Lung medical image processing method and device, storage medium and computer equipment

Abstract

The present application provides a method and apparatus for processing lung medical images, a storage medium and a computer device, the method comprising: performing segmentation processing on the lung medical image to determine the detection region of interest image; inputting the detection region of interest image into the pulmonary embolism detection model to determine the embolism region in the detection region of interest image; determining the lung structure region of interest in the detection region of interest image, the lung structure region of interest being associated with lung anatomical structure information; performing embolism positioning processing on the lung structure region of interest according to the embolism region and lung anatomical structure information to determine the structural position information of the embolism region in the lung structure region of interest; and associating and displaying the embolism region and the structural position information in the lung medical image. The above method can intuitively obtain the detailed coordinates and morphological data of the lung anatomical structure where the pulmonary embolism is located, while taking into account the accuracy and efficiency of pulmonary embolism detection.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method and apparatus for processing lung medical images, a storage medium, and a computer device. Background Art

[0002] Pulmonary embolism (PE) is caused by emboli in the lower limb venous system migrating to the pulmonary artery, resulting in pulmonary artery blood flow obstruction, which in turn endangers the patient's health and life. Since the clinical manifestations of PE lack specificity, missed diagnosis and misdiagnosis are prone to occur. Effective detection methods are of great significance for the diagnosis and standardized treatment of PE.

[0003] In the related art, a pulmonary vascular tree is constructed based on Computed Tomography Pulmonary Arteriography (CTPA) to track the pulmonary artery vessels to detect the location of pulmonary embolism. However, due to the low contrast of the embolic pulmonary artery and the influence of factors such as angiography image acquisition and noise, it is difficult to ensure the complete extraction of the pulmonary artery, which easily leads to the loss of the pulmonary artery and pulmonary embolism areas. At the same time, based on the diverse morphology, size, density and distribution of embolism, it is difficult to cover various types of embolism, which leads to the omission of pulmonary embolism areas and low accuracy of pulmonary embolism detection. Moreover, this scheme uses the overall vascular tree of the lungs as the basis for detection, which not only lacks the constraints of the structural information of the local partitions of the lungs, but also easily introduces some false positive emboli, and it is difficult to feedback refined positioning information for the local partitions of the lungs, which is not conducive to pulmonary embolism analysis. Summary of the invention

[0004] In view of this, the present application provides a method and apparatus for processing lung medical images, a storage medium and a computer device, which realizes the precise positioning of pulmonary embolism through the lung anatomical structure information of the region of interest of the lung structure, thereby improving the sensitivity of pulmonary embolism detection and providing reliable data support for the diagnosis of pulmonary embolism.

[0005] According to one aspect of the present application, a method for processing a lung medical image is provided, comprising:

[0006] Perform segmentation processing on lung medical images to determine the image of the region of interest for detection;

[0007] Inputting the detection region of interest image into the pulmonary embolism detection model to determine the embolic region in the detection region of interest image;

[0008] Determine a lung structure region of interest in the detection region of interest image, where the lung structure region of interest is associated with lung anatomical structure information;

[0009] According to the embolism area and the lung anatomical structure information, the embolism positioning processing is performed on the region of interest of the lung structure to determine the structural position information of the embolism area in the region of interest of the lung structure;

[0010] The embolism area and structure location information are displayed in association in lung medical images.

[0011] Optionally, segmenting the lung medical image to determine the detection region of interest image specifically includes:

[0012] Inputting the lung medical image into the lung parenchyma segmentation model to determine the lung parenchyma image;

[0013] According to the lung parenchyma image and the preset external size, the lung medical image is segmented to obtain an external bounding box image;

[0014] The external bounding box image is input into the false positive suppression region segmentation model to determine the false positive suppression image;

[0015] The lung parenchyma image and the false positive suppression image are merged to generate a detection region of interest image.

[0016] Optionally, before merging the lung parenchyma image and the false positive suppression image to generate a detection region of interest image, the lung medical image processing method further includes:

[0017] Performing blood vessel recognition processing on the lung parenchyma image to determine the intrapulmonary blood vessel tree area in the lung parenchyma image;

[0018] The background area except the pulmonary vascular tree area in the lung parenchyma image is deleted.

[0019] Optionally, determining the lung structure region of interest in the detection region of interest image specifically includes:

[0020] According to the granularity information of the detection region of interest image, the pulmonary embolism region of interest is divided into multiple lung anatomical structure regions;

[0021] According to preset structural feature information, a lung structure region of interest is screened from multiple lung anatomical structure regions.

[0022] Optionally, according to the embolism area, embolism positioning processing is performed on the pulmonary structure region of interest to determine the structural position information of the embolism area in the pulmonary structure region of interest, specifically including:

[0023] determining a first coordinate range of the embolism region in the coordinate system of the lung medical image;

[0024] Comparing the first coordinate range with the second coordinate range, the second coordinate range being a coordinate range in the lung anatomical structure information associated with the lung structure region of interest;

[0025] Based on the first coordinate range being within the second coordinate range, the structural position information is determined according to the lung anatomical structure information associated with the first coordinate range and the lung structure region of interest.

[0026] Optionally, the lung medical image processing method further includes:

[0027] According to the embolism area, feature detection processing is performed on the image of the detection area of ​​interest to determine the feature information of the embolism area;

[0028] Correlate and display embolism area and characteristic information in lung medical images;

[0029] In response to an evaluation instruction for a lung medical image, outputting pulmonary embolism evaluation information of the lung medical image according to the feature information and the structural position information;

[0030] The characteristic information includes at least one of the following: embolism volume, CT value of embolism area, embolism center position, embolism number, embolism size and embolism direction.

[0031] Optionally, the characteristic information includes the embolism volume; according to the embolism area, feature detection processing is performed on the image of the detection region of interest to determine the characteristic information of the embolism area, specifically including:

[0032] The product of the number of at least one pixel in the pulmonary embolism region of the detection region of interest image and the unit pixel volume is calculated to obtain the embolism volume.

