Breathing feature extraction method and device for chest and abdomen body surface point cloud partitioning by considering lung physiological structure, readable storage medium and computer program product

Abstract

The invention belongs to the field of lung medical data processing and radiotherapy, and discloses a respiratory feature extraction method for chest and abdomen body surface point cloud partitioning by considering a lung physiological structure. After the body surface point cloud image is differentiated, correlation coefficient calculation is carried out according to a lung lobe model reconstructed by a lung 4DCT image, a body surface area highly correlated with a lung lobe structure is selected to carry out motion information dimension reduction recombination, and finally body surface global features of breathing are represented in a three-dimensional vector form. Compared with the prior art, the method has the advantages that the influence of the lung physiological structure on the lung tumor movement mode is creatively considered, meanwhile, the advantages of being noninvasive and free of extra dosage are kept, a new thought can be provided for development of the radiotherapy image guide technology, and further research in the field of thoracic and abdominal activity tumor radiotherapy is facilitated.

Description

Technical Field

[0001] The present invention belongs to the field of lung medical data processing and radiotherapy, and more specifically, relates to a respiratory feature extraction method, device, readable storage medium and computer program product for partitioning chest and abdominal surface point clouds taking into account the physiological structure of the lungs. Background Art

[0002] In recent years, the number of malignant tumors and deaths worldwide has continued to rise. Cancer and tumors pose a serious obstacle to further development of global health and well-being. Global cancer data released by the World Health Organization indicate that in 2022, nearly 20 million new cases and approximately 9.7 million deaths will occur worldwide, with lung cancer leading the way in both incidence and mortality. Radiotherapy is currently the main treatment option for thoracic and abdominal cancers such as lung cancer. However, due to respiratory motion, the target volume of active thoracic and abdominal tumors can shift in real time, causing beam deviation and, in turn, dose deviation, significantly reducing radiotherapy efficacy and increasing the risk of complications. To prevent this, existing technologies primarily rely on image-guided techniques, including the gold standard method, for positioning correction. Commonly used image-guided techniques include X-rays and optical imaging. X-rays can capture anatomical images but impose additional dose on the patient. Optical imaging, while radiation-free, cannot monitor the position of active thoracic and abdominal tumor targets and is therefore only suitable for guiding radiotherapy for superficial tumors or rigid areas. Non-invasive, radiation-free monitoring of active thoracic and abdominal targets during radiotherapy is currently a key focus of radiotherapy research and development.

[0003] Some studies have proposed that by establishing a surface-to-body motion correlation model, the position trajectory of active targets in the body can be predicted by monitoring surface motion. Extracting and representing surface motion information is crucial for establishing a motion correlation model between surface motion and tumor targets.

[0004] Initially, researchers used a single marker on the chest and abdomen to attempt to match its displacement to tumor motion, but the results showed that a single marker was insufficient. The CyberKnife system uses a fabric vest with three optical markers to capture motion at three locations on the body surface, achieving for the first time continuous alignment of the radiation beam with the tumor target. Subsequent studies have explored the possibility of using a larger number of surface markers, and a general rule has been established: the more surface markers, the easier it is to extract respiratory signal features, but the completeness of the respiratory signal is limited by the distribution of the markers.

[0005] The effectiveness of surface marker extraction is generally positively correlated with the number of points, but too many points can complicate the treatment process and easily lead to errors in point placement. A study has proposed a method for establishing a tumor-chest and abdominal surface association model based on a three-dimensional point cloud, validating the feasibility of using point cloud data modeling instead of external markers. Therefore, researchers have taken this approach and reconstructed the chest and abdominal point cloud data captured by a depth camera into a voxel model. They then selected surface regions with a high correlation with tumor motion for dimensionality reduction, resulting in an effective one-dimensional representation of the surface region.

