Lung image processing method and device, electronic equipment and storage medium

Abstract

The present disclosure relates to a lung image processing method, and relates to the technical field of medical image processing, comprising: training a first segmentation model by using a first number of first lung images and corresponding lung blood vessel label images; segmenting lung blood vessels of a second number of second lung images based on the trained first segmentation model to obtain corresponding first lung blood vessel segmentation images; selecting the corresponding first lung blood vessel segmentation images to obtain selected first lung blood vessel segmentation images; training a second segmentation model by using the first number of first lung images and corresponding lung blood vessel label images, the selected first lung blood vessel segmentation images and corresponding second lung images; and segmenting lung blood vessels of a third number of second lung images remaining after selection based on the trained second segmentation model to obtain corresponding second lung blood vessel segmentation images. The embodiment of the present disclosure can realize segmentation of blood vessels.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of medical image processing, and particularly relates to a lung image processing method and device, electronic equipment and storage medium. BACKGROUND

[0002] Medical image analysis is an important field in medical research and clinical application in recent years. Among various medical image processing techniques, image segmentation, as a fundamental and key technology, has always been a research hotspot. Image segmentation divides an image into multiple regions with relevant characteristics, enabling doctors to more clearly identify and analyze structures and features in the image. In this context, it is particularly important to accurately segment the vascular structure from computed tomography (CT) images. Blood vessels are an important part of the human circulatory system, and vascular segmentation through CT images can enable doctors to visually observe the shape, thickness, branching, and relationship with surrounding structures of blood vessels. This has a profound significance for the diagnosis of circulatory system diseases such as coronary artery disease, arteriosclerosis, and cerebrovascular disease. The accuracy of vascular segmentation directly affects the accuracy and timeliness of diagnosis. Interventional surgery often involves highly delicate and complex operations, and the structure and location of blood vessels must be very accurate. Through vascular segmentation, doctors can clearly see the three-dimensional structure of blood vessels, which is crucial for surgical path planning, tool selection, and potential risk assessment. At the same time, vascular segmentation not only plays a role in preliminary diagnosis, but also plays a key role in disease monitoring and evaluation during treatment. By comparing vascular images of patients at different time points, doctors can observe changes in blood vessels and evaluate treatment effectiveness and adjust treatment plans.

[0003] However, there are several difficulties and challenges in using CT images for blood vessel annotation and segmentation.(1) The resolution of CT images may limit the visibility of blood vessel structures, especially for microvessels. In lower resolution images, small blood vessels may not be clearly visible, making accurate annotation and segmentation difficult.(2) CT images often come with noise, especially in low-dose scans. Noise can obscure or blur the edges of blood vessels, making it more difficult to accurately annotate and segment blood vessels.(3) In CT images, the contrast between blood vessels and surrounding tissues (such as muscles, bones, or other organs) may not be obvious enough. This makes it difficult for automatic segmentation algorithms to accurately identify the boundaries of blood vessels.(4) CT images usually provide information in three dimensions. The complex arrangement and variation of blood vessels in three spatial dimensions make the task of annotation and segmentation more difficult and time-consuming.(5) For some complex or ambiguous cases, manual annotation of blood vessels may be required. However, manual annotation is a very time-consuming and labor-intensive task, and may be limited by the operator's experience and skills.(6) In CT images, artifacts may occur due to various reasons (such as metal implants, patient movement, etc.). These artifacts can negatively affect image quality, further increasing the difficulty of blood vessel annotation and segmentation.

[0004] In recent years, deep learning, especially convolutional neural networks (CNNs), has shown significant advantages in medical image analysis. By training with a large amount of labeled data, deep learning models can learn to recognize complex blood vessel structures and resist the interference of noise and artifacts. However, the acquisition of labeled data is often time-consuming and expensive, and medical images often involve privacy issues, so only a limited amount of labeled data can be obtained. Therefore, semi-supervised learning has received widespread attention and application in medical image analysis. The main aspects are as follows:(1) Medical image segmentation: By using unlabeled image data to train the model, the accuracy of the medical image segmentation task can be improved. For example, semi-supervised learning can be used to train a lung segmentation model to help doctors better diagnose lung cancer and other diseases.(2) Medical image classification: Using unlabeled data, more samples can be provided for the medical image classification task, thereby improving the accuracy of the model. For example, semi-supervised learning can be used to train a breast X-ray image classifier to help doctors diagnose breast cancer and other diseases.(3) Medical image reconstruction: Through semi-supervised learning, unlabeled data can be used to train medical image reconstruction models, thereby improving the quality of medical images. For example, semi-supervised learning can be used to train an MRI reconstruction model to improve the resolution and quality of MRI images. In summary, the application of semi-supervised learning in the field of medical images can improve the accuracy and efficiency of doctors' diagnosis, thereby better serving the health of patients.

[0005] It is a challenging task to segment lung vessel tree accurately and efficiently. Due to the complex structure, different diameters and many bifurcations of the lung vessel tree, especially in the labeling of small blood vessels, it is very difficult to label data, resulting in a small amount of labeled data and low quality. Further, the classification accuracy of chronic obstructive pulmonary disease (COPD) due to the influence of lung blood vessels needs to be improved. SUMMARY

[0006] The present disclosure provides a lung image processing method and device, an electronic device and a storage medium technical solution.

[0007] According to an aspect of the present disclosure, a lung image processing method is provided, comprising:

[0008] A first segmentation model is trained using a first number of first lung images and their corresponding lung vessel label images; and based on the trained first segmentation model, a second number of second lung images are segmented for lung vessels to obtain corresponding first lung vessel segmentation images;

[0009] The corresponding first lung vessel segmentation images are selected to obtain selected first lung vessel segmentation images; and a second segmentation model is trained using the first number of first lung images and their corresponding lung vessel label images, the selected first lung vessel segmentation images and their corresponding second lung images; and based on the trained second segmentation model, a third number of second lung images remaining after selection are segmented for lung vessels to obtain corresponding second lung vessel segmentation images.

[0010] Preferably, the processing method further comprises: obtaining a performance indicator of the second segmentation model;

[0011] If the performance indicator is lower than a set performance indicator, the corresponding second lung vessel segmentation images are selected to obtain selected second lung vessel segmentation images;

[0012] A third segmentation model is trained using the first number of first lung images and their corresponding lung vessel label images, the selected second lung vessel segmentation images and their corresponding second lung images;

[0013] And based on the trained third segmentation model, a fourth number of second lung images remaining after selection are segmented for lung vessels to obtain corresponding second lung vessel segmentation images;

[0014] The above process is repeated until the performance indicator of the final segmentation model is higher than or equal to the set performance indicator; and / or,

[0015] The method of selecting the corresponding first lung vessel segmentation images to obtain selected first lung vessel segmentation images comprises:

[0016] obtaining a segmentation index corresponding to each first lung vessel segmentation image and a first set segmentation index;

[0017] selecting the corresponding first lung vessel segmentation image based on the segmentation index corresponding to each first lung vessel segmentation image and the first set segmentation index, to obtain a selected first lung vessel segmentation image; and / or,

[0018] The method for selecting the corresponding first lung vessel segmentation image based on the segmentation index corresponding to each first lung vessel segmentation image and the first set segmentation index, to obtain a selected first lung vessel segmentation image, comprises:

[0019] If the segmentation index corresponding to the first lung vessel segmentation image is greater than or equal to the first set segmentation index, the first lung vessel segmentation image is determined as the selected first lung vessel segmentation image; and / or,

[0020] The method for selecting the corresponding second lung vessel segmentation image, to obtain a selected second lung vessel segmentation image, comprises:

[0021] obtaining a segmentation index corresponding to each second lung vessel segmentation image and a second set segmentation index;

[0022] selecting the corresponding second lung vessel segmentation image based on the segmentation index corresponding to each second lung vessel segmentation image and the second set segmentation index, to obtain a selected second lung vessel segmentation image; and / or,

[0023] The method for selecting the corresponding second lung vessel segmentation image based on the segmentation index corresponding to each second lung vessel segmentation image and the second set segmentation index, to obtain a selected second lung vessel segmentation image, comprises:

[0024] If the segmentation index corresponding to the second lung vessel segmentation image is greater than or equal to the second set segmentation index, the second lung vessel segmentation image is determined as the selected second lung vessel segmentation image.

[0025] Preferably, the method for determining the first number of first lung image corresponding lung vessel label images comprises:

[0026] using the trained preset lung vessel convolutional segmentation model to perform first lung vessel segmentation on the first number of first lung images respectively, to obtain corresponding first lung vessel label images;

[0027] performing second lung vessel segmentation on the first number of first lung images respectively by using the machine learning segmentation model to obtain corresponding second lung vessel label images; wherein a tube diameter of lung vessels in the second lung vessel label images is smaller than a tube diameter of lung vessels in the first lung vessel label images;

[0028] performing fusion on the first lung vessel label images and the second lung vessel label images respectively to obtain lung vessel label images corresponding to the first number of first lung images; and / or,

[0029] The method of performing fusion on the first lung vessel label images and the second lung vessel label images respectively to obtain lung vessel label images corresponding to the first number of first lung images comprises:

[0030] performing position superposition on the first lung vessel label images and the second lung vessel label images respectively to obtain lung vessel label images corresponding to the first number of first lung images; and / or,

[0031] The method of performing second lung vessel segmentation on the first number of first lung images respectively by using the machine learning segmentation model to obtain corresponding second lung vessel label images comprises:

[0032] performing multi-scale representation on the first number of first lung images respectively to obtain corresponding multi-scale lung images;

[0033] performing feature extraction on the multi-scale lung images respectively, and performing classification on the extracted features by using a preset classifier to obtain corresponding second lung vessel label images.

[0034] Preferably, in the process of training the segmentation model, a loss of the segmentation model is calculated, and a network parameter of the segmentation model is adjusted by using the loss; and / or,

[0035] The method of calculating the loss of the segmentation model comprises: obtaining a Dice loss function and a cross-entropy loss function;

[0036] first loss values and second loss values of the Dice loss function and the cross-entropy loss function are calculated respectively;

[0037] a sum of the first loss values and the second loss values is configured as a loss value of the segmentation model; and / or,

[0038] The method of calculating the loss function of the segmentation model further comprises: setting loss adjustment of the Dice loss function and the cross-entropy loss function;

[0039] determining whether the segmentation model calculates loss by selecting the Dice loss function and the cross-entropy loss function based on the loss adjustment; and / or,

[0040] The method of setting the loss adjustment of the Dice loss function and the cross-entropy loss function comprises:

[0041] calculating a plurality of difference values of segmentation pixels between the lung blood vessel segmentation image corresponding to the lung image to be segmented by the segmentation model output and the lung blood vessel label image corresponding thereto; and calculating the mean value between the plurality of difference values;

[0042] If the mean value is less than a set value, the segmentation model does not calculate loss; otherwise, the segmentation model calculates loss by selecting the Dice loss function and the cross-entropy loss function.

[0043] Preferably, before the first segmentation model is trained by using the first number of first lung images and the lung blood vessel label images corresponding thereto, the method further comprises: obtaining a first number of first lung images and the lung blood vessel label images corresponding thereto, and a second number of second lung images; and / or,

[0044] Before the first segmentation model is trained by using the first number of first lung images and the lung blood vessel label images corresponding thereto, the method further comprises: performing lung field segmentation on the first number of first lung images and the second number of second lung images respectively to obtain corresponding first lung field images and second lung field images;

[0045] Further, the first segmentation model is trained by using the first number of first lung field images and the lung blood vessel label images corresponding thereto; and the lung blood vessels of the second number of second lung field images are segmented based on the trained first segmentation model to obtain corresponding first lung blood vessel segmentation images;

[0046] The corresponding first lung blood vessel segmentation images are selected to obtain selected first lung blood vessel segmentation images; and the second segmentation model is trained by using the first number of first lung field images and the lung blood vessel label images corresponding thereto, the selected first lung blood vessel segmentation images and the corresponding second lung field images; and the lung blood vessels of the remaining third number of second lung field images after selection are segmented based on the trained second segmentation model to obtain corresponding second lung blood vessel segmentation images.

[0047] Preferably, the processing method further comprises:

[0048] Obtaining an inhalation phase image, an exhalation phase image to be diagnosed, and a plurality of set threshold intervals;

[0049] The inhalation phase image and the exhalation phase image are segmented by using the above processing method to obtain corresponding lung blood vessel images;

[0050] determining a corresponding parameter response map based on the inhalation phase image, the exhalation phase image and a plurality of set threshold intervals;

[0051] performing pulmonary vessel and / or set airway exclusion correction on the parameter response map using the corresponding pulmonary vessel image, and performing COPD typing diagnosis of chronic obstructive pulmonary disease.