[0033] Optionally, the characteristic information includes a CT value of the embolism region; according to the embolism region, feature detection processing is performed on the image of the detection region of interest to determine the characteristic information of the embolism region, specifically including:

[0034] Calculating the sum of the CT values ​​of at least one pixel in the embolism area of ​​the detection region of interest image to obtain a total CT value;

[0035] The CT value of the embolism area is obtained by calculating the quotient of the total CT value and the number of at least one pixel.

[0036] According to another aspect of the present application, a lung medical image processing device is provided, comprising:

[0037] A segmentation module is used to perform segmentation processing on the lung medical image and determine the image of the region of interest for detection;

[0038] A detection module, used for inputting the detection region of interest image into the pulmonary embolism detection model to determine the embolic region in the detection region of interest image;

[0039] A determination module is used to determine a lung structure region of interest in the detection region of interest image, where the lung structure region of interest is associated with lung anatomical structure information;

[0040] A positioning module is used to perform embolism positioning processing on the region of interest of the lung structure according to the embolism region and the lung anatomical structure information, and determine the structural position information of the embolism region in the region of interest of the lung structure;

[0041] The display module is used for correlating and displaying embolism area and structure position information in a lung medical image.

[0042] Optionally, the segmentation module specifically includes:

[0043] A parenchymal segmentation module, used for inputting a lung medical image into a lung parenchymal segmentation model to determine a lung parenchymal image;

[0044] A bounding box segmentation module is used to segment the lung medical image according to the lung parenchyma image and a preset external size to obtain an external bounding box image;

[0045] A false positive suppression segmentation module is used to input the external bounding box image into the false positive suppression region segmentation model to determine the false positive suppression image;

[0046] The region merging module is used to merge the lung parenchyma image and the false positive suppression image to generate a detection region of interest image.

[0047] Optionally, the lung medical image processing device further includes:

[0048] A blood vessel recognition module is used to perform blood vessel recognition processing on the lung parenchyma image and determine the intrapulmonary blood vessel tree area in the lung parenchyma image;

[0049] The deletion module is used to delete the background area except the intrapulmonary vascular tree area in the lung parenchyma image.

[0050] Optionally, the determination module is specifically used to divide the pulmonary embolism region of interest into multiple lung anatomical structure regions based on the granularity information of the detected region of interest image; and screen the lung structure region of interest from the multiple lung anatomical structure regions according to preset structural feature information.

[0051] Optionally, the determination module is specifically used to determine a first coordinate range of the embolism area in the coordinate system of the lung medical image; compare the first coordinate range with a second coordinate range, the second coordinate range being a coordinate range in the lung anatomical structure information associated with the lung structure region of interest; based on the first coordinate range being within the second coordinate range, determine the structural position information according to the first coordinate range and the lung anatomical structure information associated with the lung structure region of interest.

[0052] Optionally, the lung medical image processing device further includes:

[0053] A feature detection module is used to perform feature detection processing on the image of the detection region of interest according to the embolism region to determine the feature information of the embolism region;

[0054] The display module is also used to associate and display the embolism area and characteristic information in the lung medical image;

[0055] an output module, in response to an evaluation instruction for a lung medical image, for outputting pulmonary embolism evaluation information of the lung medical image according to the feature information and the structural position information;

[0056] The characteristic information includes at least one of the following: embolism volume, CT value of embolism area, embolism center position, embolism number, embolism size and embolism direction.

[0057] Optionally, the feature information includes the embolism volume; the feature detection module is specifically used to calculate the product of the number of at least one pixel in the pulmonary embolism area of ​​the detection region of interest image and the unit pixel volume to obtain the embolism volume.

[0058] Optionally, the feature information includes the CT value of the embolism area; the feature information includes the embolism volume; the feature detection module is specifically used to calculate the sum of the CT values ​​of at least one pixel in the embolism area of ​​the detection area of ​​interest image to obtain the total CT value; calculate the quotient of the total CT value and the number of at least one pixel to obtain the CT value of the embolism area.

[0059] According to another aspect of the present application, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned lung medical image processing method is implemented.

[0060] According to another aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor implements the above-mentioned lung medical image processing method when executing the computer program.

[0061] By means of the above technical solution, the detection region of interest image where pulmonary embolism may occur is first segmented from the lung medical image through segmentation processing. Then, the embolism area containing pulmonary embolism in the detection region of interest image is detected by the pulmonary embolism detection model. At the same time, the detection region of interest image is partitioned according to the anatomical structure of the lung, and the region of interest of the lung structure required by the user is determined. The structural position information of the embolism area is located for the region of interest of the lung structure, and finally the embolism area and the structural position information are associated and displayed in the lung medical image. On the one hand, the user can intuitively obtain the detailed coordinates and morphological data of the anatomical structure of the lung where the pulmonary embolism is located through the processed lung medical image, which is helpful for analyzing the pulmonary embolism situation. On the other hand, the anatomical structure of the region of interest of the lung structure can be used as a positioning constraint for pulmonary embolism, so as to reduce the possibility of identifying false positive embolism as much as possible and improve the accuracy of pulmonary embolism detection. On the other hand, the use of a pulmonary embolism detection model for pulmonary embolism detection can greatly reduce the detection error caused by differences in the size and morphology of the embolism area. The sensitivity of pulmonary embolism detection is higher and the false positive embolism detection rate is lower, further improving the accuracy of pulmonary embolism detection and providing reliable data support for the diagnosis of pulmonary embolism.

