to remove the noise of the pulmonary lobe segmentation results and obtain the final pulmonary lobe segmentation results.

[0116] See attached Figure 4 and attached Figure 5 , Figure 4is a schematic diagram of two-dimensional comparison between a pulmonary fissure region and a pulmonary lobe region without and with three-dimensional coordinate correction of the pulmonary fissure region according to an implementation manner of an embodiment of the present invention; Figure 5 is with Figure 4 The three-dimensional comparison diagram corresponding to the two-dimensional comparison diagram of . Among them, Figure 4 The upper left picture in the middle is the uncorrected pulmonary fissure area; the upper right picture is the corrected pulmonary fissure area; the lower left picture is the pulmonary lobe area obtained by performing pulmonary lobe segmentation based on the uncorrected pulmonary fissure area; the lower right picture is the pulmonary lobe area obtained by performing pulmonary lobe segmentation based on the corrected pulmonary fissure area. Figure 5 The upper left picture is a 3D display of the uncorrected fissure area; the upper right picture is a 3D display of the corrected fissure area; the lower left picture is a 3D display of the lobe area obtained by segmenting the lobes according to the uncorrected fissure area; the lower right picture is a 3D display of the lobe area obtained by segmenting the lobes according to the corrected fissure area. Figure 4 and Figure 5 As shown, the method of segmenting the lobe region using the corrected pulmonary fissure region is more accurate in obtaining the lobe region.

[0117] One implementation method, see the attached Figure 6 , Figure 6 FIG. 1 is a flow chart showing the main steps of a method for segmenting a lung lobe region in a lung image according to an implementation of an embodiment of the present invention. Figure 6 As shown, the method for segmenting the lung lobe region in the lung image may include the following steps S201 to S205:

[0118] Step S201: Acquire a lung region from a CT image of the lung;

[0119] Step S202: Acquire a pulmonary fissure region from the lung region;

[0120] In this embodiment, the methods of step S201 and step S202 are similar to the method described in the aforementioned step S101, and are not described again for simplicity.

[0121] Step S203: applying a point cloud neural network to correct the three-dimensional coordinates of the acquired pulmonary fissure region;

[0122] In this embodiment, the method of step S203 is similar to the method described in the aforementioned step S102, and for the sake of simplicity, it will not be repeated here.

[0123] Step S204: performing lung lobe segmentation on the CT image of the lungs;

[0124] In this embodiment, the method described in step S204 is similar to the method described in the aforementioned step S1031, and for the sake of simplicity, it is not repeated here.

[0125] Step S205: Post-processing of the lung lobe segmentation results.

[0126] In this embodiment, the method described in step S205 is similar to the methods described in the aforementioned steps S1032 and S1033, and for the sake of simplicity, they are not repeated here.

[0127] It should be pointed out that although the various steps in the above embodiments are described in a specific order, those skilled in the art can understand that in order to achieve the effects of the present invention, different steps do not have to be performed in such an order. They can be performed simultaneously (in parallel) or in other orders. These changes are within the scope of protection of the present invention.

[0128] Furthermore, the present invention also provides a system for segmenting lung lobe regions in lung images.

[0129] See attached Figure 7 , Figure 7 FIG. 1 is a main structural block diagram of a segmentation system for a lung lobe region in a lung image according to an embodiment of the present invention. Figure 7 As shown, the segmentation system of the lung lobe region in the lung image in the embodiment of the present invention may include a lung fissure region acquisition module, a lung fissure region correction module and a lung lobe region segmentation module. In this embodiment, the lung fissure region acquisition module may be configured to acquire the lung region in the lung image and acquire the lung fissure region in the lung region. The lung fissure region correction module may be configured to use a point cloud neural network model to correct the three-dimensional coordinates of the lung fissure region in the lung image to remove the noise of the lung fissure region and complete the missing lung fissure region. The lung lobe region segmentation module may be configured to perform lung lobe segmentation on the lung image according to the lung region and the corrected lung fissure region to obtain the lung lobe region.

[0130] In one embodiment, the lung fissure region correction module may include a point cloud neural network training unit, a z coordinate prediction value acquisition unit, a z coordinate setting unit, and a z coordinate replacement unit. In this embodiment, the point cloud neural network training unit may be configured to use training samples and sample labels to perform network training on the point cloud neural network model, wherein the training samples include the coordinate values ​​of the x, y, and z coordinates in the three-dimensional coordinates of the first type of lung fissure points without missing coordinate data, and the coordinate values ​​of the x and y coordinates in the three-dimensional coordinates of the second type of lung fissure points, and the sample labels are the coordinate values ​​of the z coordinates in the three-dimensional coordinates of the second type of lung fissure points; the second lung fissure point acquisition method is: randomly sampling the first type of lung fissure points to obtain a first type of lung fissure point as the first target lung fissure point; according to the first target lung fissure point, obtain multiple first type of lung fissure points with continuous coordinate values ​​in three-dimensional space with the first target lung fissure point from other first type of lung fissure points as the second target lung fissure point; the first target lung fissure point and the second target lung fissure point form a second type of lung fissure point. The z-coordinate prediction value acquisition unit can be configured to use the trained point cloud neural network model to respectively obtain the coordinate prediction value of the z coordinate in the three-dimensional coordinates of each lung fissure point in the lung fissure area. The z-coordinate setting unit can be configured to, for each lung fissure point, if the current lung fissure point lacks the coordinate value of the z coordinate, set the coordinate prediction value to the coordinate value of the z coordinate of the current lung fissure point. The z-coordinate replacement unit can be configured to replace the coordinate actual value with the coordinate prediction value if the current lung fissure point does not lack the coordinate value of the z coordinate and the coordinate actual value of the z coordinate is inconsistent with the coordinate prediction value.

