Method and apparatus for segmenting pulmonary vessels, electronic device, and storage medium

Abstract

The present disclosure relates to a method and apparatus for segmenting pulmonary blood vessels, an electronic device, and a storage medium, and relates to the technical field of pulmonary blood vessel segmentation. The method for segmenting pulmonary blood vessels includes: obtaining an enhanced lung image and its corresponding pulmonary blood vessel image, a first non-enhanced lung image, and a preset first segmentation model; performing a registration operation on the enhanced lung image and the first non-enhanced lung image to obtain a lung registration image corresponding to the first non-enhanced lung image; performing lung region segmentation on the lung registration image to obtain a lung region image; training the preset first segmentation model based on the pulmonary blood vessel image and the lung region image; and performing pulmonary blood vessel segmentation on a second non-enhanced lung image based on the trained preset first segmentation model. Embodiments of the present disclosure can implement non-enhanced lung images of non-enhanced lung images.

Description

Technical Field

[0001] The present disclosure relates to the technical field of pulmonary vascular segmentation, and in particular, to a method and apparatus for segmenting pulmonary vessels, an electronic device, and a storage medium. Background Art

[0002] The lungs are an important organ of the human respiratory system and are the site of gas exchange. Due to reasons such as smoking and air pollution, lung diseases have been on the rise in recent years, seriously threatening people's physical health. High-spatial-resolution chest CT [4] images for lung examinations can diagnose lung diseases and provide good features of the lesion location and size.

[0003] Accurately segmenting pulmonary vessels from chest CT images is necessary and complex. It can be used for the screening, diagnosis, and treatment of pulmonary embolism, pulmonary hypertension, or pulmonary nodules. In addition, the segmented pulmonary vessels can be used as an auxiliary guide for the resection of lung lobes and segments in lung cancer. However, due to the complex and variable morphology of pulmonary vessels, the uneven thickness of vessel diameters, the uneven gray values of vessels, and the similarity of gray values to surrounding tissues, it brings certain difficulties to the accurate segmentation of vessels.

[0004] Many rule-based methods have been proposed to segment pulmonary vessels. Methods based on Hessian matrix enhancement mostly show good performance. However, the enhancement effect of Hessian eigenvalues on vessels with non-regular tubular structures is not good, and further segmentation is required to extract the final vascular tree. Rule-based methods also include graph-based models such as graph cuts. The graph cut method first models the vessels and then finds the optimal cut on the entire graph by using an optimization method. Graph-based methods require appropriate preprocessing and parameter adjustment in combination with prior knowledge.

[0005] Deep learning methods rely on labeled data, and the quality of labeling directly determines the performance of the model. Due to the complex anatomical structure and CT values similar to other tissues (airway walls and nodules), especially in non-enhanced CT (NCCT) lung images, the labeling of pulmonary vessels is a time-consuming and laborious process, and the labeled samples are scarce and inaccurate. By intravenous injection of a contrast agent, contrast-enhanced CT (CECT) lung images can show pulmonary vessels. However, for some patients with contrast agent contraindications and patients without CECT lung images, it is necessary to segment pulmonary vessels from NCCT lung images. Summary of the Invention

[0006] The present disclosure provides a technical solution for a method and apparatus for segmenting pulmonary vessels, an electronic device, and a storage medium.

[0007] According to one aspect of the present disclosure, there is provided a method for segmenting pulmonary vessels, including:

[0008] Obtain an enhanced lung image and its corresponding pulmonary vascular image, a first non-enhanced lung image, and a preset first segmentation model;

[0009] Perform a registration operation on the enhanced lung image and the first non-enhanced lung image to obtain a lung registration image corresponding to the first non-enhanced lung image;

[0010] Perform lung region segmentation on the lung registration image to obtain a lung region image;

[0011] Train the preset first segmentation model based on the pulmonary vascular image and the lung region image;

[0012] Perform pulmonary vascular segmentation on a second non-enhanced lung image based on the trained preset first segmentation model.

[0013] Preferably, before obtaining the enhanced lung image and its corresponding pulmonary vascular image, perform vascular segmentation and lung region segmentation on the obtained enhanced lung image respectively to obtain a vascular image and a lung region image;

[0014] Obtain the pulmonary vascular image corresponding to the enhanced lung image according to the vascular image and the lung region image.

[0015] Preferably, the method for performing vascular segmentation and lung region segmentation on the obtained enhanced lung image respectively includes: obtaining a preset vascular segmentation model and a first preset lung region segmentation model;

[0016] Perform vascular segmentation and lung region segmentation on the obtained enhanced lung image respectively based on the preset vascular segmentation model and the first preset lung region segmentation model to obtain a vascular image and a lung region image;

[0017] And / or,

[0018] The method for obtaining the pulmonary vascular image corresponding to the enhanced lung image according to the vascular image and the lung region image includes:

[0019] Perform a multiplication operation on the vascular image and the lung region image to obtain the pulmonary vascular image corresponding to the enhanced lung image.

[0020] Preferably, the method for performing lung region segmentation on the lung registration image to obtain a lung region image includes:

[0021] Obtain a second preset lung region segmentation model;

[0022] Perform lung region segmentation on the lung registration image based on the second preset lung region segmentation model to obtain a lung region image;

[0023] And / or,

[0024] The first segmentation model includes: a U-Net backbone network, which obtains multiple feature maps after each convolution operation; performs feature map normalization on the multiple feature maps respectively; and activates the normalized multiple feature maps by using an activation function.

[0025] And / or

[0026] The method for training the preset first segmentation model based on the pulmonary vascular image and the pulmonary region image includes:

[0027] During the decoding process of the first segmentation model, calculate the loss of the corresponding feature map after each decoding; and obtain the total loss during the training process according to the loss of the corresponding feature map after each decoding.

[0028] According to one aspect of the present disclosure, there is provided a method for segmenting pulmonary vessels, including:

[0029] Obtain an enhanced lung image and its corresponding pulmonary vascular image, a first non-enhanced lung image, and a preset first segmentation model;

[0030] Segment the enhanced lung image and the first non-enhanced lung image respectively to obtain a first pulmonary region image and a second pulmonary region image;

[0031] Perform a registration operation on the first pulmonary region image and the second pulmonary region image to obtain a pulmonary region registration image corresponding to the second pulmonary region image;

[0032] Train the preset first segmentation model based on the pulmonary region registration image and the pulmonary vascular image;

[0033] Segment the pulmonary vessels of the second non-enhanced lung image based on the trained preset first segmentation model.

[0034] Preferably, before obtaining the enhanced lung image and its corresponding pulmonary vascular image, perform vascular segmentation and pulmonary region segmentation on the obtained enhanced lung image respectively to obtain a vascular image and a pulmonary region image;

[0035] Obtain the pulmonary vascular image corresponding to the enhanced lung image according to the vascular image and the pulmonary region image;

[0036] And / or

[0037] The method for respectively performing vascular segmentation and pulmonary region segmentation on the obtained enhanced lung image includes: obtaining a preset vascular segmentation model and a first preset pulmonary region segmentation model;

[0038] Based on the preset blood vessel segmentation model and the first preset lung region segmentation model respectively, perform blood vessel segmentation and lung region segmentation on the obtained enhanced lung image to obtain a blood vessel image and a lung region image;

[0039] and / or

[0040] The method for obtaining the lung blood vessel image corresponding to the enhanced lung image according to the blood vessel image and the lung region image includes:

[0041] Perform a multiplication operation on the blood vessel image and the lung region image to obtain the lung blood vessel image corresponding to the enhanced lung image.

[0042] Preferably, the method for respectively segmenting the enhanced lung image and the first non-enhanced lung image to obtain a first lung region image and a second lung region image includes:

[0043] Obtain a third preset lung region segmentation model and a fourth preset lung region segmentation model;

[0044] Based on the third preset lung region segmentation model and the fourth preset lung region segmentation model respectively, segment the enhanced lung image and the first non-enhanced lung image to obtain a first lung region image and a second lung region image;

[0045] and / or

[0046] The first segmentation model includes: a U-Net backbone network. After each convolution operation of the U-Net backbone network, multiple feature maps are obtained; the multiple feature maps are respectively subjected to feature mapping normalization; and the normalized multiple feature maps are activated by using an activation function;

[0047] and / or

[0048] The method for training the preset first segmentation model based on the lung blood vessel image and the lung region image includes:

[0049] During the decoding process of the first segmentation model, calculate the loss of the corresponding feature map after each decoding; according to the loss of the corresponding feature map after each decoding, obtain the total loss during the training process.

[0050] According to an aspect of the present disclosure, there is provided a lung blood vessel segmentation device, including:

[0051] A first acquisition unit, configured to acquire an enhanced lung image and its corresponding lung blood vessel image, a first non-enhanced lung image, and a preset first segmentation model;

[0052] The first registration unit is configured to perform a registration operation on the enhanced lung image and the first non-enhanced lung image to obtain a lung registration image corresponding to the first non-enhanced lung image;

[0053] The first lung region segmentation unit is configured to perform lung region segmentation on the lung registration image to obtain a lung region image;

[0054] The first training unit is configured to train the preset first segmentation model based on the lung vascular image and the lung region image;

[0055] The first lung vascular segmentation unit is configured to perform lung vascular segmentation on a second non-enhanced lung image based on the trained preset first segmentation model.

[0056] According to one aspect of the present disclosure, there is provided a method for segmenting lung vessels, including:

[0057] The second acquisition unit acquires an enhanced lung image and its corresponding lung vascular image, a first non-enhanced lung image, and a preset first segmentation model;

[0058] The second lung region segmentation unit respectively segments the enhanced lung image and the first non-enhanced lung image to obtain a first lung region image and a second lung region image;

[0059] The second registration unit performs a registration operation on the first lung region image and the second lung region image to obtain a lung region registration image corresponding to the second lung region image;

[0060] The second training unit trains the preset first segmentation model based on the lung region registration image and the lung vascular image;

[0061] The second lung vascular segmentation unit performs lung vascular segmentation on a second non-enhanced lung image based on the trained preset first segmentation model.

[0062] According to one aspect of the present disclosure, there is provided an electronic device, including:

[0063] A processor;

[0064] A memory for storing instructions executable by the processor;

[0065] Wherein, the processor is configured to: execute the above method for segmenting lung vessels.

[0066] According to one aspect of the present disclosure, there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the above method for segmenting lung vessels is implemented.

[0067] In the embodiments of the present disclosure, the technical problems to be solved are as follows: Due to the complex anatomical structure and values similar to other tissues (airway wall and nodules), especially in non-enhanced lung images, the annotation of pulmonary vessels is a time-consuming and laborious process, with scarce and inaccurate annotation samples; By intravenous injection of a contrast agent, enhanced lung images can show pulmonary vessels. However, for some patients with contrast agent contraindications and patients without enhanced lung images, it is necessary to segment pulmonary vessels from non-enhanced lung images.

[0068] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and do not limit the present disclosure.

[0069] Other features and aspects of the present disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification. These drawings illustrate embodiments consistent with the present disclosure and, together with the specification, are used to explain the technical solutions of the present disclosure.

[0071] Figure 1 A flowchart showing a method for segmenting pulmonary vessels according to an embodiment of the present disclosure;

[0072] Figure 2 A flowchart showing a method for segmenting pulmonary vessels according to another embodiment of the present disclosure;

[0073] Figure 3 Showing in accordance with an embodiment of the present disclosure Figure 1 The implementation process of the corresponding method for segmenting pulmonary vessels;

[0074] Figure 4 A network schematic diagram showing a first segmentation model according to an embodiment of the present disclosure;

[0075] Figure 5 A block diagram showing an electronic device according to an exemplary embodiment;

[0076] Figure 6 A block diagram showing an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0077] The following will detail various exemplary embodiments, features, and aspects of the present disclosure with reference to the accompanying drawings. The same reference numerals in the drawings denote elements having the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings do not have to be drawn to scale unless otherwise specified.