[0062] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0064] Figure 1 One of the flowcharts of the lung medical image processing method provided in the embodiment of the present application is shown;

[0065] Figure 2 The second flowchart of the method for processing lung medical images provided in the embodiment of the present application is shown;

[0066] Figure 3 One of the schematic diagrams of a lung medical image showing a lung medical image processing method provided by an embodiment of the present application;

[0067] Figure 4 One of the schematic diagrams of detecting a region of interest image in the method for processing a lung medical image provided by an embodiment of the present application is shown;

[0068] Figure 5The second schematic diagram of detecting an image of a region of interest in the method for processing a lung medical image provided by an embodiment of the present application is shown;

[0069] Figure 6 A second schematic diagram of a lung medical image showing a lung medical image processing method provided in an embodiment of the present application;

[0070] Figure 7 A schematic diagram of a pulmonary embolism detection model provided by an embodiment of the present application is shown;

[0071] Figure 8 A structural block diagram of a lung medical image processing device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0072] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0073] In this embodiment, a method for processing lung medical images is provided. Figure 1 As shown, the method includes:

[0074] Step 101, segmenting the lung medical image to determine the detection region of interest image;

[0075] In this embodiment, a detection region of interest image where pulmonary embolism may occur is segmented from a lung medical image through segmentation processing. The detection region of interest image includes a lung parenchyma region and a false positive suppression region, and the false positive suppression region includes the pulmonary artery outside the lung parenchyma. Thus, the pulmonary embolism region is extracted based on the segmentation method to provide more accurate embolism information, and the false positive suppression region can be combined to remove false positive embolism outside the pulmonary artery, thereby improving the comprehensiveness of pulmonary embolism detection.

[0076] It is understandable that the lung medical image can be a three-dimensional CT (Computed Tomography) image or a digital radiography (DR) image, etc. The three-dimensional CT or DR image includes the three-dimensional structure of the entire lung. The human lung is mainly composed of lung parenchyma, middle mediastinum, pulmonary blood vessels and soft tissue, wherein the lung parenchyma includes the left lung and the right lung, and the middle mediastinum is the mediastinum between the left and right lungs. The pulmonary blood vessels can be divided into pulmonary artery vessels and pulmonary vein vessels. The lung medical image can also be a plurality of two-dimensional tomographic lung images obtained by segmenting the three-dimensional CT or DR image in a specified three-dimensional direction.

[0077] Specifically, the lung medical image is segmented and the steps of determining the detection of the image of the region of interest specifically include:

[0078] Step 101 - 1 , inputting a lung medical image into a lung parenchyma segmentation model to determine a lung parenchyma image;

[0079] The lung parenchyma segmentation model is trained based on the lung sample images and the lung parenchyma labels corresponding to the lung sample images. The lung parenchyma image output by the lung parenchyma segmentation model is a lung parenchyma binary image. For example, a pixel value of 1 indicates that the pixel belongs to the lung parenchyma area, and a value of 0 indicates that the pixel belongs to the non-lung parenchyma area.

[0080] In this embodiment, since pulmonary embolism is mainly based on the blockage of the pulmonary artery and its branches by various emboli in the systemic circulation, the pulmonary artery and its branches are the key areas for the occurrence of pulmonary embolism. The pulmonary artery and its branches are widely distributed in the lung parenchyma. Therefore, the lung parenchyma image is segmented by a pre-trained lung parenchyma segmentation model based on a convolutional neural network to facilitate the determination of the image of the region of interest for detection. The model can not only quickly extract the lung parenchyma area from the lung medical image and output the lung parenchyma image, thereby improving the efficiency of lung parenchyma segmentation, but also more accurately obtain various types of lung parenchyma areas, thereby improving work efficiency and providing technical support for subsequent pulmonary embolism detection.

[0081] For example, Figure 7As shown in the figure, the lung parenchyma segmentation model contains 7 convolutional modules. The first three modules (convolutional modules 1, 2, and 3) are downsampling modules, and each module is composed of two convolutional layers, an activation function, and a downsampling layer connected together. After passing through each downsampling module, the size of its feature map becomes 1 / 2 of the original, and the number of channels also increases accordingly. Convolutional modules 4 to 7 are upsampling modules, and each module is composed of two convolutional layers, an activation function, and an upsampling layer connected together. After passing through each upsampling module, the size of the feature map becomes 2 times the original, and the number of channels becomes 1 / 2 of the original. The output connection of convolutional module 3 and convolutional module 4 serves as the input of convolutional module 5, the output connection of convolutional module 2 and convolutional module 5 serves as the input of convolutional module 6, and the output connection of convolutional module 1 and convolutional module 6 serves as the input of convolutional module 7. Among them, taking the lung parenchyma segmentation model based on a two-dimensional network as an example, the input of the lung parenchyma segmentation model is a two-dimensional pulmonary medical image with a width and height of w×h (w≤W, h≤H, where W and H are the standardized width and height sizes), and the output is a binary image of the lung parenchyma of the pulmonary medical image. It should be noted that when w<W or h<H, it means that the standardized pulmonary medical image is divided into blocks and then input into the convolutional network, and the binary images of the lung parenchyma output from the divided blocks are stitched into an image of size W×H. Taking the lung parenchyma segmentation model based on a three-dimensional network as an example, the input of the model is changed to a three-dimensional image block with a width, height, and depth of w×h×d (w≤W, h≤H, d≤D, where W, H, and D are the standardized width, height, and depth sizes), and the output is a segmentation image of the same size. When w<W, h<H, or d<D, it means that the input three-dimensional pulmonary medical image is divided into blocks and input into the network, and the outputs of the divided blocks are stitched into a size of W×H×D, and a binary image of the lung parenchyma can also be obtained. Among them, w, W, h, H, d, and D are all greater than 0.

[0082] Step 101-2: According to the lung parenchyma image and the preset circumscribed size, segment the pulmonary medical image to obtain a circumscribed bounding box image;

[0083] Among them, the preset circumscribed size is used to extract the bounding box where the pulmonary artery, including the pulmonary trunk, is located, and it can be reasonably set according to the human body structure. As Figure 3 shown in the figure, the image frame is the boundary of the circumscribed bounding box image, and this circumscribed bounding box image includes the lung parenchyma region and the non-lung parenchyma region within the preset circumscribed size of the lung parenchyma image.