[0131] In one embodiment, the lobe region segmentation module may include a lobe region and lobe label acquisition unit, an image connected region acquisition unit, and an image connected region processing unit. In this embodiment, the lobe region and lobe label acquisition unit may be configured to use a convolutional neural network and perform lobe segmentation on the lung image based on the lung region and the corrected fissure region to obtain the lobe region of each lobe in the lung image and determine the lobe label of each lobe region. The image connected region acquisition unit may be configured to respectively obtain the largest image connected region in each lobe region, and obtain other image connected regions in the lung region except the largest image connected region. The image connected region processing unit may be configured to determine the largest image connected region closest to the current other image connected region for each other image connected region, use the current other image connected region as a part of the lobe region to which the largest image connected region belongs, and set a corresponding lobe label for the current other image connected region.

[0132] In one embodiment, the point cloud neural network model can be a model based on the Point Net network.

[0133] The segmentation system of the lung lobe region in the lung image is used to perform Figure 1The embodiment of the method for segmenting the lung lobe area in the lung image shown in the figure has similar technical principles, technical problems solved and technical effects produced. Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working process and related instructions of the system for segmenting the lung lobe area in the lung image can refer to the contents described in the embodiment of the method for segmenting the lung lobe area in the lung image, which will not be repeated here.

[0134] It is understood by those skilled in the art that the present invention implements all or part of the processes in the method of the above embodiment, and can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium that can carry the computer program code. It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electric carrier signals and telecommunication signals.

[0135] Furthermore, the present invention also provides a control device. In an embodiment of a control device according to the present invention, the control device includes a processor and a storage device. The storage device can be configured to store a program for executing the method for segmenting the lobe region in the lung image of the above-mentioned method embodiment, and the processor can be configured to execute the program in the storage device, which includes but is not limited to executing the program for segmenting the lobe region in the lung image of the above-mentioned method embodiment. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The control device can be a control device device formed by various electronic devices.

[0136] Furthermore, the present invention also provides a computer-readable storage medium. In a computer-readable storage medium embodiment according to the present invention, the computer-readable storage medium can be configured to store a program for executing the method for segmenting the lobe region in the lung image of the above-mentioned method embodiment, and the program can be loaded and run by the processor to implement the above-mentioned method for segmenting the lobe region in the lung image. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium can be a storage device including various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present invention is a non-temporary computer-readable storage medium.

[0137] Further, it should be understood that since the setting of each module is only for illustrating the functional units of the device of the present invention, the physical devices corresponding to these modules may be the processor itself, or a part of the software in the processor, a part of the hardware, or a part of the combination of software and hardware. Therefore, the number of each module in the figure is only schematic.

[0138] It is understood by those skilled in the art that the modules in the device can be adaptively split or merged. Such splitting or merging of specific modules will not cause the technical solution to deviate from the principle of the present invention, and therefore, the technical solutions after splitting or merging will fall within the protection scope of the present invention.

[0139] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A method for segmenting a lung lobe region in a lung image, characterized in that: The method comprises: Acquire a lung region in a lung image and acquire a lung fissure region in the lung region; Using a point cloud neural network model to correct the three-dimensional coordinates of the pulmonary fissure region in the lung image to remove noise in the pulmonary fissure region and complete the missing pulmonary fissure region; Performing lung lobe segmentation on the lung image according to the lung region and the corrected lung fissure region to obtain a lung lobe region; The step of "correcting the three-dimensional coordinates of the pulmonary fissure region in the lung image using a point cloud neural network model to remove noise from the pulmonary fissure region and complete the missing pulmonary fissure region" specifically includes: The point cloud neural network model is trained using training samples and sample labels, wherein the training samples include the coordinate values ​​of the x, y and z coordinates in the three-dimensional coordinates of the first type of lung fissure points without missing coordinate data, and the coordinate values ​​of the x and y coordinates in the three-dimensional coordinates of the second type of lung fissure points, and the sample labels are the coordinate values ​​of the z coordinates in the three-dimensional coordinates of the second type of lung fissure points; the second type of lung fissure points are obtained by randomly sampling the first type of lung fissure points to obtain a first type of lung fissure point as a first target lung fissure point; according to the first target lung fissure point, a plurality of first type of lung fissure points whose coordinate values ​​are continuous with the first target lung fissure point in three-dimensional space are obtained from other first type of lung fissure points as second target lung fissure points; the first target lung fissure point and the second target lung fissure point form the second type of lung fissure point; Using the trained point cloud neural network model, respectively obtain the coordinate prediction value of the z coordinate in the three-dimensional coordinates of each pulmonary fissure point in the pulmonary fissure area; For each lung fissure point, if the coordinate value of the z coordinate of the current lung fissure point is missing, the coordinate prediction value is set to the coordinate value of the z coordinate of the current lung fissure point; If the current lung fissure point does not lack the coordinate value of the z coordinate and the actual coordinate value of the z coordinate is inconsistent with the predicted coordinate value, the actual coordinate value is replaced with the predicted coordinate value.

