Mathematical morphology-based CT image lung nodule automatic segmentation method and device

Abstract

The present application relates to the technical field of medical image processing, in order to solve the technical problem of high false positive caused by the existing detection segmentation method, the present application discloses a CT image lung nodule automatic segmentation method and device based on mathematical morphology, including the steps of obtaining original data, lung parenchyma segmentation, mathematical morphology preprocessing, high signal region extraction, mathematical morphology processing, false positive nodule exclusion and three-dimensional reconstruction. The mathematical morphology processing step performs a mathematical morphology opening operation on the high signal region with a structure element radius similar to the maximum intrapulmonary vessel radius, obtaining a potential nodule image region. The false positive nodule exclusion step removes false positive nodules from the potential nodule image region through geometric features and threshold processing. Mathematical morphology operation is fast, and can realize fast lung nodule detection and segmentation operation. Geometric features and threshold processing can effectively remove most false positive objects.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, and in particular to a CT image lung nodule automatic segmentation method and device based on mathematical morphology. BACKGROUND

[0002] A lung nodule refers to a circular or irregular lesion in the lung with a diameter less than or equal to 3 cm, which is manifested as a density-increased shadow in imaging. With the rapid development of CT scanning technology, the imaging resolution is getting higher and higher, and the volume of lung nodules that can be found in the reconstructed image is getting smaller and smaller, and the data volume of the image is rapidly increasing. Obviously, it is very difficult and time-consuming to rely on the naked eye to observe and find lung nodules. Assisting doctors to detect and segment lung nodules from CT images has gradually become the focus of people's research and attention.

[0003] The existing machine learning and deep learning detection and segmentation lung nodule method needs to rely on a large amount of standard data for training, and the traditional lung nodule detection and segmentation method relies on single features, resulting in too high false positives. SUMMARY

[0004] The purpose of the present application is to provide a CT image lung nodule automatic segmentation method and device based on mathematical morphology to solve the technical problem of too high false positives caused by the existing detection and segmentation method.

[0005] To achieve the above-mentioned purpose, the specific technical scheme of the CT image lung nodule automatic segmentation method based on mathematical morphology of the present application is as follows:

[0006] As a first aspect of the present application, a CT image lung nodule automatic segmentation method based on mathematical morphology is provided, comprising the following steps:

[0007] Step S1, obtaining original data;

[0008] Step S2, lung parenchyma segmentation, extracting the lung parenchyma region by threshold segmentation algorithm to obtain a lung parenchyma binary image;

[0009] Step S3, mathematical morphology preprocessing, performing a corrosion operation with a structure element radius of 1 on the lung parenchyma binary image, and then traversing the original data according to the original data and the lung parenchyma binary image to take out the pixels in the binary image that are not 0 to obtain lung parenchyma image data;

[0010] Step S4, high signal region extraction, using a mathematical morphology dilation reconstruction method to extract the high signal region in the lung parenchyma image data; the high signal region in the lung parenchyma image data is extracted by the mathematical morphology dilation reconstruction method, which limits the range of suspected lung nodules;

[0011] Step S5, mathematical morphology processing, a mathematical morphology open operation with a structural element radius similar to the maximum radius of the blood vessels in the lung is performed on the high signal region to obtain a potential nodule image region; this step effectively excludes most non-pulmonary nodule regions and reduces the interference of screening for true nodules;

[0012] Step S6, false positive nodule exclusion, false positive nodules are removed from the potential nodule image region through geometric features and threshold processing to generate a nodule binary image;

[0013] Step S7, three-dimensional reconstruction, a FlyingEdge three-dimensional reconstruction method is used to perform three-dimensional reconstruction on the pulmonary nodule region detected and segmented from the nodule binary image to obtain a surface model of the nodule.

