A chest CT image lung nodule microvessel quantitative analysis method and device

Abstract

This invention belongs to the field of medical image processing technology, specifically relating to a method and apparatus for quantitative analysis of microvessels in pulmonary nodules in chest CT images. The analysis method of this invention includes the following steps: Step 1, preprocessing the chest CT image; Step 2, segmenting the chest CT image using a convolutional neural network model to obtain images of the pulmonary airways, pulmonary vessels, and lung tissue; Step 3, detecting pulmonary nodules in the chest CT image using the convolutional neural network model, extracting lesions, and extracting microvessels around the lesions and within the pulmonary nodules; Step 4, calculating the microvessel density value based on the extraction results of Step 3. This invention also provides an apparatus suitable for the above analysis method. This invention can non-invasively, rapidly, and accurately display the microvessel density value in pulmonary nodules and around lesions in three dimensions, and provide precise quantitative data. It has significant clinical and socioeconomic value.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing technology, specifically relating to a method and apparatus for quantitative analysis of microvessels in pulmonary nodules in chest CT images. Background Technology

[0002] Lung cancer is the leading cause of death among malignant tumors today, and improving its diagnosis and treatment is one of the most important issues in cancer treatment. Computed tomography (CT) has become one of the main non-invasive and routine methods for examining lung cancer. CT imaging technology features fast imaging speed, high chest image resolution, and low-dose scanning. Post-processing techniques can provide rich information for clinical use. In recent years, in particular, with the rapid development of artificial intelligence technology, methods for automatically detecting lung nodules in CT images and determining their benign or malignant nature using deep learning have been continuously improved, greatly promoting the clinical diagnosis and further in-depth research of lung nodules.

[0003] Chest CT images can provide information on the size, location, morphology, and relationship of pulmonary nodules to surrounding tissues, offering clinical information for differential diagnosis. However, accurate differential diagnosis of pulmonary nodules is difficult based solely on this visually visible morphological information. The development and progression of lung cancer are accompanied by abnormal angiogenesis within tissues; therefore, detecting and monitoring changes in the number and size of microvessels in pulmonary nodule tissue is of significant clinical value for differential diagnosis and evaluation of lung cancer treatment. While contrast-enhanced CT can obtain vascular information in pulmonary nodules, current methods involve cumbersome procedures to obtain only qualitative analysis parameters. Radiomics can extract vascular features from pulmonary nodules, but it cannot directly provide morphological and precise quantitative information about pulmonary nodule vessels. Some molecular probes targeting vascular endothelial cells can visualize small blood vessels in vivo, but molecular imaging techniques cannot provide vascular morphological information and are limited by cost and the availability of molecular probes, hindering widespread application.

[0004] Microvessel density (MVD) is a parameter that quantitatively characterizes the distribution density of microvessels. Currently, MVD results for lung nodules are obtained from histopathology. However, it is difficult to obtain whole-body lung cancer tissue during surgery, and obtaining tumor histological information during treatment monitoring is even more challenging. Therefore, rapidly and accurately obtaining visualized and precisely quantified MVD of lung nodules presents significant technical and methodological challenges.

[0005] Machine learning technology is now widely used in the medical field, assisting doctors in disease diagnosis and treatment selection. Research has been conducted on tasks such as blood vessel segmentation and processing in medical images. Chinese invention patent application "CN113205524A Blood Vessel Image Segmentation Method, Apparatus, and Device Based on U-Net" discloses a method for segmenting blood vessels in medical images using a convolutional neural network model. However, current machine learning methods cannot segment and quantify microvessels in medical images. Therefore, how to use appropriate models and methods to achieve quantitative analysis of microvessels and calculate parameters such as MVD remains a pressing problem to be solved in this field. Summary of the Invention

[0006] To address the problems of existing technologies, this invention provides a method and device for quantitative analysis of microvessels in pulmonary nodules in chest CT images, aiming to achieve non-invasive, rapid, and accurate three-dimensional visualization of pulmonary nodules (MVD).

[0007] A method for precise quantitative analysis of microvessels in pulmonary nodules in chest CT images based on convolutional neural networks includes the following steps:

[0008] Step 1: Preprocess the chest CT images;

[0009] Step 2: The chest CT image is segmented using a convolutional neural network model to obtain images of the lung airways, pulmonary vessels, and lung tissue;

[0010] Step 3: Use the convolutional neural network model to detect lung nodules in the chest CT image and extract the lesions, and extract the microvessels around the lesions and within the lung nodules;

[0011] Step 4: Calculate the microvascular density value based on the extraction results of Step 3.

