Image processing apparatus and image processing method

Abstract

An image processing apparatus and an image processing method, the image processing apparatus including: a volume data acquisition unit that acquires volume data of an object; a volume data segmentation unit that segments the volume data to obtain base tubular data; a fine tubular data acquisition unit that acquires fine tubular data from the volume data; a tubular data generation unit that generates tubular data based on the fine tubular data and the base tubular data; and a data output unit that outputs the tubular data.

Description

Technical Field

[0001] This invention relates to an image processing apparatus and an image processing method capable of accurately segmenting tubular objects. Background Technology

[0002] In the field of medical imaging, accurate segmentation and separation of blood vessels are of great value in the diagnosis and treatment of vascular diseases. For example, in 3D-guided intersegmental nodule resection surgery, detecting subsegments and lower-level vessels helps to better determine the boundaries of subsegmental resection.

[0003] In the existing technology, there are a variety of methods for segmenting blood vessels.

[0004] Traditional blood vessel segmentation methods often use thresholding or multi-scale tubular detection to segment blood vessels, or combine methods such as tracking to track blood vessels along the centerline.

[0005] Moreover, there are deep learning-related algorithms that introduce clDice (Centerline Dice) to improve detection connectivity, and use two networks, segmentation and separation, together to ensure the connectivity of arteriovenous separation through post-processing techniques such as graph-cut (graph segmentation algorithm).

[0006] In addition, there is a CNN (Convolutional Neural Networks) + tracking method. In this method, the initial segmentation is performed by CNN, and then the blood vessel is tracked based on the assumptions of the initial segmentation centerline and path direction to achieve the detection of distal blood vessels.

[0007] However, in the aforementioned existing vascular segmentation methods, for distal small blood vessels, due to their lower contrast, smaller structure, and distance from the root of the vessel, as well as the lack of information about the entire vessel, it is difficult to accurately segment distal small blood vessels to achieve arteriovenous separation.

[0008] Specifically, in traditional blood vessel segmentation methods, it is difficult to remove interference from lesions and other areas with similar grayscale values ​​to blood vessels when performing multi-scale blood vessel detection. Moreover, due to the complex and varied shapes of blood vessels and the grayscale changes from the root to the tip, traditional algorithms are difficult to apply.

[0009] While deep learning algorithms have made some improvements in ensuring the connectivity of blood vessels, they cannot significantly address the issues of low probability of small blood vessels at the distal end of the network's output probability map and the easy confusion between arteries and veins. Furthermore, it is difficult to create a complete vascular ground truth (GT).

[0010] The CNN+tracking method is suitable for situations with fewer vascular branches, thicker vessels, and obvious vascular contrast. However, it is difficult to handle situations with more vascular branches in the distal part of the vessel and different directions of vascular branches.

[0011] As mentioned above, the existing vascular segmentation methods have problems such as loss of small vascular branches, absence of distal vessels, and incorrect classification and disconnection of distal arteries and veins.

[0012] Moreover, similar problems exist in the segmentation of other tubular structures such as the trachea, as in the segmentation of blood vessels. Summary of the Invention

[0013] Therefore, in view of the above, the present invention provides an image processing apparatus and an image processing method capable of accurately detecting small tubular objects and thus accurately segmenting tubular objects.

[0014] The image processing apparatus of the present invention comprises: a volume data acquisition unit for acquiring volume data of a subject; a volume data segmentation unit for segmenting the volume data to obtain basic tubular data; a small tubular data acquisition unit for acquiring small tubular data from the volume data; a tubular data generation unit for generating tubular data based on the small tubular data and the basic tubular data; and a data output unit for outputting the tubular data.

[0015] Alternatively, the tubular data generation unit may perform connectivity analysis on the small tubular data and the basic tubular data, and connect the small tubular data that meets the connectivity conditions to the basic tubular data.

[0016] Alternatively, the tubular data generation unit may perform connectivity analysis on the small tubular data and the basic tubular data to obtain partial small tubular data that meets the connectivity conditions and is included in the small tubular data, and connect the obtained partial small tubular data with the basic tubular data to generate updated basic tubular data.

[0017] Alternatively, the tubular data generation unit may further perform connectivity analysis on the remaining small tubular data after removing the aforementioned portion of the small tubular data from the aforementioned small tubular data and the aforementioned updated basic tubular data, further obtaining a second portion of small tubular data within the remaining small tubular data that meets the connectivity conditions, and connecting the obtained second portion of small tubular data with the aforementioned updated basic tubular data to generate further updated basic tubular data, thereby generating the aforementioned tubular data.

