Methods for automated segmentation of an arteriovenous malformation

Abstract

The computer-implemented method according to the invention for processing image data for the automated segmentation of an arteriovenous malformation is based on the following steps known from the prior art: - Obtaining a graph representation of a vascular system that includes all nodes and connections, - S5) Grouping nodes of the graph representation, - S6) Determining a node or aggregate node with the maximum number of directly connected nodes as the origin node, - S7) Grouping nodes directly connected to the origin node. The invention improves the segmentation determined by the preceding steps through the subsequent further steps: - S8) Merge a branch node connected to the origin node with the origin node if it satisfies the following two conditions: - the branching node is directly connected to exactly three nodes, and - the branching node is directly or indirectly connected to another branching node which is directly connected to at least four nodes, wherein the indirect connection comprises one or more nodes which are each directly connected to exactly two nodes, - Repeat the procedure starting with step S7) if nodes are grouped with the origin node in step S8).

Description

[0001] The invention relates to a computer-implemented method for the automated segmentation of an arteriovenous malformation as well as a device configured to carry out the method, a corresponding computer program product and a corresponding computer-readable storage medium.

[0002] An arteriovenous malformation (AVM), also known as an arteriovenous malformation, is a malformation of the blood vessels in which arteries are directly connected to veins. It is therefore a vascular malformation consisting of a network of connections between the blood-supplying arteries and the blood-draining veins. It is often not possible to precisely demarcate the arteries and veins in the area of ​​an AVM. Likewise, it is often not possible to precisely distinguish an AVM from the surrounding, non-malformed vascular system. Due to the optical structure of the malformation, an AVM often appears in medical images as a tangle of arterial and venous blood vessels. The center of the tangle of vessels, or AVM, is also referred to in medicine as the nidus. For ease of description, the abbreviations AVM and AVM nidus will be used below.

[0003] Medical images are based on a medical image dataset, which may include, for example, a two-dimensional image or a three-dimensional image. Furthermore, a four-dimensional image dataset may be included, which has three spatial and one temporal dimensions. Furthermore, the medical image dataset may include a plurality of individual medical images.

[0004] A medical image dataset consists of image data, for example, in the form of a two- or three-dimensional array of pixels or voxels. Such arrays of pixels or voxels can represent color, intensity, absorption, or other parameters as a function of two- or three-dimensional position. They can be obtained, for example, by appropriate processing of measurement signals from a medical imaging modality or image scanning device.

[0005] A medical image dataset can comprise a plurality of images or image slices. Each slice shows a cross-sectional view of the image volume. The slices can comprise a two-dimensional array of pixels or voxels as image data. The arrangement of the slices in the medical image dataset can be determined by the imaging modality or by any post-processing applied to the image dataset. Furthermore, slices can be artificially defined in the image volume spanned by the medical image dataset. Optionally, this can be done as a function of the image data contained in the medical image dataset to optimally prepare the medical image dataset for the subsequent diagnostic workflow.

[0006] The medical image dataset may be a radiological image dataset depicting a patient's body part. Accordingly, it may contain two- or three-dimensional image data of the patient's body part. The medical image may be representative of an image volume or a cross-section through the image volume. The patient's body part may be contained within the image volume. The depicted patient's body part will generally include a variety of anatomies and / or organs. Taking a chest image as an example, the medical image may show lung tissue, the rib cage, lymph nodes, and others.

[0007] The medical image dataset can be an angiographic image dataset. Angiographic image data is image data depicting a patient's vascular system or a patient's body part. An angiographic image dataset can contain two-, three-, or four-dimensional image data of the vascular system.

[0008] The medical image dataset may be stored in a standard image format such as the Digital Imaging and Communications in Medicine (DICOM) format and in a storage or computer storage system such as a Picture Archiving and Communication System (PACS), a Radiology Information System (RIS), and the like. Whenever DICOM is mentioned herein, it should be understood that this refers to the Digital Imaging and Communications in Medicine (DICOM) standard, for example, according to the DICOM PS3.1 2020c standard (or a later or earlier version of this standard).