[0014] Further, the step S6 includes the following steps:

[0015] Step S61, the potential nodule image region is divided into single objects according to connected regions, and each object corresponds to a suspected nodule;

[0016] Step S62, each object is traversed, and a first round of exclusion is performed based on geometric shapes;

[0017] Step S63, the center point of the suspected nodule region is extracted as a seed point for the remaining objects after the first round of traversal, a radius is preset, a threshold is used for inflation, the number of pixels of the suspected nodule region after inflation is counted, and if the ratio of the pixel number increment after inflation to the pixel number of the suspected nodule region before inflation is greater than a preset threshold range, the suspected nodule is determined to be a false positive nodule; this step uses the threshold inflation method to exclude the false positive nodule region at the intersection of the blood vessels formed after the open operation;

[0018] Step S64, the average value of all pixel CT values in the object is counted for the remaining objects after the second round of traversal, and if the average value is greater than -420 and / or less than -800, the suspected nodule is determined to be a false positive nodule.

[0019] Further, the step S1 includes the following steps:

[0020] Step S11, if the original data exist in stages, generating staging data according to the original data,

[0021] Step S12, selecting the required staging data for processing.

[0022] As a second aspect of the present application, a CT image pulmonary nodule automatic segmentation device based on mathematical morphology is provided, characterized in that it comprises the following modules:

[0023] An acquisition module is configured to acquire original data;

[0024] The segmentation module is configured to extract a lung parenchyma region by a threshold segmentation algorithm to obtain a lung parenchyma binary image.

[0025] The preprocessing module is configured to perform a corrosion operation with a structural element radius of 1 on the lung parenchyma binary image according to mathematical morphology, and then traverse the original data according to the original data and the lung parenchyma binary image, and take out pixels other than 0 in the binary image to obtain lung parenchyma image data.

[0026] The extraction module is configured to extract a high signal region in the lung parenchyma image data by using a mathematical morphology dilation reconstruction method.

[0027] The processing module is configured to perform a mathematical morphology opening operation with a structural element radius close to a maximum intrapulmonary vessel radius on the high signal region according to mathematical morphology to obtain a potential nodule image region.

[0028] The elimination module is configured to eliminate false positive nodules from the potential nodule image region by geometric features and threshold processing to generate a nodule binary image.

[0029] The three-dimensional reconstruction module is configured to perform three-dimensional reconstruction on the lung nodule region detected and segmented by using a FlyingEdge three-dimensional reconstruction method for the nodule binary image to obtain a surface model of the nodule.

[0030] Further, the elimination module comprises:

[0031] The division module is configured to divide the potential nodule image region into single objects according to connected regions, and each object corresponds to a suspected nodule.

[0032] The first judgment module is configured to traverse each object and perform a first round of elimination based on geometric shapes.

[0033] The second judgment module is configured to traverse the remaining objects in the first round, extract a center point of the suspected nodule region as a seed point, preset a radius, perform dilation by using a threshold, and count a pixel number of the suspected nodule region after the dilation. If a ratio of an increment of the pixel number after the dilation to a pixel number of the suspected nodule region before the dilation is greater than a preset threshold range, the suspected nodule is determined as a false positive nodule.

[0034] The third judgment module is configured to traverse the remaining objects in the second round, count an average value of all pixel CT values in the object, and if the average value is in a region greater than -420 and / or less than -800, the suspected nodule is determined as a false positive nodule.

[0035] Further, the acquisition module comprises:

[0036] The generation module is configured to generate staging data according to the original data if the original data exists.

[0037] Select the module and choose the required installment data for processing.

[0038] The automatic lung nodule segmentation method based on mathematical morphology in CT images provided by this invention has the following advantages:

[0039] A mathematical morphological dilation reconstruction method is employed to extract high-signal regions from lung parenchyma image data, limiting the range of suspicious pulmonary nodules. Based on mathematical morphology, an opening operation is performed on the high-signal regions with a structuring element radius similar to the radius of the largest intrapulmonary vessel, effectively excluding most non-pulmonary nodule areas and reducing interference in the screening of true nodules. This solves the problem of excessively high false positives caused by the reliance on single features in traditional pulmonary nodule detection and segmentation methods. Furthermore, a threshold dilation method is used to further exclude false positive nodule regions at vascular intersections formed after the opening operation, thus obtaining true positive nodules. Attached Figure Description