[0006] The method proposed in the present invention solves the technical bottlenecks of current respiratory motion feature extraction methods, such as insufficient anatomical basis. Based on chest and abdominal surface point cloud data, and taking the physiological structure of the lungs as the starting point, the correlation between the lung lobes and the body surface is constructed by fusing the three-dimensional point cloud of the body surface with the anatomical division information of the lungs. Based on the body surface deformation data, a method is used to extract the global three-dimensional feature vector of the chest and abdomen that characterizes respiratory motion with lung lobe spatial specificity. This method echoes the law that the position of the tumor in the lungs affects the tumor motion pattern, and can reflect the specificity of the region where the tumor is located. After characterizing the body surface motion through this method, an association model can be further established with the tumor target area in the body to guide the offset compensation of radiotherapy. Summary of the Invention

[0007] In response to the above-mentioned defects or improvement needs of the prior art, the purpose of this application is to propose a respiratory feature extraction method, device, readable storage medium and computer program product that considers the physiological structure of the lungs to perform chest and abdominal surface point cloud partitioning, which can more comprehensively and flexibly characterize the surface respiratory motion characteristic information.

[0008] To achieve the above objectives, according to one aspect of the present invention, a respiratory feature extraction method for performing chest and abdominal surface point cloud partitioning taking into account the physiological structure of the lungs is provided, comprising the following steps:

[0009] Acquire multi-respiratory phase images of the patient's lung region, delineate the lung lobes and tumor target areas in each respiratory phase, and reconstruct a three-dimensional lung lobe model; synchronously acquire dynamic point cloud data of the human chest and abdominal surface to obtain multi-frame point cloud images in a time series, and uniformly grid-segment the point cloud images of the chest and abdominal surface to obtain point cloud sub-regions; wherein the point cloud images of the patient's chest and abdominal surface are synchronized with the multi-respiratory phase images of the patient's lung region;

[0010] Based on the physiological structural characteristics of the three-dimensional lung lobe model, point cloud sub-regions with high spatial motion correlation with different lung lobes are selected to establish a body surface-lung lobe registration relationship, and point cloud sub-regions with biomechanical coupling relationships with different lung lobes are selected and combined into corresponding high-correlation body surface regions;

[0011] The feature space of each highly correlated body surface area is reduced in dimension, and physiological weights are assigned according to the anatomical location of the tumor target area. The reduced feature vectors are weightedly fused to generate a global motion feature vector that represents the respiratory motion cycle.

[0012] Furthermore, the step of uniformly gridding the chest and abdomen surface point cloud image comprises:

[0013] The three-dimensional point cloud image is evenly divided into multiple regular square areas within the projection range of the XOY plane. Each point cloud sub-area is evenly divided on the X and Y axes based on the square area projection of the XOY plane, and the length in the Z axis direction is adaptively adjusted according to the points within the square area, and should just surround all points in the sub-area.

[0014] Furthermore, the method for synchronizing the patient's chest and abdomen surface point cloud image with the patient's lung area multi-respiration phase image includes:

[0015] Set up a scanning area covering the chest and abdomen, and obtain body surface depth data through structured light camera encoding;

[0016] Preprocess the raw depth data, including downsampling, denoising, and statistical outlier filtering;

[0017] Output standardized point cloud data files.

[0018] Furthermore, the combination is a corresponding high-correlation body surface area, including:

[0019] For each gridded point cloud sub-region, calculate its spatial coordinate mean vector;

[0020] Calculate the center of mass motion trajectory for each reconstructed lung lobe model;

[0021] Extract the mean Z-axis displacement of the point cloud sub-region and the Z-axis displacement of the lung lobe center of mass in each respiratory phase;

[0022] Calculate the Pearson correlation coefficient between each point cloud sub-region and the displacement sequence of different lung lobes, and use the Pearson correlation coefficient as the correlation coefficient of the point cloud sub-region to the corresponding lung lobe;

[0023] According to the anatomical divisions of the right upper lobe, right middle lobe, right lower lobe, left upper lobe, and left lower lobe, point cloud sub-regions with high correlation coefficients were selected for spatial clustering to form high-correlation surface regions of the five lobes.

[0024] Furthermore, the value of the Pearson correlation coefficient r i,j The calculation formula is as follows:

[0025]

[0026] Among them, r i,j is the Pearson correlation coefficient between body surface patch i and lung lobe j, Indicates that the surface patch i is in the The Z-axis mean coordinate of the respiratory phase, represents the Z-axis mean of the surface patch i in all phases, Indicates that the lung lobe is in the The Z-axis centroid coordinate of the respiratory phase, represents the mean Z-axis centroid coordinate of lung lobe j in all respiratory phases.