[0078] As used herein, the term "exemplary" means "serving as an example, embodiment, or illustration". Any embodiment described as "exemplary" herein should not necessarily be construed as superior to or better than other embodiments.

[0079] As used herein, the term "and / or" is merely a description of the relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent three cases: A exists alone, A and B exist simultaneously, and B exists alone. Additionally, the term "at least one" as used herein means any one of a plurality or any combination of at least two of a plurality. For example, including at least one of A, B, and C can represent including any one or more elements selected from the set consisting of A, B, and C.

[0080] In addition, for a better illustration of the present disclosure, numerous specific details are given in the following detailed description. Those skilled in the art should understand that the present disclosure can still be implemented without certain specific details. In some instances, methods, means, elements, and circuits well-known to those skilled in the art are not described in detail to highlight the gist of the present disclosure.

[0081] It can be understood that, without violating the principle logic, the above-mentioned various method embodiments mentioned in the present disclosure can be combined with each other to form a combined embodiment. Due to space limitations, the present disclosure will not elaborate further.

[0082] In addition, the present disclosure also provides a pulmonary vascular segmentation device, an electronic device, a computer-readable storage medium, and a program, all of which can be used to implement any one of the pulmonary vascular segmentation methods provided by the present disclosure. For the corresponding technical solutions and descriptions, reference can be made to the corresponding records in the method section, which will not be elaborated further.

[0083] Figure 1 A flowchart showing the pulmonary vascular segmentation method according to an embodiment of the present disclosure is presented. At the same time, Figure 3 shows in the embodiment according to the present disclosure Figure 1 the implementation process of the corresponding pulmonary vascular segmentation method. As Figure 1 well as Figure 3As shown, the method for segmenting pulmonary vessels includes: Step S101: Obtain an enhanced lung image, its corresponding pulmonary vessel image, a first non-enhanced lung image, and a preset first segmentation model; Step S102: Perform a registration operation on the enhanced lung image and the first non-enhanced lung image to obtain a lung registration image corresponding to the first non-enhanced lung image; Step S103: Perform pulmonary region segmentation on the lung registration image to obtain a pulmonary region image; Step S104: Train the preset first segmentation model based on the pulmonary vessel image and the pulmonary region image; Step S105: Perform pulmonary vessel segmentation on a second non-enhanced lung image based on the trained preset first segmentation model. This solves the problem that for some patients with contrast agent contraindications and patients without enhanced lung images, it is impossible to accurately segment pulmonary vessels from non-enhanced lung images.

[0084] Step S101: Obtain an enhanced lung image, its corresponding pulmonary vessel image, a first non-enhanced lung image, and a preset first segmentation model.

[0085] In an embodiment of the present disclosure, the obtained enhanced lung image and the first non-enhanced lung image are the first non-enhanced lung images of the same patient.

[0086] In the embodiments of the present disclosure and other possible embodiments, one or several of conventional imaging devices such as CT, PET, MR, ultrasound, and DR can be used to perform chest imaging on the patient to obtain corresponding lung images, such as a chest CT enhanced lung image and the first non-enhanced lung image of the same patient, a chest PET enhanced lung image and the first non-enhanced lung image of the same patient, a chest MR enhanced lung image and the first non-enhanced lung image of the same patient, a chest ultrasound enhanced lung image and the first non-enhanced lung image of the same patient, a chest PET-CT enhanced lung image and the first non-enhanced lung image of the same patient, and a chest DR enhanced lung image and the first non-enhanced lung image of the same patient. Among them, by injecting a contrast agent into the patient, the above-mentioned enhanced lung image can be obtained; at the same time, the image corresponding to the non-injected contrast agent is the non-enhanced lung image.

[0087] After injecting the contrast agent into the blood vessels, the intensity difference between the vascular tissue and other tissues is increased. In the enhanced lung image, the CT value of the pulmonary vessels is significantly greater than that of other surrounding tissues (such as the airway wall). Compared with the non-enhanced lung image, it is easier to visually distinguish it from the enhanced lung image. Marking the pulmonary vessels on the enhanced lung image effectively removes other non-vascular tissues and ensures the accuracy of the marking.

[0088] Non-enhanced lung images are widely used clinically and play an important role in disease screening and preliminary assessment. Usually, enhanced lung images are not obtainable or not required. For example, patients with contraindications and severe cardiopulmonary diseases cannot undergo chest enhanced lung image scans. Some patients with respiratory diseases usually do not require enhanced lung images but need to analyze pulmonary vascular changes (such as COPD patients). COPD patients exhibit pulmonary vascular pruning, and the total vascular volume less than 5 mm2 (BV5) is significantly correlated with respiratory function parameters. Finally, contrast agents in chest CT are usually not required to detect parenchymal lung lesions.

[0089] In an embodiment of the present disclosure, before obtaining the enhanced lung image and its corresponding pulmonary vascular image, vascular segmentation and lung region segmentation are respectively performed on the obtained enhanced lung image to obtain a vascular image and a lung region image; according to the vascular image and the lung region image, the pulmonary vascular image corresponding to the enhanced lung image is obtained.

[0090] In an embodiment of the present disclosure and other possible embodiments, two experienced radiologists annotated the blood vessels of 17 patients in the original enhanced lung images. Two patients were annotated in the original defined non-enhanced lung images to evaluate our CE-NC-VesselSegNet (preset first segmentation model). First, Mimics 21.0 (Materialise Corp, Belgium) was used to semi-automatically label the enhanced lung images in the sagittal plane. Secondly, repeated verification and modification were performed on the reconstructed three-dimensional pulmonary vessels. Finally, the blood vessels were labeled 1, and other tissues were labeled 0 after binarization. The first radiologist drew the annotations, and the second radiologist made minor modifications and confirmations to the results to obtain the final vascular annotation image.

[0091] In an embodiment of the present disclosure, the method for respectively performing vascular segmentation and lung region segmentation on the obtained enhanced lung image includes: obtaining a preset vascular segmentation model and a first preset lung region segmentation model; respectively performing vascular segmentation and lung region segmentation on the obtained enhanced lung image based on the preset vascular segmentation model and the first preset lung region segmentation model to obtain a vascular image and a lung region image.

[0092] In an embodiment of the present disclosure and other possible embodiments, the preset vascular segmentation model can be a preset vascular segmentation model based on a traditional algorithm or a preset vascular segmentation model based on deep learning. Obviously, those skilled in the art can use the above-mentioned vascular annotation images to train the vascular segmentation model to obtain the preset vascular segmentation model. Among them, the vascular segmentation model can be a U-Net convolutional neural network or a convolutional neural network improved therefrom.

[0093] Among them, the method for performing vascular segmentation and lung region segmentation on the obtained enhanced lung image based on the preset vascular segmentation model to obtain a vascular image includes: based on the preset vascular segmentation model, respectively obtaining N to-be-processed vascular images corresponding to the enhanced lung image under N views; fusing the N to-be-processed vascular images to obtain a vascular image. Among them, N≥2. In the embodiments of the present disclosure and other possible embodiments, the N views may be at least 2 views among the sagittal plane, coronal plane, and transverse plane views, and may also be at least 2 views among the sagittal plane, coronal plane, transverse plane views, and views at any angle. For example, the views at any angle may be 45° views at various possible orientations.

[0094] In the embodiments of the present disclosure and other possible embodiments, the method for determining the N views during the training process of the preset vascular segmentation model includes: based on the preset vascular segmentation model, respectively obtaining M to-be-processed vascular images corresponding to the enhanced lung image under M views; determining the N views according to the segmentation effects of the M to-be-processed vascular images. Similarly, the M views may be at least 2 views among the sagittal plane, coronal plane, and transverse plane views, and may also be at least 2 views among the sagittal plane, coronal plane, transverse plane views, and views at any angle. For example, the views at any angle may be 30°, 45°, 60° views at various possible orientations.

[0095] In the embodiments of the present disclosure and other possible embodiments, the method for determining the N views according to the segmentation effects of the M to-be-processed vascular images includes: respectively calculating M segmentation effects of the M to-be-processed vascular images; determining the N views from the segmentation effects of the M to-be-processed vascular images based on the M segmentation effects and a preset segmentation effect. Those skilled in the art can configure the preset segmentation effect according to actual needs. For example, the preset segmentation effect can select the Dice coefficient, and the Dice coefficient can be configured as 85%.

[0096] In the embodiments of the present disclosure and other possible embodiments, the segmentation effect includes at least one or several of the ratio of intersection over union (IoU), Dice coefficient (Dice), sensitivity, and precision, and each segmentation effect corresponds to a corresponding preset segmentation effect (preset value).

[0097] In embodiments of the present disclosure and other possible embodiments, the method for fusing the N vascular images to be processed to obtain a vascular image includes: respectively projecting the N vascular images to be processed in N view directions to obtain a plurality of projected vascular images; respectively performing mean processing on the projected vascular images and the vascular images to be processed at corresponding positions in the same view to obtain vascular images in N views. Specifically, the method for respectively performing mean processing on the projected vascular images and the vascular images to be processed at corresponding positions in the same view includes:

[0098] Obtain the corresponding positions of the projected vascular images and the vascular images to be processed in the same view; calculate the mean value of the positions. For example, the corresponding position points of the projected vascular image and the vascular image to be processed in the same view are (x1, y1, z1) and (x2, y2, z2) respectively, and the mean value of this position point is ((x1 + x2) / 2, (y1 + y2) / 2, (z1 + x2) / 2).

[0099] For example, N = 2, which are the vascular image A and the vascular image B to be processed under the sagittal plane view and the coronal plane view respectively; respectively project the 2 vascular images to be processed in the sagittal plane view and the coronal plane view directions to obtain the projected vascular image A1 under the coronal plane view and the projected vascular image B1 under the sagittal plane view; respectively perform mean processing on the vascular image A under the sagittal plane view and the projected vascular image B1 under the sagittal plane view at corresponding positions to obtain the vascular image under the sagittal plane view; at the same time, respectively perform mean processing on the vascular image B under the coronal plane view and the projected vascular image A1 under the coronal plane view at corresponding positions to obtain the vascular image under the coronal plane view.

[0100] In embodiments of the present disclosure and other possible embodiments, before respectively performing mean processing on the projected vascular images and the vascular images to be processed at corresponding positions in the same view, respectively register the projected vascular images and the vascular images to be processed in the same view to obtain the corresponding pixel points of the projected vascular images and the vascular images to be processed in the same view, and perform mean processing on the positions of the corresponding pixel points to obtain vascular images in N views.

[0101] In embodiments of the present disclosure and other possible embodiments, the operation of registering the projected vascular images and the vascular images to be processed in the same view can be completed by using the registration (Elastix) module in 3D Slicer (www.slicer.org). At the same time, those skilled in the art can also adopt other registration methods, such as SIFT registration method, 3DSIFT registration method or SURF registration method, etc.

[0102] In the embodiments of the present disclosure and other possible embodiments, the first preset lung region segmentation model may be a trained lung region (lung parenchyma) segmentation model. For example, the lung lobe segmentation method, device, and storage medium disclosed in Application No.: 202010534722.0 may be adopted to obtain the lung lobes of the left lung or the right lung; all the lung lobes of the left lung are spliced according to the left lung anatomical structure to obtain the left lung parenchyma; all the lung lobes of the right lung are spliced according to the right lung anatomical structure to obtain the right lung parenchyma. For subjects without lung lobe resection, there are 2 lung lobes in the left lung and 3 lung lobes in the right lung. The 2 lung lobes of the left lung are spliced according to the left lung anatomical structure to obtain the left lung parenchyma; the 3 lung lobes of the right lung are spliced according to the right lung anatomical structure to obtain the right lung parenchyma. Among them, in the embodiments of the present disclosure and other possible embodiments, the lung parenchyma includes peripheral airways and pulmonary blood vessels. Alternatively, the lung parenchyma (the lung parenchyma of the left lung and the right lung) image may also be directly obtained by using the lung region (lung parenchyma) segmentation model.

