A method for generating lung CT density difference map

Abstract

The present invention discloses a method for generating a lung CT density difference map, which includes obtaining a baseline CT image and a non-baseline CT image; registering the baseline CT image; extracting the lung contour of the original baseline CT image and deforming it as the lung contour of the non-baseline CT image; generating a preliminary lung CT density difference map; calculating the maximum density value change caused by the change in air content in the lung as a threshold; and generating a final lung CT density difference map. The present invention does not require the collection of a large data set or the participation of imaging experts, and can quickly and accurately automatically analyze areas in lung CT images where density values ​​have changed and calculate the rate of change of the corresponding density values.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a method for generating a lung CT density difference map, which is suitable for quickly and accurately automatically analyzing areas in lung CT images where density values ​​have changed and calculating the change rate of the corresponding density values, and does not require the participation of imaging experts during the analysis process. Background Art

[0002] Pulmonary computed tomography (CT) is currently an important imaging method for observing the anatomical structure of the lung. Among them, the rapid and accurate analysis of the density value information of the lung voxels in the lung CT image can provide relevant quantitative information of the lung CT image (such as the size of the area where the density value has changed, the rate of change of the density value, the proportion of certain density values ​​in the whole lung, etc.), which is of great significance.

[0003] There are two main methods for analyzing the voxel density value information in lung CT images. (1) Manual analysis, in which imaging experts read the lung CT images of the subjects and manually analyze the lung voxel density value information contained in the images. Manual analysis is time-consuming and labor-intensive, and is limited by the scarcity of imaging experts [Shi F, Wang J, Shi J, et al. IEEE Rev Biomed Eng 2021; 14: 4–15]. (2) Image post-processing, which can be divided into two types: conventional methods and deep learning methods. Conventional methods automatically analyze the lung voxel density value information by analyzing the grayscale, edge and other information in the lung CT images of the subjects. However, because some areas in the lungs may have irregular shapes, fuzzy boundaries, low contrast, etc., the analysis accuracy of conventional methods is usually low [Pu J, Leader JK, Bandos A, et al. Eur Radiol 2021; 31(1): 436–446]. Deep learning methods are a method that has emerged in recent years. It uses convolutional neural networks to learn lung CT image features and output results. It has a high accuracy in analyzing lung voxel density value information. However, deep learning methods rely on large data sets and require imaging experts to provide a large number of labels in advance for network training [Roberts M, Driggs D, Thorpe M, et al. Nat Mach Intell 2021; 3(3): 199–217].

[0004] In summary, current methods for analyzing voxel density value information in lung CT images (such as manual analysis or deep learning methods) usually require the participation of imaging experts and rely on imaging experts to provide manual labels, which imposes a heavy workload on imaging experts. In addition, current methods only use lung images of subjects at a single time point, while in actual situations, subjects may have scanned lung images at multiple time points, and existing methods do not fully utilize all the imaging information of the subjects. Therefore, it is necessary to develop a method that fully utilizes all the imaging information of the subjects and does not require the participation of imaging experts, and can quickly and accurately automatically analyze the relevant density value information in lung CT images. Summary of the invention

[0005] In view of the deficiency that the current methods for analyzing voxel density value information in lung CT images (such as manual analysis or deep learning methods) need to rely on the participation of imaging experts, the present invention provides a method for generating a lung CT density difference map, which utilizes the progress information of the image itself without the participation of imaging experts. That is, it can quickly and accurately automatically analyze the areas in the lung CT images where the density values ​​have changed and calculate the change rate of the corresponding density values.

[0006] The above-mentioned object of the present invention is achieved by the following technical solutions:

[0007] A method for generating a lung CT density difference map comprises the following steps:

[0008] Step 1, reading the 3D lung CT images of the subject at multiple time points, the lung CT image of the subject at the first time point is called the baseline CT image, that is, the original baseline CT image, and the lung CT images at the remaining time points are called non-baseline CT images;

[0009] Step 2: registering the original baseline CT image to obtain the deformed baseline CT image;

[0010] Step 3: extract the lung contour of the original baseline CT image and deform it to obtain a deformed lung contour, and use the deformed lung contour as the lung contour of the baseline CT image and the non-baseline CT image after deformation in step 2;

[0011] Step 4: filtering the non-baseline CT image and the deformed baseline CT image, and generating a preliminary lung CT density difference map;

[0012] Step 5, calculating the relative change of lung volume between the non-baseline CT image and the original baseline CT image, and calculating the maximum density value change caused by the change of air content in the lung as a threshold;

[0013] Step 6: The gray values ​​of all lung voxels whose gray values ​​are less than the threshold in the preliminary lung CT density difference map are set to 0, thereby generating a final lung CT density difference map.