[0027] Furthermore, the weighted feature fusion of the feature vector after dimensionality reduction specifically includes the following steps:

[0028] a) performing principal component analysis on the three-dimensional point cloud data of each associated body surface area, extracting the principal component vector corresponding to the maximum eigenvalue to represent the point cloud distribution of each area;

[0029] b) determining a weight coefficient based on the contribution of respiratory motion of the lung lobe where the tumor target is located;

[0030] c) performing linear summation on the corresponding weight coefficients and the principal component vectors of the highly correlated body surface areas to extract the global three-dimensional feature vector representing the chest and abdomen during respiratory motion.

[0031] According to another aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the respiratory feature extraction method as described in any one of the preceding items.

[0032] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for extracting respiratory features as described in any of the above items is implemented.

[0033] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, which implements the respiratory feature extraction method as described in any one of the preceding items when executed by a processor.

[0034] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0035] 1. Starting with point cloud images of the human chest and abdomen, this method considers the influence of lung physiological structure and tumor location on tumor motion patterns. It selects regions with high surface-to-lung lobe correlation and combines the dimensionality-reduced features of these highly correlated regions to generate global respiratory features. This global feature vector is then used to establish a surface-to-body motion correlation model. This method overcomes the limitations of previous methods for extracting surface respiratory features, which often ignore the influence of lung physiological structure and tumor location.

[0036] 2. The dimensionality reduction method of the present invention using the Z-axis mean of the point cloud sub-region as its motion information is very simple. It can not only avoid a lot of calculations, but also the Z-axis movement is the direction with the largest amplitude of respiratory movement on the chest and abdomen surface, and large amplitude changes are more conducive to reflecting motion characteristics.

[0037] 3. For the movement pattern of lung tumors, the influence of the physiological structure of the lungs is more essential. The present invention considers the distribution of lung lobes and the location of tumors, which is more in line with physical laws.

[0038] 4. The adaptive adjustment of the weight of the high correlation region representation vector of the present invention can make the global respiratory feature more flexible and more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A flowchart of a preferred embodiment of the present invention;

[0040] Figure 2 Reconstructed 3D model of lung lobes

[0041] Figure 3 It is a point cloud image of the chest and abdomen surface;

[0042] Figure 4 This is the effect of uniformly cutting the chest and abdomen point cloud;

[0043] Figure 5 This is an example diagram of the characteristic change of the Z-axis mean of the point cloud sub-region over time;

[0044] Figure 6 It is a point cloud image of a highly correlated body surface area. DETAILED DESCRIPTION

[0045] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0046] A preferred respiratory feature extraction method for partitioning chest and abdominal surface point clouds by considering the physiological structure of the lungs adopted by the present invention includes the following steps:

[0047] The 4DCT imaging system is used to obtain multi-respiratory phase images of the patient's lung area, delineate the lung lobes and tumor target areas in each respiratory phase, and reconstruct a three-dimensional lung lobe model.

[0048] A structured optical body surface imaging system is used to synchronously collect dynamic point cloud data of the human chest and abdomen surface to obtain multi-frame point cloud images in a time series, and the chest and abdomen surface point cloud images are uniformly gridded to obtain point cloud sub-regions;

[0049] The data acquisition of the structured optical body surface imaging system is synchronized with the respiratory phase of 4DCT imaging;

[0050] Considering the physiological structural characteristics of the three-dimensional lung lobe model, point cloud sub-regions with high spatial motion correlation with different lung lobes are selected, and a body surface-lung lobe registration relationship is established. Point cloud sub-regions with biomechanical coupling relationships with different lung lobes are selected and combined into their corresponding high-correlation body surface regions;

[0051] The feature space of each highly correlated body surface area is reduced in dimension, and physiological weights are assigned considering the anatomical position of the tumor target area. The feature vectors after dimensionality reduction are weighted fused to generate a global motion feature vector that represents the respiratory motion cycle.