[0084] Step 101-3: Input the circumscribed bounding box image into the false positive suppression region segmentation model to determine the false positive suppression image;

[0085] The false positive inhibition region segmentation model is trained based on the lung sample image and the false positive inhibition region of interest labels corresponding to the lung sample image. The structure, training and prediction methods of the false positive inhibition region segmentation model can be the same as those of the lung parenchyma segmentation model, and other commonly used segmentation models can also be used.

[0086] It is worth mentioning that when the training conditions meet the requirements, if the lung sample image is annotated with both the lung parenchyma label and the false positive suppression region of interest label, the lung parenchyma and pulmonary aorta (false positive suppression region of interest) labels can be merged, and a network can be used to train a pulmonary embolism region of interest segmentation model. The model can be used to output the pulmonary aorta and lung parenchyma regions at one time and obtain the bounding box of the pulmonary embolism detection region of interest, thereby improving the segmentation efficiency of the detection region of interest image.

[0087] Step 101 - 4 , merging the lung parenchyma image and the false positive suppression image to generate a detection region of interest image.

[0088] In this embodiment, considering that the pulmonary artery is also connected to structures such as the heart in the human body structure, that is, the pulmonary artery also exists in other non-pulmonary parenchyma areas outside the pulmonary parenchyma. To this end, the pulmonary artery is first segmented from the pulmonary medical image according to the preset circumscribed size using the traditional threshold segmentation and morphological operation, with the pulmonary parenchyma image as a reference. The circumscribed bounding box image is then input into the false positive suppression region segmentation model to extract the pulmonary artery region (false positive suppression region) in other non-pulmonary parenchyma regions, and the corresponding false positive suppression image is output. Finally, the pulmonary parenchyma image and the false positive suppression image are merged to generate a detection region of interest image. At this time, the detection region of interest image includes the pulmonary parenchyma region and the false positive suppression region. Thus, by using pulmonary parenchyma segmentation and false positive suppression segmentation, the areas that may cause false positives and the areas where pulmonary embolism will not occur are removed from the pulmonary medical image, so that the subsequent detection of pulmonary embolism by detecting the region of interest image is more accurate.

[0089] like Figure 4 As shown, the detection region of interest image includes the lung parenchyma and the pulmonary artery (false positive suppression region) other than the lung parenchyma region.

[0090] For example, after obtaining the pulmonary parenchyma binary image through the real-time lung segmentation model, since the pulmonary parenchyma area has been obtained, it is only necessary to segment the pulmonary artery area of ​​other areas. Specifically, the pulmonary parenchyma area is used to calculate the external bounding box with a width, height and depth of W', H', and D'. Then, the intersection of the pulmonary parenchyma area and the predicted pulmonary artery segmentation area is calculated in pixels as the final pulmonary embolism positive area, and the detection area of ​​interest image is segmented from the distribution medical image according to the pulmonary embolism area of ​​interest.

[0091] Furthermore, before the lung parenchyma image and the false positive suppression image are merged to generate the detection region of interest image, it also includes: performing blood vessel recognition processing on the lung parenchyma image to determine the intrapulmonary vascular tree area in the lung parenchyma image; and deleting the background area in the lung parenchyma image except the intrapulmonary vascular tree area.

[0092] In this embodiment, before merging the lung parenchyma image and the false positive suppression image, the pulmonary vascular tree region in the lung parenchyma image is determined by a recognition algorithm, and the background region other than the pulmonary vascular tree region in the lung parenchyma image is deleted, so that only the pulmonary vascular region is retained in the lung parenchyma image. Thus, the lung parenchyma is segmented more carefully, so that the region in the lung parenchyma where pulmonary embolism will not occur is removed from the lung medical image, and the size of the region of interest of pulmonary embolism is reduced as much as possible, which not only reduces the data processing amount of subsequent pulmonary embolism detection, reduces the computing requirements of the processor, and improves the detection efficiency, but also eliminates the interference of non-pulmonary parenchyma or non-pulmonary artery regions during the detection process, which helps to improve the false positive suppression ability of pulmonary embolism detection, improves the detection accuracy, and meets the needs of users in many aspects.

[0093] It should be noted that the intrapulmonary vascular tree includes both intrapulmonary arterial vessels and intrapulmonary venous vessels. In order to further reduce the size of the region of interest of pulmonary embolism, the intrapulmonary arterial vessels can also be identified from the intrapulmonary vascular tree to delete the intrapulmonary venous vessels in the lung parenchyma image.

[0094] Furthermore, since there is soft tissue in the lung, the lung parenchyma image also includes the soft tissue area. It can be understood that if the lung parenchyma image is used to extract the pulmonary vascular tree, what is obtained is a pulmonary vascular tree to be cleaned up, which includes both pulmonary vessels and soft tissue. Therefore, before performing vessel recognition processing on the lung parenchyma image, the lung parenchyma image can be filtered to clean up the part where the soft tissue in the lung is located, so as to obtain a lung parenchyma image that can accurately represent the pulmonary vessels.

[0095] In some possible embodiments, a morphological closing operation may be performed on the binary image of the lung parenchyma first, that is, a dilation operation is performed first and then an erosion operation is performed, so as to fill the holes in the binary image of the lung parenchyma, that is, the pulmonary blood vessels. Then, the filled binary image of the lung parenchyma may be subtracted from the binary image of the lung parenchyma before filling to obtain the pulmonary blood vessel tree region. Among them, the erosion operation makes the target area "smaller" and the dilation operation makes the target area "larger". For details, please refer to the erosion formula and the dilation formula in the related art.

[0096] Step 102, inputting the detection region of interest image into the pulmonary embolism detection model to determine the embolic region in the detection region of interest image;

[0097] Among them, the pulmonary embolism detection model is trained based on pulmonary sample images and the corresponding pulmonary embolism labels of the pulmonary sample images. The pulmonary embolism detection model is used to distinguish each pixel point in the image of the region of interest to be detected as an embolism region or a non-embolism region.