[0103] In the embodiments of the present disclosure and other possible embodiments, subjects after lung lobe resection are also considered, where at least one of the upper left lung lobe, lower left lung lobe, upper right lung lobe, middle right lung lobe, and lower right lung lobe of the subject after lung lobe resection is resected.

[0104] Based on the above, in the embodiments of the present disclosure and other possible embodiments, lung segmentation may include: a left lung segmentation model, a right lung segmentation model, a left lung lobe missing segmentation model, and a right lung lobe missing segmentation model; therefore, in the present disclosure, a technical solution for separately segmenting the left lung and the right lung is proposed to separately segment the left lung and the right lung. Among them, the left lung segmentation model, the right lung segmentation model, the left lung lobe missing segmentation model, and the right lung lobe missing segmentation model may be lung segmentation models based on traditional segmentation algorithms or lung segmentation models based on deep learning, such as lung segmentation models based on U-Net or U-ResNet. The training method of the model is a commonly used technical means for those skilled in the art, and the present disclosure will not elaborate on it here. However, it should be noted that the method of separately segmenting the left lung and the right lung is proposed for the case of missing lung lobes, and there is currently no method for segmenting the remaining lung after lung lobe resection. Therefore, the method of separately segmenting the left lung and the right lung is not a commonly used technical means for those skilled in the art and requires corresponding creative labor from those skilled in the art.

[0105] In an embodiment of the present disclosure, the method for obtaining the pulmonary vascular image corresponding to the enhanced lung image based on the vascular image and the lung region image includes: performing a multiplication operation on the vascular image and the lung region image to obtain the pulmonary vascular image corresponding to the enhanced lung image. Among them, the pulmonary vascular image corresponding to the enhanced lung image only includes the blood vessels within the lung region (lung parenchyma), and removes the cardiovascular vessels outside the lung region (lung parenchyma).

[0106] Step S102: Perform a registration operation on the enhanced lung image and the first non-enhanced lung image to obtain the lung registration image corresponding to the first non-enhanced lung image.

[0107] In the embodiments of the present disclosure and other possible embodiments, the registration between the enhanced lung image and the first non-enhanced lung image can be completed by using the registration (Elastix) module in 3D Slicer (www.slicer.org). At the same time, those skilled in the art can also adopt other registration methods, for example, SIFT registration method, 3DSIFT registration method or SURF registration method, etc.

[0108] In the embodiments of the present disclosure and other possible embodiments, the enhanced lung image is configured as a fixed image, and the first non-enhanced lung image is configured as a moving image. Using a registration algorithm, the lung registration image corresponding to the first non-enhanced lung image is obtained.

[0109] Step S103: Perform lung region segmentation on the lung registration image to obtain a lung region image.

[0110] In an embodiment of the present disclosure, the method for performing lung region segmentation on the lung registration image to obtain a lung region image includes: obtaining a second preset lung region segmentation model; based on the second preset lung region segmentation model, performing lung region segmentation on the lung registration image to obtain a lung region image.

[0111] In embodiments of the present disclosure and other possible embodiments, the second preset lung region segmentation model may be a trained lung region (lung parenchyma) segmentation model. For example, the lung lobe segmentation method, device, and storage medium disclosed in Application No.: 202010534722.0 may be adopted to obtain the lung lobes of the left lung or the right lung; all the lung lobes of the left lung are spliced according to the left lung anatomical structure to obtain the left lung parenchyma; all the lung lobes of the right lung are spliced according to the right lung anatomical structure to obtain the right lung parenchyma. For subjects who have not undergone lung lobe resection, there are 2 lung lobes in the left lung and 3 lung lobes in the right lung. The 2 lung lobes of the left lung are spliced according to the left lung anatomical structure to obtain the left lung parenchyma; the 3 lung lobes of the right lung are spliced according to the right lung anatomical structure to obtain the right lung parenchyma. Among them, in the embodiments of the present disclosure and other possible embodiments, the lung parenchyma includes peripheral airways and pulmonary blood vessels. Alternatively, the lung parenchyma (the lung parenchyma of the left lung and the right lung) image may also be directly obtained by using the lung region (lung parenchyma) segmentation model.

[0112] In embodiments of the present disclosure and other possible embodiments, subjects after lung lobe resection are also considered. Among them, at least one lung lobe of the subject after lung lobe resection is resected, including at least the left upper lobe, left lower lobe, right upper lobe, right middle lobe, and right lower lobe.

[0113] Based on the above, in embodiments of the present disclosure and other possible embodiments, lung segmentation may include: a left lung segmentation model, a right lung segmentation model, a left lung lobe absence segmentation model, and a right lung lobe absence segmentation model. Therefore, in the present disclosure, a technical solution for separately segmenting the left lung and the right lung is proposed, and the left lung and the right lung are segmented separately. Among them, the left lung segmentation model, the right lung segmentation model, the left lung lobe absence segmentation model, and the right lung lobe absence segmentation model may be lung segmentation models based on traditional segmentation algorithms or lung segmentation models based on deep learning, such as lung segmentation models based on U-Net or U-ResNet. The training method of the model is a commonly used technical means for those skilled in the art, and will not be described in detail herein. However, it should be noted that the method of separately segmenting the left lung and the right lung is proposed for the case of lung lobe absence, and there is currently no method for segmenting the remaining lung after lung lobe resection. Therefore, the method of separately segmenting the left lung and the right lung is not a commonly used technical means for those skilled in the art and requires corresponding creative labor from those skilled in the art.

[0114] In the embodiments of the present disclosure and other possible embodiments, a method for separately segmenting the left lung and the right lung includes: obtaining the lung registration image, determining the position of the main bronchus (primary bronchus) in the lung registration image, and dividing the lung registration image into a left lung image and a right lung image according to the position of the main bronchus; separately determining whether there is a missing lung lobe in the left lung image and the right lung image; if there is a missing lung lobe, determining whether the missing lung lobe is in the left lung or the right lung; if the missing lung lobe is in the left lung, obtaining a left lung lobe missing segmentation model and a right lung segmentation model, and separately using the left lung lobe missing segmentation model and the right lung segmentation model to perform lung parenchyma segmentation on the left lung with the missing lobe and the right lung without the missing lobe; if the missing lung lobe is in the right lung, obtaining a right lung lobe missing segmentation model and a left lung segmentation model, and separately using the right lung lobe missing segmentation model and the left lung segmentation model to perform lung parenchyma segmentation on the right lung with the missing lobe and the left lung without the missing lobe; splicing the segmented left lung parenchyma and right lung parenchyma according to the anatomical structure to obtain the above-mentioned lung parenchyma. Wherein, the main bronchus is the trachea from the larynx to the hilum of the lung.

[0115] For example, if only the upper left lung lobe or the lower left lung lobe exists in the lung registration image, obtain a left lung lobe missing segmentation model and a right lung segmentation model, and separately use the left lung lobe missing segmentation model and the right lung segmentation model to perform lung parenchyma segmentation on the left lung with the missing lobe and the right lung without the missing lobe; finally, splice the segmented left lung parenchyma and right lung parenchyma according to the anatomical structure to obtain the above-mentioned lung parenchyma.

[0116] In the embodiments of the present disclosure and other possible embodiments, the method for determining the position of the main bronchus (primary bronchus) in the lung registration image and dividing the lung registration image into a left lung image and a right lung image according to the position of the main bronchus includes: obtaining an airway segmentation model, performing airway segmentation on the lung registration image to obtain an airway tree; determining the main bronchus in the airway tree and calculating the centerline of the main bronchus; dividing the lung registration image into a left lung image and a right lung image according to the centerline. At the same time, the airway segmentation model can select an existing airway segmentation model, and the airway segmentation only needs to be able to segment the main bronchus, and it is not necessary to perform fine segmentation on the airway. For example, the airway segmentation model used in the registration method and device, electronic device and storage medium based on lung lobes and tracheal trees disclosed in Application No. 202010540322.0.

[0117] Step S104: Train the preset first segmentation model based on the lung vascular image and the lung region image.

[0118] In the embodiments of the present disclosure, a first segmentation model CE-NC-VesselSegNet is proposed. Figure 4A network diagram of the first segmentation model according to an embodiment of the present disclosure is shown. As Figure 4 shown, the first segmentation model includes: a U-Net backbone network, which obtains a plurality of feature maps after each convolution operation; performs feature map normalization on the plurality of feature maps respectively; and activates the normalized plurality of feature maps using an activation function. In an embodiment of the present disclosure, feature map normalization is performed on the plurality of feature maps respectively to accelerate the convergence of the first segmentation model and maintain the independence between each image instance (feature map). Meanwhile, the Leaky ReLU (lReLU) activation function can be used to activate the normalized plurality of feature maps.

[0119] In an embodiment of the present disclosure, the method for training the preset first segmentation model based on the pulmonary vascular image and the pulmonary region image includes: calculating the loss of the corresponding feature map after each decoding during the decoding process of the first segmentation model; and obtaining the total loss during the training process according to the loss of the corresponding feature map after each decoding.

[0120] To implement this first segmentation model, a training-validation-testing strategy is adopted. In dataset D1, for CE-NC-VesselSegNet and CE-NC-VesselSegNet with fine-tuning, 12 pieces of data with two epochs are used as the training and validation datasets, and two non-enhanced lung images are used as the test set. For CE-VesselSegNet, 12 cases of data divided into two epochs are used for training and validation, and 5 enhanced lung images are used for testing.

[0121] Due to limited GPU memory, it is impractical to input the entire lung image into the first segmentation model for training. Therefore, the image input into the first segmentation model is cropped into a cuboid with a size of 270×186×210 and used to train the network. The number of training epochs is set to 1000. The network is optimized using the stochastic gradient descent method with a Nesterov momentum of 0.99. The initial learning rate is 0.01, and the learning rate decays according to poly-LR (polynomial learning rate).

[0122]

[0123] For the stability of the first segmentation model training and the accuracy of segmentation, nnU-Net uses the sum of the dice and cross-entropy loss functions as the loss function.

[0124] Loss=Loss dice+Loss cross-entropy 。

[0125] Meanwhile, the first segmentation type adopts a deep supervision strategy to better train the network and effectively utilize the hidden layer information. The sliding window method is used to predict the data. The sliding window size is the size of the trained patch, and the stride is half of the patch size. In addition, Gaussian weighting is used to suppress edge errors. The training of the first segmentation model is carried out using Python 3.9 in the CentOS operating system, implemented using PyTorch 1.9, and run using an Nvidia GeForce RTX 2080Ti GPU with 11GB of memory.

[0126] In the embodiments of the present disclosure and other possible embodiments, the method for training the preset first segmentation model based on the pulmonary vascular image and the pulmonary region image further includes: respectively training the preset first segmentation model based on the enhanced pulmonary images of the pulmonary region images and their corresponding pulmonary vascular images under multiple views to obtain first segmentation models under multiple views; based on the performance of the first segmentation models under the multiple views, selecting at least two first segmentation models for pulmonary vascular segmentation of non-enhanced pulmonary images from the first segmentation models under the multiple views.

[0127] For example, respectively training the preset first segmentation model based on the enhanced pulmonary images of the pulmonary region images and their corresponding pulmonary vascular images under a views to obtain first segmentation models under a views. Among them, the a views can be at least two views among the sagittal plane, coronal plane, and transverse plane views, and can also be at least two views among the sagittal plane, coronal plane, transverse plane views, and views at any angle. For example, the views at any angle can be 30°, 45°, 60° views at various possible orientations.