[0040] Figure 1 A schematic diagram illustrating the threshold expansion method for excluding false positives in lung nodules provided by the present invention;

[0041] Figure 2 This is a cross-sectional view of a CT image provided by the present invention;

[0042] Figure 3 This is a sagittal view of a CT image provided by the present invention;

[0043] Figure 4 This is a coronal view of a CT image provided by the present invention;

[0044] Figure 5 This invention provides a fusion display of high-signal regions extracted using a mathematical morphological dilation method and CT images.

[0045] Figure 6 A three-dimensional view of a binary image of a high-signal region extracted using a mathematical morphological dilation method, provided by this invention.

[0046] Figure 7 This is a view of the result after performing an opening operation on a high-signal region, as provided by the present invention.

[0047] Figure 8 This is a cross-sectional view of a lung nodule after three rounds of false positive exclusion, provided by the present invention.

[0048] Figure 9 In response to Figure 8 A magnified view of a portion of the text;

[0049] Figure 10 Binary image reconstruction diagram of lung nodule segmentation provided by the present invention;

[0050] Figure 11A CT image lung nodule automatic segmentation method flow chart provided by the present application;

[0051] Figure 12 A mathematical morphology processing flow chart provided by the present application. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application, and are not used to limit the present application.

[0053] Referring to Figures 1 to 12 The present application provides a CT image lung nodule automatic segmentation method based on mathematical morphology, comprising the following steps:

[0054] Step S1, data preparation: import the chest sequence CT image conforming to the DICOM (Digital Imaging and Communications in Medicine, medical digital imaging and communication) 3.0 standard into the program, referring to Figures 2 to 4 The imported image data can be viewed from three directions of transverse, sagittal and coronal, and displayed in three orthogonal views of transverse, sagittal and coronal. If the obtained data contains multiple stages, the data is first staged, and then the appropriate staged data is selected for processing. Define the input original data as ImageData.

[0055] Step S2, lung parenchyma segmentation: since lung nodules can only exist in the lung, in order to exclude the influence of other image regions on the lung nodule detection and segmentation process, after importing the CT image data, a threshold segmentation algorithm is used to extract the lung parenchyma region, exclude the non-lung region, limit the range of subsequent lung nodule detection and segmentation, and obtain a lung parenchyma binary image (image binarization processing is obtained, the gray value of the point on the image is converted to 0 or 1), recorded as LungMask. In the process of lung nodule detection and segmentation, the lung parenchyma region is first segmented, which can effectively reduce the interference of non-lung region on lung nodule detection, and improve the operation efficiency.

[0056] Step S3, mathematical morphology preprocessing: in order to reduce the influence of the surface pixels of the lung parenchyma image LungMask extracted in the previous step on the subsequent lung nodule detection and segmentation, a corrosion operation with a structure element radius of 1 is performed on the binary image LungMask of the lung parenchyma; then according to the original data ImageData and the lung parenchyma binary image LungMask, the ImageData data is traversed, and the pixels in LungMask that are not 0 are taken out to obtain the lung parenchyma image data, recorded as LungImage.

[0057] Referring toFigure 5 and Figure 6 The high signal region is extracted from the lung parenchyma image data based on a mathematical morphology dilation reconstruction method.

[0058] The operation can not only exclude the influence of non-lung regions in the original CT image, but also improve the calculation speed, so that the subsequent lung nodule detection and segmentation is only performed in the lung parenchyma.

[0059] Step S4, high signal region extraction: According to the performance of lung nodules in radiology, the lung nodule region on the CT image is a high-density shadow relative to the surrounding pixels. Based on this, after the lung parenchyma region is extracted, a mathematical morphology dilation reconstruction method is used to extract the high signal region in the lung parenchyma image data, and the extracted high signal region is marked as MaximaImage.