[0098] It should be noted that the number of embolism regions can be one or more. Each embolism region has unique identification information, so as to facilitate subsequent positioning of the corresponding embolism region through the identification information. For example, if the image of the region of interest to be detected contains i embolism regions in total, their numbers (identification information) are successively denoted as P1, P2, ……, Pi-1, Pi, where i is a positive integer greater than 0.

[0099] It is worth mentioning that when using multiple two-dimensional images as model inputs, the embolism regions in the multiple two-dimensional images of the region of interest to be detected output by the model are only cross-sectional regions of the three-dimensional pulmonary embolism and cannot represent the whole pulmonary embolism. Therefore, three-dimensional connected component analysis is performed on the embolism regions in multiple different images to obtain several independent three-dimensional embolism regions, so as to accurately display pulmonary embolisms of different morphologies. Among them, a connected component is an image region composed of foreground pixel points with the same pixel value or satisfying specific similarity criteria in the image and adjacent in position.

[0100] In this embodiment, the pulmonary embolism detection takes the image of the region of interest to be detected as the input, and realizes the extraction of the pulmonary embolism region through the pre-constructed pulmonary embolism detection model, and outputs a binary classification image of the region of interest to be detected divided by pixels, in which at least one embolism region is shown. Through the model, not only can the embolism region be quickly extracted from the image of the region of interest to be detected, improving the efficiency of pulmonary embolism detection, but also it is not sensitive to the size and morphological differences of the embolism regions, with high sensitivity for detecting pulmonary embolisms and lower false positive embolism detections, achieving the purpose of accurately extracting pulmonary embolisms from pulmonary medical images.

[0101] Exemplarily, the structure of the pulmonary embolism detection model can be similar to the aforementioned pulmonary parenchyma segmentation model structure, the difference being that the training sample data is pulmonary embolism label data and the output is a binary classification image of the region of interest to be detected containing the embolism region. As Figure 7 shown, both the input and output of the pulmonary embolism detection model are two-dimensional images of size w×h (h≤H, w≤W). To improve the detection accuracy of the model, a three-dimensional network model with a similar structure can also be used. The input of the model is changed to a three-dimensional image block of w×h×d (w≤W', h≤H', d≤D'), and the output is an image of the same size. When W<W', h<H' or d<D', it means that the input image is divided into blocks and input into the network, and the outputs after block division are spliced into an image of size W'×H'×D'.

[0102] Step 103, determining a lung structure region of interest in the region of interest image;

[0103] The lung structure region of interest is associated with lung anatomical structure information, which includes coordinates of the lung anatomical structure region in the coordinate system of the lung medical image, and information such as the name, number or features of the lung anatomical structure region.

[0104] In this embodiment, the detection region of interest image is partitioned according to the pulmonary anatomical structure, and the pulmonary structure region of interest where pulmonary embolism may occur is determined. This is to facilitate the location of the structural position information of the embolic region for different pulmonary structure regions of interest. This allows the user to obtain the detailed coordinates and morphological data of the pulmonary anatomical structure where the pulmonary embolism is located, which is helpful for analyzing the pulmonary embolism situation. In addition, the anatomical structure of the pulmonary structure region of interest can be used as a positioning constraint for the pulmonary embolism, which can reduce the possibility of identifying false positive embolism as much as possible, improve the accuracy of pulmonary embolism detection, and provide reliable data support for pulmonary embolism diagnosis.

[0105] Specifically, the step of determining and detecting a lung structure region of interest in the region of interest image specifically includes:

[0106] Step 103-1, dividing the pulmonary embolism region of interest into a plurality of lung anatomical structure regions according to the granularity information of the detected region of interest image;

[0107] The anatomical structure of the lungs includes the left lung, right lung, and mediastinum. The left lung and right lung can be further divided into lobes, segments, or subsegments. Figure 5 As shown, different colors are used to represent different lung anatomical structural areas after division.

[0108] In this embodiment, the pulmonary embolism region of interest is divided into a plurality of lung anatomical structure regions according to the size of the positioning granularity, so as to facilitate the positioning of pulmonary embolism occurring in different structural regions.

[0109] Specifically, according to the granularity information of the image of the region of interest, the pulmonary embolism region of interest can be divided into multiple lung anatomical structure regions using a pre-trained structural segmentation model, which needs to be trained based on a granularity training data set of anatomical structure annotated images. Taking lung lobe positioning as an example, a lung lobe segmentation model based on a convolutional neural network needs to be constructed, and the model output is multiple different lung lobe regions and a background region.

[0110] Step 103 - 2 , screening a pulmonary structure region of interest where pulmonary embolism may occur from multiple pulmonary anatomical structure regions according to preset structural feature information.

[0111] In this embodiment, the preset structural feature information is used as a screening condition to screen a pulmonary structure region of interest from multiple pulmonary anatomical structure regions, so as to further refine the location of pulmonary embolism by using the anatomical structure of the pulmonary structure region of interest, thereby reducing the possibility of identifying false positive embolism as much as possible, improving the accuracy of pulmonary embolism detection, and further achieving the purpose of accurately extracting pulmonary embolism from pulmonary medical images.

[0112] Step 104, performing embolism positioning processing on the region of interest of the lung structure according to the embolism region and the lung anatomical structure information, and determining the structural position information of the embolism region in the region of interest of the lung structure;

[0113] The structural position information includes the coordinates of the embolism area and the lung anatomical structure information associated with the lung structure region of interest;

[0114] In this embodiment, the embolic region and the lung anatomical structure information are combined to perform positioning processing for different lung structure regions of interest, so as to locate the structural position information of the embolic region existing in each lung structure region of interest. This allows the user to intuitively obtain the detailed coordinates and morphological data of the lung anatomical structure where the pulmonary embolism is located, which is helpful for analyzing the pulmonary embolism situation. In addition, the anatomical structure of the lung structure region of interest can be used as a positioning constraint for the pulmonary embolism, which reduces the possibility of identifying false positive embolism as much as possible, and achieves the purpose of accurately extracting pulmonary embolism from the lung medical image.