[0014] As mentioned above, step 2 includes the following steps:

[0015] Step 2.1, perform affine transformation on the original baseline CT image, take the non-baseline CT image as a fixed image, take the similarity measure as the optimization function, iteratively calculate the optimal affine transformation matrix and the corresponding baseline CT image after affine transformation;

[0016] Step 2.2, perform deformation registration on the baseline CT image after affine transformation, take the non-baseline CT image as the fixed image, the B-spline transformation model as the spatial transformation model, and the similarity measure as the optimization function, iteratively calculate the optimal displacement vector field and the corresponding deformed baseline CT image.

[0017] As mentioned above, step 3 includes the following steps:

[0018] Step 3.1, using the seed region growing method to segment the lung contour on the original baseline CT image;

[0019] Step 3.2: Apply the affine transformation matrix and the displacement vector field to deform the lung contour of the original baseline CT image to obtain a deformed lung contour, and use the deformed lung contour as the lung contour of the baseline CT image and the non-baseline CT image after deformation in step 2.

[0020] Step 4 as described above includes the following steps:

[0021] Step 4.1, retaining the density value of the area within the lung contour in the non-baseline CT image and setting all the density values ​​of the area outside the lung contour to 0; retaining the density value of the area within the lung contour of the deformed baseline CT image and setting all the density values ​​of the area outside the lung contour to 0;

[0022] Step 4.2, using a median filter to perform image filtering on the non-baseline CT image obtained in step 4.1 and the deformed baseline CT image;

[0023] Step 4.3: Subtract the density value of the lung voxel on the corresponding deformed baseline CT image from the density value of the lung voxel on the non-baseline CT image obtained by filtering in step 4.2 to generate a preliminary lung CT density difference map.

[0024] Step 5 as described above includes the following steps:

[0025] Step 5.1: Multiply the area of ​​the lung contour of the original baseline CT image obtained in step 3.1 by the image resolution to obtain the lung volume V of the original baseline CT image.baseline ; According to the area of ​​the lung contour of the non-baseline CT image obtained in step 3.2, multiply it by the image resolution to obtain the lung volume V of the non-baseline CT image current ; Calculate the relative change in lung volume FV between the non-baseline CT image and the original baseline CT image;

[0026] Step 5.2: The mathematical relationship between the change in air content and density value within the lung voxel is as follows:

[0027]

[0028] Among them, HU current and HU baseline are the density values ​​of lung voxels in non-baseline CT images and baseline CT images, respectively, and the density values ​​of lung voxels in non-baseline CT images are HU current The value range is [-950HU, -700HU], based on the FV and the density value HU of the lung voxels of the non-baseline CT image current , calculate the density value HU of the lung voxel of the corresponding baseline CT image baseline ;

[0029] The density value HU of the lung voxels in the non-baseline CT image is set current and the density value HU of the lung voxels of the baseline CT image baseline The difference between them is taken as the maximum density change caused by the change of air content in the lungs and is recorded as the threshold.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. The present invention can automatically and quickly analyze the areas in the lung CT images where the density values ​​have changed and calculate the change rates of the corresponding density values ​​without the participation of imaging experts, thereby saving a lot of time and energy required for manual analysis by imaging experts and reducing the workload of imaging experts.

[0032] 2. The present invention is physiologically interpretable. Automatic analysis of areas in lung CT images where density values ​​have changed is based on calculating the density value change of each lung voxel and excluding the density value change caused by breathing, and is interpretable.

[0033] 3. The present invention does not need to rely on large data sets, does not require cumbersome processing procedures, is simple and easy to promote, and can be quickly popularized and applied. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a structural diagram of the device of the present invention.

[0035] Figure 2It is a specific embodiment diagram of the present invention.

[0036] Figure 3 This is the lung CT density difference map finally generated by the present invention. DETAILED DESCRIPTION

[0037] In order to facilitate those skilled in the art to understand and implement the present invention, the present invention is further described in detail below in conjunction with embodiments. It should be understood that the implementation examples described herein are only used to illustrate and explain the present invention and are not used to limit the present invention.