[0018] It can also be configured to further include a small tubular data filtering unit, which removes the portion common to the basic tubular data from the small tubular data obtained by the small tubular data acquisition unit, and removes tubular data outside the threshold range, thereby obtaining filtered small tubular data.

[0019] Alternatively, the aforementioned small tubular object data acquisition unit may be configured to acquire the small tubular object data using a small-scale tubular object detection operator based on the structure and grayscale features of the small tubular objects.

[0020] Alternatively, it can be configured such that, in the above-mentioned small-scale tubular object detection operator, the eigenvalue enhancement of the Hessian matrix of the small-scale tubular object is calculated.

[0021] Alternatively, the volume data segmentation unit can segment the volume data using a deep learning initial segmentation network to obtain basic tubular data, and generate a foreground probability map for the tubular data using the deep learning initial segmentation network. The small tubular data acquisition unit can then threshold the foreground probability map and acquire the regions in the foreground probability map that fall within a specified threshold range as small tubular data.

[0022] Alternatively, the configuration may involve the user selecting tubular branch data of a tubular structure from the aforementioned basic tubular data, and the tubular data generation unit automatically displaying multiple small tubular data adjacent to the tubular branch data. The user then selects the small tubular data to be connected from the multiple small tubular data, and the tubular data generation unit connects the small tubular data with the tubular branch data to generate tubular data.

[0023] Alternatively, the aforementioned tubular data could be configured as vascular data.

[0024] Alternatively, the aforementioned vascular data can be configured to include both venous and arterial vascular data.

[0025] The image processing method of the present invention comprises: a volume data acquisition step, acquiring volume data of a subject; a volume data segmentation step, segmenting the volume data to obtain basic tubular data; a fine tubular data acquisition step, acquiring fine tubular data from the volume data; a tubular data generation step, generating tubular data based on the fine tubular data and the basic tubular data; and a data output step, outputting the tubular data.

[0026] Invention Effects

[0027] The image processing apparatus and image processing method according to the present invention can accurately segment tubular objects. Attached Figure Description

[0028] Figure 1 This is a structural block diagram of an image processing apparatus according to an embodiment of the present invention.

[0029] Figure 2 (a) is the basic venous data obtained by segmenting the lung body data. Figure 2 (b) is the basic arterial data obtained by segmenting the lung body data.

[0030] Figure 3 The diagram shows the relationship between points in the shape of a structure in three-dimensional volume data and their corresponding eigenvalues ​​of the Hessian matrix.

[0031] Figure 4 (a) shows fine vascular data obtained from lung body data. Figure 4 (b) shows from Figure 4 The small vessel data in (a) after removing the portion common to the baseline arteriovenous data.

[0032] Figure 5 This is a diagram illustrating the process of connecting small blood vessel data with basic arteriovenous data.

[0033] Figure 6 (a) shows the basic arterial data for the lungs. Figure 6 (b) shows data on an artery connected to multiple small blood vessels.

[0034] Figure 7 (a) shows basic venous data of the lungs. Figure 7 (b) shows data on veins connected to multiple small blood vessels.

[0035] Figure 8 This is a flowchart of an image processing method according to an embodiment of the present invention.

[0036] Figure 9 It is a graph synthesized from basic arterial data and basic venous data.

[0037] Figure 10 This is a schematic diagram illustrating the foreground probability map of arteriovenous data.

[0038] Figure 11 This is a flowchart of an image processing method according to a variation of an embodiment of the present invention, Example 1.

[0039] Figure 12 This is a flowchart of an image processing method according to a modified example 2 of the embodiments of the present invention. Detailed Implementation

[0040] The image processing apparatus and image processing method of the present invention will now be described with reference to the accompanying drawings.

[0041] The image processing apparatus of the present invention comprises multiple functional modules. It can be installed as software in a standalone computer or other device with a CPU (central processing unit) and memory, or it can be distributed across multiple devices, with a processor executing each functional module of the image processing apparatus stored in memory. Alternatively, it can be implemented in hardware as a circuit capable of executing the various functions of the image processing apparatus. The circuit implementing the image processing apparatus can transmit and receive data or acquire data via a network such as the Internet.

[0042] Furthermore, the image processing device of the present invention can be installed at the site of medical image acquisition to perform image segmentation processing on-site. Alternatively, the image processing device can also be directly installed within a medical image acquisition device, such as a CT scanner or an MRI scanner.

[0043] In the following description, a medical image captured with pulmonary blood vessels as the target is used as an example to illustrate the segmentation process of the image processing device on the medical image. However, the present invention is not limited to this, and can also be applied to the segmentation of other tubular structures such as trachea present in medical images.