[0052] Preferably, when the exhalation phase image corresponding to the inhalation phase image to be diagnosed is missing, or the inhalation phase image corresponding to the exhalation phase image to be diagnosed is missing, a preset synthesizer is used to synthesize the inhalation phase image into a corresponding synthetic exhalation phase image or to synthesize the exhalation phase image into a corresponding synthetic inhalation phase image;

[0053] performing vessel segmentation on the inhalation phase image and its corresponding synthetic exhalation phase image using the above processing method to obtain a corresponding first pulmonary vessel image; or performing vessel segmentation on the exhalation phase image and its corresponding synthetic inhalation phase image using the above processing method to obtain a corresponding second pulmonary vessel image;

[0054] determining a first parameter response map based on the inhalation phase image, the corresponding synthetic exhalation phase image and a plurality of set threshold intervals; or determining a second parameter response map based on the exhalation phase image and the corresponding synthetic inhalation phase image, and a plurality of set threshold intervals;

[0055] performing pulmonary vessel and / or set airway exclusion correction on the first parameter response map using the corresponding first pulmonary vessel image to obtain a corrected first parameter response map; or performing pulmonary vessel and / or set airway exclusion correction on the second parameter response map using the corresponding second pulmonary vessel image to obtain a corrected second parameter response map;

[0056] performing COPD typing diagnosis of chronic obstructive pulmonary disease based on the corrected first parameter response map; or performing COPD typing diagnosis of chronic obstructive pulmonary disease based on the corrected second parameter response map.

[0057] According to an aspect of the present disclosure, a lung image processing device is provided, comprising:

[0058] a first processing unit configured to train a first segmentation model using a first number of first lung images and corresponding pulmonary vessel label images, and perform pulmonary vessel segmentation on a second number of second lung images based on the trained first segmentation model to obtain corresponding first pulmonary vessel segmentation images;

[0059] The second processing unit is configured to select the corresponding first lung blood vessel segmentation image to obtain a selected first lung blood vessel segmentation image, train a second segmentation model by using the first number of first lung images and the corresponding lung blood vessel label images, the selected first lung blood vessel segmentation image and the corresponding second lung image, and perform lung blood vessel segmentation on the third number of second lung images remaining after the selection based on the trained second segmentation model to obtain a corresponding second lung blood vessel segmentation image.

[0060] According to an aspect of the present disclosure, an electronic device is provided, comprising:

[0061] a processor;

[0062] a memory for storing processor-executable instructions;

[0063] The processor is configured to execute the lung image processing method.

[0064] According to an aspect of the present disclosure, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the lung image processing method.

[0065] In the embodiments of the present disclosure, a lung image processing method and device, an electronic device and a storage medium are provided to solve the problem that the structure of a lung blood vessel tree is complex, the diameters of the lung blood vessels are different, and the lung blood vessels have many branches, especially in the labeling of small blood vessels, which makes it very difficult to label, resulting in a small amount of labeled data and low quality, and further providing a basis for accurate determination of COPD phenotypes.

[0066] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, but not limiting the present disclosure.

[0067] Other features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0068] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present disclosure and, together with the specification, serve to explain the technical solutions of the present disclosure.

[0069] Figure 1 A flowchart of a lung image processing method according to an embodiment of the present disclosure is shown;

[0070] Figure 2 A specific flow and network architecture of a lung image processing method according to an embodiment of the present disclosure are shown;

[0071] Figure 3A schematic diagram showing fusion of a first lung blood vessel label image and a second lung blood vessel label image according to an embodiment of the present disclosure is shown.

[0072] Figure 4 A flow chart showing a method of synthesizing the inhalation phase image into a corresponding synthetic exhalation phase image or synthesizing the exhalation phase image into a corresponding synthetic inhalation phase image using a preset synthesizer according to an embodiment of the present disclosure is shown.

[0073] Figure 5 A network structure diagram corresponding to the preset synthesizer according to an embodiment of the present disclosure is shown.

[0074] Figure 6 is a block diagram of an electronic device 800 according to an exemplary embodiment;

[0075] Figure 7 is a block diagram of an electronic device 1900 according to an exemplary embodiment. DETAILED DESCRIPTION

[0076] Various exemplary embodiments, features, and aspects of the present disclosure will be explained in detail below with reference to the accompanying drawings. Like reference numerals in the drawings denote like elements or elements having a similar function. Although various aspects of the embodiments are illustrated in the drawings, the drawings are not necessarily drawn to scale unless specifically indicated.

[0077] The term "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any implementation described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other implementations.

[0078] The term "and / or" in this document is used to describe and / or connect related objects in a manner that any one of the related objects could be present or any combination of the related objects could be present. In addition, the term "at least one of" in this document is used to describe and / or connect related objects in a manner that any one of the related objects could be present or any combination of the related objects could be present, for example, including at least one of A, B, and C means that any one of A, B, and C could be included or any combination of A, B, and C could be included.

[0079] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following detailed description. It should be understood by those skilled in the art that the present disclosure can also be implemented without some specific details. In some examples, methods, means, elements, and circuits that are well known to those skilled in the art are not described in detail in order to highlight the main idea of the present disclosure.

[0080] It can be understood that the above-mentioned various method embodiments of the present disclosure can be combined with each other to form a combined embodiment without deviating from the principle logic. Due to the limited space, the present disclosure will not be described again.

[0081] Further, the disclosure also provides a lung image processing apparatus, an electronic device, a computer readable storage medium, and a program, which can be used to implement any of the lung image processing methods provided by the disclosure. The corresponding technical solutions and descriptions are described in the method section and are not repeated here.

[0082] Figure 1 A flow chart of a lung image processing method according to an embodiment of the disclosure is shown; 2 shows the corresponding specific process and network architecture of the lung image processing method according to an embodiment of the disclosure. As shown in the figure, Figures 1-2 The lung image processing method comprises the following steps: S101: training a first segmentation model using a first number of first lung images and their corresponding lung blood vessel label images; and performing lung blood vessel segmentation on a second number of second lung images based on the trained first segmentation model to obtain corresponding first lung blood vessel segmentation images; S102: selecting the corresponding first lung blood vessel segmentation images to obtain selected first lung blood vessel segmentation images; and training a second segmentation model using the first number of first lung images and their corresponding lung blood vessel label images, the selected first lung blood vessel segmentation images, and their corresponding second lung images; and performing lung blood vessel segmentation on a third number of second lung images remaining after selection based on the trained second segmentation model to obtain corresponding second lung blood vessel segmentation images. This solves the problem that the structure of the lung blood vessel tree is complex, the diameters are of different sizes, and there are many branches, especially in the labeling of small blood vessels, which makes it very difficult to label data, resulting in a small amount of labeled data and low quality, and further provides a basis for accurate determination of COPD phenotypes.

[0083] In the disclosed embodiments and other possible embodiments, the first lung images and the second lung images can be configured as CT images, DR images, MRI images, ultrasound images, PET images, CT-PET images, or other medical images. Further, the CT images, DR images, MRI images, ultrasound images, PET images, CT-PET images, or other medical images can be configured as CT images, DR images, MRI images, ultrasound images, PET images, CT-PET images, or other medical images of the chest (lungs) in the inhale state or / and exhale state. In addition, the CT images, DR images, MRI images, ultrasound images, PET images, CT-PET images, or other medical images of the chest (lungs) in the inhale state or / and exhale state can be configured as CT images, DR images, MRI images, ultrasound images, PET images, CT-PET images, or other medical images of the chest (lungs) in the deep inhale state or / and deep exhale state.

[0084] In the disclosed embodiments, before the first segmentation model is trained using the first number of first lung images and their corresponding lung blood vessel label images, the following steps are further included: obtaining the first number of first lung images and their corresponding lung blood vessel label images, and the second number of second lung images.

[0085] In the disclosed embodiments, before the first segmentation model is trained using the first number of first lung images and their corresponding lung blood vessel label images, the method further comprises: performing lung field segmentation on the first number of first lung images and the second number of second lung images respectively to obtain corresponding first lung field images and second lung field images; then, training the first segmentation model using the first number of first lung field images and their corresponding lung blood vessel label images; and performing lung blood vessel segmentation on the second number of second lung field images based on the trained first segmentation model to obtain corresponding first lung blood vessel segmentation images; selecting the corresponding first lung blood vessel segmentation images to obtain selected first lung blood vessel segmentation images; and training the second segmentation model using the first number of first lung field images and their corresponding lung blood vessel label images, the selected first lung blood vessel segmentation images, and their corresponding second lung field images; and performing lung blood vessel segmentation on the remaining third number of second lung field images after selection based on the trained second segmentation model to obtain corresponding second lung blood vessel segmentation images.

[0086] In the disclosed embodiments and other possible embodiments, the method of performing lung field segmentation on the first number of first lung images and the second number of second lung images respectively to obtain corresponding first lung field images and second lung field images comprises: obtaining a preset lung field segmentation model, and performing lung field segmentation on the first number of first lung images and the second number of second lung images respectively using the preset lung field segmentation model to obtain corresponding first lung field images and second lung field images. Meanwhile, after performing lung field segmentation on the first number of first lung images and the second number of second lung images respectively to obtain corresponding first lung field images and second lung field images, the first lung images and the second lung images in all subsequent processing methods are configured as corresponding first lung field images and second lung field images respectively.

[0087] In the disclosed embodiments and other possible embodiments, the preset lung field segmentation model can be configured based on a U-Net convolutional neural network, or a UNETR convolutional neural network, or a Swin UNETR convolutional neural network, or a nnU-Net convolutional neural network, or an improved preset lung field segmentation model thereof.

[0088] Step S101: training the first segmentation model using the first number of first lung images and their corresponding lung blood vessel label images; and performing lung blood vessel segmentation on the second number of second lung images based on the trained first segmentation model to obtain corresponding first lung blood vessel segmentation images.

[0089] For example, the first number is configured as 12 cases, and the second number is configured as 168 cases. That is, 12 cases of first lung images with lung blood vessel label images; 168 cases of second lung images without lung blood vessel label images.

[0090] In the disclosed embodiments and other possible embodiments, the first lung images of 12 cases are from the data set provided by the VESSEL12 (https: / / vessel12.grand-challenge.org / ) challenge. The challenge provides some chest CT scans for download. Each downloaded file contains CT scans stored in Meta (or MHD / RAW) format. This format stores images as an ASCII-readable header file with the extension.mhd and a separate binary file of image data with the extension.raw. Among them, about half of the scan images of the first lung images of 12 cases contain abnormalities such as emphysema, nodules or pulmonary embolism, and the maximum slice interval is 1 millimeter. Three scans with labels are downloaded. For each lung region in these scans, the mask and blood vessel annotation csv files are included in the download. Each annotation CSV file contains a list of labeled points for a single scan. Each point has been independently labeled by three annotators. Only points on which the three annotators agree are included. The annotation file is provided in csv format. The format of each point is "x, y, z, label", where (x, y, z) represents the position information of the label in each first lung image, and the voxel with the label configured as 1 represents a blood vessel, and the voxel with the label configured as 0 is classified as a non-blood vessel (the lung blood vessel label image corresponding to the first lung image of 12 cases).

[0091] In the disclosed embodiments and other possible embodiments, the second lung images of 168 cases are from the First Affiliated Hospital of Guangzhou Medical University, configured as non-enhanced CT images, and the thickness of all CT scans is 1.0 mm, the slice size is 512x512, and the storage format is dicom.

[0092] In the disclosed embodiments and other possible embodiments, Figure 2 In the disclosed embodiments and other possible embodiments, The first number (12 cases) of first lung images and the corresponding lung blood vessel label images are used to perform full-supervised training on the first segmentation model (Teacher Model, teacher model) to obtain an initial teacher model (Teacher Model trained first segmentation model). Then, Figure 2 In the disclosed embodiments and other possible embodiments, Based on the trained first segmentation model, the second number (168 cases) of second lung images are segmented for lung blood vessels to obtain the corresponding first lung blood vessel segmentation images (168 cases of first pseudo labels).

[0093] The embodiments of the present disclosure use 12 cases of lung CT scan data and corresponding labels for full supervision training (use a first number of first lung images and their corresponding lung blood vessel label images to train a first segmentation model), and use 10 cases of data for testing. First, the lung area (lung field) is automatically segmented from each CT image (first lung image) to obtain the corresponding first lung field image, and the blood vessel gold standard (lung blood vessel label image) in the lung area is obtained. In order to solve the problem of class imbalance, the data in the smallest bounding box of the lung area in the first lung field image is retained.

[0094] In the embodiments of the present disclosure, the first number of first lung images or first lung field images are respectively cropped to a set size. For example, the set sizes of the 12 training data after cropping are (234, 269, 336), (301, 266, 343), (320, 235, 357), (238, 277, 323), (315, 270, 383), (270, 239, 345), (250, 259, 368), (282, 264, 403), (323, 286, 426), (296, 295, 418), (286, 254, 372), (314, 258, 399), and then the first lung images or first lung field images cropped to the set size are resampled to the median inter-voxel spacing (1x0.74x0.74mm3) of all first lung images or first lung field images. Then, the first lung images or first lung field images have a set cuboid size of 128x112x160 and are used to train the network. The number of training rounds can be initially set to 1000, and the number of input samples processed by the network at a time is 2 during each training iteration. The initial learning rate can be set to 0.01, the optimizer can use the SGD optimizer, the momentum can be set to 0.99, and the weight decay can be set to 3e-5. After full supervision training, a teacher model is obtained, which is used for subsequent semi-supervised iterative training.

[0095] Step S102: Figure 2 The corresponding first lung blood vessel segmentation images are selected to obtain selected first lung blood vessel segmentation images; and a second segmentation model is trained using the first number of first lung images and their corresponding lung blood vessel label images, the selected first lung blood vessel segmentation images and their corresponding second lung images; and based on the trained second segmentation model, lung blood vessel segmentation is performed on the remaining third number of second lung images after selection to obtain corresponding second lung blood vessel segmentation images. The above process is repeated until the performance indicator of the final segmentation model is higher than or equal to the set performance indicator.