[0128] In the embodiments of the present disclosure and other possible embodiments, the performance of the first segmentation models under the multiple views at least includes one or more of the ratio of intersection over union (IoU), dice coefficient (Dice), sensitivity, and precision.

[0129] In embodiments of the present disclosure and other possible embodiments, the method of selecting first segmentation models for pulmonary vascular segmentation of non-enhanced lung images from the first segmentation models under at least two views based on the performance of the first segmentation models under the multiple views includes: calculating the corresponding performances of the first segmentation models under the multiple views respectively; sorting the corresponding performances of the first segmentation models under the multiple views from large to small, and selecting first segmentation models for pulmonary vascular segmentation of non-enhanced lung images from the sorted first segmentation models under the multiple views according to the obtained number; where the number is at least 2.

[0130] In embodiments of the present disclosure and other possible embodiments, the method of sorting the corresponding performances of the first segmentation models under the multiple views from large to small includes: obtaining the set performances corresponding to the first segmentation models under the multiple views; normalizing / standardizing the set performances corresponding to the first segmentation models under the multiple views; summing the normalized / standardized set performances to obtain an evaluation performance; and sorting the corresponding performances of the first segmentation models under the multiple views from large to small based on the evaluation performance. Wherein, the set performances at least include one or more of the ratio of intersection over union (IoU), dice coefficient (Dice), sensitivity, and precision.

[0131] Step S105: Perform pulmonary vascular segmentation on the second non-enhanced lung image based on the trained preset first segmentation model.

[0132] In embodiments of the present disclosure and other possible embodiments, the second non-enhanced lung image is the non-enhanced lung image corresponding to the patient to be subjected to pulmonary vascular segmentation.

[0133] In embodiments of the present disclosure and other possible embodiments, the method of performing pulmonary vascular segmentation on the second non-enhanced lung image based on the trained preset first segmentation model includes: obtaining the first segmentation models for pulmonary vascular segmentation of non-enhanced lung images under the at least two views; projecting the second non-enhanced lung image according to the at least two views of the first segmentation models for pulmonary vascular segmentation of non-enhanced lung images to obtain corresponding second non-enhanced lung projection images; respectively using the first segmentation models for pulmonary vascular segmentation of non-enhanced lung images under the at least two views to segment the second non-enhanced lung projection images to obtain corresponding lung vascular segmentation images to be fused; and fusing the lung vascular segmentation images to be fused to obtain the lung vascular image corresponding to the second non-enhanced lung image.

[0134] For example, at least two views of the first segmentation model for pulmonary vascular segmentation of non-enhanced lung images are respectively the coronal view and the transverse view. The second non-enhanced lung image is projected according to the coronal view and the transverse view to obtain corresponding second non-enhanced lung projection images (second non-enhanced lung coronal view image and second non-enhanced lung transverse view image); the first segmentation model for pulmonary vascular segmentation of non-enhanced lung images under the coronal view is used to segment the second non-enhanced lung coronal view image to obtain a corresponding pulmonary vascular segmentation coronal view image to be fused; at the same time, the first segmentation model for pulmonary vascular segmentation of non-enhanced lung images under the transverse view is used to segment the second non-enhanced lung transverse view image to obtain a corresponding pulmonary vascular segmentation transverse view image to be fused; the pulmonary vascular segmentation coronal view image to be fused and the pulmonary vascular segmentation transverse view image to be fused are fused to obtain a pulmonary vascular image corresponding to the second non-enhanced lung image.

[0135] In the embodiments of the present disclosure and other possible embodiments, the method for fusing the pulmonary vascular segmentation images to be fused to obtain a pulmonary vascular image corresponding to the second non-enhanced lung image includes: determining a projection direction; based on the projection direction, respectively projecting the pulmonary vascular segmentation images to be fused to obtain pulmonary vascular segmentation projection images; registering the pulmonary vascular segmentation images to be fused to obtain corresponding registration point pairs; performing mean processing on the positions corresponding to the registration points to obtain a pulmonary vascular image corresponding to the second non-enhanced lung image. Specifically, the method for performing mean processing on the positions corresponding to the registration points to obtain a pulmonary vascular image corresponding to the second non-enhanced lung image includes: calculating the mean of the positions corresponding to the registration points to obtain a pulmonary vascular image corresponding to the second non-enhanced lung image. The position points are respectively (x1, y1, z1) and (x2, y2, z2), and the mean of the position points is ((x1 + x2) / 2, (y1 + y2) / 2, (z1 + x2) / 2).

[0136] In the embodiments of the present disclosure and other possible embodiments, the projection direction can be configured as the view direction corresponding to the pulmonary vascular segmentation images to be fused. For example, for the pulmonary vascular segmentation coronal view image to be fused and the pulmonary vascular segmentation transverse view image to be fused, the projection direction can be configured as the direction of the coronal view or the direction of the pulmonary vascular segmentation transverse view.

[0137] In the embodiments of the present disclosure and other possible embodiments, the operation of registering the lung vessel segmentation images to be fused can be completed using the registration (Elastix) module in 3D Slicer (www.slicer.org). Meanwhile, those skilled in the art can also adopt other registration methods, such as SIFT registration method, 3DSIFT registration method, or SURF registration method, etc.

[0138] Figure 2 The flowchart showing the segmentation method of lung vessels according to another embodiment of the present disclosure is as follows. As Figure 2 shown, step S201: Obtain an enhanced lung image and its corresponding lung vessel image, a first non-enhanced lung image, and a preset first segmentation model; step S202: Respectively segment the enhanced lung image and the first non-enhanced lung image to obtain a first lung region image and a second lung region image; step S203: Perform a registration operation on the first lung region image and the second lung region image to obtain a lung region registration image corresponding to the second lung region image; step S204: Based on the lung region registration image and the lung vessel image, train the preset first segmentation model; step S205: Based on the trained preset first segmentation model, perform lung vessel segmentation on a second non-enhanced lung image. To solve the problem that for some patients with contrast agent contraindications and patients without enhanced lung images, the lung vessels cannot be accurately segmented from the non-enhanced lung images.

[0139] Step S201: Obtain an enhanced lung image and its corresponding lung vessel image, a first non-enhanced lung image, and a preset first segmentation model.

[0140] In the embodiments of the present disclosure, before obtaining the enhanced lung image and its corresponding lung vessel image, perform vessel segmentation and lung region segmentation on the obtained enhanced lung image respectively to obtain a vessel image and a lung region image; according to the vessel image and the lung region image, obtain the lung vessel image corresponding to the enhanced lung image. The specific method can be seen in the detailed description of step S101: Obtain an enhanced lung image and its corresponding lung vessel image, a first non-enhanced lung image, and a preset first segmentation model.

[0141] In the embodiments of the present disclosure, the method for respectively performing vessel segmentation and lung region segmentation on the obtained enhanced lung image includes: obtaining a vessel segmentation model and a first preset lung region segmentation model; respectively based on the vessel segmentation model and the first preset lung region segmentation model, perform vessel segmentation and lung region segmentation on the obtained enhanced lung image to obtain a vessel image and a lung region image. The specific method can be seen in the detailed description of step S101: Obtain an enhanced lung image and its corresponding lung vessel image, a first non-enhanced lung image, and a preset first segmentation model.

[0142] In an embodiment of the present disclosure, the method for obtaining the pulmonary vascular image corresponding to the enhanced lung image according to the vascular image and the lung region image includes: performing a multiplication operation on the vascular image and the lung region image to obtain the pulmonary vascular image corresponding to the enhanced lung image. For the specific method, refer to step S101 for details: obtaining the enhanced lung image and its corresponding pulmonary vascular image, the first non-enhanced lung image, and a detailed description of the preset first segmentation model.

[0143] Step S202: Segment the enhanced lung image and the first non-enhanced lung image respectively to obtain a first lung region image and a second lung region image.

[0144] In an embodiment of the present disclosure, the method for segmenting the enhanced lung image and the first non-enhanced lung image respectively to obtain a first lung region image and a second lung region image includes: obtaining a third preset lung region segmentation model and a fourth preset lung region segmentation model; segmenting the enhanced lung image and the first non-enhanced lung image respectively based on the third preset lung region segmentation model and the fourth preset lung region segmentation model to obtain a first lung region image and a second lung region image. That is, segmenting the enhanced lung image based on the third preset lung region segmentation model to obtain a first lung region image; segmenting the first non-enhanced lung image based on the fourth preset lung region segmentation model to obtain a second lung region image. Among them, the third preset lung region segmentation model and the fourth preset lung region segmentation model can be segmentation networks based on deep learning, such as a U-Net convolutional neural network or its improved convolutional neural network. The third preset lung region segmentation model and the fourth preset lung region segmentation model are trained using the enhanced lung image and the non-enhanced lung image respectively. The training process is a conventional technical means for those skilled in the art and will not be described in detail here.

[0145] In the embodiments of the present disclosure and other possible embodiments, the third preset lung region segmentation model and the fourth preset lung region segmentation model may also be pre-trained lung region (lung parenchyma) segmentation models. For example, the lung lobe segmentation method, device, and storage medium disclosed in Application No. 202010534722.0 may be adopted to obtain the lung lobes of the left lung or the right lung. All the lung lobes of the left lung are spliced according to the anatomical structure of the left lung to obtain the left lung parenchyma. All the lung lobes of the right lung are spliced according to the anatomical structure of the right lung to obtain the right lung parenchyma. For subjects who have not undergone lung lobe resection, there are 2 lung lobes in the left lung and 3 lung lobes in the right lung. The 2 lung lobes of the left lung are spliced according to the anatomical structure of the left lung to obtain the left lung parenchyma. The 3 lung lobes of the right lung are spliced according to the anatomical structure of the right lung to obtain the right lung parenchyma. Among them, in the embodiments of the present disclosure and other possible embodiments, the lung parenchyma includes peripheral airways and pulmonary blood vessels. Alternatively, the lung parenchyma (the lung parenchyma of the left lung and the right lung) image may be directly obtained by using the lung region (lung parenchyma) segmentation model.

[0146] In the embodiments of the present disclosure and other possible embodiments, subjects after lung lobe resection are also considered. Among them, at least one lung lobe of the subject after lung lobe resection is resected, including at least the left upper lobe, the left lower lobe, the right upper lobe, the right middle lobe, and the right lower lobe.

[0147] Based on the above, in the embodiments of the present disclosure and other possible embodiments, lung segmentation may include a left lung segmentation model, a right lung segmentation model, a left lung lobe absence segmentation model, and a right lung lobe absence segmentation model. Therefore, in the present disclosure, a technical solution for separately segmenting the left lung and the right lung is proposed to separately segment the left lung and the right lung. Among them, the left lung segmentation model, the right lung segmentation model, the left lung lobe absence segmentation model, and the right lung lobe absence segmentation model may be lung segmentation models based on traditional segmentation algorithms or lung segmentation models based on deep learning, such as lung segmentation models based on U-Net or U-ResNet. The training method of the model is a technical means commonly used by those skilled in the art, and will not be described in detail herein. However, it should be noted that the method of separately segmenting the left lung and the right lung is proposed for the case of lung lobe absence, and there is currently no method for segmenting the remaining lung after lung lobe resection. Therefore, the method of separately segmenting the left lung and the right lung is not a technical means commonly used by those skilled in the art and requires corresponding creative labor from those skilled in the art.