[0038] like Figure 1 As shown, a method for generating a lung CT density difference map is used to quickly and accurately automatically analyze areas in a lung CT image where density values ​​have changed and calculate the change rate of the corresponding density values, specifically comprising the following steps:

[0039] Step 1: Read the 3D lung CT images of the subject at multiple time points. The image data is obtained when the subject is in a supine position and holding his / her breath. The lung CT image of the subject at the first time point is called the baseline CT image, that is, the original baseline CT image, and the lung CT images at the remaining time points are called non-baseline CT images.

[0040] Step 2, perform image registration on the non-baseline CT image in step 1 and the original baseline CT image. The non-baseline CT image is used as a fixed image in the image registration algorithm, and the original baseline CT image is used as a floating image. The original baseline CT image is the image that needs to be registered. The image registration algorithm can be selected according to the specific usage. When selecting the image registration algorithm, the present invention first uses the affine transformation algorithm, and then uses the B-spline deformation registration algorithm, which specifically includes the following steps.

[0041] Step 2.1, perform affine transformation on the original baseline CT image, take the non-baseline CT image as the fixed image, take the original baseline CT image as the floating image, take the similarity measure as the optimization function, iteratively calculate the optimal affine transformation matrix and the corresponding baseline CT image after affine transformation. The similarity measure in this step is selected as mutual information, and the multi-resolution strategy of Gaussian pyramid structure is also adopted to progressively calculate the optimal affine transformation matrix from coarse resolution to fine resolution. There are 4 layers of multi-resolution, and the downsampling factors of each layer are 8, 4, 2, and 1 respectively. The present invention adopts gradient descent optimization algorithm for iterative calculation, and the number of iterations for each layer is 1000. This step finally outputs the affine transformation matrix and the baseline CT image after affine transformation.

[0042] Step 2.2, deform and align the baseline CT image after affine transformation, take the non-baseline CT image as the fixed image, the baseline CT image after affine transformation as the floating image, take the B-spline transformation model as the spatial transformation model, take the similarity measure as the optimization function, and iteratively calculate the optimal displacement vector field and the corresponding deformed baseline CT image. The similarity measure in this step is selected as mutual information, and the multi-resolution strategy of Gaussian pyramid structure is also adopted, with a total of 4 layers, and the downsampling factors of each layer are 8, 4, 2, and 1 respectively. The B-spline transformation model used in the present invention sets the grid spacing of the control points to 8, 4, 2, and 1 at each layer of resolution. The present invention uses a gradient descent optimization algorithm for iterative solution, and the number of iterations for each layer is 2000. This step finally outputs the displacement vector field and the deformed baseline CT image.

[0043] Step 3: extract the lung contour of the original baseline CT image and deform it to obtain a deformed lung contour. The deformed lung contour is used as the lung contour of the baseline CT image and the non-baseline CT image after deformation in step 2. Specifically, the following steps are included.

[0044] Step 3.1: Use the seed region growing method to segment the lung contour on the original baseline CT image, and set the neighborhood radius of the seed to 3. Output the lung contour of the original baseline CT image.

[0045] Step 3.2: Apply the affine transformation matrix and displacement vector field obtained in step 2 to deform the lung contour of the original baseline CT image to obtain a deformed lung contour, and use the deformed lung contour as the lung contour of the baseline CT image and the non-baseline CT image after deformation in step 2.2.

[0046] Step 4: Filter the non-baseline CT image and the deformed baseline CT image, and generate a preliminary lung CT density difference map, such as Figure 2 As shown, the specific steps include the following steps.

[0047] Step 4.1, according to the lung contour obtained in step 3, the density value of the area within the lung contour in the non-baseline CT image is retained and the density values ​​of the area outside the lung contour are all set to 0; the density value of the area within the lung contour of the deformed baseline CT image is retained and the density values ​​of the area outside the lung contour are all set to 0;

[0048] Step 4.2: Use a median filter to filter the non-baseline CT image and the deformed baseline CT image obtained in step 4.1. The window size of the median filter can be adjusted according to the resolution of the actual image. The present invention uses a median filter with a window size of 7×7 to filter each 2D layer (cross section) of the non-baseline CT image and the deformed baseline CT image, and traverses and filters all 2D layers in the non-baseline CT image and the deformed baseline CT image in turn.