[0052] Furthermore, the “synchronous acquisition of dynamic point cloud data of the human chest and abdominal surface using a structured optical body surface imaging system” includes:

[0053] A structured optical body surface imaging system is used to collect body surface depth information, including operations such as setting the scanning area, downsampling, denoising, and removing outliers, to obtain three-dimensional dynamic information that can fully represent human breathing.

[0054] Furthermore, the “uniform grid segmentation of the point cloud image” includes:

[0055] The 3D point cloud image is evenly divided into multiple regular square regions within the projection range of the XOY plane. Each point cloud sub-region is evenly divided on the X and Y axes based on the projection of the square region on the XOY plane, and the length along the Z axis is adaptively adjusted according to the points within the region, which should just enclose all points in the sub-region.

[0056] Furthermore, the step of “selecting and combining point cloud sub-regions having biomechanical coupling relationships with different lung lobes into corresponding highly correlated body surface regions” includes:

[0057] Calculate the mean coordinates of all points in each point cloud sub-region;

[0058] Calculate the center of mass motion trajectory for each reconstructed three-dimensional lung lobe model;

[0059] Take the Z-axis value in the mean coordinate of the point cloud sub-region, divide the Z-axis value in each cycle according to the 4DCT respiratory phase, and take the mean Z-axis value in the same respiratory phase as the motion information of the sub-region in that phase;

[0060] The Z-axis value of the centroid coordinate of the lung lobe model is taken as its motion information;

[0061] For each lung lobe, the Pearson Correlation Coefficient (PCC) between the motion information of the point cloud sub-region and the motion information of the lung lobe is calculated, and the coefficient value is used as the correlation coefficient of the point cloud sub-region to the lung lobe;

[0062] According to the anatomical divisions of the right upper lobe, right middle lobe, right lower lobe, left upper lobe, and left lower lobe, point cloud sub-regions with high correlation coefficients were selected for spatial clustering to form high-correlation surface regions of the five lobes.

[0063] Furthermore, the “reducing the dimension of the feature space of each highly correlated body surface area” includes:

[0064] Principal Component Analysis (PCA) was performed on the five highly correlated body surface areas, and the eigenvector corresponding to the maximum eigenvalue was used to represent the point cloud distribution of each area.

[0065] Furthermore, the “performing weighted feature fusion on the feature vector after dimensionality reduction to generate a global motion feature vector representing the respiratory motion cycle” includes:

[0066] The weight coefficient is determined based on the respiratory motion contribution of the lung lobe where the tumor target is located. The closer the lung lobe is to the tumor, the higher the weight of the highly correlated body surface area.

[0067] The representation vectors of the highly correlated body surface areas are weightedly summed according to the obtained weights to obtain a three-dimensional vector to represent the global respiratory characteristics.

[0068] The present invention is further described below with a more specific example.

[0069] Reference Figure 1 As shown, the present invention discloses a method based on chest and abdominal surface point cloud data, taking the physiological structure of the lung as the starting point, fusing the body surface three-dimensional point cloud with the lung anatomical partition information, constructing the association relationship between the lung lobe and the body surface, and extracting the chest and abdomen global three-dimensional feature vectors representing the respiratory motion with lung lobe spatial specificity based on the body surface deformation data, including the following steps:

[0070] The patient's lung area was imaged using a 4DCT machine. The lung lobes and tumor target areas were delineated at different respiratory phases (generally 10 phases) of the 4DCT images. The delineated lung lobe areas were 3D reconstructed using Slicer software to obtain a set of lung lobe motion models of the complete respiratory process. The single respiratory phase reconstruction model was as follows: Figure 2 shown.

[0071] A structured optical body surface imaging system is used to continuously collect multi-frame three-dimensional point cloud data of the human chest and abdomen. After point cloud preprocessing, multi-frame body surface point cloud images in time series are obtained. The preprocessing includes setting the scanning area, downsampling, denoising, and removing outliers. The center reference point of the point cloud image is set to the xiphoid process. The final point cloud image is as follows Figure 3 shown.