[0096] ​In the disclosed embodiments and other possible embodiments, the corresponding first lung vessel segmentation images are selected to obtain selected first lung vessel segmentation images (e.g., 40 first pseudo-labels); and the first number of first lung images and their corresponding lung vessel label images (12 examples), the selected first lung vessel segmentation images and their corresponding second lung images (40 examples) are used to train a second segmentation model (Student Model); and based on the trained second segmentation model (Student becomes the new teacher), the remaining third number of second lung images are segmented for lung vessels to obtain corresponding second lung vessel segmentation images.

[0097] In the disclosed embodiments and other possible embodiments, the first segmentation model and / or the second segmentation model can be configured as a preset vessel segmentation model based on a U-Net convolutional neural network, or a UNETR convolutional neural network, or a Swin UNETR convolutional neural network, or a nnU-Net convolutional neural network, or an improved preset vessel segmentation model.

[0098] The first segmentation model and / or the second segmentation model (teacher model and / or student model) can be configured as a preset vessel segmentation model as shown. Figure 2 The preset vessel segmentation model is shown. Figure 2 The specific flow and network architecture of the lung image processing method according to the embodiments of the present disclosure are shown, and the preset vessel segmentation model (network architecture) is configured as an nnFomer model. Among them, the nFormer model can mainly be divided into 3 blocks: an encoder module (encoding module), a bottleneck module (bottleneck module), and a decoder module (decoding module), which retains the U-Net structure. The nnFormer combines a hybrid model of convolution and self-attention mechanism, fully utilizes their advantages, and proposes a computationally efficient method to capture inter-slice dependencies. In the encoder of nnFormer, a lightweight convolutional embedding layer is added to encode spatial information at the pixel level into low-level but high-resolution 3D features. Then, after the embedding block, transformer and convolution downsampling blocks are used alternately to fully mix long-term dependencies and high-level, hierarchical object concepts, thereby improving the generalization ability and robustness of the learned representation. In addition, nnFormer introduces V-MSA to learn the representation on 3D local, and then aggregates it to produce predictions for the whole data.

[0099] In the disclosed embodiments and other possible embodiments, the encoder module includes, in sequence or in skip connection, an embedding layer, two local self-attention layers, a down-sampling, two local self-attention layers, a down-sampling, a bottleneck module including two global self-attention layers, a down-sampling, two global self-attention layers, an up-sampling, two global self-attention layers, an up-sampling, two global self-attention layers, a decoder module including an up-sampling, two local self-attention layers, an up-sampling, two local self-attention layers, and an expending layer.

[0100] Specifically, the preset blood vessel segmentation model (nnFormer model) is respectively configured with an embedding layer (Embedding Layer) and an expending layer (Expending Layer) at the head and tail thereof, and further comprises: two local self-attention layers (Local Self-attention Layer) connected in sequence or in a skip connection manner, a down-sampling (Down-sampling), two local self-attention layers (Local Self-attention Layer), a down-sampling (Down-sampling), two global self-attention layers (Global Self-attention Layer), a down-sampling (Down-sampling), two global self-attention layers (Global Self-attention Layer), an up-sampling (Up-sampling), two global self-attention layers (Global Self-attention Layer), an up-sampling (Up-sampling), two global self-attention layers (Global Self-attention Layer), an up-sampling (Up-sampling), two local self-attention layers (Local Self-attention Layer), an up-sampling (Up-sampling), and two local self-attention layers (Local Self-attention Layer).

[0101] In the disclosed embodiment, the processing method further comprises: obtaining a performance index of the second segmentation model; if the performance index is lower than a set performance index, selecting the corresponding second lung blood vessel segmentation image to obtain a selected second lung blood vessel segmentation image; training a third segmentation model by using the first number of first lung images and the corresponding lung blood vessel label images, the selected second lung blood vessel segmentation image and the corresponding second lung image; and performing lung blood vessel segmentation on the remaining fourth number of second lung images after the selection based on the trained third segmentation model to obtain corresponding second lung blood vessel segmentation images; and repeating the above process until the performance index of the final segmentation model is higher than or equal to the set performance index.

[0102] In the disclosed embodiments and other possible embodiments, the performance indicators of the second segmentation model can be configured as one or more of the corresponding Dice values, Iou values, sensitivity (Sensitivity), precision (Precision), and other performance indicators. At the same time, the performance indicators can be configured as one or more of the corresponding set Dice values, set Iou values, set sensitivity (Sensitivity), set precision (Precision), and other set performance indicators. At the same time, the values of the set Dice values, set Iou values, set sensitivity (Sensitivity), set precision (Precision), and other set performance indicators can be configured by the person skilled in the art as needed.

[0103] In the disclosed embodiments and other possible embodiments, if the performance indicators are lower than the set performance indicators, the corresponding second lung blood vessel segmentation images are selected to obtain selected second lung blood vessel segmentation images (40 second pseudo-labels). Further, the first number of first lung images and their corresponding lung blood vessel label images (12 examples), the selected first lung blood vessel segmentation images and their corresponding second lung images (40 examples) are used to train a second segmentation model (Student Model, student model); and based on the trained second segmentation model (Student becomes the new teacher, student model is converted into a new teacher model) to segment the lung blood vessels of the remaining third number of second lung images (128 examples-40 examples=68 examples) after selection, to obtain the corresponding second lung blood vessel segmentation images.

[0104] For example, 128 of the corresponding second lung blood vessel segmentation images are selected to obtain 40 selected second lung blood vessel segmentation images; the first number (12 examples) of first lung images and their corresponding lung blood vessel label images, the 40 selected second lung blood vessel segmentation images and their corresponding second lung images are used to train a third segmentation model or retrain the first segmentation model; and based on the trained third segmentation model or the retrained first segmentation model, the fourth number (68 examples) of second lung images after selection are segmented for lung blood vessels to obtain the corresponding second lung blood vessel segmentation images; repeat the above process until the performance indicators of the final segmentation model are higher than or equal to the set performance indicators.

[0105] In the disclosed embodiments and other possible embodiments, the third segmentation model and the segmentation model in the repeated process can be configured as a preset blood vessel segmentation model based on a U-Net convolutional neural network, or a UNETR convolutional neural network, or a Swin UNETR convolutional neural network, or a nnU-Net convolutional neural network, or an improved preset blood vessel segmentation model, or a preset blood vessel segmentation model as shown in 2.

[0106] In the disclosed embodiments, the method of selecting the corresponding first lung vessel segmentation image to obtain the selected first lung vessel segmentation image comprises: obtaining a segmentation index corresponding to each first lung vessel segmentation image and a first set segmentation index; and selecting the corresponding first lung vessel segmentation image based on the segmentation index corresponding to each first lung vessel segmentation image and the first set segmentation index to obtain the selected first lung vessel segmentation image.

[0107] In the disclosed embodiments and other possible embodiments, the segmentation index can be configured as one or more of the corresponding Dice value, Iou value, sensitivity (Sensitivity), precision (Precision), and other performance indicators. Meanwhile, the first set segmentation index, the second set segmentation index, or the set segmentation index corresponding to the segmentation model in the above-mentioned process can be configured as one or more of the corresponding set Dice value, set Iou value, set sensitivity (Sensitivity), set precision (Precision), and other set performance indicators. Meanwhile, a person skilled in the art can configure the numerical values of the set Dice value, set Iou value, set sensitivity (Sensitivity), set precision (Precision), and other set performance indicators according to actual needs.

[0108] In the disclosed embodiments, the method of selecting the corresponding first lung vessel segmentation image based on the segmentation index corresponding to each first lung vessel segmentation image and the first set segmentation index to obtain the selected first lung vessel segmentation image comprises: if the segmentation index corresponding to the first lung vessel segmentation image is greater than or equal to the first set segmentation index, determining this first lung vessel segmentation image as the selected first lung vessel segmentation image.

[0109] For example, the segmentation index is configured as precision (Precision), the first set segmentation index is configured as a set precision (Precision) value, and the selected first lung vessel segmentation image is determined from the first lung vessel segmentation image in which the precision or average set precision is greater than the set precision value or average set precision value. Wherein, a person skilled in the art can configure the set precision value or average set precision value according to actual needs, for example, 0.9.

[0110] More specifically, the first iteration uses precision as the basis for selection, selects the top 40 pseudo-labels with average precision value > 0.9 (set precision value or average set precision value) as reliable pseudo-labels (selected first lung blood vessel segmentation images), so that the ratio of labels to pseudo-labels is close to 1 to 4. Finally, 12 labeled images (the first number of first lung images and their corresponding lung blood vessel label images) and 40 unlabeled images and their pseudo-labels (selected first lung blood vessel segmentation images and their corresponding second lung images) are used for retraining to obtain a student model, completing the first iteration training.

[0111] Similarly, in the disclosed embodiments, the method of selecting the corresponding second lung blood vessel segmentation image to obtain a selected second lung blood vessel segmentation image includes: obtaining a segmentation index corresponding to each second lung blood vessel segmentation image and a second set segmentation index; and selecting the corresponding second lung blood vessel segmentation image based on the segmentation index corresponding to each second lung blood vessel segmentation image and the second set segmentation index to obtain a selected second lung blood vessel segmentation image.

[0112] Similarly, in the disclosed embodiments, the method of selecting the corresponding second lung blood vessel segmentation image based on the segmentation index corresponding to each second lung blood vessel segmentation image and the second set segmentation index to obtain a selected second lung blood vessel segmentation image includes: if the segmentation index corresponding to the second lung blood vessel segmentation image is greater than or equal to the second set segmentation index, determining this second lung blood vessel segmentation image as a selected second lung blood vessel segmentation image.

[0113] In the disclosed embodiments and other possible embodiments, the segmentation index corresponding to the second lung blood vessel segmentation image can be configured as precision or average set precision and Dice value or average Dice value. At this time, the second set segmentation index corresponding to the segmentation index is configured as a set precision value or an average set precision value and a set Dice value or an average set Dice value. Among them, the set precision value or the average set precision value can be configured by those skilled in the art according to actual needs, for example, 0.95. Similarly, the set Dice value or the average set Dice value can be configured by those skilled in the art according to actual needs, for example, 0.85.

[0114] For example, the student model (second segmentation model) trained in the first iteration is taken as the teacher model (first segmentation model) to predict the second lung images of the remaining 128 cases (a third number of remaining cases), obtaining 128 pseudo labels (second lung vessel segmentation images). As in the first iteration process, the reliable pseudo labels (selected second lung vessel segmentation images) are selected based on the accuracy or average set accuracy > 0.95 (set accuracy value or average set accuracy value) and the Dice value or average Dice value > 0.85 (set Dice value or average set Dice value). 40 reliable pseudo labels (selected second lung vessel segmentation images) are selected, and the first segmentation model is retrained or the third segmentation model is trained, completing the second iteration and obtaining the best segmentation effect (preset segmentation index).

[0115] The above embodiment only performs twice iteration, and in the disclosed embodiment and other possible embodiments, multiple iterations can be performed according to the above method by those skilled in the art to obtain a more optimal segmentation effect (preset segmentation index).

[0116] In the disclosed embodiment and other possible embodiments, the strategy of selecting reliable pseudo labels (second pseudo labels or selected first lung vessel segmentation images) is added in step S102. In step S101, the teacher model is obtained after full supervision training, and the teacher model is used to predict 168 cases of data, obtaining 168 pseudo labels (selected first lung vessel segmentation images). Then, the reliable pseudo labels (selected second pseudo labels or selected first lung vessel segmentation images) are selected, and the specific method is as follows: checkpoints are saved once every 100 rounds in the full supervision training process, and the checkpoint with the highest Dice value in the validation set is saved as the Best model (best model). After training, for each unlabeled CT scan image (a second number of second lung images), the evaluation index (performance index) of the prediction result of the pseudo label (selected first lung vessel segmentation image) and the Best model is calculated as the selection standard. The larger the average evaluation index, the higher the coincidence of the predicted pseudo label, that is, the more stable the pseudo label in the training process, and the more reliable the quality. Since the segmentation task in this study is relatively fine and aims to be applied to the quantitative index analysis of COPD disease, it is necessary to ensure that the prediction result is more real blood vessels, that is, the fewer false positives, the better.

[0117] In the disclosed embodiments, the method for determining the lung blood vessel label images corresponding to the first number of first lung images comprises: using a trained preset lung blood vessel convolutional segmentation model to perform first lung blood vessel segmentation on the first number of first lung images respectively to obtain corresponding first lung blood vessel label images; using a machine learning segmentation model to perform second lung blood vessel segmentation on the first number of first lung images respectively to obtain corresponding second lung blood vessel label images; wherein the tube diameter of the lung blood vessels in the second lung blood vessel label images is smaller than the tube diameter of the lung blood vessels in the first lung blood vessel label images; and fusing the first lung blood vessel label images and the second lung blood vessel label images respectively to obtain the lung blood vessel label images corresponding to the first number of first lung images.

[0118] In the disclosed embodiments, the method for using a machine learning segmentation model to perform second lung blood vessel segmentation on the first number of first lung images respectively to obtain corresponding second lung blood vessel label images comprises: performing multi-scale representation on the first number of first lung images respectively to obtain corresponding multi-scale lung images; performing feature extraction on the multi-scale lung images respectively and using a preset classifier to classify the extracted features to obtain corresponding second lung blood vessel label images.