[0049] Step 4.3: Subtract the density value of the lung voxel on the corresponding deformed baseline CT image from the density value of the lung voxel on the non-baseline CT image obtained by filtering in step 4.2 to generate a preliminary lung CT density difference map.

[0050] Step 5: Calculate the maximum density value change that can be caused by the change of air content in the lungs between the lung CT images at multiple time points. First, calculate the relative change of lung volume based on the lung contour; then, calculate the maximum density value change that can be caused by the reduction of air content in the lungs based on the density value range of the normal lungs and the relative change of lung volume. Specifically, the following steps are included.

[0051] Step 5.1: multiply the area of ​​the lung contour of the original baseline CT image obtained in step 3.1 by the image resolution to obtain the lung volume of the original baseline CT image (denoted as V baseline ), according to the area of ​​the lung contour of the non-baseline CT image obtained in step 3.2, multiplied by the image resolution, the lung volume of the non-baseline CT image (denoted as V current ). Based on V baseline With V current , calculate the relative change in lung volume between the non-baseline CT image and the original baseline CT image (denoted as FV):

[0052]

[0053] Step 5.2: The change in air content in a certain lung voxel between two CT scans (i.e., between a non-baseline CT image and a baseline CT image) (i.e., the relative change in lung volume FV) will cause a change in the density value of the voxel. The mathematical relationship between the change in air content in a lung voxel and its density value can be described by the following formula:

[0054]

[0055] Where FV is the relative change in lung volume between the non-baseline CT image and the original baseline CT image, which has been obtained in step 5.1. current and HU baselinerepresents the density value of the lung voxel in the non-baseline CT image and the baseline CT image. According to the density value range of the normal lung area in the human body is [-950HU, -700HU], in this embodiment, HU current Combine the FV obtained in step 5.1 with the HU set in step 5.2 current (-700HU), the corresponding HU can be calculated according to this formula baseline .

[0056] Then calculate HU baseline After that, HU current (-700HU) and HU baseline The difference between the values ​​is the maximum density change that can be caused by the change in air content in the lungs (denoted as ΔHU air ):

[0057] ΔHU air =-700-HU baseline

[0058] Step 6: Perform thresholding on the preliminary lung CT density difference map and output the final lung CT density difference map. air As a threshold, the preliminary lung CT density difference map obtained in step 4 is thresholded, and the grayscale value in the preliminary lung CT density difference map is less than ΔHU air The grayscale values ​​of all lung voxels are set to 0, and the final lung CT density difference map can be generated. Based on the lung voxel areas with grayscale values ​​greater than 0 in the final lung CT density difference map, the areas where the density values ​​in the lung CT image have changed can be quickly and accurately automatically analyzed, and the change rates of the density values ​​of these areas are calculated based on the density values ​​corresponding to these areas on the non-baseline CT image and the density values ​​corresponding to these areas on the baseline CT image. The processing time for the present invention to generate a final lung CT density difference map is less than 3 minutes.

[0059] Figure 3 Four examples of lung CT density difference maps generated based on the present invention are shown. The first column is a non-baseline CT image, the second column is a deformed baseline CT image, the third column is a lung contour of the non-baseline CT image, the fourth column is a lung contour of the deformed baseline CT image, and the fifth column is the final generated lung CT density difference map.

[0060] A device for generating a lung CT density difference map, comprising:

[0061] An input module for implementing step 1;

[0062] An image registration module, used to implement step 2;

[0063] An image segmentation module, used to implement step 3;

[0064] Image self-labeling module, used to implement step 4;

[0065] A ventilation function analysis module, used to implement step 5;

[0066] Output module, used to implement step 6.

[0067] Among them, the image registration module includes:

[0068] Affine transformation module, used to implement step 2.1;

[0069] B-spline deformable registration module, used to implement step 2.2.

[0070] The image segmentation module includes:

[0071] Baseline CT image lung contour segmentation module, used to implement step 3.1;

[0072] Lung contour transfer module, used to implement step 3.2.

[0073] The image self-labeling module includes:

[0074] A lung density value retention module, used to implement step 4.1;

[0075] Image filtering module, used to implement step 4.2;

[0076] A preliminary lung CT density difference map generation module is used to implement step 4.3.

[0077] The ventilation function analysis module includes:

[0078] A lung volume relative change calculation module, used to implement step 5.1;

[0079] The maximum density value change calculation module is used to implement step 5.2.