[0115] Specifically, according to the embolism area, the region of interest of the lung structure is identified and processed to determine the structural position information of the embolism area in the region of interest of the lung structure, which specifically includes:

[0116] Step 104 - 1 , determining a first coordinate range of the embolism region in the coordinate system of the lung medical image;

[0117] Specifically, a coordinate system is constructed in advance for the lung medical image, and the first coordinate range is a coordinate set of pixels at the boundary of the embolism region in the coordinate system.

[0118] Step 104-2, comparing the first coordinate range and the second coordinate range;

[0119] The second coordinate range is a coordinate range in the lung anatomical structure information associated with the lung structure region of interest, and the second coordinate range is a coordinate set of the boundary of the lung structure region of interest, which can be determined according to the human anatomical structure.

[0120] Step 104 - 3 , based on the first coordinate range being within the second coordinate range, determine the structural position information according to the lung anatomical structure information associated with the first coordinate range and the lung structure region of interest.

[0121] In this embodiment, if it is detected that the first coordinate range is within the second coordinate range, that is, the coordinates of each pixel in the embolism area are within the second coordinate range, it is determined that there is a pulmonary embolism in the pulmonary structure region of interest. Then, according to the pulmonary anatomical structure information associated with the first coordinate range and the pulmonary structure region of interest, the structural position information is determined. In order to facilitate the user to clearly know the detailed coordinates of the pulmonary anatomical structure where the pulmonary embolism is located through the structural position information, the pulmonary embolism positioning information is more comprehensive, and the difficulty of pulmonary embolism analysis is greatly reduced.

[0122] For example, the detection region of interest image contains j lung structure regions of interest, which are numbered S1, S2, ..., Sj-1, Sj. Traverse i embolic regions and calculate the overlap between Pi and each lung structure region of interest Sj. If overlap occurs, the lung structure region of interest Sj is counted as containing a pulmonary embolism. After completing all statistics, the position of each embolic region in the anatomical structure can be obtained.

[0123] Step 105 : Correlate and display the embolism region and the structural position information in the lung medical image.

[0124] In this embodiment, the embolic region is superimposed on the lung medical image according to the coordinates, so that each embolic region is highlighted on the lung medical image according to a preset display mode. At the same time, the structural position information of each embolic region is marked on the lung medical image accordingly, so that the user can intuitively know the detailed coordinates and morphological data of the lung anatomical structure where the pulmonary embolism is located.

[0125] The preset display mode may be to display in a specified color, such as Figure 6 As shown, an indication frame of the embolism area may also be added, and the embodiments of the present application are not listed one by one here.

[0126] It is understandable that the lung medical image can be displayed in different three-dimensional orientations to accurately know the shape and size of the pulmonary embolism, for example, displaying three-dimensional visualized lung medical images in three orientations: transverse, sagittal, and coronal.

[0127] The present application implements the method for processing lung medical images, which first segments the detection region of interest image where pulmonary embolism may occur from the lung medical image through segmentation processing. Then, the embolism area containing pulmonary embolism in the detection region of interest image is detected by the pulmonary embolism detection model. At the same time, the detection region of interest image is partitioned according to the anatomical structure of the lung, and the region of interest of the lung structure required by the user is determined. The structural position information of the embolism area is located for the region of interest of the lung structure, and finally the embolism area and the structural position information are associated and displayed in the lung medical image. On the one hand, the user can intuitively obtain the detailed coordinates and morphological data of the anatomical structure of the lung where the pulmonary embolism is located through the processed lung medical image, which is helpful for analyzing the pulmonary embolism situation. On the other hand, the anatomical structure of the region of interest of the lung structure can be used as a positioning constraint for pulmonary embolism, so as to reduce the possibility of identifying false positive embolism as much as possible and improve the accuracy of pulmonary embolism detection. On the other hand, the use of a pulmonary embolism detection model for pulmonary embolism detection can greatly reduce the detection error caused by differences in the size and morphology of the embolism area. The sensitivity of pulmonary embolism detection is higher and the false positive embolism detection rate is lower, further improving the accuracy of pulmonary embolism detection and providing reliable data support for the diagnosis of pulmonary embolism.

[0128] In the embodiments of the present application, further, Figure 2 As shown, the method for processing lung medical images further includes:

[0129] Step 201, performing feature detection processing on the image of the region of interest according to the embolism region to determine feature information of the embolism region;

[0130] The characteristic information includes at least one of the following: embolism volume, CT value of embolism area, embolism center position, embolism number, embolism size and embolism direction.

[0131] In actual application scenarios, the characteristic information includes the embolism volume; according to the embolism area, feature detection processing is performed on the detection area of ​​interest image to determine the characteristic information of the embolism area, specifically including: calculating the product of the number of at least one pixel in the pulmonary embolism area of ​​the detection area of ​​interest image and the unit pixel volume to obtain the embolism volume.

[0132] In actual application scenarios, the characteristic information includes the CT value of the embolism area; according to the embolism area, feature detection processing is performed on the detection area of ​​interest image to determine the characteristic information of the embolism area, specifically including: calculating the sum of the CT values ​​of at least one pixel in the embolism area of ​​the detection area of ​​interest image to obtain the total CT value; calculating the quotient of the total CT value and the number of at least one pixel to obtain the CT value of the embolism area.

[0133] Step 202, associating and displaying the embolism region and feature information in the lung medical image;

[0134] In this embodiment, the user can not only determine the structural position information of the embolic area through the lung medical image, but also understand the characteristic information of each embolic area, such as the volume, CT value, and center position, etc. This helps the user determine the morphology, density, and other information of the pulmonary embolism, helps to reduce the workload of pulmonary embolism analysis, reduces the difficulty of analysis, and thus improves the quality of pulmonary embolism diagnosis.

[0135] Step 203: In response to the instruction for evaluating the lung medical image, output pulmonary embolism evaluation information of the lung medical image according to the feature information and the structural position information.