[0119] In the disclosed embodiments and other possible embodiments, the trained preset lung blood vessel convolutional segmentation model can be configured as the previously proposed lung blood vessel convolutional segmentation model CE-NC-VesselSegNet, and the coarse blood vessels are segmented using the lung blood vessel convolutional segmentation model CE-NC-VesselSegNet to obtain the first lung blood vessel label images. At the same time, first, the Gaussian pyramid is used to perform multi-scale representation on the first number of first lung images respectively, and the scale is zoomed to 6 scales, and then a feature extraction model (for example, k-means) is used to extract features from the multi-scale lung images, and a certain number (for example, 34) of features are retained from the extracted features; then, based on the certain number of retained features, a feature vector is calculated for each pixel, and the calculated feature vector corresponding to each pixel is input into a classifier (for example, logistic regression) to perform blood vessel probability segmentation, to obtain the probability of all pixel points and segment the complete fine blood vessels to obtain the corresponding second lung blood vessel label images.

[0120] Figure 3 A schematic diagram of fusing the first lung blood vessel label images and the second lung blood vessel label images according to the embodiments of the present disclosure is shown. As shown in FIG. 4, the first lung blood vessel label images and the second lung blood vessel label images are fused to obtain the lung blood vessel label images corresponding to the first number of first lung images. Figure 3As shown, in the disclosed embodiment, the method of fusing the first lung vessel label image and the second lung vessel label image respectively to obtain the lung vessel label image corresponding to the first number of first lung images comprises: superimposing the positions of the first lung vessel label image and the second lung vessel label image respectively to obtain the lung vessel label image corresponding to the first number of first lung images. Wherein, the first lung vessel label image and the second lung vessel label image come from the same first lung image.

[0121] In the disclosed embodiment and other possible embodiments, the complete fine blood vessel result (second lung vessel label image) of each first lung image segmentation is fused with the coarse blood vessels segmented from each first lung image by the model CE-NC-VesselSegNet. After fusion, it can be ensured that the coarse blood vessel part does not contain non-blood vessel tissue such as airway wall, and then the lung vessel label image corresponding to the first number of first lung images is obtained. In addition, the lung vessel convolution segmentation model can be configured as other existing lung vessel segmentation models according to the needs of those skilled in the art.

[0122] In the disclosed embodiment, during the training of the segmentation model, the loss of the segmentation model is calculated, and the network parameters of the segmentation model are adjusted by using the loss.

[0123] In the disclosed embodiment, the method of calculating the loss of the segmentation model comprises: obtaining a Dice loss function and a cross-entropy loss function; calculating a first loss value and a second loss value of the Dice loss function and the cross-entropy loss function respectively; configuring the sum of the first loss value and the second loss value as the loss value of the segmentation model; and / or, the method of calculating the loss function of the segmentation model further comprises: setting the loss adjustment of the Dice loss function and the cross-entropy loss function; based on the loss adjustment, determining whether the segmentation model calculates the loss by selecting the Dice loss function and the cross-entropy loss function.

[0124] In the disclosed embodiment, the method of setting the loss adjustment of the Dice loss function and the cross-entropy loss function comprises: calculating a plurality of difference values of the segmentation pixel points between the lung vessel segmentation image corresponding to the lung image to be segmented by the segmentation model and the lung vessel label image corresponding thereto; and calculating the mean value between the plurality of difference values; if the mean value is less than a set value, the segmentation model does not calculate the loss; otherwise, the segmentation model calculates the loss by selecting the Dice loss function and the cross-entropy loss function.

[0125] In the disclosed embodiment and other possible embodiments, the loss function is defined as follows:

[0126] Loss = I(yn, y^n) Loss Dice + I(yn, y^n) Loss cross entropy

[0127] The disclosure adds a calculation condition (loss adjustment) in front of the Dice loss function and the cross-entropy loss function (CE loss), yn represents the predicted value or segmentation value of the segmented pixel point (pulmonary vessel segmentation image), y^n represents the gold standard of the segmented pixel point (pulmonary vessel label image), if the difference value I = |yn, y^n| between the pulmonary vessel segmentation image corresponding to the pulmonary image to be segmented by the segmentation model and the pulmonary vessel label image corresponding thereto is less than a set value T, I = 0, otherwise the set value I = 1. The set value T of the disclosure can be configured as 0.1. By adding loss adjustment, the network (segmentation model) pays more attention to difficult points, that is, the end points and branch points of the pulmonary vessels, thereby improving the segmentation accuracy.

[0128] In the disclosed embodiments and other possible embodiments, the 2-time iteration process is divided into three stages in total: 1. Supervised pre-training, an initial teacher model (trained first segmentation model) is completely trained on a first number of first lung images and their corresponding pulmonary vessel label images. 2. Generate pseudo-labels, use the initial teacher model (trained first segmentation model) to segment the pulmonary vessels of all unlabeled second number of second lung images to obtain corresponding first pulmonary vessel segmentation images (first pseudo-labels). 3. Retraining. Mix the labeled images and unlabeled images and their pseudo-labels (the first number of first lung images and their corresponding pulmonary vessel label images, selected first pulmonary vessel segmentation images and their corresponding second lung images), and retrain a student model (second segmentation model) thereon, and iterate again according to the above method until the best effect is achieved.

[0129] In order to find a full-supervised training network more suitable for vessel segmentation, the network to be used in this paper is compared with classic and latest segmentation networks, including UNETR, Swin UNETR, nnU-Net.

[0130] UNETR is a method that uses ViT as its encoder, without relying on CNN-based feature extractors. The architecture uses pure Transformers as encoders to learn sequential representations of inputs and effectively capture global multi-scale information. At the same time, it also follows the successful “U-shaped” network design of encoders and decoders, and the Transformers encoder is directly connected to the decoder through different resolution skip connections to calculate the final semantic segmentation output. In different medical image segmentation tasks, UNETR shows better accuracy and efficiency.

[0131] Swin Transformers were proposed as a hierarchical visual Transformer that computes self-attention in an efficient shift window partitioning scheme. Therefore, Swin Transformers are suitable for various downstream tasks in which multi-scale features extracted can be further processed. Then, Swin unetr was proposed, which utilizes a U-shaped network with Swintransformer as an encoder and connects it to a CNN-based decoder with different resolutions through a skip connection. This network validated the effectiveness of the method in the multi-modal 3D brain tumor segmentation task of the 2021 Multi-modal Brain Tumor Segmentation Challenge (BraTS).

[0132] In Table 1, the performance of UNETR, Swin UNETR, nnU-Net, and nnFormer is compared. All the compared models are evaluated using the same training, validation, and test sets as nnFormer full supervision. nnFormer significantly outperforms UNETR, Swin UNETR, and nnU-Net in terms of performance.

[0133] Table 1. Segmentation performance of nnFormer and UNETR, Swin UNETR, nnU-Net networks

[0134]

[0135] The semi-supervised iterative training is completed for 2 times, the first iteration model is denoted as Semi1, the second iteration model is denoted as Semi2, and the full supervision model is denoted as Full. The test sets of the three models are the same, which are the pulmonary vessel annotation data of 10 COPD patients. The evaluation indexes are the same as those in Chapter 2 for coarse blood vessel segmentation, i.e., Precision, Dice, Iou, and Sensitivity.

[0136] Table 2 shows the segmentation performance of the three models. Semi-supervised iterative training aims to improve the accuracy of segmentation results and reduce false positive segmentation. It can be seen that the Precision of Semi2 reaches 0.903, which is improved by nearly 0.2 compared with 0.8802 of Full and by nearly 0.1 compared with Semi1, achieving the expected effect. After the Precision index is improved, the Sensitivity will decrease, which is a normal phenomenon. Due to the complex structure of pulmonary vessels, the terminal diameter is too small, and the diameter size is not uniform, so the segmentation result does not fully contain the gold standard, which is less important for analysis indicators. What is important is that the segmentation result should not contain non-vascular regions as much as possible. In addition, the Dice and IOU indexes of the three models are very close.

[0137] Table 2 Performance of the fully supervised model and the two semi-supervised models

[0138]

[0139] In addition, in the disclosed embodiments, the processing method further includes: obtaining an inspiration phase image, an expiration phase image, and a plurality of set threshold intervals to be diagnosed; performing blood vessel segmentation on the inspiration phase image and the expiration phase image respectively by using the processing method to obtain corresponding lung blood vessel images; determining a corresponding parameter response map based on the inspiration phase image, the expiration phase image, and the plurality of set threshold intervals; performing lung blood vessel and / or set airway removal correction on the parameter response map by using the corresponding lung blood vessel image respectively, and performing COPD typing diagnosis of chronic obstructive pulmonary disease.

[0140] In the disclosed embodiments, when the expiration phase image corresponding to the inspiration phase image to be diagnosed is missing, or the inspiration phase image corresponding to the expiration phase image to be diagnosed is missing, a preset synthesizer is used to synthesize the inspiration phase image into a corresponding synthesized expiration phase image or to synthesize the expiration phase image into a corresponding synthesized inspiration phase image.

[0141] The inspiration phase image and the corresponding synthesized expiration phase image are segmented by using the processing method respectively to obtain a corresponding first lung blood vessel image; or, the expiration phase image and the corresponding synthesized inspiration phase image are segmented by using the processing method respectively to obtain a corresponding second lung blood vessel image.

[0142] A first parameter response map is determined based on the inspiration phase image, the corresponding synthesized expiration phase image, and the plurality of set threshold intervals; or, a second parameter response map is determined based on the expiration phase image, the corresponding synthesized inspiration phase image, and the plurality of set threshold intervals.

[0143] The first parameter response map is corrected by using the corresponding first lung blood vessel image to obtain a corrected first parameter response map; or, the second parameter response map is corrected by using the corresponding second lung blood vessel image to obtain a corrected second parameter response map.

[0144] COPD typing diagnosis of chronic obstructive pulmonary disease is performed based on the corrected first parameter response map; or, COPD typing diagnosis of chronic obstructive pulmonary disease is performed based on the corrected second parameter response map.

[0145] In the disclosed embodiments and other possible embodiments, the plurality of set threshold intervals are configured as attenuation values corresponding to the inspiratory phase images or expiratory phase images. For example, the inspiratory phase images or expiratory phase images can be configured as inspiratory CT phase images or expiratory CT phase images. At this time, the plurality of set threshold intervals corresponding to the inspiratory CT phase images or synthetic inspiratory CT phase images can be configured as one or more of greater than -950HU and / or greater than -856HU, respectively; the plurality of set threshold intervals corresponding to the expiratory CT phase images or synthetic expiratory CT phase images can be configured as one or more of greater than -950HU and / or greater than -856HU, respectively. Meanwhile, for those skilled in the art, the plurality of set threshold intervals can also be configured according to the types (CT images, DR images, MRI images, ultrasound images, PET images, CT-PET images or other medical images) corresponding to the inspiratory phase images or expiratory phase images.

[0146] In the disclosed embodiments and other possible embodiments, this study has been approved by the Ethics Committee of the First Affiliated Hospital of Guangzhou Medical University, in accordance with the ethical standards of the Declaration of Helsinki of 1964 and its subsequent amendments, or similar ethical standards. All participants have obtained informed consent. From August 2017 to April 2021, 558 pairs of inspiratory and expiratory CT images were collected from the First Affiliated Hospital of Guangzhou Medical University to construct Dataset 1.

[0147] All participants underwent pulmonary function tests according to the guidelines of the American Thoracic Society and the European Respiratory Society. For those whose percentage of FVC to FEV1 was less than 0.7, an additional bronchodilator test was performed within 20 minutes after administration of 180 grams of albuterol. Participants were divided into five categories according to the results of the pulmonary function test: normal, GOLD 1-4, which represents mild to severe COPD. All participants underwent chest CT scans at maximum inhalation and exhalation, respectively, to obtain corresponding inspiratory phase images and / or expiratory phase images. Information such as gender, age, and smoking status was also recorded. Patient information and acquisition parameters are listed in Table 1. Low-dose CT scans used a Siemens device with a tube voltage of 110 kVp and a slice thickness of 1 mm. The CTDIvol was 2.21 mGy. DICOM data were collected and converted to 3D NIfTI image format. The matrix size of each layer of the inspiratory phase images or expiratory phase images was 512x512, and the number of slices was 280-400.

[0148] In the disclosed embodiments and other possible embodiments, before acquiring the inhalation phase image and / or the exhalation phase image and the plurality of set threshold intervals, a set pre-processing threshold interval is acquired; the inhalation phase image and / or the exhalation phase image are pre-processed based on the set pre-processing threshold interval, and then the pre-processed inhalation phase image or the pre-processed exhalation phase image is subjected to airway segmentation by using a preset airway segmentation model to obtain a corresponding inhalation phase airway image or exhalation phase airway image; the pre-processed inhalation phase image is synthesized into a corresponding synthetic exhalation phase image or the pre-processed exhalation phase image is synthesized into a corresponding synthetic inhalation phase image by using a preset synthesizer; a first parameter response map is determined according to the pre-processed inhalation phase image and the corresponding synthetic exhalation phase image, the plurality of set threshold intervals; or a second parameter response map is determined according to the pre-processed exhalation phase image and the corresponding synthetic inhalation phase image, the plurality of set threshold intervals; the first parameter response map is subjected to lung blood vessel and / or set airway removal correction by using the pre-processed inhalation phase airway image or the exhalation phase airway image, and then a COPD phenotype is determined; or the second parameter response map is subjected to lung blood vessel and / or set airway removal correction by using the pre-processed inhalation phase airway image or the exhalation phase airway image, and then a COPD phenotype is determined.