[0012] Preferably, the chest CT image is a plain scan image or an enhanced image.

[0013] Preferably, in step 1, the preprocessing method includes at least one of cropping, resampling, and noise reduction processing.

[0014] Preferably, in step 2, after segmenting the airways, blood vessels, and lung tissue, cavity filling, erosion, and expansion processes are performed to obtain images of the airways, blood vessels, and lung tissue.

[0015] Preferably, in steps 2 and 3, the convolutional neural network model is selected from the Dense-Vnet neural network.

[0016] Preferably, in step 4, the method for calculating the microvascular density value includes the following steps:

[0017] Step 4.1: Perform radiomics processing on the lung nodule images obtained in Step 3 to obtain first-order voxel features;

[0018] Step 4.2: In the segmented lung tissue, the location of the nodules is obtained, and the region of interest corresponding to the lung nodules is segmented according to the seed points;

[0019] Step 4.3: Based on the solidity ratio of lung nodule tissue, select a threshold to perform calculations on the pulmonary vessels, lung nodules, and first-order voxel image features of lung nodules to obtain the microvascular density values ​​around the lung nodules and / or the microvascular density values ​​within the lung nodules.

[0020] Preferably, in step 4.2, the location of the nodule is calculated using a 3D Deep Leaky Noisy model.

[0021] Preferably, step 4 further includes the following steps:

[0022] Step 4.4: Based on the calculation results of Step 4.3, obtain the average MVD value, central MVD value, peripheral MVD value, and edge MVD value of the lung nodules.

[0023] This invention also provides a device for precise quantitative analysis of microvessels in pulmonary nodules in chest CT images based on convolutional neural networks, used to implement the above analysis method, comprising:

[0024] The first acquisition unit is used to acquire image data of a chest CT scan from a CT device or a medical unit network system;

[0025] The first execution unit is used for preprocessing chest CT images;

[0026] The second execution acquisition unit is used to segment the chest CT image using a convolutional neural network model to obtain images of pulmonary vessels, pulmonary nodules and lesions.

[0027] The third execution unit is used to calculate the microvascular density value.

[0028] The present invention also provides a computer-readable storage medium having a computer program stored thereon for implementing the above-described analysis method.

[0029] In this invention, the meanings of the parameter names are as follows:

[0030] "Average MVD of lung nodules" represents the average MVD value of the actual lesion volume.

[0031] "Central MVD value" and "Peripheral MVD value" for lung nodules: If the nodule diameter is less than 20 mm, the edges of the periphery and center will be located at the middle of the nodule radius, i.e., at 1 / 4 of the diameter; if the nodule diameter is greater than or equal to 20 mm, the edges of the periphery and center will be located 10 mm inward from the nodule boundary. The central and peripheral MVD values ​​are expressed as percentages and sum to 100%.

[0032] "MVD edge value" refers to the portion of the nodule extending 2mm outwards plus the portion minus the center of the nodule.

[0033] This invention provides a method for analyzing chest CT images based on neural network technology. The specific steps include preprocessing, segmentation of chest anatomical structures (airways, blood vessels, and lung tissue), detection and segmentation of pulmonary nodules, extraction of microvessels from pulmonary nodules, and calculation of microvessel density (MVD) of pulmonary nodules. This method is applicable to all multi-slice spiral CT scanners and can be used for plain or enhanced CT images. It enables non-invasive, rapid, and accurate three-dimensional visualization of pulmonary nodule MVD and provides precise quantitative data. It has significant clinical and socioeconomic value.

[0034] Obviously, based on the above description of the present invention, and according to common technical knowledge and conventional methods in the field, various other modifications, substitutions or alterations can be made without departing from the basic technical concept of the present invention.

[0035] The following detailed embodiments further illustrate the above-described content of the present invention. However, this should not be construed as limiting the scope of the present invention to the following examples. All technologies implemented based on the above-described content of the present invention fall within the scope of the present invention. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the analysis method in Embodiment 1 of the present invention;

[0037] Figure 2 This is a flowchart and an example of the segmentation results for lung vessels based on a convolutional neural network in Embodiment 1 of the present invention;

[0038] Figure 3 This is a schematic diagram of the microvascular extraction process for pulmonary nodules in Embodiment 1 of the present invention, along with an example of the extraction results.

[0039] Figure 4 This is a schematic diagram of the MVD center value and perimeter value calculation method in Embodiment 1 of the present invention.