[0136] In this embodiment, when the user needs to perform an overall evaluation of pulmonary embolism appearing in a pulmonary medical image, an evaluation instruction can be issued by operating the computer device. The computer device responds to the evaluation instruction, uses the feature information and structural position information to evaluate the severity of the pulmonary embolism in the pulmonary medical image, and outputs the pulmonary embolism evaluation information of the pulmonary medical image. This allows the user to quickly learn the severity of pulmonary embolism based on the pulmonary medical image, which helps in the early diagnosis and pre-judgment of pulmonary embolism, facilitates subsequent clinical research on pulmonary embolism, reduces the workload of pulmonary embolism analysis, and significantly reduces the risk of pulmonary embolism.

[0137] In the embodiment of the present application, further, before segmenting the lung medical image, the lung medical image processing method further includes: preprocessing the lung medical image.

[0138] In actual application scenarios, preprocessing includes at least one of the following: truncation processing, grayscale processing, normalization processing, window transformation processing, clipping processing, and resampling processing.

[0139] In this embodiment, the lung medical image is transformed into an image that meets the detection requirements through preprocessing, thereby reducing the interference of non-critical information in the lung medical image and improving the accuracy, stability and consistency of subsequent pulmonary embolism detection.

[0140] For example, the CT values ​​of the lung medical image are truncated and normalized between -1000 and 1000, and the obtained image values ​​are distributed between 0 and 1, and are re-interpolated in the three-dimensional direction at intervals of 0.6 mm in pixel size to obtain a standardized lung medical image with a width, height and depth of W×H×D.

[0141] Furthermore, as a specific implementation of the above-mentioned lung medical image processing method, the present application embodiment provides a lung medical image processing device, such as Figure 8 As shown, the lung medical image processing device includes: a segmentation module, a detection module, a determination module, a positioning module and a display module.

[0142] A segmentation module is used to perform segmentation processing on the lung medical image and determine the image of the region of interest for detection;

[0143] A detection module, used for inputting the detection region of interest image into the pulmonary embolism detection model to determine the embolic region in the detection region of interest image;

[0144] A determination module is used to determine a lung structure region of interest in the detection region of interest image, where the lung structure region of interest is associated with lung anatomical structure information;

[0145] A positioning module is used to perform embolism positioning processing on the region of interest of the lung structure according to the embolism region and the lung anatomical structure information, and determine the structural position information of the embolism region in the region of interest of the lung structure;

[0146] The display module is used for correlating and displaying embolism area and structure position information in a lung medical image.

[0147] Furthermore, the segmentation module specifically includes:

[0148] A parenchymal segmentation module, used for inputting a lung medical image into a lung parenchymal segmentation model to determine a lung parenchymal image;

[0149] A bounding box segmentation module is used to segment the lung medical image according to the lung parenchyma image and a preset external size to obtain an external bounding box image;

[0150] A false positive suppression segmentation module is used to input the external bounding box image into the false positive suppression region segmentation model to determine the false positive suppression image;

[0151] The region merging module is used to merge the lung parenchyma image and the false positive suppression image to generate a detection region of interest image.

[0152] Furthermore, the lung medical image processing device further includes:

[0153] A blood vessel recognition module is used to perform blood vessel recognition processing on the lung parenchyma image and determine the intrapulmonary blood vessel tree area in the lung parenchyma image;

[0154] The deletion module is used to delete the background area except the intrapulmonary vascular tree area in the lung parenchyma image.

[0155] Furthermore, a determination module is specifically used to divide the pulmonary embolism region of interest into multiple lung anatomical structure regions based on the granularity information of the detected region of interest image; and to screen the lung structure region of interest from the multiple lung anatomical structure regions according to preset structural feature information.

[0156] Furthermore, a determination module is specifically used to determine a first coordinate range of the embolism area in the coordinate system of the lung medical image; compare the first coordinate range with a second coordinate range, the second coordinate range being a coordinate range in the lung anatomical structure information associated with the lung structure region of interest; based on the first coordinate range being within the second coordinate range, determine the structural position information according to the first coordinate range and the lung anatomical structure information associated with the lung structure region of interest.

[0157] Furthermore, the lung medical image processing device further includes:

[0158] A feature detection module is used to perform feature detection processing on the image of the detection region of interest according to the embolism region to determine the feature information of the embolism region;

[0159] The display module is also used to associate and display the embolism area and characteristic information in the lung medical image;

[0160] an output module, in response to an evaluation instruction for a lung medical image, for outputting pulmonary embolism evaluation information of the lung medical image according to the feature information and the structural position information;

[0161] The characteristic information includes at least one of the following: embolism volume, CT value of embolism area, embolism center position, embolism number, embolism size and embolism direction.

[0162] Furthermore, the feature information includes the embolism volume; the feature detection module is specifically used to calculate the product of the number of at least one pixel in the pulmonary embolism area of ​​the detection region of interest image and the unit pixel volume to obtain the embolism volume.

[0163] Furthermore, the characteristic information includes the CT value of the embolism area; the characteristic information includes the embolism volume; the characteristic detection module is specifically used to calculate the sum of the CT values ​​of at least one pixel in the embolism area of ​​the detection area of ​​interest image to obtain the total CT value; calculate the quotient of the total CT value and the number of at least one pixel to obtain the CT value of the embolism area.

[0164] It should be noted that for other corresponding descriptions of the functional modules involved in the lung medical image processing device provided in the embodiment of the present application, reference can be made to Figure 1 and Figure 2 The corresponding description will not be repeated here.

[0165] Based on the above Figure 1 and Figure 2 The method shown in the embodiment of the present application accordingly provides a storage medium on which a computer program is stored, and when the program is executed by a processor, the above-mentioned Figure 1 and Figure 2 A method for processing a lung medical image is shown.

[0166] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product or a hardware product or a combination of software and hardware. The software product can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of the present application.