[0044] Below, refer to Figures 1 to 8 The embodiments of the present invention will be described.

[0045] Figure 1 This is a structural block diagram of the image processing apparatus of the present invention.

[0046] like Figure 1 As shown, the image processing device 10 includes a volume data acquisition unit 11, a volume data segmentation unit 12, a small tubular object data acquisition unit 13, a small tubular object data filtering unit 14, a tubular object data generation unit 15, and a data output unit 16.

[0047] Body data acquisition unit 11 acquires body data of the subject.

[0048] Before examining or treating a patient, a three-dimensional scan is usually performed to obtain clear volumetric data of the examination site with a good anatomical environment, in order to fully understand the condition of the site. This volumetric data can be any type of volumetric data, such as three-dimensional CT (computed tomography) volumetric data or three-dimensional MR (magnetic resonance) volumetric data.

[0049] The examination site can be an organ such as the lungs, heart, or liver, or a body part such as the chest, which includes the lungs.

[0050] In this embodiment, the lungs will be used as an example for explanation.

[0051] The volume data segmentation unit 12 segments the volume data to obtain basic tubular data.

[0052] In this embodiment, arteries and veins are used as examples to illustrate the tubular structure. Figure 2 As shown, the body data segmentation unit 12 performs initial segmentation on the body data related to the lungs to obtain basic venous data (see reference). Figure 2 (a) and baseline arterial data (refer to) Figure 2 (b)

[0053] Basic arterial data pertains to the larger arteries in the lungs, but lacks data on smaller arteries. Similarly, basic venous data pertains to the larger veins in the lungs, but lacks data on smaller veins.

[0054] The initial segmentation method can be any method, such as a deep learning-based initial segmentation network or a traditional segmentation method based on thresholding. The key is that the segmentation method can obtain the basic arterial and venous data of the lungs.

[0055] The small tubular data acquisition unit 13 acquires small tubular data from the volume data.

[0056] In this embodiment, the small tubular data acquisition unit 13 acquires small blood vessel data of the lungs and distal blood vessel data from the body data of the lungs. Here, "distal" means the end located far from the basic large blood vessels, and can also be called "terminal". Typically, distal blood vessels are also extremely small blood vessels.

[0057] In this embodiment, the small tubular data acquisition unit 13 uses a small-scale tubular detection operator to detect and acquire small blood vessel data in the volume data of the lungs based on the structure and grayscale features of small blood vessels.

[0058] In the small-scale tubular object detection operator, the feature enhancement value of the Hessian matrix of the small-scale tubular objects is calculated. The specific detection method for the small blood vessel data obtained by the small tubular object data acquisition unit 13 is described below.

[0059] The Hessian matrix is ​​a second-order partial derivative matrix of a multidimensional variable function, which can be used to detect vascular structures based on the properties of its eigenvalues.

[0060] Figure 3 The diagram shows the relationship between points in the shape of a structure in three-dimensional volume data and their corresponding eigenvalues ​​of the Hessian matrix.

[0061] Each point in the body data of the lungs has such Figure 3 The three feature values ​​λ1, λ2, and λ3 shown are used to calculate the tubular feature enhancement value for each point.

[0062] As a method for calculating the feature enhancement value of tubular structures, for example, the known Frangi filtering algorithm described in the following formula can be used.

[0063]

[0064]

[0065]

[0066] In the above formula, v F It is the feature enhancement value of the tubular structure, and α and k are hyperparameters whose values ​​can be changed.

[0067] In the calculation of the above formula, by setting the hyperparameters α and k to appropriate values, the feature enhancement value of the small tubular structure can be calculated.

[0068] In this embodiment, the volume data of the lung, namely the feature enhancement value of each point in the small blood vessel structure in the lung region, is calculated according to the above formula, and a small blood vessel enhancement map is obtained that highlights the structure of small blood vessels and suppresses the structure of other parts.

[0069] Next, for the basic arterial data and basic vein data obtained by the volume data segmentation unit 12, the image grayscale distribution range (I) is statistically analyzed. lower I higher For small blood vessels located in distal areas, a reference threshold is set, for example, the reference threshold is set to I. lower +σ, where σ is a constant value that can be positive or negative.

[0070] By retaining small blood vessel structures in the lung region whose feature enhancement values ​​are higher than a reference threshold and removing those whose feature enhancement values ​​are lower than the reference threshold, a vascular probability map of small blood vessels in the lung region can be obtained. In other words, a vascular probability map of small blood vessels can be derived based on the feature enhancement values ​​of small blood vessel structures and the grayscale distribution range of baseline arterial and venous data.