[0040] Figure 5This is an example of the control experiment method in Embodiment 1 of the present invention, wherein: A: Browse the entire film under low magnification, select the area with the highest blood vessel density (hotspot area) for photography, and select three hotspot areas for each slice; B: Open the hotspot area image in Photoshop, and mark the microvessels (in red) using the quick selection tool; C: Statistically analyze the marked blood vessels using IPP6.0, including the minimum area, maximum area, total value, mean, standard deviation, etc.; D: Export the statistical data of the three hotspot areas, calculate the average blood vessel content, and calculate the percentage of positive blood vessels in the tissue. Detailed Implementation

[0041] It should be noted that the algorithms for data acquisition, transmission, storage and processing steps not specifically described in the embodiments, as well as the hardware structures and circuit connections not specifically described, can all be implemented using content already disclosed in the prior art.

[0042] Example 1: Method and apparatus for quantitative analysis of microvessels in pulmonary nodules from chest CT images

[0043] The apparatus in this embodiment includes:

[0044] The first acquisition unit is used to acquire image data of a chest CT scan from a CT device or a medical unit network system;

[0045] The first execution unit is used for preprocessing chest CT images;

[0046] The second execution acquisition unit is used to segment the chest CT image using a convolutional neural network model to obtain images of pulmonary vessels, pulmonary nodules and lesions.

[0047] The third execution unit is used to calculate the microvascular density value.

[0048] The method for quantitative analysis of microvessels in pulmonary nodules using the above-mentioned device is as follows: Figure 1 As shown, it includes the following steps:

[0049] Step one: Obtain a chest CT image through a single scan and transmit the image directly or via network to the image processing device. The image processing device preprocesses the chest CT image, performing cropping, resampling, and noise reduction. The specific process is as follows:

[0050] 1) Crops the obtained chest images into standardized chest CT images according to the specified requirements; image cropping can be done manually or using a neural network;

[0051] 2) Reuse the cropped image according to the voxel size (1.0mm×1.0mm×1.0mm);

[0052] 3) Then, Gaussian filtering is used to reduce noise in the image.

[0053] Step two involves using a neural network model to segment anatomical structures in the chest CT images, including segmentation of the airways, pulmonary vessels, and lung tissue, as well as detecting and extracting pulmonary nodules. After anatomical segmentation, the segmented tissues undergo cavity filling, erosion, and expansion processing to improve the accuracy of the segmentation. For example... Figure 2 , 3 As shown, the specific process is as follows:

[0054] 1) The Dense-Vnet neural network was used to segment lung tissue, pulmonary blood vessels and airways, and the traditional bilateral threshold segmentation method was used to segment lung nodules.

[0055] The Dense-Vnet neural network was used for segmentation of thoracic anatomy and pulmonary vessels. For thoracic anatomy training, the original image and drawn labels were directly resampled into a 200×200×200 pixel size and input into the Dense-Vnet network for training. However, due to the smaller voxels of the pulmonary vessels, direct resampling would lead to poor segmentation accuracy. Therefore, the lung region was cropped into 64×64×64 patches before being trained with Dense-Vnet. After obtaining the anatomical structures and vessels, the segmented tissues were filled in, and a dual threshold was used to further eliminate noise and improve the segmentation accuracy of thoracic anatomy.

[0056] 2) Use a threshold method to fill the cavities in the segmented organs and tissues; further improve the segmented organs and tissues.

[0057] Step 3: Calculate the parameters related to MVD based on the extraction results, such as... Figure 3 As shown, the specific process is as follows:

[0058] 1) Perform radiomics processing on pulmonary nodules to obtain first-order voxel features, so as to accurately obtain information on the vascular wall and vascular lumen, thereby improving the accuracy of microvascular segmentation.

[0059] 2) In the segmented lung tissue, the location of nodules is obtained using a 3D Deep Leaky Noisy model, and the corresponding regions of interest (ROI) of the lung nodules are segmented according to the seed points;

[0060] 3) Based on the solidity ratio of lung nodules, a threshold is selected to calculate the MVD value of the first-order voxel image features of pulmonary vessels and lung nodules. The "average MVD of lung nodules" represents the average MVD value of the actual lesion volume. "Central and peripheral MVD values ​​of lung nodules": If the nodule diameter is less than 20 mm, the edges of the periphery and center will be located at the middle of the nodule radius, i.e., at 1 / 4 of the diameter; if the nodule diameter is greater than or equal to 20 mm, the edges of the periphery and center will be located 10 mm inward from the nodule boundary. The central and peripheral MVD values ​​are expressed as percentages, summing to 100%. The "MVD edge value" refers to the portion extending 2 mm outward from the nodule plus the portion minus the nodule center. 4) Based on the above calculations, the final visualized and precisely quantitative average MVD value, central MVD value, peripheral MVD value, and edge MVD value of lung nodules are obtained.