2. The method for segmenting the lung lobe region in the lung image according to claim 1, characterized in that: The step of "segmenting the lung image according to the lung region and the corrected lung fissure region to obtain the lung lobe region" specifically includes: Using a convolutional neural network and based on the lung area and the corrected lung fissure area, perform pulmonary lobe segmentation on the lung image to obtain the pulmonary lobe area of ​​each pulmonary lobe in the lung image and determine the pulmonary lobe label of each pulmonary lobe area; respectively obtaining the largest image connected area in each lung lobe region, and obtaining other image connected areas in the lung region except the largest image connected area; For each of the other image connected areas, determine the largest image connected area that is closest to the current other image connected area, take the current other image connected area as a part of the lung lobe area to which the largest image connected area belongs, and set a corresponding lung lobe label for the current other image connected area.

3. The method for segmenting the lung lobe region in the lung image according to claim 1, characterized in that: The point cloud neural network model is a model based on the PointNet network.

4. A system for segmenting lung lobe regions in lung images, characterized in that: The system comprises: A lung fissure region acquisition module, configured to acquire a lung region in a lung image and acquire a lung fissure region in the lung region; A lung fissure region correction module, which is configured to use a point cloud neural network model to correct the three-dimensional coordinates of the lung fissure region in the lung image to remove noise in the lung fissure region and complete the missing lung fissure region; A lung lobe region segmentation module, configured to perform lung lobe segmentation on the lung image according to the lung region and the corrected lung fissure region to obtain a lung lobe region; The lung fissure area correction module includes: A point cloud neural network training unit, configured to use training samples and sample labels to perform network training on a point cloud neural network model, wherein the training samples include coordinate values ​​of x, y and z coordinates in the three-dimensional coordinates of a first type of lung fissure point without missing coordinate data, and coordinate values ​​of x and y coordinates in the three-dimensional coordinates of a second type of lung fissure point, and the sample labels are coordinate values ​​of z coordinates in the three-dimensional coordinates of the second type of lung fissure point; a method for acquiring the second type of lung fissure point is as follows: randomly sampling the first type of lung fissure point to obtain a first type of lung fissure point as a first target lung fissure point; according to the first target lung fissure point, obtaining multiple first type of lung fissure points whose coordinate values ​​are continuous with the first target lung fissure point in three-dimensional space from other first type of lung fissure points as second target lung fissure points; and combining the first target lung fissure point and the second target lung fissure point to form the second type of lung fissure point; A z-coordinate prediction value acquisition unit is configured to respectively acquire a coordinate prediction value of a z-coordinate in the three-dimensional coordinates of each pulmonary fissure point in the pulmonary fissure region using a trained point cloud neural network model; A z-coordinate setting unit is configured to set the coordinate prediction value to the coordinate value of the z-coordinate of the current lung fissure point for each lung fissure point if the current lung fissure point lacks the coordinate value of the z-coordinate; A z-coordinate replacement unit is configured to replace the actual coordinate value with the predicted coordinate value if the current lung fissure point does not lack the coordinate value of the z-coordinate and the actual coordinate value of the z-coordinate is inconsistent with the predicted coordinate value.

5. The segmentation system of lung lobe region in lung image according to claim 4, characterized in that: The lung lobe region segmentation module includes: a lobe region and lobe label acquisition unit, configured to use a convolutional neural network and perform lobe segmentation on the lung image according to the lung region and the corrected fissure region, obtain the lobe region of each lobe in the lung image and determine the lobe label of each lobe region; An image connected region acquisition unit, configured to respectively acquire the largest image connected region in each lung lobe region, and acquire other image connected regions in the lung region except the largest image connected region; An image connected area processing unit is configured to determine, for each of the other image connected areas, the largest image connected area that is closest to the current other image connected area, take the current other image connected area as a part of the lobe area to which the largest image connected area belongs, and set a corresponding lobe label for the current other image connected area.

6. The segmentation system of lung lobe region in lung image according to claim 4, characterized in that: The point cloud neural network model is a model based on the PointNet network.

7. A control device, comprising a processor and a storage device, wherein the storage device is suitable for storing a plurality of program codes, characterized in that: The program code is suitable for being loaded and run by the processor to execute the method for segmenting the lung lobe region in the lung image according to any one of claims 1 to 3.

8. A computer-readable storage medium storing a plurality of program codes, characterized in that: The program code is suitable for being loaded and run by a processor to execute the method for segmenting the lung lobe region in a lung image according to any one of claims 1 to 3.

Images