[0149] For example, in the disclosed embodiments and other possible embodiments, the set pre-processing threshold interval can be configured as [-1000HU, 0HU], the Hounsfield unit (HU) value is limited in the range of [-1000HU, 0HU], the pixel value in the inhalation phase image and / or the exhalation phase image that is less than the lower limit value -1000HU of the pre-processing threshold interval is set to -1000HU, and the pixel value that is greater than the upper limit value 0HU of the pre-processing threshold interval is set to 0HU, to obtain the pre-processed inhalation phase image and / or the exhalation phase image.

[0150] In addition, the inhalation phase image or the exhalation phase image can be configured as a 2D image or a 3D image, and the 3D image can be converted into a 2D DICOM slice. Among them, 558 samples are randomly divided into a training set and a test set, wherein 449 samples contain a total of 158,455 slices for model training, and 109 samples contain 38,167 DICOM slices for testing the feasibility of the model.

[0151] At the same time, in order to verify the applicability of our model, we collected data of 62 external verification cases (data set 2). The external verification data were collected from three hospitals, using different acquisition devices, and were all conventional dose CT.

[0152] In the disclosed embodiments, a preset airway segmentation model is used to segment the airway of the inhalation phase image or the exhalation phase image to obtain a corresponding inhalation phase airway image or exhalation phase airway image.

[0153] In the disclosed embodiments and other possible embodiments, because the airway can affect the results generated by the PRM, the airway can be extracted from the fixed image (inhalation phase image or exhalation phase image) used for the registration process. The preset airway segmentation model can use an existing U-Net model based on a two-stage 3D context transformer for airway segmentation proposed by the team, which is trained using CT images. The two-stage 3D context transformer U-Net model uses, including two stages: initial airway segmentation and fine airway segmentation, the two-stage model shares the same subnetwork, and uses different airway masks as input. Using this model, the airway of the inhalation phase image or the exhalation phase image is segmented to obtain a corresponding inhalation phase airway image or exhalation phase airway image (airway tree image).

[0154] In the disclosed embodiments and other possible embodiments, the preset airway segmentation model can also be configured based on a U-Net convolutional neural network, a nnU-Net convolutional neural network, or an improved preset airway segmentation model thereof.

[0155] In the disclosed embodiments, a preset synthesizer is used to synthesize the inhalation phase image into a corresponding synthetic exhalation phase image or to synthesize the exhalation phase image into a corresponding synthetic inhalation phase image.

[0156] In the disclosed embodiments, the method of using a preset synthesizer to synthesize the inhalation phase image into a corresponding synthetic exhalation phase image includes: training the preset synthesizer using the inhalation phase image for training and its corresponding exhalation phase image; and synthesizing the inhalation phase image into a corresponding synthetic exhalation phase image based on the trained preset synthesizer.

[0157] In the disclosed embodiments, the method of training the preset synthesizer using the inhalation phase image for training and its corresponding exhalation phase image includes: using a first synthesizer G I in the preset synthesizer to synthesize the inhalation phase image for training into a corresponding first synthetic exhalation phase image; and using a second synthesizer G Econvert the first synthetic expiratory phase image into a first synthetic inspiratory phase image; calculate a cycle consistency loss between the inspiratory phase image for training and the first synthetic inspiratory phase image; and perform convolution processing on the inspiratory phase image for training and the first synthetic inspiratory phase image respectively to obtain a corresponding inspiratory phase feature map and a first synthetic inspiratory phase feature map; calculate a perceptual loss between the inspiratory phase feature map and the first synthetic inspiratory phase feature map; and use a first preset discriminator D I to determine whether the first synthetic expiratory phase image is a real image or a synthetic image; and calculate an adversarial loss based on the result of the first preset discriminator D I ; based on the cycle consistency loss, the perceptual loss, and the adversarial loss, adjust network parameters of a first synthesizer G I and a second synthesizer G E in the preset synthesizer to complete training of the preset synthesizer.

[0158] In the disclosed embodiments, the method of synthesizing the inspiratory phase image into a corresponding synthetic expiratory phase image based on the trained preset synthesizer includes: obtaining a first synthesizer G I in the trained preset synthesizer; based on the first synthesizer G I in the trained preset synthesizer, performing convolution processing on the inspiratory phase image to synthesize the inspiratory phase image into a corresponding synthetic expiratory phase image.

[0159] Similarly, in the disclosed embodiments, the method of synthesizing the expiratory phase image into a corresponding synthetic inspiratory phase image using a preset synthesizer includes: training the preset synthesizer using an expiratory phase image for training and its corresponding inspiratory phase image; and synthesizing the expiratory phase image into a corresponding synthetic inspiratory phase image based on the trained preset synthesizer.

[0160] In the disclosed embodiments, the method of training the preset synthesizer using an expiratory phase image for training and its corresponding inspiratory phase image includes: using a second synthesizer G E in the preset synthesizer to synthesize the expiratory phase image for training into a corresponding first synthetic inspiratory phase image; using a first synthesizer G I, the first synthetic inhalation phase image is converted into a first synthetic exhalation phase image; a cycle consistency loss between the exhalation phase image for training and the first synthetic exhalation phase image is calculated; and the exhalation phase image for training and the first synthetic exhalation phase image are respectively subjected to convolution processing to obtain corresponding exhalation phase feature maps and a first synthetic exhalation phase feature map; a perception loss between the exhalation phase feature map and the first synthetic exhalation phase feature map is calculated; and a second preset discriminator D E is used to determine whether the first synthetic exhalation phase image is a real image or a synthetic image; and an adversarial loss is calculated based on a result of the second preset discriminator D E ; based on the cycle consistency loss, the perception loss and the adversarial loss, network parameters of a first synthesizer G I and a second synthesizer G E in the preset synthesizer are adjusted to complete training of the preset synthesizer.

[0161] In the disclosed embodiments, the method of synthesizing the exhalation phase image into a corresponding synthetic inhalation phase image based on the trained preset synthesizer includes: obtaining a second synthesizer G E in the trained preset synthesizer; and performing convolution processing on the exhalation phase image based on the second synthesizer G E in the trained preset synthesizer to synthesize the exhalation phase image into a corresponding synthetic inhalation phase image.

[0162] Figure 4 A flowchart of a method of synthesizing the inhalation phase image into a corresponding synthetic exhalation phase image or synthesizing the exhalation phase image into a corresponding synthetic inhalation phase image using a preset synthesizer according to embodiments of the present disclosure is shown. As shown in Figure 4 , in the disclosed embodiments and other possible embodiments, a CycleGAN-based visual loss image synthesizer named PCycleGAN Figure 2 is proposed. PCycleGAN is based on two generators and two discriminators G I (the first synthesizer or generator), G E (the second synthesizer or generator), D I (the first preset discriminator), and D E (the second preset discriminator). Wherein, I represents the exhalation phase image, and E represents the exhalation phase image; wherein G I learns to map from the inhalation phase CT to the exhalation phase CT, and the second generator G E learns to map from the exhalation phase CT to the inhalation phase CT. The first discriminator D I and the second discriminator D E are responsible for determining whether an image is a real inhalation phase or a real exhalation phase CT image.E attempt to distinguish the first synthesizer G I (I) is real. Here, embodiments of the present disclosure introduce an adversarial loss. Wherein the adversarial loss aims to make the generator generate or synthesize high-quality CT images (synthetic expiratory phase images or synthetic inspiratory phase images) to fool the discriminator. Wherein the first synthesizer G I generate a fake expiratory CT image (synthetic expiratory phase image) G I (I), which is then input to the second generator G E to generate a fake inspiratory CT image (synthetic inspiratory phase image) G E (G I (I)). This image is similar to the real inspiratory phase CT. Wherein the cycle-consistency loss is computed to ensure that the generated image can be reliably recovered from the source image. To obtain the perceptual loss, the inspiratory phase image I and G E (G I (I)) are input into a convolutional model (e.g., a VGG network or other convolutional neural network model, etc.) and the difference between the high-dimensional features is computed. This process also works in the opposite direction Figure 2 (lower part).

[0163] In the disclosed embodiments and other possible embodiments, the key to the success of GANs is an adversarial loss that makes the generated photo indistinguishable from the real target image. The function of the loss is as follows:

[0164] L GAN (G,D Y ,X,Y)=E Y [logD Y (y)]+E X [log(1-D Y (G(x)))] (1)

[0165] where E denotes the expected value. It can be seen that D attempts to maximize the adversarial loss, while G attempts to minimize it. To further increase the stability of training, we replace the negative log-likelihood cost of the adversarial loss with a squared loss function (Equation 2). Wherein X represents the inspiratory phase CT image or the synthetic inspiratory phase CT image I, and Y represents the expiratory phase CT image or the synthetic expiratory phase CT image.

[0166] L GAN (D,G,X,Y)=-E Y [(D(y)-1) 2 ]-E X [D(G(x)) 2 ] (2)

[0167] In the disclosed embodiments and other possible embodiments, although the network can map the same input image to any image in the target domain, the adversarial loss cannot guarantee that a single input x i can be accurately mapped to y i . Since there are no paired images in the training, an important loss function (Formula 3) is introduced in the CycleGAN framework to ensure the consistency between the input and output images: cycle consistency loss.

[0168] L cyc (G X ,G Y )=E X [‖G Y (G X (x))-x‖1]+E Y [‖G X (G Y (y))-y‖1] (3)

[0169] In the disclosed embodiments and other possible embodiments, at the same time, CycleGAN also introduces an additional loss function (discrimination loss function, Formula 4) to ensure that the generator will not arbitrarily change the color of the input image, and to preserve its useful features by taking real samples in the target domain as the input of the generator.

[0170] L idt (G X ,G Y )=E X [‖G Y (x)-x‖1]+E Y [‖G X (y)-y‖1] (4)

[0171] In Formulas (3) and (4), ‖‖1 represents 1-norm.

[0172] The loss function L idt corresponding to the introduction of the discrimination loss or perception loss is introduced to ensure that there is no unwanted change between the inhalation and exhalation CT images, and to ensure that there is no change between the inhalation and exhalation CT images (between the inhalation phase image and the corresponding synthesized exhalation phase image, between the exhalation phase image and the corresponding synthesized inhalation phase image).

[0173] In the disclosed embodiments and other possible embodiments, the introduction of the perceptual loss function can effectively improve the visual quality and texture features of the generated images. Compared with the loss function comparing the pixel difference, the perceptual loss function extracts high-dimensional information from the images to show the high-dimensional feature difference between the real and output images. It can be assumed that there are some flow field features in the lung CT that cannot be obtained by the pixel-level loss function. The perceptual loss function more effectively shows the differences and trends between different volumes. These representations are usually obtained by inputting the images into a pre-trained VGG network. After being processed by multiple convolutions and other operations, the size of the image is reduced, while the dimension is increased, thereby representing new features. The formula is as follows:

[0174] L perc (G X ,G Y ,X)=E X,Y [‖V(G Y (G X (x)))-V(x)‖1] (5)

[0175] In formula (5), V represents a convolution model, for example, the convolution model adopts a VGG network model or other convolution neural network model.

[0176] Based on formulas (2), (3), (4) and (5), the model of the disclosed embodiments is named PCycleGAN, and its loss function is defined as follows:

[0177] L PCycleGAN =L GAN (D Y ,G X ,X,Y)+L GAN (D X ,G Y ,Y,X)+λL CYC (G X ,G Y )+L idt (G X ,G Y )+βL perc (G X ,G Y ,X)+βL perc (G Y ,G x ,Y) (6)

[0178] In formula (6), the first coefficient λ=0.2 and the second coefficient β=1. In addition, the first coefficient and the second coefficient can be configured according to actual needs by those skilled in the art.

[0179] Figure 5A network structure diagram corresponding to a preset synthesizer (PCycleGAN) according to an embodiment of the present disclosure is shown; as shown, the preset synthesizer includes two generators and two discriminators G Figure 5 I (a first synthesizer or generator), G E (a second synthesizer or generator), D I (a first preset discriminator), and D E (a second preset discriminator); wherein, Figure 5 (a) represents the first synthesizer or generator and / or the second synthesizer or generator; Figure 5 (b) represents D I (a first preset discriminator) and / or D E (a second preset discriminator); Figure 5 (c) represents a convolution model (for example, VGG16). In addition, the preset synthesizer shown can also be configured as a CycleGAN model, a Pix2Pix model, a ResVit model, a CUT model, etc.

[0180] In the Figure 5 , the real image represents the inspiration phase image or the expiration phase image; the synthetic image represents the expiration phase image or the synthetic inspiration phase image; the feature map represents the corresponding inspiration phase feature map and the first synthetic inspiration phase feature map obtained by performing convolution processing on the inspiration phase image for training and the first synthetic inspiration phase image, or the corresponding expiration phase feature map and the first synthetic expiration phase feature map obtained by performing convolution processing on the expiration phase image for training and the first synthetic expiration phase image, respectively.