[0148] In embodiments of the present disclosure and other possible embodiments, a method for separately segmenting the left lung and the right lung includes: obtaining the enhanced lung image or the first non-enhanced lung image, determining the position of the main bronchus (level 1 trachea) in the enhanced lung image or the first non-enhanced lung image, and dividing the enhanced lung image or the first non-enhanced lung image into a left lung image and a right lung image according to the position of the main bronchus; separately determining whether there is a missing lung lobe in the left lung image and the right lung image; if there is a missing lung lobe, determining whether the missing lung lobe is in the left lung or the right lung; if the missing lung lobe is in the left lung, obtaining a left lung lobe missing segmentation model and a right lung segmentation model, and separately using the left lung lobe missing segmentation model and the right lung segmentation model to perform lung parenchyma segmentation on the left lung with the missing lobe and the right lung without the missing lobe; if the missing lung lobe is in the right lung, obtaining a right lung lobe missing segmentation model and a left lung segmentation model, and separately using the right lung lobe missing segmentation model and the left lung segmentation model to perform lung parenchyma segmentation on the right lung with the missing lobe and the left lung without the missing lobe; splicing the segmented left lung parenchyma and right lung parenchyma according to the anatomical structure to obtain the above-mentioned lung parenchyma. Wherein, the main bronchus is the trachea from the larynx to the hilum of the lung.

[0149] For example, if only the upper left lung lobe or the lower left lung lobe exists in the enhanced lung image or the first non-enhanced lung image, obtain a left lung lobe missing segmentation model and a right lung segmentation model, and separately use the left lung lobe missing segmentation model and the right lung segmentation model to perform lung parenchyma segmentation on the left lung with the missing lobe and the right lung without the missing lobe; finally, splice the segmented left lung parenchyma and right lung parenchyma according to the anatomical structure to obtain the above-mentioned lung parenchyma.

[0150] In embodiments of the present disclosure and other possible embodiments, the method for determining the position of the main bronchus (level 1 trachea) in the enhanced lung image or the first non-enhanced lung image and dividing the enhanced lung image or the first non-enhanced lung image into a left lung image and a right lung image according to the position of the main bronchus includes: obtaining an airway segmentation model, performing airway segmentation on the enhanced lung image or the first non-enhanced lung image to obtain an airway tree; determining the main bronchus in the airway tree and calculating the center line of the main bronchus; dividing the enhanced lung image or the first non-enhanced lung image into a left lung image and a right lung image according to the center line. At the same time, the airway segmentation model can select an existing airway segmentation model, and the airway segmentation only needs to segment out the main bronchus, and does not require fine segmentation of the airway. For example, the airway segmentation model used in the registration method and device, electronic device and storage medium based on lung lobes and tracheal trees disclosed in Application No. 202010540322.0.

[0151] Step S203: Perform a registration operation on the first lung region image and the second lung region image to obtain a lung region registration image corresponding to the second lung region image.

[0152] In the embodiments of the present disclosure and other possible embodiments, the registration operation on the first lung region image and the second lung region image can be completed by using the registration (Elastix) module in 3D Slicer (www.slicer.org). At the same time, those skilled in the art can also adopt other registration methods, for example, SIFT registration method, 3DSIFT registration method or SURF registration method, etc.

[0153] The non-enhanced lung images before and after registration are covered by the pulmonary vessels extracted from the enhanced lung images. Before registration, there is an obvious deviation between the pulmonary vessels drawn in the enhanced lung image and the pulmonary vessels on the non-enhanced lung image, but they can be accurately covered after registration. After registration, the MAE between the enhanced lung image and the non-enhanced lung image decreased significantly from 13.561 to 5.196 (p < 0.001), the RMSE decreased significantly from 75.995 to 26.986 (p < 0.001), the SSIM increased significantly from 0.853 to 0.879 (p = 0.001), and the PSNR increased significantly from 31.831 to 37.505 (p = 0.059).

[0154] Step S204: Train the preset first segmentation model based on the lung region registration image and the pulmonary vessel image.

[0155] In the embodiments of the present disclosure, a first segmentation model CE-NC-VesselSegNet is proposed. Figure 4 The network schematic diagram of the first segmentation model according to the embodiments of the present disclosure is shown.

[0156] In the embodiments of the present disclosure and other possible embodiments, the losses and dice during the training of CE-NC-VesselSegNet are given. The dice value during the training process increases continuously and reaches about 0.87 after 1000 rounds. At the same time, as the number of rounds increases, the losses of training and validation steadily decrease to about -0.88. There is no obvious difference in the losses between the training and validation processes, indicating that there is no serious overfitting.

[0157] In an embodiment of the present disclosure, the first segmentation model includes: a U-Net backbone network. After each convolution operation of the U-Net backbone network, a plurality of feature maps are obtained; the plurality of feature maps are respectively subjected to feature mapping normalization; and the normalized plurality of feature maps are activated by using an activation function. For the specific method, refer to step S104: a detailed description of training the preset first segmentation model based on the pulmonary vascular image and the lung region image (lung region registration image).

[0158] In an embodiment of the present disclosure, the method for training the preset first segmentation model based on the lung region registration image and the lung region image includes: during the decoding process of the first segmentation model, calculating the loss of the feature map corresponding to each decoding; and obtaining the total loss during the training process according to the loss of the feature map corresponding to each decoding. For the specific method, refer to step S104: a detailed description of training the preset first segmentation model based on the pulmonary vascular image and the lung region image (lung region registration image).

[0159] In an embodiment of the present disclosure and other possible embodiments, the method for training the preset first segmentation model based on the pulmonary vascular image and the lung region registration image further includes: respectively training the preset first segmentation model based on the enhanced lung image in the lung region registration images under a plurality of views and their corresponding plurality of pulmonary vascular images to obtain first segmentation models under a plurality of views; and selecting, based on the performance of the first segmentation models under the plurality of views, at least two first segmentation models for pulmonary vascular segmentation of non-enhanced lung images from the first segmentation models under the plurality of views.

[0160] For example, the preset first segmentation model is respectively trained based on the enhanced lung image in the lung region registration images under a views and their corresponding a pulmonary vascular images to obtain first segmentation models under a views. Among them, the a views can be at least two views among the sagittal plane, the coronal plane, and the transverse plane views, and can also be at least two views among the sagittal plane, the coronal plane, the transverse plane views, and views at any angle. For example, the views at any angle can be 30°, 45°, 60° views at various possible orientations.

[0161] In an embodiment of the present disclosure and other possible embodiments, the performance of the first segmentation models under the plurality of views at least includes one or more of the ratio of intersection over union (IoU), the dice coefficient (Dice), sensitivity, and precision.

[0162] In embodiments of the present disclosure and other possible embodiments, the method of selecting first segmentation models for pulmonary vascular segmentation of non-enhanced lung images from at least two views among the first segmentation models under the multiple views based on the performance of the first segmentation models under the multiple views includes: calculating the performance corresponding to the first segmentation models under the multiple views respectively; sorting the performance corresponding to the first segmentation models under the multiple views from large to small, and selecting at least two first segmentation models for pulmonary vascular segmentation of non-enhanced lung images from the sorted first segmentation models under the multiple views according to the obtained number; wherein the number is at least 2.

[0163] In embodiments of the present disclosure and other possible embodiments, the method of sorting the performance corresponding to the first segmentation models under the multiple views from large to small includes: obtaining the set performance corresponding to the first segmentation models under the multiple views; normalizing / standardizing the set performance corresponding to the first segmentation models under the multiple views; summing the normalized / standardized set performance to obtain an evaluation performance; sorting the performance corresponding to the first segmentation models under the multiple views from large to small based on the evaluation performance. Wherein, the set performance at least includes one or more of the ratio of intersection over union (IoU), dice coefficient (Dice), sensitivity, and precision.

[0164] Step S205: Perform pulmonary vascular segmentation on a second non-enhanced lung image based on the trained preset first segmentation model.

[0165] In embodiments of the present disclosure and other possible embodiments, the second non-enhanced lung image is the non-enhanced lung image corresponding to the patient to be subjected to pulmonary vascular segmentation.

[0166] In the embodiments of the present disclosure and other possible embodiments, the method for segmenting pulmonary vessels from a second non-enhanced lung image based on the preset first segmentation model obtained by training includes: obtaining the first segmentation model for segmenting pulmonary vessels from non-enhanced lung images under at least two views; projecting the second non-enhanced lung image according to at least two views of the first segmentation model for segmenting pulmonary vessels from non-enhanced lung images to obtain corresponding second non-enhanced lung projection images; respectively using the first segmentation model for segmenting pulmonary vessels from non-enhanced lung images under at least two views to segment the second non-enhanced lung projection images to obtain corresponding pulmonary vessel segmentation images to be fused; and fusing the pulmonary vessel segmentation images to be fused to obtain a pulmonary vessel image corresponding to the second non-enhanced lung image.

[0167] For example, at least two views of the first segmentation model for segmenting pulmonary vessels from non-enhanced lung images are respectively a coronal view and a transverse view. The second non-enhanced lung image is projected according to the coronal view and the transverse view to obtain corresponding second non-enhanced lung projection images (a second non-enhanced lung coronal view image and a second non-enhanced lung transverse view image); the first segmentation model for segmenting pulmonary vessels from non-enhanced lung images under the coronal view is used to segment the second non-enhanced lung coronal view image to obtain a corresponding pulmonary vessel segmentation coronal view image to be fused; at the same time, the first segmentation model for segmenting pulmonary vessels from non-enhanced lung images under the transverse view is used to segment the second non-enhanced lung transverse view image to obtain a corresponding pulmonary vessel segmentation transverse view image to be fused; and the pulmonary vessel segmentation coronal view image to be fused and the pulmonary vessel segmentation transverse view image to be fused are fused to obtain a pulmonary vessel image corresponding to the second non-enhanced lung image.

[0168] In the embodiments of the present disclosure and other possible embodiments, the method for fusing the pulmonary vessel segmentation images to be fused to obtain a pulmonary vessel image corresponding to the second non-enhanced lung image includes: determining a projection direction; based on the projection direction, respectively projecting the pulmonary vessel segmentation images to be fused to obtain pulmonary vessel segmentation projection images; registering the pulmonary vessel segmentation images to be fused to obtain corresponding registration point pairs; and performing a mean value process on the positions corresponding to the registration points to obtain a pulmonary vessel image corresponding to the second non-enhanced lung image. Specifically, the method for performing a mean value process on the positions corresponding to the registration points to obtain a pulmonary vessel image corresponding to the second non-enhanced lung image includes: calculating the mean value of the positions corresponding to the registration points to obtain a pulmonary vessel image corresponding to the second non-enhanced lung image. The position points are respectively (x1, y1, z1) and (x2, y2, z2), and the mean value of the position points is ((x1 + x2) / 2, (y1 + y2) / 2, (z1 + z2) / 2).

[0169] In embodiments of the present disclosure and other possible embodiments, the projection direction may be configured as the view direction corresponding to the lung vascular segmentation image to be fused. For example, for the lung vascular segmentation coronal view image to be fused and the lung vascular segmentation cross-sectional view image to be fused, the projection direction may be configured as the direction of the coronal view or the direction of the lung vascular segmentation cross-sectional view.

[0170] In embodiments of the present disclosure and other possible embodiments, the operation of registering the lung vascular segmentation image to be fused can be completed using the registration (Elastix) module in 3D Slicer (www.slicer.org). Meanwhile, those skilled in the art can also adopt other registration methods, such as the SIFT registration method, the 3DSIFT registration method, or the SURF registration method, etc.

[0171] In embodiments of the present disclosure and other possible embodiments, dice coefficient (Dice), the ratio of intersection over union (IoU), sensitivity, and precision are used to evaluate the performance of the preset first segmentation model corresponding to different segmentation methods. Dice and IoU are used to measure the similarity between the network segmentation result and the gold standard. The value range is 0–1, where 1 represents the best performance. Sensitivity represents the number of actual positive samples predicted as positive. Precision is the correct proportion of positive samples predicted by the model. In this study, the positive sample region is the labeled lung vessels, and the negative sample region represents other tissues. Using the prediction result and the gold standard, TP (true positive), TN (true negative), FP (false positive), and FN (false negative) are calculated. We can calculate the four metrics through these four regions using the following formulas.