[0071] As described above, a method for filtering small blood vessel structures is shown by setting a reference threshold for small blood vessels based on the grayscale distribution range of basic arterial and basic venous data. However, various methods can be used for setting the reference threshold and filtering small blood vessel structures, and the method described above is not the only one.

[0072] Next, adaptive threshold segmentation is performed on the vascular probability map to obtain the volume data of the lungs, namely the small vascular region within the lungs.

[0073] Specifically, for example, a gray-level histogram of the vascular probability map can be calculated, and then the Otsu method can be used to perform threshold segmentation on the vascular probability map to obtain the small vascular region in the lung.

[0074] The above example illustrates a method for adaptive threshold segmentation of vascular probability maps, but other methods can also be used for adaptive threshold segmentation of vascular probability maps.

[0075] The small tubular data filtering unit 14 removes the portion common to the basic tubular data from the small tubular data obtained by the small tubular data acquisition unit 13, and removes tubular data outside the threshold range, thereby obtaining filtered small tubular data.

[0076] Figure 4 (a) shows the small blood vessel data obtained from the body data of the lungs by the small tubular data acquisition unit 13. The small tubular data filtering unit 14 removes the portion of the small blood vessel data that is common (intersecting) with the basic arterial data and basic vein data, thus obtaining the small blood vessel connected regions of the small blood vessel data other than the basic arterial data and basic vein data (see reference). Figure 4 (b)

[0077] For example, the removal method can be to superimpose the small blood vessel data obtained by the small tubular data acquisition unit 13 with the basic arterial data and basic vein data obtained by the volume data segmentation unit, and remove the common (intersection) part of the two data from the small blood vessel data.

[0078] Then, for the small blood vessel data after removing the common parts mentioned above, the size of the connected region of each blood vessel is calculated, and blood vessel data whose connected region size is outside the preset threshold range is removed from the small blood vessel data. In other words, blood vessel data that are too large or too small are removed from the small blood vessel data.

[0079] In addition, during the screening of small blood vessel data, the shape characteristics of each vessel can be calculated to remove non-tubular data from the small blood vessel data. This is because, through various processing steps performed on the small blood vessel data, some data in the small blood vessel data may no longer be tubular data, and the initially obtained small blood vessel data may also include non-vascular shapes.

[0080] Through the above processing, data on small blood vessels were obtained.

[0081] The tubular data generation unit 15 generates tubular data based on small tubular data and basic tubular data.

[0082] Specifically, the tubular data generation unit 15 performs connectivity analysis on the small tubular data and the basic tubular data, and connects the small tubular data that meets the connectivity conditions to the basic tubular data.

[0083] The tubular data generation unit 15 performs connectivity analysis on the small tubular data and the basic tubular data, obtains partial small tubular data that meets the connectivity conditions and is included in the small tubular data, and connects the obtained partial small tubular data with the basic tubular data to generate updated basic tubular data.

[0084] The tubular data generation unit 15 further repeatedly performs connectivity analysis on the remaining small tubular data after removing some small tubular data from the small tubular data and the updated basic tubular data, and further obtains a second part of small tubular data that meets the connectivity conditions within the remaining small tubular data. The obtained second part of small tubular data is then connected with the updated basic tubular data to generate further updated basic tubular data, thereby generating tubular data.

[0085] Below, refer to Figure 5 The processing procedure of the tubular data generation unit 15 is explained.

[0086] First, the tubular data generation unit 15 performs connectivity analysis on the small blood vessel data (hereinafter, sometimes referred to as small blood vessels) and the basic artery data and basic vein data (hereinafter, sometimes referred to as basic artery and basic vein) obtained from the small tubular data filtering unit 14.

[0087] Connectivity analysis can be performed, for example, in the following manner.

[0088] The connection points between small blood vessels and basic arteries and veins are defined. These connection points can be, for example, the terminals of the basic arteries and veins, the main trunks of the basic arteries and veins, or the middle of the vessels. When a small blood vessel is located at one of these connection points, it is determined that the small blood vessel is connected to the basic artery or vein.

[0089] The method involves determining whether the direction and angle of a small blood vessel are approximately the same as those of the underlying arteries and veins. For example, when a small blood vessel connects to a underlying artery, if both the small blood vessel and the underlying artery run downwards, it is determined that the small blood vessel and the underlying artery have the same direction and angle, thus indicating that the small blood vessel is connected to the underlying artery. Conversely, if the small blood vessel runs downwards while the underlying artery runs to the left, it is determined that the small blood vessel and the underlying artery have different direction and angle, thus indicating that the small blood vessel is not connected to the underlying artery.