[0167] Based on the above Figure 1 and Figure 2 The method shown, and Figure 8 In order to achieve the above-mentioned purpose, the embodiment of the present application further provides a computer device, which can be a personal computer, a server, a network device, etc. The computer device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-mentioned Figure 1 and Figure 2 A method for processing a lung medical image is shown.

[0168] Optionally, the computer device may further include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), etc.

[0169] Those skilled in the art will appreciate that the computer device structure provided in this embodiment does not limit the computer device, and may include more or fewer components, or a combination of certain components, or different component arrangements.

[0170] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages and saves the hardware and software resources of the computer device, and supports the operation of information processing programs and other software and / or programs. The network communication module is used to realize the communication between the various controls inside the storage medium, and the communication with other hardware and software in the physical device.

[0171] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform, or by hardware.

[0172] Those skilled in the art will appreciate that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the units or processes in the accompanying drawings are not necessarily necessary for implementing the present application. Those skilled in the art will appreciate that the units in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the description of the implementation scenario, or can be changed accordingly and located in one or more devices different from the present implementation scenario. The units of the above-mentioned implementation scenarios can be combined into one unit, or can be further split into multiple sub-units.

[0173] The above serial numbers of this application are only for description and do not represent the advantages and disadvantages of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of this application, but this application is not limited to them, and any changes that can be thought of by technicians in this field should fall within the scope of protection of this application.

Claims

1. A method for processing lung medical images, characterized in that: The method comprises: Segmenting the lung medical image to determine an image of a region of interest for detection; Inputting the detection region of interest image into a pulmonary embolism detection model to determine an embolic region in the detection region of interest image; Determine a lung structure region of interest in the detection region of interest image, wherein the lung structure region of interest is associated with lung anatomical structure information; According to the embolism area and the lung anatomical structure information, performing embolism positioning processing on the lung structure region of interest to determine the structural position information of the embolism area in the lung structure region of interest; Associatively displaying the embolism region and the structural position information in the lung medical image; The step of performing embolism positioning processing on the pulmonary structure region of interest according to the embolism region and the lung anatomical structure information to determine the structural position information of the embolism region in the pulmonary structure region of interest includes: Determining a first coordinate range of the embolic region in the coordinate system of the lung medical image; Comparing the first coordinate range with a second coordinate range, the second coordinate range being a coordinate range in the lung anatomical structure information associated with the lung structure region of interest; Based on the fact that the first coordinate range is within the second coordinate range, the structure position information is determined according to the lung anatomical structure information associated with the first coordinate range and the lung structure region of interest.

2. The method according to claim 1, characterized in that The segmenting process of the lung medical image to determine the detection region of interest image includes: Inputting the lung medical image into a lung parenchyma segmentation model to determine a lung parenchyma image; Segmenting the lung medical image according to the lung parenchyma image and a preset circumscribed size to obtain an circumscribed bounding box image; Inputting the external bounding box image into a false positive suppression region segmentation model to determine a false positive suppression image; The lung parenchyma image and the false positive suppression image are merged to generate the detection region of interest image.

3. The method according to claim 2, characterized in that Before merging the lung parenchyma image and the false positive suppression image to generate the detection region of interest image, the method further includes: Performing blood vessel recognition processing on the lung parenchyma image to determine an intrapulmonary blood vessel tree region in the lung parenchyma image; The background area other than the intrapulmonary vascular tree area in the lung parenchyma image is deleted.

4. The method according to claim 1, characterized in that The step of determining the lung structure region of interest in the detection region of interest image comprises: Dividing the pulmonary embolism region of interest into a plurality of lung anatomical structure regions according to the granularity information of the detected region of interest image; The lung structure region of interest is screened from the multiple lung anatomical structure regions according to preset structural feature information.

5. The method according to any one of claims 1 to 4, characterized in that The method further comprises: According to the embolism area, performing feature detection processing on the detection region of interest image to determine feature information of the embolism area; Associatively displaying the embolism region and the characteristic information in the lung medical image; In response to an evaluation instruction for the pulmonary medical image, outputting pulmonary embolism evaluation information of the pulmonary medical image according to the feature information and the structural position information; The characteristic information includes at least one of the following: embolism volume, CT value of embolism area, embolism center position, embolism number, embolism size and embolism direction.

6. The method according to claim 5, characterized in that The characteristic information includes the embolism volume; the step of performing characteristic detection processing on the detection region of interest image according to the embolism region to determine the characteristic information of the embolism region includes: Calculating the product of the number of at least one pixel in the pulmonary embolism region of the detection region of interest image and the unit pixel volume to obtain the embolism volume; and / or The characteristic information includes a CT value of the embolism region; and the step of performing characteristic detection processing on the detection region of interest image according to the embolism region to determine the characteristic information of the embolism region includes: Calculating the sum of the CT values ​​of at least one pixel in the embolism region of the detection region of interest image to obtain a total CT value; The quotient of the total CT value and the number of the at least one pixel is calculated to obtain the CT value of the embolism area.

7. A lung medical image processing device, characterized in that: The device comprises: A segmentation module is used to perform segmentation processing on the lung medical image to determine and detect an image of a region of interest; A detection module, used for inputting the detection region of interest image into a pulmonary embolism detection model to determine an embolic region in the detection region of interest image; A determination module, configured to determine a lung structure region of interest in the detection region of interest image, wherein the lung structure region of interest is associated with lung anatomical structure information; A positioning module, configured to perform embolism positioning processing on the pulmonary structure region of interest according to the embolism region, and determine structural position information of the embolism region in the pulmonary structure region of interest; A display module, configured to display the embolism region and the structural position information in association with each other in the lung medical image; The determination module is specifically used to determine a first coordinate range of the embolism area in the coordinate system of the lung medical image; compare the first coordinate range with a second coordinate range, the second coordinate range being a coordinate range in the lung anatomical structure information associated with the lung structure region of interest; based on the first coordinate range being within the second coordinate range, determine the structural position information according to the first coordinate range and the lung anatomical structure information associated with the lung structure region of interest.

8. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

Images