[0181] In the Figure 5 , k in the convolution Cov and the transposed convolution Transposed Cov represents the number of convolution kernels, and s represents the convolution step size. Among them, the activation function Leaky ReLU can also be configured as other existing activation functions according to needs, for example, one or several of ReLU / LeakyReLU / PReLU / ELU or tanh. Among them, the batch normalization Bath Norm and the pooling are selected and configured by those skilled in the art; for example, the batch normalization Bath Norm can be selected to be used or not to be used, and at the same time, the maximum pooling Max Pooling can be used. Among them, those skilled in the art can select the number of each module; for example, ×6 represents that the number of modules is configured as 6.

[0182] In the Figure 5 , in the PCycleGAN, U-Net can be used as the backbone network of the generator. The network encoding and decoding structure used is as shown in Figure 5 ​(a) is shown, where each pair of symmetric connections (skip connections) from the encoder appears in the corresponding layer of the decoder. The input is a corresponding inhaled phase image or exhaled phase image of a 512x512 2D array with C channels. Wherein C can be set to 1, which means that the input is a single-channel array. Wherein the encoder consists of a 4x4 convolutional layer with a stride of 2, followed by six groups of LeakyReLU layers, a combination of a 4x4 convolutional layer with a stride of 2 and a normalization layer, and a group of 4x4 convolutional layers with a stride of 2 and a combination of LeakyReLU layers. The decoder consists of seven groups of ReLU layers, a combination of 4x4 transposed convolutional layers with a stride of 2 and a normalization layer, and a combination of a 4x4 up-convolutional layer with a stride of 2 and a Tanh activation function. During the downsampling process, the feature map size is halved, and the number of channels is increased after each module.

[0183] In the embodiments and other possible embodiments of the present disclosure, a Markov PatchGAN discriminator can be used to distinguish between real images and fake images and provide adversarial feedback. PatchGAN attempts to classify image patches. Figure 5 (b) shows a PatchGAN with five convolutional layers, where the input channel can be configured to 1 and the output channel can be configured to 1. For an input image, it generates a patch, and each patch point corresponds to a region in the input image. Compared with the results of PixelGAN, using PatchGAN as a discriminator has advantages in generating image details and capturing features, can reduce the number of parameters by classifying each pixel, and is easier to train.

[0184] In order to consider perceptual loss, the generator and the real image can be input into a VGG network to extract their high-dimensional feature maps, and then the loss is calculated. Further, a VGG-16 model pre-trained on an ImageNet dataset can be used, which is commonly used for classification tasks. The network structure is shown in (c). Figure 5 (c). Wherein the present disclosure uses the part before the second pooling layer in the VGG-16 model, including the first 9 convolutional layers, each of which consists of two 3x3 convolutional layers and ReLU layers, followed by a max pooling layer, and then two 3x3 convolutional layers and ReLU layers. Wherein the input image has 1 channel and the output image has 128 channels.

[0185] In the disclosed embodiments, a first parameter response map is determined according to the inhaled phase image and the corresponding synthesized exhaled phase image, a plurality of set threshold intervals; or, a second parameter response map is determined according to the exhaled phase image and the corresponding synthesized inhaled phase image, a plurality of set threshold intervals.

[0186] In the disclosed embodiments, the method for determining the first parameter response map according to the inhalation phase image and the corresponding synthetic exhalation phase image and a plurality of set threshold intervals includes: performing a registration operation on the inhalation phase image and the corresponding synthetic exhalation phase image to obtain a corresponding first registration image (pair); and determining a first parameter response map (first PRM) based on the first registration image (pair) and the plurality of set threshold intervals.

[0187] In the disclosed embodiments, the method for determining the second parameter response map according to the exhalation phase image and the corresponding synthetic inhalation phase image and a plurality of set threshold intervals includes: performing a registration operation on the exhalation phase image and the corresponding synthetic inhalation phase image to obtain a corresponding second registration image (pair); and determining a second parameter response map (second PRM) based on the second registration image (pair) and the plurality of set threshold intervals.

[0188] In the disclosed embodiments and other possible embodiments, the method for determining the first parameter response map (first PRM) based on the first registration image and the plurality of set threshold intervals includes: performing a double-threshold operation on the inhalation phase image and the corresponding synthetic exhalation phase image (first registration image) based on the plurality of set threshold intervals, respectively, to determine the first parameter response map (first PRM). Meanwhile, the method for determining the second parameter response map (second PRM) based on the second registration image (pair) and the plurality of set threshold intervals includes: performing a double-threshold operation on the exhalation phase image and the corresponding synthetic inhalation phase image (second registration image) based on the plurality of set threshold intervals, respectively, to determine the second parameter response map (second PRM).

[0189] For example, by setting different multiple threshold value intervals for the inspiratory and expiratory phase CT images, and based on the two registered images (the inspiratory phase image and the corresponding synthetic expiratory phase image; or, the expiratory phase image and the corresponding synthetic inspiratory phase image), each position in the two registered images can be classified into four categories, thereby determining the first parameter response map (first PRM) and / or the second parameter response map (second PRM). Each category corresponds to a phenotype of COPD, which are emphysema region, functional small airway to disease (fSAD) region, no feature region, and normal region (no emphysema and fSAD region) respectively; wherein, the emphysema region, the fSAD region, and the normal region are represented by red, yellow, and green respectively. Green represents normal, the attenuation value in the inspiratory phase CT image or the synthetic inspiratory phase CT image is greater than -950HU, and the voxel value (attenuation value) in the expiratory phase CT image or the synthetic expiratory phase CT image is greater than -856HU. Red represents emphysema, the attenuation value in the inspiratory phase CT image or the synthetic inspiratory phase CT image is less than -950HU, and the attenuation value in the expiratory phase CT image or the synthetic expiratory phase CT image is less than -856HU. Yellow represents the fSAD region, the attenuation value in the inspiratory phase CT image or the synthetic inspiratory phase CT image is greater than -950HU, and the attenuation value in the expiratory phase CT image or the synthetic expiratory phase CT image is less than -856HU. Further, the proportion value of each phenotype can be obtained by calculating the proportion of the number of voxels in each category to the total number of voxels in the lung field (left lung and / or right lung), thereby quantitatively analyzing COPD.

[0190] In the disclosed embodiments, before determining the first parameter response map according to the inspiratory phase image and the corresponding synthetic expiratory phase image, multiple threshold value intervals, or determining the second parameter response map according to the expiratory phase image and the corresponding synthetic inspiratory phase image, multiple threshold value intervals, a lung field segmentation model is used to perform lung field segmentation on the inspiratory phase image or the expiratory phase image, to obtain the corresponding inspiratory phase lung field image and the corresponding synthetic expiratory phase lung field image, or the expiratory phase lung field image and the corresponding synthetic inspiratory phase lung field image; and then, the first parameter response map is determined according to the inspiratory phase lung field image and the corresponding synthetic expiratory lung field image, multiple threshold value intervals; or, the second parameter response map is determined according to the expiratory phase lung field image and the corresponding synthetic inspiratory phase lung field image, multiple threshold value intervals.

[0191] In the disclosed embodiments and other possible embodiments, the method of determining a first parameter response map (first PRM) based on the first registration image and the plurality of set threshold intervals further includes: performing lung field segmentation on the inspiratory phase image or the synthesized expiratory phase image using a lung field segmentation model to obtain a first lung field image; and performing double-threshold operation on the inspiratory phase image and the corresponding synthesized expiratory phase image (first registration image) based on the plurality of set threshold intervals in the first lung field image to determine a first parameter response map (first PRM) corresponding to the first lung field image.

[0192] Meanwhile, in the disclosed embodiments and other possible embodiments, the method of determining a second parameter response map (second PRM) based on the second registration image and the plurality of set threshold intervals further includes: performing lung field segmentation on the expiratory phase image or the synthesized inspiratory phase image using a lung field segmentation model to obtain a second lung field image; and performing double-threshold operation on the expiratory phase image and the corresponding synthesized inspiratory phase image (second registration image) based on the plurality of set threshold intervals in the second lung field image to determine a second parameter response map (second PRM) corresponding to the second lung field image.

[0193] In the disclosed embodiments and other possible embodiments, the lung field segmentation model used to perform lung field segmentation on the inspiratory phase image or the synthesized expiratory phase image to obtain a first lung field image, or the lung field segmentation model used to perform lung field segmentation on the expiratory phase image or the synthesized inspiratory phase image to obtain a second lung field image, includes lung segmentation (lung field segmentation) and marking operation. The lung segmentation is to exclude the influence of external factors on the generated image effect and facilitate the subsequent registration process. Hofmanninger et al. proposed a model for segmenting the lung region and obtaining a lung marker, and the marking operation is performed using this model to extract the lung region (lung field region) from the original image (the inspiratory phase image or the synthesized expiratory phase image or the expiratory phase image or the synthesized inspiratory phase image).

[0194] In the disclosed embodiments and other possible embodiments, the lung field segmentation model can also be configured based on a U-Net convolutional neural network, a nnU-Net convolutional neural network, or an improved lung field segmentation model thereof.

[0195] In the disclosed embodiments and other possible embodiments, the registration process is performed using the Elastix tool. The registration mainly includes two steps, the first is affine transformation, which allows the translation, rotation, scaling and tilting of the image to be registered (floating image), and the second is non-rigid B-spline transformation. The B-spline transformation is modeled as a weighted sum of B-spline basis functions, which are placed on a uniform grid of control points; at the same time, the B-spline basis functions have local support, which is beneficial for fast computation. Among them, the image to be registered (moving image) can be configured as a synthetic expiratory phase image or a synthetic inspiratory phase image, and at this time the fixed image is configured as an inspiratory phase image or an expiratory phase image. Similarly, the image to be registered (floating image) can be configured as an inspiratory phase image or an expiratory phase image, and at this time the fixed image is configured as a synthetic expiratory phase image or a synthetic inspiratory phase image.

[0196] In the disclosed embodiments and other possible embodiments, other existing registration algorithms can also be selected by those skilled in the art, such as the SIFT registration algorithm, to implement the registration operation on the inspiratory phase image and the corresponding synthetic expiratory phase image or the registration operation on the expiratory phase image and the corresponding synthetic inspiratory phase image.

[0197] In the disclosed embodiments, in the setting airway pruning correction, the method for determining the setting airway includes: determining the corresponding tube wall diameter of the airway affected by COPD, and determining the setting airway based on the tube wall diameter.

[0198] In the disclosed embodiments and other possible embodiments, the tube wall diameter can be configured as 2mm, and the airway with a diameter greater than 2mm of the tube wall diameter is configured as the setting airway. Meanwhile, those skilled in the art can also configure the tube wall diameter according to actual needs.

[0199] In the disclosed embodiments, the method for determining the setting airway based on the tube wall diameter includes: respectively measuring the airway tube wall diameter in the inspiratory phase airway image or the expiratory phase airway image; and determining the airway with a tube wall diameter greater than or equal to the tube wall diameter as the setting airway.

[0200] The execution subject of the lung image processing method can be a lung image processing device, for example, the lung image processing method can be executed by a terminal device or a server or other processing device, wherein the terminal device can be a user equipment (User Equipment, UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital processing (Personal Digital Assistant, PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementation manners, the lung image processing method can be realized by a processor calling computer readable instructions stored in a memory.

[0201] Those skilled in the art can understand that, in the above-mentioned lung image processing method of the specific implementation, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process, and the specific execution order of each step should be determined according to its function and possible internal logic.

[0202] Meanwhile, the disclosure also proposes a lung image processing device, the lung image processing device comprises: a first processing unit, configured to train a first segmentation model by using a first number of first lung images and corresponding lung blood vessel label images of the first lung images; and perform lung blood vessel segmentation on a second number of second lung images based on the trained first segmentation model to obtain corresponding first lung blood vessel segmentation images; a second processing unit, configured to select the corresponding first lung blood vessel segmentation images to obtain selected first lung blood vessel segmentation images; and train a second segmentation model by using the first number of first lung images and corresponding lung blood vessel label images of the first lung images, the selected first lung blood vessel segmentation images and corresponding second lung images of the selected first lung blood vessel segmentation images; and perform lung blood vessel segmentation on a third number of second lung images remaining after the selection based on the trained second segmentation model to obtain corresponding second lung blood vessel segmentation images.

[0203] In the disclosed embodiments, the lung image processing device further comprises: a third processing unit, configured to obtain an inspiration phase image, an expiration phase image and a plurality of set threshold intervals to be diagnosed; perform blood vessel segmentation on the inspiration phase image and the expiration phase image respectively by using the above-mentioned processing method to obtain corresponding lung blood vessel images; determine corresponding parameter response maps based on the inspiration phase image, the expiration phase image and the plurality of set threshold intervals; and perform lung blood vessel and / or set airway removal correction on the parameter response maps respectively by using the corresponding lung blood vessel images to perform COPD typing diagnosis of chronic obstructive pulmonary disease.

[0204] In some embodiments, the device provided by the embodiments of the disclosure has functions or contains modules that can be used to execute the methods described in the above method embodiment, and the specific implementation can refer to the description of the above lung image processing method embodiment. For brevity, it will not be described here.

[0205] The embodiments of the disclosure also propose a computer-readable storage medium having computer program instructions stored thereon, wherein the computer program instructions are executed by a processor to implement the above-mentioned lung image processing method. The computer-readable storage medium can be a non-volatile computer-readable storage medium.

[0206] The embodiments of the disclosure also propose an electronic device, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the above-mentioned lung image processing method. The electronic device can be provided as a terminal, a server or other forms of devices.