[0090] The system determines whether the ratio of the diameter of a small blood vessel to that of a basic artery and vein exceeds a predetermined threshold. For example, when a small blood vessel is connected to a basic artery, the diameters of the small blood vessel and the basic artery at the connection point are calculated, and then the ratio of their diameters is calculated. This ratio is compared with a pre-set threshold. If the ratio is greater than the threshold, meaning the difference between the diameters of the small blood vessel and the basic artery is too large, it is determined that the small blood vessel and the basic artery are not connected. If the ratio is less than the threshold, it is determined that the small blood vessel and the basic artery are connected.

[0091] Furthermore, the connection location, direction of travel, and diameter ratio of the blood vessels can be considered comprehensively to determine whether all three conditions are met. If all three conditions are met, the small blood vessel is determined to be connected to the basic artery or vein. This effectively removes data on small blood vessels with a low probability of connection.

[0092] Furthermore, when a small blood vessel is connected to both a basic artery and a basic vein, connectivity is determined based on the directional angles of the small blood vessel and the basic artery and vein. For example, if the directional angle of the small blood vessel is closer to that of the basic vein than that of the basic artery, then the small blood vessel is considered to be connected to the basic vein.

[0093] The above illustrates several methods for analyzing vascular connectivity. When determining vascular connectivity, one or more methods can be used. Furthermore, it is not limited to the methods described above; other methods can also be used.

[0094] Next, when it is determined that a small blood vessel can connect with a basic artery or basic vein (meeting the connection condition), the small blood vessel is incorporated into the basic artery or basic vein, that is, the small blood vessel is connected and linked to the basic artery or basic vein to form a new basic artery or basic vein. At the same time, the small blood vessel data is removed from the filtered small blood vessel data to update the set of filtered small blood vessel data.

[0095] Next, it is determined whether the basic arterial data and basic venous data (hereinafter sometimes referred to as basic arteriovenous data) have been updated. As mentioned above, when small vessels are incorporated into the basic arterial or basic venous data, it is determined that the basic arteriovenous data has been updated. Then, connectivity analysis is performed on the remaining small vessel data after removing those already connected to the basic arterial or basic venous data from the screened small vessel data, and compared with the updated basic arteriovenous data. When it is determined that a small vessel can be connected to the updated basic arterial or basic venous data (meeting the connectivity condition), the small vessel is connected and linked to the updated basic arterial or basic venous data, forming a further updated basic arterial or basic venous data. At the same time, the small vessel data is removed from the screened small vessel data to update the set of screened small vessel data. The update of the screened small vessel data and the basic arteriovenous data are repeated in the above manner until no more small vessel data meeting the connectivity condition is generated. At this point, it is determined that the basic arteriovenous data has not been updated, and the processing ends.

[0096] Therefore, in response to such Figure 6 The baseline arterial data shown in (a) generates the following: Figure 6 (b) shows arterial data connected to multiple small blood vessels, and, for example... Figure 7 The baseline venous data shown in (a) is used to generate data such as... Figure 7 (b) shows the vein data connected to multiple small blood vessels.

[0097] In the above embodiment, the example described is that the tubular data generation unit 15 performs connectivity analysis on the small blood vessel data, basic artery data, and basic vein data obtained by the small tubular data screening unit 14. However, it can also be configured such that the tubular data generation unit 15 performs connectivity analysis on the small blood vessel data, basic artery data, and basic vein data obtained by the small tubular data acquisition unit 13.

[0098] Data output unit 16 outputs tubular data.

[0099] In this embodiment, specifically, the data output unit 16 outputs arterial data and venous data generated by the tubular data generation unit 15.

[0100] Below, refer to Figure 8 The image processing method of this embodiment will be described.

[0101] In the image processing method, we will take the formation of arterial and venous data of the lungs as an example for explanation.

[0102] In step S1, input any volume data such as three-dimensional CT volume data or three-dimensional MR volume data of the lungs of the subject to obtain the volume data of the lungs of the subject.

[0103] In step S2, the lung body data is segmented using any volume data segmentation method to obtain basic arterial data and basic venous data.

[0104] In step S3, based on the structure and grayscale features of small blood vessels, a small-scale tubular detection operator is used to obtain small blood vessel data from the lung volume data.

[0105] In step S4, the small blood vessel data is filtered by removing the portion of the small blood vessel data that is common to the basic arteriovenous data, and removing non-tubular data and tubular data that are too large or too small and are outside the threshold range.

[0106] In step S5, arteriovenous data is generated based on the selected small vessel data and the basic arteriovenous data. Specifically, it is analyzed whether the selected small vessel data can be connected to the basic arteriovenous data, and the connected small vessel data are connected one by one to the basic arterial data or basic venous data, thereby generating arteriovenous data.