[0009] A medical imaging modality corresponds to a system used to generate or produce medical image data. For example, a medical imaging modality can be a computed tomography (CT) system, a magnetic resonance imaging (MR) system, an angiography X-ray system, a C-arm X-ray system, a positron emission tomography (PET) system, or similar.

[0010] In particular, computed tomography is a widely used imaging method and uses X-ray beams that are generated and acquired by an instrument that spatially rotates around the object of imaging, i.e., the patient. This can be a so-called cone beam computed tomography (CBCT), which is based on images of a cone-shaped X-ray beam, such as those generated by a C-arm X-ray system, or a fan beam computed tomography, which is based on images of a fan-shaped X-ray beam, such as those generated by a computed tomography system. The resulting absorption data (also called raw data) is processed by computer-assisted analysis software that reconstructs detailed images of the internal structure of the patient's body parts.The image data sets produced are called CT scans, which can display multiple series of sequential images to depict the internal anatomical structures in cross-sections perpendicular to the axis of the human body.

[0011] Magnetic resonance imaging (MRI), to give another example, is an advanced medical imaging technique that uses the effect of a magnetic field on the movement of protons. In MRI machines, the detectors are antennas, and the signals are analyzed by a computer, creating detailed images of the internal structures in every part of the human body.

[0012] A computer is basically a system comprising a memory, a main memory, one or more microprocessors and one or more input / output data interfaces.

[0013] Storage can be defined as a device capable of storing data and information. This can be any type of storage medium, including, but not limited to, hard drives, solid-state drives, optical drives, magnetic drives, flash memory, and other forms of non-volatile or volatile memory.

[0014] Random Access Memory (RAM) is a temporary storage unit capable of temporarily storing program elements, data, and information accessed during the execution of a program or process. This can be any type of RAM, including, but not limited to, DRAM, SRAM, SDRAM, and other forms of volatile memory.

[0015] A microprocessor can be defined as a unit capable of executing instructions and managing processes. This can include any type of processor, including but not limited to single-core processors, multi-core processors, graphics processors, digital signal processors, and other forms of processors or microcontrollers.

[0016] An input / output data interface (or I / O data interface) can be understood as a unit capable of transferring data and information between different components or systems. This can include any type of data transmission interface, including but not limited to USB, HDMI, VGA, DVI, Ethernet, Wi-Fi, Bluetooth, and other forms of wired or wireless data transmission interfaces.

[0017] A graph is a term from graph theory. It is an abstract structure that represents a set of objects along with the connections between these objects. The mathematical abstractions of the objects are called nodes. The pairwise connections between nodes are called edges. In this sense, an edge is a direct connection between two nodes. In the following description, the term "direct connection between nodes" is used interchangeably with the term "edge". A graph can be represented, for example, by representing the nodes as points and the edges as lines.

[0018] In medical imaging, segmentation is a process by which specific structures or regions within an image are identified or determined. Segmentation can be used, for example, to determine image regions that represent specific anatomical features, such as vessels, bones, or internal organs. Segmentation can be useful, for example, in a vascular examination in angiography to trace the course of blood vessels or measure vessel diameter.

[0019] The segmentation of an arteriovenous malformation (AVM), including the AVM nidus, is of significant importance for the diagnosis and treatment of AVMs. For example, the AVM nidus can disrupt normal blood flow, thus impairing tissue perfusion and oxygenation. Segmenting the AVM nidus in medical image data is challenging because AVMs have complex and irregular morphologies. Furthermore, the appearance of AVMs varies from patient to patient.

[0020] Accurate segmentation of an AVM nidus is crucial for treatment planning. It helps determine its size and location and identify the feeding arteries and draining veins. This information helps determine the most appropriate treatment strategy, such as surgery, radiosurgery, or endovascular embolization.

[0021] In particular, AVM Nidus segmentation can be used for: • Planning of the region to be treated surgically or radiosurgically (AVM nidus region) • Planning of endovascular embolization, in which the AVM nidus, supplying blood vessels and draining blood vessels must be determined in detail • 3D overlay during embolization to determine whether the entire AVM nidus has been embolized (and the associated risk of rupture)

[0022] A particular challenge in segmenting AVMs is caused by the high variability in their size, shape, and appearance. This variability, coupled with the relatively small number of cases (usually no more than 50 per clinic per year), complicates the application of machine learning algorithms.