[0060] Step S5, mathematical morphology processing: The high signal region extracted in the previous step includes blood vessels and lung nodules and other high signal regions in the lung. In order to further eliminate other non-lung nodule regions, refer to Figure 7 , a mathematical morphology opening operation with a structural element radius similar to the maximum lung vessel radius is performed on the high signal region MaximaImage, and a potential nodule image region is obtained, which is marked as PotentialNodules.

[0061] Step S6, false positive nodule elimination: The potential nodule image region PotentialNodules output in the previous step includes real lung nodules and false positive nodules formed by the intersection of blood vessels after the opening operation. The real lung nodules need to be excluded. The false positive lung nodule elimination is performed according to the following steps:

[0062] Step S61, the potential nodule image region PotentialNodules is divided into single objects according to connected regions, and each object is a suspicious nodule;

[0063] Step S62, each suspicious nodule object is traversed to perform the first round of elimination. Since the nodule is generally a circular structure, first, based on the geometric shape, according to the aspect ratio, length-to-short-axis ratio, circularity rate, circularity diameter size (more than 3cm cannot be defined as a nodule), flatness and pixel number of the suspicious nodule object, the objects that do not meet the requirements are excluded.

[0064] Step S63, for the objects remaining after the first round of elimination, each suspicious nodule object is traversed to perform the second round of elimination. First, the center point of the suspicious nodule region is extracted as a seed point; then a radius is preset, and a threshold dilation method is used, refer to Figure 1, the left side of the figure is a true nodule, and the right side of the figure is a false nodule. When the left side of the figure is a true nodule, the object processed by the threshold inflation method has limited inflation pixels, and if it is a false positive case in the right side of the figure, due to the similar CT values of the pixels in the blood vessels, a considerable number of pixels will be inflated in the preset inflation area. Based on this observation, we can exclude the false positive nodules on the blood vessels generated by the opening operation of step S5; the high signal area extracted from the pulmonary parenchyma image data by the mathematical morphology inflation reconstruction method can accurately contain the area of the true nodule, and the mathematical morphology operation is fast, which can realize the fast lung nodule detection and segmentation operation;

[0065] Step S64, for the objects left after the second round of exclusion, each suspicious nodule is traversed again for the third round of exclusion. Mainly according to the average value of the pixel CT value in the nodule to exclude. For each suspicious nodule object, the average value of all pixel CT values in the object is calculated, and if the average value is greater than -420 and less than -800, it is determined as a false positive nodule and can be directly excluded, and other objects are retained. Through the three rounds of false positive exclusion method, most of the false positive objects can be effectively removed, especially by using the different results of the threshold inflation of different regions to exclude false positive lung nodules.

[0066] Referring to Figure 8 and Figure 9 After three rounds of false positive nodule exclusion, the final nodule binary image is generated, denoted as NodulesMask.

[0067] Step S7, three-dimensional reconstruction: for the nodule binary image NodulesMask, the FlyingEdge three-dimensional reconstruction method is used to perform three-dimensional reconstruction on the lung nodule region detected and segmented, to obtain the surface model of the nodule, denoted as NodulesMesh.

[0068] Referring to Figure 10 , the left part of the figure is the reconstructed lung nodule model, and the right part of the figure from top to bottom is the transverse, sagittal and coronal views.

[0069] At this point, the operation of the CT image automatic lung nodule detection and segmentation method based on mathematical morphology is completed, and the NodulesMesh object can be visualized or saved as an STL file output.

[0070] The application also provides a CT image lung nodule automatic segmentation device based on mathematical morphology, characterized by comprising the following modules:

[0071] The acquisition module is used to acquire the original data;

[0072] The segmentation module is used to extract the lung parenchyma region by a threshold segmentation algorithm to obtain a lung parenchyma binary image;

[0073] a preprocessing module configured to perform a corrosion operation with a radius of 1 on a lung parenchyma binary image according to mathematical morphology, and then traverse the original data according to the original data and the lung parenchyma binary image, and take out pixels other than 0 in the binary image to obtain lung parenchyma image data;