[0061] To verify the consistency between the results obtained by the method provided in this embodiment and the actual pathological results, pathological sections from 10 patients with pulmonary nodules were selected for routine hematoxylin-eosin staining and immunohistochemical staining (using CD34 antibody to label blood vessels), and the percentage of positive blood vessels in the tissue (blood vessel area / total area) was calculated. An example of this process is shown below. Figure 5 As shown in Table 1, the mean MVD value of the nodules was calculated based on thin-section high-resolution CT scans of the chest, following the method of this embodiment. The results for 10 patients with pulmonary nodules are shown in Table 1. Pearson correlation analysis was performed on the mean MVD value and the percentage of positive blood vessels in pathological staining. The results showed that the percentage of positive blood vessels in the pathological tissue was in good agreement with the MVD value of the imaging, with a Pearson correlation coefficient of 0.63. This indicates that the method of this embodiment is reliable and highly accurate.

[0062] Table 1

[0063] patient Average MVD value Percentage of positive blood vessel staining in pathological staining (%)* 1 0.112 1.19 2 0.266 3.91 3 0.25 3.20 4 0.187 0.91 5 0.135 2.26 6 0.209 1.40 7 0.216 1.37 8 0.094 0.79 9 0.275 2.43 10 0.204 0.82

[0064] *Pathological staining area of ​​blood vessels ÷ Total area of ​​lesions

[0065] As can be seen from the above embodiments, the present invention constructs a novel analytical method and apparatus capable of quantitatively calculating MVD after segmenting the microvessels around the lesion and within the pulmonary nodules. The present invention has advantages such as being non-invasive, rapid, accurate, and capable of three-dimensional visualization, and has excellent application prospects in clinical practice.

Claims

1. A method for precise quantification of pulmonary nodule microvessels in chest CT images based on convolutional neural networks, characterized in that, Includes the following steps: Step 1: Preprocess the chest CT images; Step 2: The chest CT image is segmented using a convolutional neural network model to obtain images of the lung airways, pulmonary vessels, and lung tissue; Step 3: Use the convolutional neural network model to detect lung nodules in the chest CT image and extract the lesions, and extract the microvessels around the lesions and within the lung nodules; Step 4: Calculate the microvascular density value based on the extraction results of Step 3; Step 4, the method for calculating the microvascular density value includes the following steps: Step 4.1: Perform radiomics processing on the lung nodule images obtained in Step 3 to obtain first-order voxel features; Step 4.2: In the segmented lung tissue, the location of the nodules is obtained, and the region of interest corresponding to the lung nodules is segmented according to the seed points; Step 4.3: Based on the solidity ratio of lung nodule tissue, select a threshold to perform calculations on the pulmonary vessels, lung nodules, and first-order voxel image features of lung nodules to obtain the microvascular density values ​​around the lung nodules and / or the microvascular density values ​​within the lung nodules.

2. The analysis method according to claim 1, characterized in that: The chest CT image is either a plain scan image or an enhanced image.

3. The method of analysis according to claim 1, characterized in that: In step 1, the preprocessing method includes at least one of cropping, resampling, and noise reduction processing.

4. The method of analysis according to claim 1, characterized in that: In step 2, after segmenting the airways, blood vessels, and lung tissue, cavity filling, erosion, and expansion processes are performed to obtain images of the airways, blood vessels, and lung tissue.

5. The analytical method according to claim 1, characterized in that: In steps 2 and 3, the convolutional neural network model is selected from the Dense-Vnet neural network.

6. The analytical method according to claim 1, characterized in that: In step 4.2, the location of the nodule is calculated using the 3D Deep Leaky Noisy model.

7. The analytical method according to claim 1, characterized in that: Step 4 also includes the following steps: Step 4.4: Based on the calculation results of Step 4.3, obtain the average MVD value, central MVD value, peripheral MVD value, and edge MVD value of the lung nodules.

8. A device for precise quantitative analysis of microvessels in pulmonary nodules in chest CT images based on convolutional neural networks, used to implement the analysis method described in any one of claims 1-7, characterized in that, include: The first acquisition unit is used to acquire image data of a chest CT scan from a CT device or a medical unit network system; The first execution unit is used for preprocessing chest CT images; The second execution acquisition unit is used to segment the chest CT image using a convolutional neural network model to obtain images of pulmonary vessels, pulmonary nodules and lesions. The third execution unit is used to calculate the microvascular density value.

9. A computer-readable storage medium, characterized in that: It stores a computer program for implementing the analytical method according to any one of claims 1-7.

Images