[0023] A method for the automatic segmentation of an AVM is known from the publication Babin et al. 2018 Skeletonization method for vessel delineation of arteriovenous malformation. Computers in Biology and Medicine 93 (2018) 93-105. Babin et al first create a graph from a medical image dataset depicting a blood vessel system. The nodes of the graph are analyzed with regard to the number of their respective edges. The number of edges of a node is also referred to as the degree, i.e. a node with five edges is referred to as a fifth-degree node. The node or the node grouped from a plurality of nodes with the most edges, i.e. the node of the highest degree, is regarded as the AVM nidus. A region of the graph with 5 or more edges and a connection to the node with the most edges is regarded as the nidus region.

[0024] Furthermore, work is currently underway on methods for automatic segmentation using deep learning algorithms.

[0025] However, the accuracy of segmentation of the AVM nidus region using current methods is not sufficient for diagnostic and therapeutic purposes.

[0026] It is therefore an object of the invention to provide a computer-implemented automatic segmentation of an arteriovenous malformation (AVM) including the AVM nidus in a 3D image dataset that has improved accuracy. A further object of the invention is to provide such an automatic segmentation that can be performed efficiently and quickly using a computer.

[0027] The invention solves these problems by the method according to claim 1. The invention also solves these problems by a device for processing image data, by an imaging device, by a computer program product and by a computer-readable storage medium according to one of the independent claims.

[0028] The inventive computer-implemented image processing method for automated segmentation of an arteriovenous malformation comprises the following steps: - Obtaining a graph representation of a vascular system comprising nodes and connections, where obtaining can be understood as, for example, determining based on an image data set or receiving from a data storage or a cloud or receiving from a device for processing image data, - S5) Combining all directly connected nodes of the graph representation that are directly connected to more than two nodes into one combined node, - S6) Determining the maximum number of direct connections to other nodes that have nodes or aggregated nodes of the graph representation, and determining a node or aggregated node with the maximum number of directly connected nodes as the originating node, where the maximum number is understood to mean the maximum number of direct connections that occur at nodes of the graph representation, - S7) Combining all nodes directly connected to the source node that satisfy one of the following conditions with the source node: - the nodes are directly connected to exactly two nodes, or - the nodes are directly connected to at least four nodes.

[0029] According to the invention, the method comprises the following further steps: - S8) Merging a branch node connected to the source node with the source node if it satisfies both of the following conditions: - the branching node is directly connected to exactly three nodes, and - the branching node is directly or indirectly connected to another branching node which is directly connected to at least four nodes, whereby the indirect connection comprises one or more nodes which are each directly connected to exactly two nodes, - Repeating the process starting with step S7) if nodes are merged with the original node in step S8).

[0030] Subsequently, the graph representation determined in step S8 is provided for display by a display device and / or for carrying out further image processing steps

[0031] By combining branching nodes with the source node in step S8, additional nodes are combined with the source node, the AVM nidus, compared to the prior art. It is a finding of the invention that the condition tested in this step of the method better accounts for the anatomical characteristics of blood vessel systems with arteriovenous malformations. Combining nodes with the AVM nidus according to this step of the method therefore results in a better match of the segmentation with the actual AVM nidus, which takes the anatomical conditions into account better than is the case with methods known in the prior art.

[0032] Furthermore, the term "grouping" refers to the assignment of nodes to one another. The assignment can be made in the graph by replacing all nodes with a single node that represents the nodes assigned to one another. The assignment can also be made by viewing the nodes assigned to one another as a single area whose connections to other nodes are treated as if the area represented a single node. In any case, connections that exist between the nodes of a grouped node or between a grouped node and other nodes are treated as if the grouped node were a single node. For the sake of simplicity of description, grouped nodes will be represented by a single node in the following.

[0033] In an advantageous development of the method, in a step S9 following step S8, the nodes of an end component connected to the source node are combined with the source node. An end component comprises an end node that is directly connected to exactly one other node. An end component can additionally comprise one or more nodes that are directly connected to exactly two other nodes.