[0074] an extraction module configured to extract a high signal region in the lung parenchyma image data by using a mathematical morphology dilation reconstruction method;

[0075] a processing module configured to perform a mathematical morphology opening operation with a radius of a structural element similar to a maximum intrapulmonary vessel radius on the high signal region according to mathematical morphology to obtain a potential nodule image region;

[0076] a rejection module configured to reject false positive nodules from the potential nodule image region by geometric features and threshold processing to generate a nodule binary image;

[0077] a three-dimensional reconstruction module configured to perform three-dimensional reconstruction on the lung nodule region detected and segmented by using a FlyingEdge three-dimensional reconstruction method to obtain a surface model of the nodule for the nodule binary image.

[0078] Further, the rejection module comprises:

[0079] a division module configured to divide the potential nodule image region into single objects according to connected regions, and each object corresponds to a suspected nodule;

[0080] a first judgment module configured to traverse each object and perform a first round of exclusion based on geometric shapes;

[0081] a second judgment module configured to traverse the remaining objects in the first round, extract a center point of the suspected nodule region as a seed point, preset a radius, perform dilation by using a threshold, and count a pixel number of the suspected nodule region after the dilation; if a ratio of an increment of the pixel number after the dilation to a pixel number of the suspected nodule region before the dilation is greater than a preset threshold range, the suspected nodule is determined as a false positive nodule;

[0082] a third judgment module configured to traverse the remaining objects in the second round, count an average value of all pixel CT values in the object, and if the average value is in a region greater than -420 and / or less than -800, the suspected nodule is determined as a false positive nodule.

[0083] Further, the acquisition module comprises:

[0084] a generation module configured to generate staging data according to the original data if the original data exists,

[0085] a selection module configured to select required staging data for processing.

[0086] The application further provides a CT image lung nodule automatic segmentation device based on mathematical morphology, comprising: a memory for storing a computer program; and a processor for implementing the steps of the CT image lung nodule automatic segmentation method based on mathematical morphology when the computer program is executed.

[0087] The application further provides a computer storage medium storing a computer program, wherein the computer program is executed by a processor to implement the steps of the CT image lung nodule automatic segmentation method based on mathematical morphology.

[0088] In conclusion, the CT image lung nodule automatic segmentation method based on mathematical morphology provided by the application can automatically detect suspicious nodule regions in the lung region, gradually eliminate false positive nodules, obtain lung nodule segmentation data, and reconstruct a lung nodule model by using a three-dimensional reconstruction method.

[0089] In the process of lung nodule detection and segmentation, the lung parenchyma region is first segmented, which can effectively reduce the interference of non-lung regions on lung nodule detection and improve the operation efficiency; the mathematical morphology dilation reconstruction method is used to extract the high signal region in the lung parenchyma image data, which can accurately contain the region of true nodules, the mathematical morphology operation is fast, and the fast lung nodule detection and segmentation operation can be realized; by using the three-round false positive elimination method, most of the false positive objects can be effectively removed, especially by using the difference between the dilation results of different regions to eliminate false positive lung nodules; the problem that the traditional lung nodule detection and segmentation method relies on single feature and causes high false positive rate is solved. Even if the method is run on an ordinary computer, the entire lung nodule detection time is less than 20 seconds, which can improve the detection efficiency and reduce the error caused by manual operation.

[0090] The above only describes the preferred embodiments of the application and should not be used to limit the application, and any modification, equivalent replacement and improvement made within the spirit and principle of the application should be included in the protection scope of the application.