[0172]

[0173]

[0174]

[0175]

[0176] In the embodiments of the present disclosure and other possible embodiments, Dataset D1: This dataset comes from the General Hospital of the Northern Theater Command and consists of 19 subjects, including 12 cases of three-phase CT images: non-enhanced phase, arterial enhancement phase, and venous enhancement phase, 5 cases of only arterial enhancement phase CT images, and 2 cases of only non-enhanced phase CT images. In this study, 12 groups of non-enhanced and arterial enhancement CT scans were used for image registration respectively. This study was approved by the General Hospital of the Northern Theater Command. All subjects signed informed consent forms in accordance with the Declaration of Helsinki (2000). The thickness of all CT scans was 1.0 mm, the number of slices was between 591 and 741, and the slice size was 512×512. Dataset D2: An open dataset from the ISICDM 2021 challenge. This dataset (http: / / www.imagecomputing.org / 2021 / cn / challenges / ) was collected and annotated by Professor Qin's team. The dataset includes 12 cases of non-enhanced lung images and 12 cases of enhanced lung images (the cases are not paired). The thickness of the enhanced lung images is between 0.625 and 1.25 mm, and the number of slices is between 285 and 533. The thickness of the non-enhanced lung images is between 0.75 and 1.5 mm, and the number of slices is between 204 and 577. The matrix size of all CT images is 512×512. All images and labels are in PNG format.

[0177] In the embodiments of the present disclosure and other possible embodiments, vascular labels are drawn in the original enhanced lung images. The lung regions of all CT images are automatically segmented, and the annotations of the blood vessels within the lung regions are obtained. Secondly, by registering the original non-enhanced lung images to the original enhanced lung images, the registered non-enhanced lung images are generated. After the lung region is segmented, the NC images within the lung region are obtained. Finally, the results of the above steps are used as inputs to train CE-NC-VesselSegNet to automatically segment the lung blood vessels from the non-enhanced lung images.

[0178] In the embodiments of the present disclosure and other possible embodiments, in Dataset D1, the lung regions are automatically segmented from each CT image case through our nnU-Net-based model. To solve the problem of class imbalance, the data within the smallest bounding box of the lung region is retained.

[0179] Since the voxel spacing and CT intensity of each case are different, spatial resampling and normalization are performed. Specifically, we resample the data to the median voxel spacing of all cases (0.50×0.74×0.74 mm 3 ). Cubic spline interpolation is used for image data, and nearest neighbor interpolation is used for labels. After statistical analysis of the intensities of all voxels in the CT images of the training dataset, the values in the first 0.5% and the last 0.5% of the CT images are removed, and their mean and variance are normalized by z-score.

[0180]

[0181] Among them, z is the normalized data, x is the original data, μ is the average value of all data, and σ is the standard deviation.

[0182] The registration of enhanced lung images and non-enhanced lung images was completed using the registration (Elastix) module in 3D Slicer (www.slicer.org). This module was initially developed by Andras Lasso (Queen’s University, PerkLab) as a front-end for the Elastix medical image registration toolbox. Specifically, the enhanced lung image was used as the fixed image, and the non-enhanced lung image was used as the moving image.

[0183] Before and after image registration, the maximum error (MAE), root mean square error (RMSE), structural similarity (SSIM)

[27] , and peak signal-to-noise ratio (PSNR) were calculated and compared.

[0184]

[0185]

[0186]

[0187] Among them, c i is the enhanced lung image, and n i is the non-enhanced lung image or the registered non-enhanced lung image. N represents the number of images used, and max(·) returns the maximum value of the set. MAX is the maximum value of the image.

[0188] The CE-NC-VesselSegNet model is based on nnU-Net

[28] and follows a 3D U-Net, which includes an encoder and a decoder. Figure 2 The encoder using ordinary convolutional layers, five-step convolution in the downsampling operation, and the decoder implementing transposed convolution for upsampling are described. Skip connections are performed to interconnect the encoder and the decoder. After each convolution operation, LeakyReLU (lReLU) is used to perform feature map normalization on the multiple feature maps respectively. In addition, other decoders adopt 1×1×1 convolution and softmax as the deep supervision strategy.

[0189] On dataset D1, we trained two comparative models: CE-VesselSegNet (a pulmonary vessel segmentation model for enhanced lung images) and CE-NC-VesselSegNet with fine-tuning. The preprocessing and network architectures of these two models are similar to CE-NC-VesselSegNet. The difference is that CE-VesselSegNet uses enhanced lung images within the lung region and vascular annotations within the lung region as inputs. CE-NC-VesselSegNet with fine-tuning uses the pre-trained CE-NC-VesselSegNet as initialization, and uses registered non-enhanced lung images within the lung region and vascular annotations within the lung region as inputs to fine-tune the model.

[0190] In dataset D2 provided by the ISICDM 2021 challenge, two models, NC-VesselSegNet (a pulmonary vessel segmentation model for non-enhanced lung images) and CE-VesselSegNet (a pulmonary vessel segmentation model for enhanced lung images), were trained for comparison. NC-VesselSegNet trained by D2 uses the original non-enhanced lung images and vascular annotations as inputs. CE-VesselSegNet uses the original enhanced lung images and their vascular annotations as inputs. Specifically, for the two comparative models, eight cases of data were used as the training set, two cases of data were used as the validation set, and two cases of data were used as the test set.

[0191] Table 1 shows the segmentation performance of different models in the internal dataset D1. Among the two models for segmenting pulmonary vessels from non-enhanced lung images, CE-NC-VesselSegNet shows similar performance to CE-NC-VesselSegNet with fine-tuning, with a dice coefficient of 0.854, IoU of 0.745, sensitivity of 0.784, and precision of 0.939. This indicates that fine-tuning does not improve the segmentation effect. For the pulmonary vessel segmentation model (CE-VesselSegNet) of enhanced (CE) lung images, the dice coefficient is 0.930, IoU is 0.869, sensitivity is 0.903, and precision is 0.958.

[0192] Table 1 Segmentation performance of different models in the internal dataset D1

[0193]

[0194] Table 2 lists the segmentation performance of different models in dataset D2. For CE-NC-VesselSegNet and NC-VesselSegNet trained by D2 (the non-enhanced lung image pulmonary vascular segmentation model NC-VesselSegNet trained by dataset D2), the test set is two non-enhanced (NC) lung images in D2. CE-NC-VesselSegNet is superior to NC-VesselSegNet trained by D2 in terms of Dice, IoU, and Precision (0.738 and 0.700, 0.605 and 0.540, 0.902 and 0.784). However, the sensitivity values of NC-VesselSegNet trained by D2 are higher (0.656 and 0.640 respectively). Through visual observation, we can find that the labels in D2 have more pulmonary vessels and finer morphology than those in D1, which may explain why the sensitivity of our model is slightly lower. In addition, our model has higher precision, indicating that more pulmonary vessels predicted by our model are correct. For CE-VesselSegNet trained by D2 (the enhanced lung image pulmonary vascular segmentation model CE-VesselSegNet trained by dataset D2) and CE-VesselSegNet, the test set is two enhanced (CE) lung images in D2. Each evaluation parameter of the model trained by our data is higher than that trained by D2 (Dice, 0.837 and 0.769; IoU, 0.727 and 0.625; Sensitivity, 0.800 and 0.718; Precision, 0.902 and 0.830).

[0195] Table 2 Segmentation performance of different models in dataset D2

[0196]

[0197] The annotation criteria for different studies are generally not the same. By comparison, it can be found that more fine terminal vessels are annotated in the ISICDM 2021 challenge than in our dataset D1. This difference in annotation criteria partly explains the decrease in Dice of CE-NC-VesselSegNet from 0.854 in dataset 1 to 0.738 in dataset 2, and the decrease in Dice of CE-VesselSegNet from 0.930 to 0.837, and the slightly lower sensitivity of CE-NC-VesselSegNet compared to NC-VesselSegNet trained by D2.

[0198] The quality of the annotation determines the performance of the segmentation model. Two annotation examples from the ISICDM 2021 challenge (dataset 2), one of a non-enhanced lung image and the other of an enhanced lung image. The lung parenchyma marked with an ellipse was mislabeled as a pulmonary vessel, and two separate vessels marked with a rectangle were mislabeled as one vessel. The airway wall marked with an ellipse was mislabeled. However, these three errors did not appear in the segmentation results of our CE-NC-VesselSegNet and CE-VesselSegNet. In our annotation of dataset D1, these similar mislabelings did not occur. Therefore, our high annotation quality in dataset D1 is considered the key reason for the excellent segmentation performance.

[0199] Finally, although the NC-VesselSegNet trained with D2 and the CE-VesselSegNet trained with D2 segmented into smaller blood vessels, these vessels were mainly intermittent and non-connected. The pulmonary vessels in the D2 dataset segmented by our model are more complete and have fewer disconnected parts.

[0200] The annotated pulmonary vessels in the enhanced lung image can be transferred to the non-enhanced lung image through a spatial registration method, which has the potential to solve the low accuracy problem of direct annotation in non-enhanced lung images. By using these high-quality annotation transfers, CE-NC-VesselSegNet can be trained to segment pulmonary vessels from non-enhanced lung images. Compared with models trained with other annotations directly drawn in non-enhanced lung images, CE-NC-VesselSegNet can segment pulmonary vessels more accurately and continuously. After validation and necessary improvements, the proposed method and model can be applied to various pulmonary vascular diseases.

[0201] In summary, automatically segmenting pulmonary vessels from CT images is of great significance. However, directly and accurately annotating pulmonary vessels in non-enhanced CT (non-enhanced lung images) is a complex and time-consuming task. Embodiments of the present disclosure aim to draw annotations using enhanced CT (enhanced lung images) and train a deep learning model for segmenting pulmonary vessels from non-enhanced lung images. We collected two datasets of 55 CT scans. Dataset 1 includes 17 cases of enhanced lung image annotations, 2 cases of non-enhanced lung image annotations, and 12 non-enhanced lung image scans. Dataset 2 consists of 12 enhanced lung images, 12 non-enhanced lung image scans, and annotations. First, the annotations drawn in the enhanced lung images (Dataset 1) are transferred to the non-enhanced lung images through a registration method. Second, a CE-NC-VesselSegNet is proposed and trained to segment pulmonary vessels from non-enhanced lung images using the transferred annotation data. Finally, CE-NC-VesselSegNet is evaluated and compared with similar studies. After registration, both the maximum error and root mean square error between the enhanced and non-enhanced lung images decrease, while the structural similarity and peak signal-to-noise ratio increase. CE-NC-VesselSegNet can accurately segment pulmonary vessels from non-enhanced lung images, with a dice of 0.854 for the segmentation evaluation metric. In the external validation using Dataset 2, the dice of CE-NC-VesselSegNet is 0.738, which is higher than that of the NC-VesselSegNet trained on Dataset 2. Visual observation shows that CE-NC-VesselSegNet can achieve more precise and continuous segmentation. The pulmonary vessel annotations drawn in the enhanced lung images can be transformed to non-enhanced lung images through spatial registration. By using these high-quality transferred annotations, CE-NC-VesselSegNet can be trained to segment pulmonary vessels from non-enhanced lung images.

[0202] The execution subject of the pulmonary vessel segmentation method can be a pulmonary vessel segmentation device. For example, the pulmonary vessel segmentation method can be executed by a terminal device, a server, or other processing devices. Among them, the terminal device can be a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementation manners, the pulmonary vessel segmentation method can be implemented by a processor invoking computer-readable instructions stored in a memory.