[0034] Components are local groups of nodes and edges within a bounded area of ​​the graph representation. Components are characterized by the fact that the nodes and edges that comprise them share a common property.

[0035] By combining unconnected end components with the source node, additional nodes are added to the AVM nidus in the graph compared to the state of the art. End components can arise, for example, when further connections are not visible in the 3D image dataset underlying the graph, are not visible correctly, or are only visible with low contrast. Accordingly, end components can arise, for example, when a vessel is very small and therefore cannot be resolved and displayed in the image, or when the image data lacks contrast, or when a stenosis narrows a vessel, or when a vessel actually ends.

[0036] It is a finding of the invention that the condition tested in this step of the method better accounts for the anatomical characteristics of blood vessel systems with arteriovenous malformations. Combining end components with the AVM nidus according to this step of the method therefore results in a better match of the segmentation with the actual AVM nidus, which better accounts for the anatomical conditions than is the case with methods known in the prior art.

[0037] In an advantageous development of the method, in a step S10 following step S8, nodes of a large component connected to the source node are combined with the source node. A connected large component consists of a plurality of interconnected nodes. The number of nodes within the large component that are directly connected to exactly three nodes must be fewer than the number of nidus connections of the large component to the source node plus two, and the nodes of the large component may only be directly connected to the source node and to other nodes of the large component.

[0038] Based on the large component condition, which is checked in step S10 of the method, local clusters are grouped together with the AVM nidus if, according to the condition, they have a sufficient number of connections to the AVM nidus relative to their three-way directly connected nodes. If a sufficient number of conditions for the AVM nidus exist, such components are not assigned an independent significance separate from the AVM nidus.

[0039] It is a finding of the invention that the condition examined in this step of the method better accounts for the anatomical characteristics of blood vessel systems with arteriovenous malformations. Combining connected large components with the AVM nidus according to the invention therefore results in a better match of the segmentation with the actual AVM nidus, which better accounts for the anatomical conditions than is the case with methods known in the prior art.

[0040] An advantageous development of the method relates to the case where a plurality of nodes or combined nodes has the maximum number of directly connected nodes. In this case, several of this plurality of nodes or combined nodes are selected as source nodes in step S6. One or more of the subsequent method steps are then performed separately for each of the selected source nodes.

[0041] If multiple nodes or clustered nodes with the maximum number of directly connected nodes are identified, it is not possible to clearly determine in advance which node is the source node, i.e., which node corresponds to the AVM nidus. By performing the procedure separately for each or at least some of the potential source nodes, a separate segmentation is automatically performed for each of the selected source nodes. The subsequent procedural steps could then either result in a common AVM nidus graph or in several different segmentations, which are then output to the user of the procedure.

[0042] In an advantageous development of the method, depending on the result of the previously executed method steps, one or more nodes or combined nodes with a number of directly connected nodes that is less than the maximum number previously determined in step S6 are selected as source nodes. Subsequently, one or more of the method steps, beginning with step S7, are performed separately for each of these selected source nodes.

[0043] According to this advantageous development of the method, for example, if the node(s) or combined node(s) with the highest number of connected nodes are not located in the AVM nidus as a result of the segmentation method, the method can be repeated with nodes or combined nodes with fewer direct connections. The result of the previously executed method steps, which gives rise to this new selection of source nodes, is therefore that the previously selected source node(s) were not located in the AVM nidus. Accordingly, incorrect results of the previously executed method steps can also give rise to this new selection of source nodes. For example, a user of the method could make a user input depending on the result, which then results in the new selection of source nodes.It is essential that if the selection of source nodes according to the original criterion of the highest degree leads to an erroneous result of the procedure, a new selection of source nodes of lower degree is made and the further procedure is repeated.

[0044] In an advantageous development of the method, the graph representation comprising nodes and connections is determined by the steps preceding step S5: - S1) Receiving a 3D image data set, - S2) Determining a representation of a vascular system in the 3D image data set, - S3) Determine a line representation of the vascular system, - S4) Determining a graph representation of the line representation comprising nodes and connections between the nodes.