[0107] In step S6, the arterial data and vein data generated in step S5 are output respectively.

[0108] Additionally, while small vessel data is filtered in step S4, this step can be omitted. As long as the basic arterial and venous data obtained in step S2 and the small vessel data obtained in step S3 can be used to generate arterial and venous data, other methods can also be employed.

[0109] In this embodiment, when segmenting the body data into blood vessels (tubular structures), it is possible to accurately detect small tubular structures of the subject, such as small blood vessel branches in the lungs, and at the same time, it is possible to reliably determine the connectivity between the small blood vessel branches and the basic arteries and veins. Thus, it is possible to detect small blood vessels that were lost in the initial blood vessel segmentation and to avoid the situation of incorrect classification of arteries and veins at the ends of blood vessels.

[0110] Variation Example 1

[0111] Below, refer to Figures 9-11 A variation of the above-described embodiment will be described.

[0112] In Variation 1, the volume data segmentation unit 12 segments the volume data obtained by the volume data acquisition unit 11 through a deep learning initial segmentation network to obtain basic tubular data, and generates a foreground probability map for the tubular data through the deep learning initial segmentation network.

[0113] The small tubular object data acquisition unit 13 thresholds the foreground probability map and acquires the area in the foreground probability map that is within the specified threshold range as small tubular object data.

[0114] Specifically, we will use the arterial and venous data that form the lungs as an example for explanation.

[0115] Body data segmentation unit 12 obtains basic venous data by segmenting lung body data through a deep learning initial segmentation network (see reference). Figure 2 (a) and baseline arterial data (refer to) Figure 2 (b) is used to generate foreground probability maps for arterial data and for vein data respectively through a deep learning initial segmentation network. Then, these two foreground probability maps are merged to obtain the foreground probability map for arteriovenous data (see [reference]). Figure 10 ).

[0116] When merging the foreground probability map for arterial data and the foreground probability map for venous data, the baseline arterial data and baseline venous data are synthesized to obtain, as shown below. Figure 9 The synthesized data shown can be used to obtain... Figure 10 The image shows a foreground probability plot for arteriovenous data. Figure 10 The diagram only schematically shows the axial plot of the foreground probability plot, but the foreground probability plot also has sagittal and coronal plots.

[0117] Next, the data acquisition unit 13 for small tubular objects... Figure 10 The foreground probability map shown is thresholded; for example, the threshold for the region indicated by symbol A is set to 1, and the threshold for the region indicated by symbol B is set to 0.6. The small tubular data acquisition unit 13 will... Figure 10 The foreground targets (regions) in the threshold range of 0.6 to 1 shown in the foreground probability map are obtained as small blood vessel data.

[0118] Below, refer to Figure 11 The image processing method of Modified Example 1 will be explained.

[0119] In the image processing method of Modified Example 1, only steps S2' and S3' are different from the image processing method of the above embodiment, while the rest of the steps are the same. Therefore, only steps S2' and S3' will be described.

[0120] In step S2', basic arterial and venous data are obtained by using a deep learning initial segmentation network. At the same time, a foreground probability map for the arterial data and a foreground probability map for the venous data are generated by the deep learning initial segmentation network.

[0121] In step S3', the two foreground probability maps generated in step S2' are merged to generate a foreground probability map for arterial and venous data. This foreground probability map is then thresholded, and foreground targets (regions) within the specified threshold range in the foreground probability map are obtained as small blood vessel data. Thus, small blood vessel data are obtained based on this foreground probability map.

[0122] In Variation Example 1, when segmenting blood vessels (tubular structures) in volume data using a deep learning initial segmentation network, it can accurately detect more and more connected small blood vessels, addressing the problem that small blood vessels are difficult to segment due to their weak contrast with the background. This avoids the loss of small blood vessels and the misclassification of arteries and veins at the ends of blood vessels.

[0123] Variation Example 2

[0124] Below, refer to Figure 12 A variation of the above-described embodiment will be described.

[0125] In Modification 2, the user selects tubular branch data of a tubular structure from the basic tubular data. The tubular data generation unit 15 automatically displays multiple small tubular data adjacent to the tubular branch data. The user selects the small tubular data to be connected from the multiple small tubular data. The tubular data generation unit 15 connects the small tubular data with the tubular branch data to generate tubular data.

[0126] Specifically, we will use the arterial and venous data that form the lungs as an example for explanation.