[0207] Figure 6 is a block diagram of an electronic device 800 according to an exemplary embodiment. The electronic device 800 can be, for example, a terminal such as a mobile phone, a computer, a digital broadcasting terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0208] Referring to Figure 6 , the electronic device 800 can include one or more of the following components: a processing component 802, a memory 804, a power supply component 806, a multimedia component 808, an audio component 810, an input / output (I / O) interface 812, a sensor component 814, and a communication component 816.

[0209] The processing component 802 usually controls overall operations of the electronic device 800, such as operations associated with displaying, making phone calls, data communications, camera operations, and recording operations. The processing component 802 can include one or more processors 820 to execute instructions to complete all or part of steps of the above-described methods. In addition, the processing component 802 can include one or more modules to facilitate interaction between the processing component 802 and other components. For example, the processing component 802 can include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0210] The memory 804 is configured to store various types of data to support operations of the electronic device 800. Examples of these data include instructions for any application or method operating on the electronic device 800, contact data, phonebook data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage devices or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0211] The power supply component 806 supplies power for the various components of the electronic device 800. The power supply component 806 can include a power supply management system, one or more power supplies, and other components associated with generating, managing and distributing power for the electronic device 800.

[0212] The multimedia component 808 includes a screen to provide an output interface between the electronic device 800 and a user. In some embodiments, the screen can include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense a touch, a slide, and a gesture on the touch panel. The touch sensor can not only sense a boundary of a touching or a sliding action, but also detect duration and intensity of the touching or sliding action. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. When the electronic device 800 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera can receive external multimedia data. Each of the front camera and the rear camera can be a fixed optical lens system or have a focal length and optical zooming capability.

[0213] The audio component 810 is configured to output and / or input an audio signal. For example, the audio component 810 includes a microphone (MIC) to receive an external audio signal when the electronic device 800 is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal can be further stored in the memory 804 or transmitted via the communication component 816. In some embodiments, the audio component 810 further includes a speaker to output an audio signal.

[0214] The I / O interface 812 provides an interface for the processing component 802 and peripheral interface modules, which can be a keypad, a click wheel, buttons, and the like. The buttons can include, but are not limited to, a home button, a volume button, a start button, and a lock button.

[0215] The sensor component 814 includes one or more sensors to provide various state assessments for the electronic device 800. For example, the sensor component 814 can detect an open / closed state of the electronic device 800, relative positioning of components, such as a display and a keypad of the electronic device 800, a change in position of the electronic device 800 or a component of the electronic device 800, presence or absence of user contact with the electronic device 800, an orientation or acceleration / deceleration of the electronic device 800, and a temperature change of the electronic device 800. The sensor component 814 can include a proximity sensor configured to detect presence of a nearby object without any physical touch. The sensor component 814 can further include a light sensor such as a CMOS or CCD image sensor for use in an imaging application. In some embodiments, the sensor component 814 can further include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0216] The communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices. The electronic device 800 can access a wireless network based on a communication standard, such as WiFi, 2G, or 3G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 further includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) techniques, infrared data association (IrDA) techniques, ultra-wideband (UWB) techniques, Bluetooth (BT) techniques, and other techniques.

[0217] In an exemplary embodiment, the electronic device 800 can be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors, or other electronic elements, for performing the above-described methods.

[0218] In an exemplary embodiment, a non-transitory computer-readable storage medium, such as the memory 804 including computer program instructions, is also provided, which can be executed by the processor 820 of the electronic device 800 to complete the above-described methods.

[0219] Figure 7 is a block diagram of an electronic device 1900 according to an exemplary embodiment. For example, the electronic device 1900 can be provided as a server. Referring to Figure 7 , the electronic device 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932, for storing instructions, such as an application program, executable by the processing component 1922. The application program stored in the memory 1932 can include one or more than one module each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute the instructions to perform the above-described methods.

[0220] The electronic device 1900 can also include a power supply component 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to a network, and an input / output (I / O) interface 1958. The electronic device 1900 can operate based on an operating system stored in the memory 1932, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, or the like.

[0221] In example embodiments, a non-transitory computer-readable storage medium, e.g., memory 1932 including computer program instructions, is also provided that can be executed by processing component 1922 of electronic device 1900 to implement the above-described methods.

[0222] The present disclosure can be a system, a method, and / or a computer program product. The computer program product can include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present disclosure.

[0223] The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium can be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium include the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or punched tape, a

[0224] The computer readable program instructions described herein can be downloaded to respective computing / processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and / or a wireless network. The network can comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing / processing device.

[0225] Computer readable program instructions for carrying out operations of the present disclosure can be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate array (FPGA), or programmable logic array (PLA) can execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present disclosure.

[0226] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0227] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0228] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus or other device to produce a computer implemented process such that the instructions which execute on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0229] The flow diagrams and the block diagrams in the drawings are presented to illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flow diagrams and the block diagrams can represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logic functions. In some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks can sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flow diagrams, and combinations thereof, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and

[0230] Embodiments of the present disclosure have been described above, and the description is intended to be illustrative of the embodiments and not restrictive. Many modifications and variations of the described embodiments are possible and are within the scope of the disclosure. The selection of the terms to be used in the description is not intended to limit the scope of the embodiments described herein, but rather to best describe the principles of the embodiments in the context of the specific application.

Claims

1. A lung image processing method, characterized by, The method comprises the following steps: training a first segmentation model by using a first number of first lung images and their corresponding lung blood vessel label images; segmenting lung blood vessels of a second number of second lung images based on the trained first segmentation model to obtain corresponding first lung blood vessel segmentation images; selecting the first lung blood vessel segmentation images corresponding to the second number of second lung images to obtain selected first lung blood vessel segmentation images, comprising: obtaining a segmentation index corresponding to each first lung blood vessel segmentation image and a first set segmentation index; if the segmentation index corresponding to the first lung blood vessel segmentation image is greater than or equal to the first set segmentation index, the first lung blood vessel segmentation image corresponding to the first set segmentation index greater than or equal to the first set segmentation index is determined as the selected first lung blood vessel segmentation image; training a second segmentation model by using the first number of first lung images and their corresponding lung blood vessel label images, the selected first lung blood vessel segmentation images and their corresponding second lung images; segmenting lung blood vessels of a third number of second lung images remaining after selecting the first lung blood vessel segmentation images based on the trained second segmentation model to obtain corresponding second lung blood vessel segmentation images; if the performance index corresponding to the second segmentation model is lower than the set performance index, selecting the second lung blood vessel segmentation images corresponding to the third number of second lung images remaining after selecting the first lung blood vessel segmentation images to obtain selected second lung blood vessel segmentation images, comprising: obtaining a segmentation index corresponding to each second lung blood vessel segmentation image and a second set segmentation index; if the segmentation index corresponding to the second lung blood vessel segmentation image is greater than or equal to the second set segmentation index, the second lung blood vessel segmentation image corresponding to the second set segmentation index greater than or equal to the second set segmentation index is determined as the selected second lung blood vessel segmentation image; training a third segmentation model by using the first number of first lung images and their corresponding lung blood vessel label images, the selected second lung blood vessel segmentation images and their corresponding second lung images; segmenting lung blood vessels of a fourth number of second lung images remaining after selecting the second lung images based on the trained third segmentation model to obtain corresponding second lung blood vessel segmentation images; repeating the above process until the performance index of the final segmentation model is higher than or equal to the set performance index.

2. The treatment method according to claim 1, characterized in that, Before the step of selecting the second lung blood vessel segmentation images corresponding to the third number of second lung images remaining after selecting the first lung blood vessel segmentation images to obtain selected second lung blood vessel segmentation images if the performance index corresponding to the second segmentation model is lower than the set performance index, the method comprises the following step:

3. The treatment method according to any one of claims 1-2, characterized in that, obtaining the performance index corresponding to the second segmentation model for segmenting lung blood vessels of the third number of second lung images remaining after selecting the first lung blood vessel segmentation images. The method for determining the lung blood vessel label images corresponding to the first number of first lung images comprises the following steps: using a trained preset lung blood vessel convolution segmentation model to perform first lung blood vessel segmentation on the first number of first lung images respectively to obtain corresponding first lung blood vessel label images; The first number of first lung images are respectively subjected to second lung blood vessel segmentation by using a machine learning segmentation model, to obtain corresponding second lung blood vessel label images; wherein, the tube diameter of lung blood vessels in the second lung blood vessel label images is smaller than the tube diameter of lung blood vessels in the first lung blood vessel label images; The first lung blood vessel label images and the second lung blood vessel label images are respectively fused to obtain lung blood vessel label images corresponding to the first number of first lung images.

4. The treatment method according to claim 3, characterized in that, The first lung blood vessel label images and the second lung blood vessel label images are respectively fused to obtain lung blood vessel label images corresponding to the first number of first lung images. The first lung blood vessel label images and the second lung blood vessel label images are respectively fused to obtain lung blood vessel label images corresponding to the first number of first lung images.

5. The treatment method of claim 3, wherein The first lung blood vessel label images and the second lung blood vessel label images are respectively fused to obtain lung blood vessel label images corresponding to the first number of first lung images. The first number of first lung images are respectively subjected to second lung blood vessel segmentation by using a machine learning segmentation model, to obtain corresponding second lung blood vessel label images, including: The first number of first lung images are respectively subjected to multi-scale representation to obtain corresponding multi-scale lung images; 6. The treatment method of claim 4, wherein Features of the multi-scale lung images are respectively extracted, and the extracted features are classified by using a preset classifier to obtain corresponding second lung blood vessel label images. The first number of first lung images are respectively subjected to multi-scale representation to obtain corresponding multi-scale lung images; Features of the multi-scale lung images are respectively extracted, and the extracted features are classified by using a preset classifier to obtain corresponding second lung blood vessel label images.

7. The treatment method according to any one of claims 1, 2, 4-6, characterized in that, In the process of training the segmentation model, the loss of the segmentation model is calculated, and the network parameters of the segmentation model are adjusted by using the loss.

8. The treatment method of claim 3, wherein In the process of training the segmentation model, the loss of the segmentation model is calculated, and the network parameters of the segmentation model are adjusted by using the loss.

9. The treatment method according to claim 7, characterized in that, The loss of the segmentation model is calculated, including: A Dice loss function and a cross-entropy loss function are obtained; First loss values and second loss values of the Dice loss function and the cross-entropy loss function are respectively calculated; The sum of the first loss values and the second loss values is configured as the loss value of the segmentation model.

10. The processing method according to claim 8, wherein The loss of the segmentation model is calculated, including: A Dice loss function and a cross-entropy loss function are obtained; First loss values and second loss values of the Dice loss function and the cross-entropy loss function are respectively calculated; The sum of the first loss values and the second loss values is configured as the loss value of the segmentation model.

11. A treatment method according to any of claims 9 or 10, characterised in that, The loss of the segmentation model is calculated, including: The loss adjustment of the Dice loss function and the cross-entropy loss function is set; Based on the loss adjustment, it is determined whether the segmentation model passes by selecting the Dice loss function and the cross-entropy loss function to calculate the loss.

12. The treatment method of claim 11, wherein, The loss adjustment of the Dice loss function and the cross-entropy loss function is set, including: Calculate a plurality of difference values of segmentation pixels between a lung vessel segmentation image corresponding to a lung image to be segmented by the segmentation model and a lung vessel label image corresponding to the lung image to be segmented, which are output by the segmentation model; Calculate a mean value between the plurality of difference values; If the mean value is less than a set value, the segmentation model does not calculate a loss; Otherwise, the segmentation model calculates the loss by selecting the Dice loss function and the cross-entropy loss function.

13. The treatment method according to any one of claims 1, 2, 4-6, 8-10, 12, characterized in that, Before the first segmentation model is trained using the first number of first lung images and the lung vessel label images corresponding to the first lung images, the method further includes: obtaining the first number of first lung images and the second number of second lung images.

14. The treatment method of claim 3, wherein, Before the first segmentation model is trained using the first number of first lung images and the lung vessel label images corresponding to the first lung images, the method further includes: obtaining the first number of first lung images and the second number of second lung images.

15. The treatment method of claim 7, wherein, Before the first segmentation model is trained using the first number of first lung images and the lung vessel label images corresponding to the first lung images, the method further includes: obtaining the first number of first lung images and the second number of second lung images.

16. The processing method of claim 11, wherein, Before the first segmentation model is trained using the first number of first lung images and the lung vessel label images corresponding to the first lung images, the method further includes: obtaining the first number of first lung images and the second number of second lung images.

17. The treatment method of any one of claims 1, 2, 4-6, 8-10, 12, 14-16, wherein, Before the first segmentation model is trained using the first number of first lung images and the lung vessel label images corresponding to the first lung images, the method further includes: Perform lung field segmentation on the first number of first lung images and the second number of second lung images respectively to obtain corresponding first lung field images and second lung field images; Train the first segmentation model using the first number of first lung field images and the lung vessel label images corresponding to the first lung field images; Perform lung vessel segmentation on the second number of second lung field images based on the trained first segmentation model to obtain corresponding first lung vessel segmentation images; Select the first lung vessel segmentation images corresponding to the second number of second lung images to obtain selected first lung vessel segmentation images; Train the second segmentation model using the first number of first lung field images and the lung vessel label images corresponding to the first lung field images, the selected first lung vessel segmentation images, and the second lung field images corresponding to the selected first lung vessel segmentation images; Perform lung vessel segmentation on the third number of second lung field images remaining after the first lung vessel segmentation images are selected based on the trained second segmentation model to obtain corresponding second lung vessel segmentation images.