[0203] Those skilled in the art can understand that in the above method of the specific implementation manner, the writing order of each step does not mean a strict execution order that constitutes any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0204] Meanwhile, an embodiment of the present disclosure provides a pulmonary vessel segmentation device, and the pulmonary vessel segmentation device includes: a first acquisition unit, configured to acquire an enhanced lung image, its corresponding pulmonary vessel image, a first non-enhanced lung image, and a preset first segmentation model; a first registration unit, configured to perform a registration operation on the enhanced lung image and the first non-enhanced lung image to obtain a lung-registered image corresponding to the first non-enhanced lung image; a first lung region segmentation unit, configured to perform lung region segmentation on the lung-registered image to obtain a lung region image; a first training unit, configured to train the preset first segmentation model based on the pulmonary vessel image and the lung region image; and a first pulmonary vessel segmentation unit, configured to perform pulmonary vessel segmentation on a second non-enhanced lung image based on the trained preset first segmentation model.

[0205] Meanwhile, another embodiment of the present disclosure provides a pulmonary vessel segmentation device, and the pulmonary vessel segmentation device includes: a second acquisition unit, configured to acquire an enhanced lung image, its corresponding pulmonary vessel image, a first non-enhanced lung image, and a preset first segmentation model; a second lung region segmentation unit, configured to perform segmentation on the enhanced lung image and the first non-enhanced lung image respectively to obtain a first lung region image and a second lung region image; a second registration unit, configured to perform a registration operation on the first lung region image and the second lung region image to obtain a lung region-registered image corresponding to the second lung region image; a second training unit, configured to train the preset first segmentation model based on the lung region-registered image and the pulmonary vessel image; and a second pulmonary vessel segmentation unit, configured to perform pulmonary vessel segmentation on a second non-enhanced lung image based on the trained preset first segmentation model.

[0206] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the pulmonary vessel segmentation method described in the above method embodiments, and its specific implementation can refer to the description of the above method embodiments. For the sake of brevity, it will not be elaborated here.

[0207] The embodiments of the present disclosure further provide a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the above-described pulmonary vessel segmentation method is implemented. The computer-readable storage medium can be a non-volatile computer-readable storage medium.

[0208] The embodiments of the present disclosure further provide an electronic device, including: a processor; and a memory for storing processor-executable instructions, wherein the processor is configured to execute the above-described pulmonary vessel segmentation method. The electronic device can be provided as a terminal, a server, or other forms of devices.

[0209] Figure 5FIG. 0 is a block diagram of an electronic device 800 shown in accordance with an exemplary embodiment. For example, the electronic device 800 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, and the like.

[0210] Referring Figure 5 , the electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power 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.

[0211] The processing component 802 generally controls the overall operation of the electronic device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations. The processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the above-described methods. In addition, the processing component 802 may include one or more modules to facilitate interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate interaction between the multimedia component 808 and the processing component 802.

[0212] The memory 804 is configured to store various types of data to support the operation of the electronic device 800. Examples of such data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, pictures, videos, and the like. The memory 804 may be implemented by any type of volatile or non-volatile storage device 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 memory, flash memory, a magnetic disk, or an optical disk.

[0213] The power component 806 provides power to the various components of the electronic device 800. The power component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device 800.

[0214] The multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user. In some embodiments, the screen may 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 input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can sense not only the boundaries of touch or swipe actions but also detect the duration and pressure associated with the touch or swipe operation. 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 zoom capabilities.

[0215] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC) that is configured to receive external audio signals 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 signals 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 for outputting audio signals.

[0216] The I / O interface 812 provides an interface between the processing component 802 and a peripheral interface module, which can be a keyboard, a click wheel, buttons, etc. These buttons can include but are not limited to: a home button, a volume button, a power button, and a lock button.

[0217] The sensor component 814 includes one or more sensors for providing an assessment of the status of various aspects of the electronic device 800. For example, the sensor component 814 can detect the on / off state of the electronic device 800, the relative positioning of components, such as the display and the keypad of the electronic device 800. The sensor component 814 can also detect a change in the position of the electronic device 800 or a component of the electronic device 800, the presence or absence of user contact with the electronic device 800, the orientation or acceleration / deceleration of the electronic device 800, and a change in the temperature of the electronic device 800. The sensor component 814 can include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor component 814 can also include a light sensor, such as a CMOS or a CCD image sensor, for use in imaging applications. In some embodiments, the sensor component 814 can further include an acceleration sensor, a gyro sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0218] The communication component 816 is configured to facilitate communication between the electronic device 800 and other devices in a wired or wireless manner. The electronic device 800 can access a communication standard-based wireless network, 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) technology, Infrared Data Association (IrDA) technology, Ultra Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0219] 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, microcontrollers, microprocessors, or other electronic components for performing the above method.

[0220] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 804 including computer program instructions, and the above computer program instructions can be executed by a processor 820 of the electronic device 800 to complete the above method.

[0221] Figure 6 is a block diagram of an electronic device 1900 shown according to an exemplary embodiment. For example, the electronic device 1900 can be provided as a server. Referring to Figure 6 , the electronic device 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by a memory 1932 for storing instructions executable by the processing component 1922, such as application programs. The application programs stored in the memory 1932 can include one or more modules each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute instructions to perform the above method.

[0222] The electronic device 1900 may further include a power 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.

[0223] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions, and the computer program instructions can be executed by a processing component 1922 of the electronic device 1900 to complete the above method.

[0224] The present disclosure may be a system, a method, and / or a computer program product. The computer program product may include a computer-readable storage medium having thereon computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0225] A computer-readable storage medium may be a tangible device that can retain and store instructions for use by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical 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: a portable computer disk, 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 disc (DVD), a memory stick, a floppy disk, a mechanically encoded device such as a punch card or raised structures in grooves having instructions stored thereon, and any suitable combination of the foregoing. The computer-readable storage medium used herein is not construed as being a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated through a waveguide or other transmission medium (e.g., an optical pulse through an optical fiber cable), or an electrical signal transmitted through a wire.

[0226] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded to an external computer or external storage device through a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network may include a copper transmission cable, an optical fiber transmission, a wireless transmission, a router, a firewall, a switch, a gateway computer, and / or an edge server. A network adapter or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing / processing device.

[0227] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine - related instructions, microcode, firmware instructions, state - setting data, or source code or object code written in any combination of one or more programming languages, including object - oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as the "C" language or similar programming languages. The computer - readable program instructions may be executed entirely on the user's computer, partially on the user's computer, executed as a stand - alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may 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 may be connected to an external computer (e.g., through the Internet using an Internet service provider). In some embodiments, by using the state information of the computer - readable program instructions to customize an electronic circuit, such as a programmable logic circuit, a field - programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can execute the computer - readable program instructions to implement various aspects of the present disclosure.

[0228] Aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer - readable program instructions.

[0229] These computer - readable program instructions can be provided to a processor of a general - purpose computer, a special - purpose computer, or other programmable data - processing apparatus to produce a machine such that the instructions, when executed by the processor of the computer or other programmable data - processing apparatus, create a means for implementing the functions / acts specified in one or more blocks of the flowchart and / or block diagram. These computer - readable program instructions can also be stored in a computer - readable storage medium, which causes a computer, a programmable data - processing apparatus, and / or other devices to operate in a particular manner. Thus, the computer - readable medium storing the instructions comprises a manufacture, which includes instructions for implementing various aspects of the functions / acts specified in one or more blocks of the flowchart and / or block diagram.

[0230] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other devices to produce a computer-implemented process such that the instructions executed on the computer, other programmable data processing apparatus, or other devices implement the functions / acts specified in one or more boxes of the flowchart and / or block diagram.

[0231] The flowcharts and block diagrams in the figures 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 flowchart or block diagram may represent a module, a segment of a program, or a portion of an instruction, which contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two consecutive blocks may in fact be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block of the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented by special-purpose hardware-based systems that perform the specified functions or acts, or by combinations of special-purpose hardware and computer instructions.

[0232] The embodiments of the present disclosure have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, the practical application, or the technical improvement of the technology in the market, or to enable other ordinary skilled persons in the art to understand the embodiments disclosed herein.

Claims

1. A method for segmenting pulmonary blood vessels, characterized in that, Including: Obtaining an enhanced lung image, a lung vascular image corresponding to the enhanced lung image, a first non-enhanced lung image corresponding to the enhanced lung image, and a preset first segmentation model; Before obtaining the enhanced lung image and the lung vascular image corresponding to the enhanced lung image, performing vascular segmentation and lung region segmentation on the enhanced lung image respectively to obtain a vascular image corresponding to the enhanced lung image and a lung region image corresponding to the enhanced lung image; obtaining the lung vascular image corresponding to the enhanced lung image according to the vascular image corresponding to the enhanced lung image and the lung region image corresponding to the enhanced lung image; Performing a registration operation on the enhanced lung image and the first non-enhanced lung image to obtain a lung registration image corresponding to the first non-enhanced lung image; Performing lung region segmentation on the lung registration image corresponding to the first non-enhanced lung image to obtain a lung region image; Training the preset first segmentation model based on transferring the lung vascular image corresponding to the enhanced lung image to the lung vascular image corresponding to the first non-enhanced lung image and the lung region image corresponding to the lung registration image; Performing lung vascular segmentation on a second non-enhanced lung image based on the trained preset first segmentation model; wherein, performing lung vascular segmentation on the second non-enhanced lung image based on the trained preset first segmentation model includes: obtaining a first segmentation model for performing lung vascular segmentation on a non-enhanced lung image under at least 2 views; projecting the second non-enhanced lung image according to the at least 2 views corresponding to the first segmentation model to obtain corresponding second non-enhanced lung projection images; respectively using the first segmentation model to segment the second non-enhanced lung projection images to obtain corresponding lung vascular segmentation images to be fused; fusing the lung vascular segmentation images to be fused to obtain a lung vascular image corresponding to the second non-enhanced lung image; projecting the lung vascular segmentation images to be fused respectively based on the projection directions to obtain lung vascular segmentation projection images; registering the lung vascular segmentation images to be fused to obtain corresponding registration point pairs; performing a mean value process on the positions corresponding to the registration points to obtain a lung vascular image corresponding to the second non-enhanced lung image.

2. The splitting method according to claim 1, wherein, The performing vascular segmentation and lung region segmentation on the enhanced lung image respectively includes: Obtaining a preset vascular segmentation model and a first preset lung region segmentation model; Respectively performing vascular segmentation and lung region segmentation on the enhanced lung image based on the preset vascular segmentation model and the first preset lung region segmentation model to obtain a vascular image corresponding to the enhanced lung image and a lung region image corresponding to the enhanced lung image.

3. The segmentation method according to claim 2, characterized in that The obtaining the lung vascular image corresponding to the enhanced lung image according to the vascular image corresponding to the enhanced lung image and the lung region image corresponding to the enhanced lung image includes: Performing a multiplication operation on the vascular image corresponding to the enhanced lung image and the lung region image corresponding to the enhanced lung image to obtain the lung vascular image corresponding to the enhanced lung image.

4. The splitting method according to any one of claims 1-3, characterized in that, The performing lung region segmentation on the lung registration image corresponding to the first non-enhanced lung image to obtain a lung region image includes: Obtaining a second preset lung region segmentation model; Based on the second preset lung region segmentation model, perform lung region segmentation on the lung registration image corresponding to the first non-enhanced lung image to obtain a lung region image.

5. The splitting method according to any one of claims 1-3, characterized in that, The first segmentation model includes: a U-Net backbone network; After each convolution operation of the U-Net backbone network, multiple feature maps are obtained; Perform feature map normalization on the multiple feature maps respectively; Use an activation function to activate the normalized multiple feature maps.