[0080] The specific embodiments of the present invention are described above. The specific embodiments described in the specification of the present invention are merely examples of the spirit of the present invention. Those skilled in the art of the present invention may make various modifications or additions to the described specific embodiments, or replace them in a similar manner, but they will not deviate from the spirit of the present invention or exceed the scope defined by the attached claims.

Claims

1. A method for generating a lung CT density difference map, characterized in that: The following steps are involved: Step 1, reading the 3D lung CT images of the subject at multiple time points, the lung CT image of the subject at the first time point is called the baseline CT image, that is, the original baseline CT image, and the lung CT images at the remaining time points are called non-baseline CT images; Step 2: registering the original baseline CT image to obtain the deformed baseline CT image; Step 3: extract the lung contour of the original baseline CT image and deform it to obtain a deformed lung contour, and use the deformed lung contour as the lung contour of the baseline CT image and the non-baseline CT image after deformation in step 2; Step 4: filtering the non-baseline CT image and the deformed baseline CT image, and generating a preliminary lung CT density difference map; Step 5, calculating the relative change of lung volume between the non-baseline CT image and the original baseline CT image, and calculating the maximum density value change caused by the change of air content in the lung as a threshold; Step 6: Set the gray values ​​of all lung voxels whose gray values ​​are less than the threshold in the preliminary lung CT density difference map to 0, thus generating the final lung CT density difference map. The step 2 comprises the following steps: Step 2.1, perform affine transformation on the original baseline CT image, take the non-baseline CT image as a fixed image, take the similarity measure as the optimization function, iteratively calculate the optimal affine transformation matrix and the corresponding baseline CT image after affine transformation; Step 2.2, perform deformation registration on the baseline CT image after affine transformation, take the non-baseline CT image as the fixed image, take the B-spline transformation model as the spatial transformation model, take the similarity measure as the optimization function, iteratively calculate the optimal displacement vector field and the corresponding deformed baseline CT image, The step 3 comprises the following steps: Step 3.1, using the seed region growing method to segment the lung contour on the original baseline CT image; Step 3.2: Apply the affine transformation matrix and the displacement vector field to deform the lung contour of the original baseline CT image to obtain a deformed lung contour, and use the deformed lung contour as the lung contour of the baseline CT image and the non-baseline CT image after deformation in step 2.

2. The method for generating a lung CT density difference map according to claim 1, characterized in that: The step 4 comprises the following steps: Step 4.1, retaining the density value of the area within the lung contour in the non-baseline CT image and setting all the density values ​​of the area outside the lung contour to 0; retaining the density value of the area within the lung contour of the deformed baseline CT image and setting all the density values ​​of the area outside the lung contour to 0; Step 4.2, using a median filter to perform image filtering on the non-baseline CT image obtained in step 4.1 and the deformed baseline CT image; Step 4.3: Subtract the density value of the lung voxel on the corresponding deformed baseline CT image from the density value of the lung voxel on the non-baseline CT image obtained by filtering in step 4.2 to generate a preliminary lung CT density difference map.

3. The method for generating a lung CT density difference map according to claim 2, characterized in that: The step 5 comprises the following steps: Step 5.1: Multiply the area of ​​the lung contour of the original baseline CT image obtained in step 3.1 by the image resolution to obtain the lung volume V of the original baseline CT image. baseline ; According to the area of ​​the lung contour of the non-baseline CT image obtained in step 3.2, multiply it by the image resolution to obtain the lung volume V of the non-baseline CT image current ; Calculate the relative change in lung volume FV between the non-baseline CT image and the original baseline CT image; Step 5.2: The mathematical relationship between the change in air content and density value within the lung voxel is as follows: Among them, HU current and HU baseline are the density values ​​of lung voxels in the non-baseline CT image and the baseline CT image, respectively, and the density value HU of the lung voxels in the non-baseline CT image current The value range is [-950HU, -700HU], based on the FV and the density value HU of the lung voxels of the non-baseline CT image current , calculate the density value HU of the lung voxel of the corresponding baseline CT image baseline ; The density value HU of the lung voxels in the non-baseline CT image is set current and the density value HU of the lung voxels of the baseline CT image baseline The difference between them is taken as the maximum density change caused by the change of air content in the lungs and is recorded as the threshold.

Images