[0072] Project the point cloud image on the XOY plane (the XOY plane coincides with the horizontal plane, and the Z axis is perpendicular to the horizontal plane), and evenly divide the projection range into multiple regular square areas. The division of each point cloud sub-area is based on the uniform intervals in the X and Y axis directions. The length of the small block in the Z axis direction is adaptively adjusted according to the points in the small block range, and should just surround all the points in the small block. That is, the X and Y axis lengths of each small block are equal (the specific length can be selected according to the actual situation, but should be controlled within 10 mm as much as possible), and the Z axis length is not necessarily equal. The effect after the division is completed is as follows Figure 4 shown.

[0073] Calculate the mean coordinates of all points in each point cloud sub-region, and calculate the centroid coordinates of each reconstructed lung lobe model. Take the Z-axis value in the mean coordinates of the point cloud sub-region, and the Z-axis coordinate changes as follows: Figure 5 As shown in the figure, the Z-axis value in each cycle is divided according to the 4DCT respiratory phase, and the mean value of the Z-axis in the same respiratory phase is taken as the motion information of the small block in the phase. The Z-axis value of the centroid coordinate of the lung lobe model is taken as its motion information. The Pearson correlation coefficient between the motion information of the point cloud sub-region and the motion information of the lung lobe is solved for each lung lobe, and the coefficient value r is used as the 4DCT respiratory phase. i,j The correlation coefficient of the point cloud sub-region to the lung lobe is as follows:

[0074]

[0075] Among them, r i,j is the Pearson correlation coefficient between body surface patch i and lung lobe j, Indicates that the surface patch i is in the The Z-axis mean coordinate of the respiratory phase, represents the Z-axis mean of the surface patch i in all phases, Indicates that the lung lobe is in the The Z-axis centroid coordinate of the respiratory phase, represents the mean Z-axis centroid coordinate of lung lobe j in all respiratory phases.

[0076] According to the anatomical divisions of the right upper lobe, right middle lobe, right lower lobe, left upper lobe, and left lower lobe, point cloud sub-regions with high correlation coefficients were selected for spatial clustering to form high-correlation surface regions of the five lobes, such asFigure 6 shown.

[0077] Principal component analysis was performed on five highly correlated body surface regions, and the eigenvector corresponding to the largest eigenvalue was used to characterize the point cloud distribution of each region. The eigenvector corresponding to the largest eigenvalue represents the direction of maximum variance in the point cloud distribution in that region, which can characterize the point cloud distribution characteristics of that region and indirectly represent the shape of the point cloud image.

[0078] The weight coefficient is determined based on the respiratory motion contribution of the lung lobe where the tumor target is located. The closer the lung lobe is to the tumor, the higher the weight corresponding to the highly correlated surface area. Based on the obtained weights, the representation vectors of the highly correlated surface areas are weighted and summed to obtain a three-dimensional vector that represents the global respiratory characteristics.

[0079] Compared with existing methods, the method proposed in this application avoids the damage to the human body caused by the gold standard method, avoids the additional dose generated by continuous X-ray exposure during image-guided radiotherapy, and reduces the negative impact of redundant areas on the accuracy of the correlation model. Furthermore, it can reflect the correlation between the location of the tumor and its movement pattern based on the different lung lobes where the tumor is located, achieving more accurate extraction of chest and abdominal surface motion feature information, which is conducive to establishing a more accurate surface-body correlation model and providing more accurate guidance for the movement of radiotherapy robots.

[0080] In general, this application proposes a method for extracting respiratory motion features based on chest and abdominal surface point cloud partitioning, taking into account the physiological structure of the lungs. After differentiating the surface point cloud image, the correlation coefficient is calculated based on the lung lobe model reconstructed from the lung 4DCT image. The surface areas with a high correlation with the lung lobe structure are selected for dimensionality reduction and reorganization of the motion information. Finally, the global surface characteristics of breathing are represented in the form of a three-dimensional vector. Compared with existing methods, this method innovatively considers the influence of the physiological structure of the lungs on the motion pattern of lung tumors, while maintaining the advantages of being non-invasive and requiring no additional dose. It can provide new ideas for the development of image-guided radiotherapy technology and contribute to further research in the field of radiotherapy for active chest and abdominal tumors.