[0127] After detecting small blood vessel data in the lungs, when the user selects one or more vascular branches from the basic arterial and venous data in the lungs, the tubular data generation unit 15 automatically displays multiple small blood vessels adjacent to the selected vascular branch. Based on their experience, the user determines which small blood vessels should be connected to the vascular branch, and the tubular data generation unit 15 connects the small blood vessels identified by the user with the corresponding arterial or venous branches to generate arterial or venous data.

[0128] Below, refer to Figure 12 The image processing method for variation example 2 will be explained.

[0129] In the image processing method of Modified Example 2, only steps S4” and S5” are different from the image processing method of the above embodiment or Modified Example 1. The remaining steps are the same. Therefore, only steps S4” and S5” will be described.

[0130] In step S4”, the user selects a vascular branch in the basic arteriovenous data, which automatically displays multiple small blood vessels adjacent to the vascular branch.

[0131] In step S5", the user selects the small blood vessel to be connected from multiple small blood vessels based on their experience, and connects the small blood vessel selected by the user with the corresponding arterial or venous branch to generate arterial or venous data.

[0132] In addition, in Modification 2, step 4 of the above-described embodiment and Modification 1, namely the screening step of small blood vessel data, can be added between step 3 and step S4”.

[0133] In variation example 2, when segmenting the volume data into blood vessels (tubular structures), a complete blood vessel GT can be quickly generated, thereby enabling more accurate detection of small blood vessels and avoiding the loss of small blood vessels and the misclassification of arteries and veins at the ends of blood vessels.

[0134] The constituent elements of each device in the above-described embodiments are functional concepts and do not necessarily need to be physically configured as shown in the illustrations. That is, the specific form of the distributed / integrated arrangement of each device is not limited to the content shown in the illustrations, and all or part of them can be configured in any unit, functionally or physically, according to various loads and usage conditions. Furthermore, all or any part of the processing functions performed in each device can be implemented by a CPU and the program parsed and executed by the CPU, or it can be implemented as hardware based on wiring logic.

[0135] Furthermore, the image processing apparatus described in the above embodiments can be implemented by executing a pre-prepared program by a computer such as a personal computer or workstation. This program can be distributed via a network such as the Internet. Additionally, the program can be recorded on a computer-readable non-volatile recording medium such as a hard disk, floppy disk (FD), CD-ROM, MO, or DVD, and then read from the recording medium by a computer for execution.

[0136] As described above, although embodiments and variations thereof of the present invention have been explained, these embodiments are shown as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other ways, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and variations thereof are included in the scope and spirit of the invention, and are included in the invention described in the technical solution and its equivalents.

Claims

1. An image processing apparatus, characterized in that, have: The body data acquisition unit acquires the body data of the subject being examined. The volume data segmentation unit segments the aforementioned volume data to obtain basic tubular data; The small tubular object data acquisition unit acquires small tubular object data from the aforementioned volume data; The small tubular object data filtering unit obtains filtered small tubular object data by superimposing the small tubular object data obtained by the small tubular object data acquisition unit and the basic tubular object data, removing the common parts of the two data and removing tubular object data outside the threshold range, and uses this as the small tubular object data. The tubular data generation unit generates tubular data based on the aforementioned small tubular data and the aforementioned basic tubular data; as well as The data output unit outputs the aforementioned tubular data.

2. The image processing apparatus as claimed in claim 1, characterized in that, The aforementioned tubular data generation unit performs connectivity analysis on the aforementioned small tubular data and the aforementioned basic tubular data, and connects the small tubular data that meets the connectivity conditions to the aforementioned basic tubular data.

3. The image processing apparatus as described in claim 2, characterized in that, The aforementioned tubular data generation unit performs connectivity analysis on the aforementioned small tubular data and the aforementioned basic tubular data, obtains partial small tubular data that meets the connectivity conditions and is included in the aforementioned small tubular data, and connects the obtained partial small tubular data with the aforementioned basic tubular data to generate updated basic tubular data.

4. The image processing apparatus as described in claim 3, characterized in that, The aforementioned tubular data generation unit further repeatedly performs connectivity analysis on the remaining small tubular data after removing the aforementioned portion of small tubular data from the aforementioned small tubular data and the aforementioned updated basic tubular data, and further obtains a second portion of small tubular data within the aforementioned remaining small tubular data that meets the connectivity conditions. The obtained second portion of small tubular data is then connected with the aforementioned updated basic tubular data to generate further updated basic tubular data, thereby generating the aforementioned tubular data.

5. The image processing apparatus as claimed in claim 1, characterized in that, The aforementioned small tubular object data acquisition unit uses a small-scale tubular object detection operator to acquire the aforementioned small tubular object data based on the structure and grayscale features of the small tubular objects.