18. The treatment method of claim 3, wherein, Before the first segmentation model is trained using the first number of first lung images and the lung vessel label images corresponding to the first lung images, the method further includes: Perform lung field segmentation on the first number of first lung images and the second number of second lung images respectively to obtain corresponding first lung field images and second lung field images; Train the first segmentation model using the first number of first lung field images and the lung vessel label images corresponding to the first lung field images; Perform lung vessel segmentation on the second number of second lung field images based on the trained first segmentation model to obtain corresponding first lung vessel segmentation images; selecting first lung vessel segmentation images corresponding to the second number of second lung images to obtain selected first lung vessel segmentation images; training a second segmentation model by using the first number of first lung field images and their corresponding lung vessel label images, the selected first lung vessel segmentation images and their corresponding second lung field images; performing lung vessel segmentation on the third number of second lung field images remaining after the first lung vessel segmentation images are selected based on the trained second segmentation model to obtain corresponding second lung vessel segmentation images.

19. The treatment method of claim 7, wherein, Before the training of the first segmentation model by using the first number of first lung images and their corresponding lung vessel label images, the method comprises: performing lung field segmentation on the first number of first lung images and the second number of second lung images respectively to obtain corresponding first lung field images and second lung field images; training a first segmentation model by using the first number of first lung field images and their corresponding lung vessel label images; performing lung vessel segmentation on the second number of second lung field images based on the trained first segmentation model to obtain corresponding first lung vessel segmentation images; selecting first lung vessel segmentation images corresponding to the second number of second lung images to obtain selected first lung vessel segmentation images; training a second segmentation model by using the first number of first lung field images and their corresponding lung vessel label images, the selected first lung vessel segmentation images and their corresponding second lung field images; performing lung vessel segmentation on the third number of second lung field images remaining after the first lung vessel segmentation images are selected based on the trained second segmentation model to obtain corresponding second lung vessel segmentation images.

20. The processing method of claim 11, wherein, Before the training of the first segmentation model by using the first number of first lung images and their corresponding lung vessel label images, the method comprises: performing lung field segmentation on the first number of first lung images and the second number of second lung images respectively to obtain corresponding first lung field images and second lung field images; training a first segmentation model by using the first number of first lung field images and their corresponding lung vessel label images; performing lung vessel segmentation on the second number of second lung field images based on the trained first segmentation model to obtain corresponding first lung vessel segmentation images; selecting first lung vessel segmentation images corresponding to the second number of second lung images to obtain selected first lung vessel segmentation images; training a second segmentation model by using the first number of first lung field images and their corresponding lung vessel label images, the selected first lung vessel segmentation images and their corresponding second lung field images; performing lung vessel segmentation on the third number of second lung field images remaining after the first lung vessel segmentation images are selected based on the trained second segmentation model to obtain corresponding second lung vessel segmentation images.

21. The treatment method of claim 13, wherein, Before the training of the first segmentation model by using the first number of first lung images and their corresponding lung vessel label images, the method comprises: performing lung field segmentation on the first number of first lung images and the second number of second lung images respectively to obtain corresponding first lung field images and second lung field images; Training a first segmentation model by using a first number of first lung field images and corresponding lung blood vessel label images thereof; Segmenting lung blood vessels of a second number of second lung field images based on the trained first segmentation model to obtain corresponding first lung blood vessel segmentation images; Selecting the first lung blood vessel segmentation images corresponding to the second number of second lung images to obtain selected first lung blood vessel segmentation images; Training a second segmentation model by using the first number of first lung field images and corresponding lung blood vessel label images thereof, the selected first lung blood vessel segmentation images and corresponding second lung field images thereof; Segmenting lung blood vessels of a third number of second lung field images remaining after the first lung blood vessel segmentation images are selected based on the trained second segmentation model to obtain corresponding second lung blood vessel segmentation images.

22. The treatment method of any one of claims 1, 2, 4-6, 8-10, 12, 14-16, 18-21, wherein, Further comprising: Obtaining an inhalation phase image, an exhalation phase image and a plurality of set threshold intervals to be diagnosed; Segmenting blood vessels of the inhalation phase image and the exhalation phase image respectively by using the lung image processing method to obtain corresponding lung blood vessel images; Determining corresponding parameter response maps based on the inhalation phase image, the exhalation phase image and the plurality of set threshold intervals; Performing lung blood vessel and / or set airway removal correction on the parameter response maps respectively by using the corresponding lung blood vessel images to perform subtyping diagnosis of chronic obstructive pulmonary disease.

23. The treatment method of claim 3, wherein, Further comprising: Obtaining an inhalation phase image, an exhalation phase image and a plurality of set threshold intervals to be diagnosed; Segmenting blood vessels of the inhalation phase image and the exhalation phase image respectively by using the lung image processing method to obtain corresponding lung blood vessel images; Determining corresponding parameter response maps based on the inhalation phase image, the exhalation phase image and the plurality of set threshold intervals; Performing lung blood vessel and / or set airway removal correction on the parameter response maps respectively by using the corresponding lung blood vessel images to perform subtyping diagnosis of chronic obstructive pulmonary disease.

24. The treatment method of claim 7, wherein, Further comprising: Obtaining an inhalation phase image, an exhalation phase image and a plurality of set threshold intervals to be diagnosed; Segmenting blood vessels of the inhalation phase image and the exhalation phase image respectively by using the lung image processing method to obtain corresponding lung blood vessel images; Determining corresponding parameter response maps based on the inhalation phase image, the exhalation phase image and the plurality of set threshold intervals; Performing lung blood vessel and / or set airway removal correction on the parameter response maps respectively by using the corresponding lung blood vessel images to perform subtyping diagnosis of chronic obstructive pulmonary disease.

25. The method of claim 11, wherein, Further comprising: Obtaining an inhalation phase image, an exhalation phase image and a plurality of set threshold intervals to be diagnosed; Segmenting blood vessels of the inhalation phase image and the exhalation phase image respectively by using the lung image processing method to obtain corresponding lung blood vessel images; Determining corresponding parameter response maps based on the inhalation phase image, the exhalation phase image and the plurality of set threshold intervals; Performing lung blood vessel and / or set airway removal correction on the parameter response maps respectively by using the corresponding lung blood vessel images to perform subtyping diagnosis of chronic obstructive pulmonary disease.

26. The treatment method of claim 13, wherein, Further comprising: Obtaining an inhalation phase image, an exhalation phase image and a plurality of set threshold intervals to be diagnosed; Segmenting blood vessels of the inhalation phase image and the exhalation phase image respectively by using the lung image processing method to obtain corresponding lung blood vessel images; Determining corresponding parameter response maps based on the inhalation phase image, the exhalation phase image and the plurality of set threshold intervals; Performing lung blood vessel and / or set airway removal correction on the parameter response maps respectively by using the corresponding lung blood vessel images to perform subtyping diagnosis of chronic obstructive pulmonary disease. determining a corresponding parameter response graph based on the inspiration phase image, expiration phase image and multiple set threshold intervals; performing lung blood vessel and / or set airway removal correction on the parameter response graph by using the corresponding lung blood vessel image, and performing chronic obstructive pulmonary disease typing diagnosis.

27. The method of claim 17, wherein, Further comprising: obtaining an inspiration phase image, expiration phase image and multiple set threshold intervals to be diagnosed; performing blood vessel segmentation on the inspiration phase image and expiration phase image by using the lung image processing method, and obtaining corresponding lung blood vessel images; determining a corresponding parameter response graph based on the inspiration phase image, expiration phase image and multiple set threshold intervals; performing lung blood vessel and / or set airway removal correction on the parameter response graph by using the corresponding lung blood vessel image, and performing chronic obstructive pulmonary disease typing diagnosis.

28. The method of claim 22, wherein, when the expiration phase image corresponding to the inspiration phase image to be diagnosed is missing, using a preset synthesizer to synthesize the inspiration phase image into a corresponding synthesized expiration phase image; performing blood vessel segmentation on the inspiration phase image and the corresponding synthesized expiration phase image by using the lung image processing method, and obtaining a corresponding first lung blood vessel image; determining a first parameter response graph based on the inspiration phase image, the corresponding synthesized expiration phase image and multiple set threshold intervals; performing lung blood vessel and / or set airway removal correction on the first parameter response graph by using the corresponding first lung blood vessel image, and obtaining a corrected first parameter response graph; performing chronic obstructive pulmonary disease typing diagnosis based on the corrected first parameter response graph.

29. The treatment method according to any one of claims 23-27, characterized by, when the expiration phase image corresponding to the inspiration phase image to be diagnosed is missing, using a preset synthesizer to synthesize the inspiration phase image into a corresponding synthesized expiration phase image; performing blood vessel segmentation on the inspiration phase image and the corresponding synthesized expiration phase image by using the lung image processing method, and obtaining a corresponding first lung blood vessel image; determining a first parameter response graph based on the inspiration phase image, the corresponding synthesized expiration phase image and multiple set threshold intervals; performing lung blood vessel and / or set airway removal correction on the first parameter response graph by using the corresponding first lung blood vessel image, and obtaining a corrected first parameter response graph; performing chronic obstructive pulmonary disease typing diagnosis based on the corrected first parameter response graph.

30. The method of claim 22, wherein, when the expiration phase image corresponding to the inspiration phase image to be diagnosed is missing, using a preset synthesizer to synthesize the inspiration phase image into a corresponding synthesized expiration phase image; performing blood vessel segmentation on the inspiration phase image and the corresponding synthesized expiration phase image by using the lung image processing method, and obtaining a corresponding first lung blood vessel image; determining a first parameter response graph based on the inspiration phase image, the corresponding synthesized expiration phase image and multiple set threshold intervals; performing lung blood vessel and / or set airway removal correction on the first parameter response graph by using the corresponding first lung blood vessel image, and obtaining a corrected first parameter response graph; performing chronic obstructive pulmonary disease typing diagnosis based on the corrected first parameter response graph.

31. The treatment method according to any one of claims 23-27, characterized by, When the inhalation phase image corresponding to the exhalation phase image to be diagnosed is absent, a preset synthesizer is used to synthesize the exhalation phase image into a corresponding synthesized inhalation phase image; Blood vessel segmentation is performed on the exhalation phase image and the corresponding synthesized inhalation phase image by using the lung image processing method to obtain a corresponding second lung blood vessel image; A second parameter response map is determined according to the exhalation phase image, the corresponding synthesized inhalation phase image, and a plurality of set threshold intervals; Pulmonary blood vessels and / or set airway removal correction is performed on the second parameter response map by using the corresponding second lung blood vessel image to obtain a corrected second parameter response map; Chronic obstructive pulmonary disease is diagnosed based on the corrected second parameter response map.

32. A lung image processing apparatus, characterized by, Comprise: A first processing unit is configured to train a first segmentation model by using a first number of first lung images and corresponding lung blood vessel label images thereof, and perform lung blood vessel segmentation on a second number of second lung images based on the trained first segmentation model to obtain corresponding first lung blood vessel segmentation images; A second processing unit is configured to select the first lung blood vessel segmentation images corresponding to the second number of second lung images to obtain selected first lung blood vessel segmentation images, including: obtaining a segmentation index corresponding to each first lung blood vessel segmentation image and a first set segmentation index; if the segmentation index corresponding to the first lung blood vessel segmentation image is greater than or equal to the first set segmentation index, the first lung blood vessel segmentation image corresponding to the first set segmentation index is determined as the selected first lung blood vessel segmentation image; a second segmentation model is trained by using the first number of first lung images and corresponding lung blood vessel label images thereof, the selected first lung blood vessel segmentation images, and corresponding second lung images thereof; lung blood vessel segmentation is performed on a third number of second lung images remaining after the first lung blood vessel segmentation images are selected based on the trained second segmentation model to obtain corresponding second lung blood vessel segmentation images; if the performance index corresponding to the second segmentation model is lower than a set performance index, the second lung blood vessel segmentation images corresponding to the third number of second lung images remaining after the first lung blood vessel segmentation images are selected are selected to obtain selected second lung blood vessel segmentation images, including: obtaining a segmentation index corresponding to each second lung blood vessel segmentation image and a second set segmentation index; if the segmentation index corresponding to the second lung blood vessel segmentation image is greater than or equal to the second set segmentation index, the second lung blood vessel segmentation image corresponding to the second set segmentation index is determined as the selected second lung blood vessel segmentation image; a third segmentation model is trained by using the first number of first lung images and corresponding lung blood vessel label images thereof, the selected second lung blood vessel segmentation images, and corresponding second lung images thereof; lung blood vessel segmentation is performed on a fourth number of second lung images remaining after the second lung images are selected based on the trained third segmentation model to obtain corresponding second lung blood vessel segmentation images; the above process is repeated until the performance index of the final segmentation model is higher than or equal to the set performance index.

33. An electronic device, comprising: Comprise: A processor; A memory for storing processor-executable instructions; The processor is configured to invoke the instructions stored in the memory to perform the lung image processing method in any one of claims 1-31.

34. A computer-readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions, when executed by the processor, implement the lung image processing method in any one of claims 1-31.

Images