Claims

1. An automatic lung nodule segmentation method based on mathematical morphology in CT images, characterized in that, Includes the following steps: Step S1: Obtain the raw data; Step S2, lung parenchyma segmentation: extract the lung parenchyma region using a threshold segmentation algorithm to obtain a binary image of the lung parenchyma; Step S3, mathematical morphology preprocessing: perform an erosion operation with a structuring element radius of 1 on the binary image of lung parenchyma, and then traverse the original data based on the original data and the binary image of lung parenchyma to extract the non-zero pixels in the binary image of lung parenchyma to obtain the lung parenchyma image data. Step S4, high signal region extraction: Mathematical morphological dilation reconstruction method is used to extract high signal regions in lung parenchyma image data; Step S5, mathematical morphological processing: Perform a mathematical morphological opening operation on the high signal region with a structuring element radius similar to the radius of the largest intrapulmonary blood vessel to obtain the potential nodule image region. Step S6: False positive nodules are eliminated. False positive nodules are removed from the potential nodule image region through geometric features and threshold processing, generating a binary image of the nodules. Step S7, 3D reconstruction: For the binary image of the nodule, the FlyingEdge 3D reconstruction method is used to perform 3D reconstruction on the detected and segmented lung nodule region to obtain the surface model of the nodule. Step S6 includes the following steps: Step S61: Divide the potential nodule image region into individual objects according to the connected regions, with each object corresponding to a suspected nodule; Step S62: Traverse each object and perform the first round of elimination based on its geometry; Step S63: Traverse the remaining objects in the first round, extract the center point of the suspicious nodule region as the seed point, preset a radius, use a threshold to expand, count the number of pixels in the suspicious nodule region after expansion, if the ratio of the increase in the number of pixels after expansion to the number of pixels in the suspicious nodule region before expansion is greater than the preset threshold range, then it is judged as a false positive nodule. Step S64: Traverse the remaining objects in the second round, calculate the average CT value of all pixels within the object, and if the average value is in the region greater than -420 and / or less than -800, it is judged as a false positive nodule.

2. The automatic lung nodule segmentation method based on mathematical morphology in CT images according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: If the original data is divided into periods, then generate the period data based on the original data. Step S12: Select the required installment data for processing.

3. An automatic lung nodule segmentation device for CT images based on mathematical morphology, characterized in that, Includes the following modules: The acquisition module is used to acquire raw data; The segmentation module is used to extract lung parenchyma regions using a threshold segmentation algorithm to obtain a binary image of the lung parenchyma; The preprocessing module performs an erosion operation with a radius of 1 on the binary image of lung parenchyma based on mathematical morphology. Then, based on the original data and the binary image of lung parenchyma, it iterates through the original data, extracts the non-zero pixels in the binary image, and obtains the lung parenchyma image data. The extraction module is used to extract high-signal regions from lung parenchyma image data using a mathematical morphological dilatation reconstruction method. The processing module is used to perform a mathematical morphological opening operation on the high-signal region based on mathematical morphology, with the structuring element radius being similar to the radius of the largest intrapulmonary blood vessel, to obtain the potential nodule image region. The elimination module removes false positive nodules from the potential nodule image region through geometric features and threshold processing, generating a binary image of the nodules. The 3D reconstruction module is used to perform 3D reconstruction of the detected and segmented lung nodule regions using the FlyingEdge 3D reconstruction method on the binary images of nodules, so as to obtain the surface model of the nodules. The rejection module includes: The zoning module is used to divide the potential nodule image region into individual objects according to the connected regions, with each object corresponding to a suspected nodule; The first judgment module is used to iterate through each object and perform the first round of elimination based on its geometric shape; The second judgment module iterates through the remaining objects from the first round, extracts the center point of the suspicious nodule region as the seed point, sets a preset radius, performs dilation using a threshold, and counts the number of pixels in the suspicious nodule region after dilation. If the number of pixels in the dilated region is... If the ratio of the increase in the number of nodules to the number of pixels in the suspicious nodule region before expansion is greater than a preset threshold range, it is judged as a false positive nodule. The third judgment module is used to traverse the remaining objects in the second round and calculate the average value of the CT values ​​of all pixels within the object. If the average value is in the region greater than -420 and / or less than -800, it is judged as a false positive nodule.

4. The automatic lung nodule segmentation device for CT images based on mathematical morphology according to claim 3, characterized in that, The acquisition module includes: The generation module, if the original data is divided into periods, generates periodic data based on the original data. Select the module and choose the required installment data for processing.

Images