6. The splitting method according to claim 4, wherein The first segmentation model includes: a U-Net backbone network; After each convolution operation of the U-Net backbone network, multiple feature maps are obtained; Perform feature map normalization on the multiple feature maps respectively; Use an activation function to activate the normalized multiple feature maps.

7. The segmentation method according to any one of claims 1-3 and 6, characterized in that, Training the preset first segmentation model includes: During the decoding process of the first segmentation model, calculate the loss of the feature map corresponding to each decoding; According to the loss of the feature map corresponding to each decoding, obtain the total loss of the first segmentation model during the training process.

8. The segmentation method according to claim 4, wherein Training the preset first segmentation model includes: During the decoding process of the first segmentation model, calculate the loss of the feature map corresponding to each decoding; According to the loss of the feature map corresponding to each decoding, obtain the total loss of the first segmentation model during the training process.

9. The segmentation method according to claim 5, characterized in that, Training the preset first segmentation model includes: During the decoding process of the first segmentation model, calculate the loss of the feature map corresponding to each decoding; According to the loss of the feature map corresponding to each decoding, obtain the total loss of the first segmentation model during the training process.

10. A method for segmenting pulmonary blood vessels, characterized in that, Includes: Obtain an enhanced lung image, the lung vascular image corresponding to the enhanced lung image, the first non-enhanced lung image corresponding to the enhanced lung image, and a preset first segmentation model; Before obtaining the enhanced lung image and the lung vascular image corresponding to the enhanced lung image, perform vascular segmentation and lung region segmentation on the enhanced lung image respectively to obtain the vascular image corresponding to the enhanced lung image and the lung region image corresponding to the enhanced lung image; according to the vascular image corresponding to the enhanced lung image and the lung region image corresponding to the enhanced lung image, obtain the lung vascular image corresponding to the enhanced lung image; Perform lung region segmentation on the enhanced lung image and the first non-enhanced lung image respectively to obtain the first lung region image corresponding to the enhanced lung image and the second lung region image corresponding to the first non-enhanced lung image; Perform a registration operation on the first lung region image of the enhanced lung image and the second lung region image of the first non-enhanced lung image to obtain a lung region registration image corresponding to the second lung region image of the first non-enhanced lung image; Based on the lung region registration image corresponding to the second lung region image and transferring the lung vascular image corresponding to the enhanced lung image to the lung vascular image corresponding to the first non-enhanced lung image, train the preset first segmentation model; Based on the pre-trained preset first segmentation model, perform pulmonary vascular segmentation on the second non-enhanced lung image; wherein, performing pulmonary vascular segmentation on the second non-enhanced lung image based on the pre-trained preset first segmentation model includes: obtaining the first segmentation model for pulmonary vascular segmentation of the non-enhanced lung image under at least two views; projecting the second non-enhanced lung image according to the at least two views corresponding to the first segmentation model to obtain the corresponding second non-enhanced lung projection image; respectively using the first segmentation model to segment the second non-enhanced lung projection image to obtain the corresponding pulmonary vascular segmentation images to be fused; fusing the pulmonary vascular segmentation images to be fused to obtain the pulmonary vascular image corresponding to the second non-enhanced lung image; respectively projecting the pulmonary vascular segmentation images to be fused based on the projection direction to obtain the pulmonary vascular segmentation projection image; registering the pulmonary vascular segmentation images to be fused to obtain the corresponding registration point pairs; performing mean processing on the positions corresponding to the registration points to obtain the pulmonary vascular image corresponding to the second non-enhanced lung image.

11. The segmentation method according to claim 10, characterized in that, The performing vascular segmentation and pulmonary region segmentation on the enhanced lung image respectively includes: Obtaining a preset vascular segmentation model and a first preset pulmonary region segmentation model; Respectively based on the preset vascular segmentation model and the first preset pulmonary region segmentation model, perform vascular segmentation and pulmonary region segmentation on the enhanced lung image to obtain the vascular image corresponding to the enhanced lung image and the pulmonary region image corresponding to the enhanced lung image.

12. The segmentation method according to claim 11, characterized in that, The obtaining the pulmonary vascular image corresponding to the enhanced lung image according to the vascular image corresponding to the enhanced lung image and the pulmonary region image corresponding to the enhanced lung image includes: Performing a multiplication operation on the vascular image corresponding to the enhanced lung image and the pulmonary region image corresponding to the enhanced lung image to obtain the pulmonary vascular image corresponding to the enhanced lung image.

13. The splitting method according to any one of claims 10-12, characterized in that, The performing pulmonary region segmentation on the enhanced lung image and the first non-enhanced lung image respectively to obtain the first pulmonary region image corresponding to the enhanced lung image and the second pulmonary region image corresponding to the first non-enhanced lung image includes: Obtaining a third preset pulmonary region segmentation model and a fourth preset pulmonary region segmentation model; Respectively based on the third preset pulmonary region segmentation model and the fourth preset pulmonary region segmentation model, perform pulmonary region segmentation on the enhanced lung image and the first non-enhanced lung image to obtain the first pulmonary region image corresponding to the enhanced lung image and the second pulmonary region image corresponding to the first non-enhanced lung image.

14. The splitting method according to any one of claims 10 to 12, characterized in that The first segmentation model includes: a U-Net backbone network; After each convolution operation of the U-Net backbone network, multiple feature maps are obtained; Respectively perform feature map normalization on the multiple feature maps; Use an activation function to activate the normalized multiple feature maps.

15. The segmentation method according to claim 13, characterized in that, The first segmentation model includes: a U-Net backbone network; After each convolution operation of the U-Net backbone network, multiple feature maps are obtained; Respectively perform feature map normalization on the multiple feature maps; Use an activation function to activate the normalized multiple feature maps.

16. The splitting method according to any one of claims 10-12 and 15, characterized in that Training the preset first segmentation model includes: During the decoding process of the first segmentation model, calculating the loss of the feature map corresponding to each decoding; Obtaining the total loss during the training process according to the loss of the feature map corresponding to each decoding.

17. The splitting method according to claim 13, characterized in that, Training the preset first segmentation model includes: During the decoding process of the first segmentation model, calculating the loss of the feature map corresponding to each decoding; Obtaining the total loss during the training process according to the loss of the feature map corresponding to each decoding.

18. The segmentation method according to claim 14, characterized in that, Training the preset first segmentation model includes: During the decoding process of the first segmentation model, calculating the loss of the feature map corresponding to each decoding; Obtaining the total loss during the training process according to the loss of the feature map corresponding to each decoding.

19. A segmentation device for pulmonary blood vessels, characterized in that, Including: A first acquisition unit, configured to acquire an enhanced lung image, the lung vascular image corresponding to the enhanced lung image, the first non-enhanced lung image corresponding to the enhanced lung image, and a preset first segmentation model; before acquiring the enhanced lung image and the lung vascular image corresponding to the enhanced lung image, respectively performing vascular segmentation and lung region segmentation on the enhanced lung image to obtain the vascular image corresponding to the enhanced lung image and the lung region image corresponding to the enhanced lung image; and obtaining the lung vascular image corresponding to the enhanced lung image according to the vascular image corresponding to the enhanced lung image and the lung region image corresponding to the enhanced lung image; A first registration unit, configured to perform a registration operation on the enhanced lung image corresponding to the first non-enhanced lung image and the first non-enhanced lung image to obtain the lung registration image corresponding to the first non-enhanced lung image; A first lung region segmentation unit, configured to perform lung region segmentation on the lung registration image corresponding to the first non-enhanced lung image to obtain a lung region image; A first training unit, configured to train the preset first segmentation model based on transferring the lung vascular image corresponding to the enhanced lung image to the lung vascular image corresponding to the first non-enhanced lung image and the lung region image corresponding to the lung registration image; The first pulmonary vascular segmentation unit is used to perform pulmonary vascular segmentation on the second non-enhanced lung image based on the trained preset first segmentation model; wherein, performing pulmonary vascular segmentation on the second non-enhanced lung image based on the trained preset first segmentation model includes: obtaining first segmentation models for pulmonary vascular segmentation of non-enhanced lung images under at least two views; projecting the second non-enhanced lung image according to the at least two views corresponding to the first segmentation model to obtain corresponding second non-enhanced lung projection images; respectively using the first segmentation model to segment the second non-enhanced lung projection images to obtain corresponding pulmonary vascular segmentation images to be fused; fusing the pulmonary vascular segmentation images to be fused to obtain a pulmonary vascular image corresponding to the second non-enhanced lung image; based on the projection direction, respectively projecting the pulmonary vascular segmentation images to be fused to obtain pulmonary vascular segmentation projection images; registering the pulmonary vascular segmentation images to be fused to obtain corresponding registration point pairs; performing mean processing on the positions corresponding to the registration points to obtain a pulmonary vascular image corresponding to the second non-enhanced lung image.

20. A pulmonary vascular segmentation device, characterized in that, Including: The second acquisition unit is used to obtain an enhanced lung image, the pulmonary vascular image corresponding to the enhanced lung image, the first non-enhanced lung image corresponding to the enhanced lung image, and a preset first segmentation model; before obtaining the enhanced lung image and the pulmonary vascular image corresponding to the enhanced lung image, respectively performing vascular segmentation and pulmonary region segmentation on the enhanced lung image to obtain a vascular image corresponding to the enhanced lung image and a pulmonary region image corresponding to the enhanced lung image; obtaining the pulmonary vascular image corresponding to the enhanced lung image according to the vascular image corresponding to the enhanced lung image and the pulmonary region image corresponding to the enhanced lung image; The second pulmonary region segmentation unit is used to perform pulmonary region segmentation on the enhanced lung image and the first non-enhanced lung image respectively to obtain a first pulmonary region image corresponding to the enhanced lung image and a second pulmonary region image corresponding to the first non-enhanced lung image; The second registration unit is used to perform a registration operation on the first pulmonary region image of the enhanced lung image and the second pulmonary region image of the first non-enhanced lung image to obtain a pulmonary region registration image corresponding to the second pulmonary region image of the first non-enhanced lung image; The second training unit is used to train the preset first segmentation model based on the pulmonary region registration image corresponding to the second pulmonary region image and transferring the pulmonary vascular image corresponding to the enhanced lung image to the pulmonary vascular image corresponding to the first non-enhanced lung image. A second pulmonary vascular segmentation unit, configured to perform pulmonary vascular segmentation on a second non-enhanced lung image based on the trained preset first segmentation model; wherein, performing pulmonary vascular segmentation on the second non-enhanced lung image based on the trained preset first segmentation model includes: obtaining first segmentation models for pulmonary vascular segmentation of non-enhanced lung images under at least two views; projecting the second non-enhanced lung image according to the at least two views corresponding to the first segmentation model to obtain corresponding second non-enhanced lung projection images; respectively using the first segmentation model to segment the second non-enhanced lung projection images to obtain corresponding pulmonary vascular segmentation images to be fused; fusing the pulmonary vascular segmentation images to be fused to obtain a pulmonary vascular image corresponding to the second non-enhanced lung image; based on the projection directions, respectively projecting the pulmonary vascular segmentation images to be fused to obtain pulmonary vascular segmentation projection images; registering the pulmonary vascular segmentation images to be fused to obtain corresponding registration point pairs; performing mean processing on the positions corresponding to the registration points to obtain a pulmonary vascular image corresponding to the second non-enhanced lung image.

21. An electronic device, characterized in that, Comprising: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the pulmonary vascular segmentation method according to any one of claims 1 to 9.

22. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the pulmonary vascular segmentation method according to any one of claims 1 to 9 is implemented.

23. An electronic device, characterized in that, Comprising: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the pulmonary vascular segmentation method according to any one of claims 10 to 18.

24. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the pulmonary vascular segmentation method according to any one of claims 10 to 18 is implemented.

Images