[0081] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A respiratory feature extraction method for chest and abdomen surface point cloud partitioning considering the physiological structure of the lungs, characterized in that: The following steps are involved: Acquire multi-respiratory phase images of the patient's lung region, delineate the lung lobes and tumor target areas in each respiratory phase, and reconstruct a three-dimensional lung lobe model; synchronously acquire dynamic point cloud data of the human chest and abdominal surface to obtain multi-frame point cloud images in a time series, and uniformly grid-segment the point cloud images of the chest and abdominal surface to obtain point cloud sub-regions; wherein the point cloud images of the patient's chest and abdominal surface are synchronized with the multi-respiratory phase images of the patient's lung region; Based on the physiological structural characteristics of the three-dimensional lung lobe model, point cloud sub-regions with high spatial motion correlation with different lung lobes are selected to establish a body surface-lung lobe registration relationship, and point cloud sub-regions with biomechanical coupling relationships with different lung lobes are selected and combined into corresponding high-correlation body surface regions; The feature space of each highly correlated body surface area is reduced in dimension, and physiological weights are assigned according to the anatomical location of the tumor target area. The reduced feature vectors are weightedly fused to generate a global motion feature vector that represents the respiratory motion cycle.

2. The respiratory feature extraction method according to claim 1, wherein The step of uniformly gridding the chest and abdomen surface point cloud image comprises: The three-dimensional point cloud image is evenly divided into multiple regular square areas within the projection range of the XOY plane. Each point cloud sub-area is evenly divided on the X and Y axes based on the square area projection of the XOY plane, and the length in the Z axis direction is adaptively adjusted according to the points within the square area, and should just surround all points in the sub-area.

3. The respiratory feature extraction method according to claim 1, wherein The method for synchronizing the patient's chest and abdomen surface point cloud image with the patient's lung area multi-respiration phase image includes: Set up a scanning area covering the chest and abdomen, and obtain body surface depth data through structured light camera encoding; Preprocess the raw depth data, including downsampling, denoising, and statistical outlier filtering; Output standardized point cloud data files.

4. The respiratory feature extraction method according to claim 1, wherein The combination corresponds to a high correlation body surface area, including: For each gridded point cloud sub-region, calculate its spatial coordinate mean vector; Calculate the center of mass motion trajectory for each reconstructed lung lobe model; Extract the mean Z-axis displacement of the point cloud sub-region and the Z-axis displacement of the lung lobe center of mass in each respiratory phase; Calculate the Pearson correlation coefficient between each point cloud sub-region and the displacement sequence of different lung lobes, and use the Pearson correlation coefficient as the correlation coefficient of the point cloud sub-region to the corresponding lung lobe; According to the anatomical divisions of the right upper lobe, right middle lobe, right lower lobe, left upper lobe, and left lower lobe, point cloud sub-regions with high correlation coefficients were selected for spatial clustering to form high-correlation surface regions of the five lobes.

5. The respiratory feature extraction method according to claim 4, wherein: The value of the Pearson correlation coefficient ri ,j The calculation formula is as follows: Among them, r i,j is the Pearson correlation coefficient between body surface patch i and lung lobe j, Indicates that the surface patch i is in the The Z-axis mean coordinate of the respiratory phase, represents the Z-axis mean of the surface patch i in all phases, Indicates that the lung lobe is in the The Z-axis centroid coordinate of the respiratory phase, represents the mean Z-axis centroid coordinate of lung lobe j in all respiratory phases.

6. The respiratory feature extraction method according to claim 1, wherein: The weighted feature fusion of the feature vector after dimensionality reduction specifically includes the following steps: a) performing principal component analysis on the three-dimensional point cloud data of each associated body surface area, extracting the principal component vector corresponding to the maximum eigenvalue to represent the point cloud distribution of each area; b) determining a weight coefficient based on the contribution of respiratory motion of the lung lobe where the tumor target is located; c) performing linear summation on the corresponding weight coefficients and the principal component vectors of the highly correlated body surface areas to extract the global three-dimensional feature vector representing the chest and abdomen during respiratory motion.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the respiratory feature extraction method according to any one of claims 1 to 6.

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

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the respiratory feature extraction method according to any one of claims 1 to 6 is implemented.

Images