6. The image processing apparatus as claimed in claim 5, characterized in that, In the small-scale tubular object detection operator described above, the eigenvalue enhancement of the Hessian matrix of the small-scale tubular object is calculated.

7. The image processing apparatus as claimed in claim 1, characterized in that, The aforementioned volume data segmentation unit segments the volume data using a deep learning initial segmentation network to obtain basic tubular data, and then generates a foreground probability map for the tubular data using the same deep learning initial segmentation network. The aforementioned small tubular object data acquisition unit thresholds the aforementioned foreground probability map, and acquires the regions in the aforementioned foreground probability map that fall within the specified threshold range as small tubular object data.

8. The image processing apparatus as claimed in claim 1, characterized in that, By having the user select tubular branch data of a tubular structure from the aforementioned basic tubular data, the tubular data generation unit automatically displays multiple small tubular data adjacent to the tubular branch data. The user then selects the small tubular data to be connected from these multiple small tubular data, and the tubular data generation unit connects the small tubular data with the tubular branch data to generate tubular data.

9. The image processing apparatus as claimed in claim 1, characterized in that, The above tubular data represents vascular data.

10. The image processing apparatus as claimed in claim 9, characterized in that, The above vascular data includes venous vascular data and arterial vascular data.

11. An image processing method, characterized in that, have: The steps for acquiring body data are as follows: Obtain body data from the subject of the examination. The volume data segmentation step involves segmenting the aforementioned volume data to obtain basic tubular data; The step of obtaining data on small tubular structures involves obtaining data on small tubular structures from the aforementioned volume data; The small tubular data screening step involves obtaining the small tubular data obtained by superimposing the small tubular data obtained in the above-mentioned small tubular data acquisition step with the above-mentioned basic tubular data, removing the common parts of the two data and removing tubular data outside the threshold range, and using the screened small tubular data as the above-mentioned small tubular data. The tubular data generation step generates tubular data based on the aforementioned small tubular data and the aforementioned basic tubular data; as well as The data output step outputs the data for the tubular structure described above.

12. The image processing method as described in claim 11, characterized in that, In the above-mentioned tubular data generation step, connectivity analysis is performed on the above-mentioned small tubular data and the above-mentioned basic tubular data, and the small tubular data that meet the connectivity conditions are connected to the above-mentioned basic tubular data.

13. The image processing method as described in claim 12, characterized in that, In the above-mentioned tubular data generation step, connectivity analysis is performed on the above-mentioned small tubular data and the above-mentioned basic tubular data to obtain partial small tubular data that meets the connectivity conditions and is included in the above-mentioned small tubular data. The obtained partial small tubular data is then connected with the above-mentioned basic tubular data to generate updated basic tubular data.

14. The image processing method as described in claim 13, characterized in that, In the above-mentioned tubular data generation step, the remaining small tubular data after removing the aforementioned portion of small tubular data from the aforementioned small tubular data is further repeatedly subjected to connectivity analysis with the aforementioned updated basic tubular data. The second portion of small tubular data that meets the connectivity conditions within the aforementioned remaining small tubular data is further obtained. The obtained second portion of small tubular data is then connected with the aforementioned updated basic tubular data to generate further updated basic tubular data, thereby generating the aforementioned tubular data.

15. The image processing method as described in claim 11, characterized in that, In the above steps for acquiring small tubular data, a small-scale tubular detection operator is used to acquire the small tubular data based on the structure and grayscale features of the small tubular objects.

16. The image processing method as described in claim 15, characterized in that, In the small-scale tubular object detection operator described above, the eigenvalue enhancement of the Hessian matrix of the small-scale tubular object is calculated.

17. The image processing method as described in claim 11, characterized in that, In the volume data segmentation step described above, the volume data is segmented using a deep learning initial segmentation network to obtain basic tubular data, and a foreground probability map for the tubular data is generated using the deep learning initial segmentation network. In the above-mentioned step of acquiring small tubular data, the foreground probability map is thresholded, and the region in the foreground probability map that is within the specified threshold range is used as small tubular data.

18. The image processing method as described in claim 11, characterized in that, By having the user select tubular branch data of a tubular structure from the aforementioned basic tubular data, multiple small tubular data adjacent to the tubular branch data are automatically displayed during the tubular data generation step. The user then selects the small tubular data to be connected from these multiple small tubular data. During the tubular data generation step, this small tubular data is connected to the tubular branch data to generate tubular data.

19. The image processing method as described in claim 11, characterized in that, The above tubular data represents vascular data.

20. The image processing method as described in claim 19, characterized in that, The above vascular data includes venous vascular data and arterial vascular data.

Images