[0045] The 3D image dataset can be any type of angiographic image dataset. For example, the 3D image dataset can be generated by a computed tomography system or a C-arm X-ray system. It can be, for example, a fan-beam computed tomography or a cone-beam computed tomography.

[0046] The representation of the vascular system can be obtained, for example, using digital subtraction angiography (DSA). In DSA, image data in which blood vessels do not contain contrast medium is subtracted from image data in which blood vessels do contain contrast medium. The result of the subtraction is image data that exclusively depicts blood vessels. The representation of the vascular system can also be obtained using another type of image data processing, for example, by segmenting an angiographic magnetic resonance imaging scan. Numerous methods are known for obtaining a representation of the vascular system.

[0047] The determination of a line representation of the vascular system can also be referred to as skeletonization. This involves converting the planar representation of a blood vessel in 2D image data or the spatially extended representation of a blood vessel in 3D image data into a line representation. The line representation therefore represents the course of a blood vessel, but not necessarily its spatial extent or diameter. The only essential thing is that the lines selected lie within the planar or spatial representation of a blood vessel. Lines are often selected in the middle of the planar or spatial representation of the blood vessels. This type of representation is called a centerline representation. Numerous methods are known for determining a line representation of a vascular system.

[0048] The determination of a graph representation comprising connections and nodes is essentially achieved by representing intersection points in the line representation of the vascular system as nodes and lines as connections, with the connections preferably being represented as straight lines. Numerous methods are known for determining a graph representation.

[0049] In an advantageous development of the method, in a subsequent step S11, those voxels of the 3D image dataset are determined that correspond to the nodes and origin nodes identified in the graph representation. For this purpose, the nearest point in the line representation of the vascular system is determined for each identified node or origin node.

[0050] The assignment of the AVM and AVM nidus segmented in the graph representation to the original 3D image dataset is necessary so that the segmentation is spatially understandable for the viewer and spatially assignable to the vascular system depicted in the 3D image dataset. In this way, the segmentation is usable for diagnosis and therapy and understandable for spatial orientation and navigation. Since a graph representation fundamentally represents a logical representation of the vascular system, it does not necessarily reflect a spatially correct assignment of nodes and connections. However, the graph representation is based on the line representation of the vascular system, as it was determined from the latter. Thus, an assignment of nodes and connections in the graph representation to intersection points and lines in the line representation is given from the outset.Exploiting the existing mapping to assign segmented features from the graph representation to the line representation requires little effort and avoids mapping errors.

[0051] In one or more subsequent steps, the segmented features of the line representation of the vascular system are then assigned to the original representation of the vascular system or to the original 3D image dataset. Since the respective assignment is given in advance, the assignment of nodes and connections can also be carried out with little effort and while avoiding assignment errors. In particular, the assignment of segmented features of the graph representation can be carried out first to the line representation and only then to an original representation of the vascular system with less effort and error than a direct assignment of features of the graph representation to the original representation of the vascular system or to the original 3D image dataset.

[0052] A further advantageous development of the invention consists in a device for processing image data comprising a computer unit configured to execute a method according to one of the preceding claims. The computer unit can be a microprocessor, a multi-core microprocessor (multi-core CPU), a distributed processor arrangement, a correspondingly configured graphics processor unit, a network resource (cloud computing), or similar conventional devices.

[0053] A further advantageous development of the invention consists in an imaging device comprising an imaging modality and a device for processing image data, designed as described above, wherein the imaging modality is designed to provide the image data. The imaging modality can be a computed tomography system, a C-arm X-ray system, another X-ray system, a magnetic resonance imaging system, or another conventional imaging modality for generating angiographic image data.

[0054] A further advantageous development of the invention consists in a computer program product comprising program elements that, when loaded into a working memory of a computer unit of a device for processing image data, cause the computer unit to execute the method steps explained above. The computer program product can comprise program instructions in source code, in compiled code (object code), or in other conventional programming languages.

[0055] A further advantageous development of the invention consists in a computer-readable storage medium on which program elements are stored that are suitable for being read and executed by a computer unit and that, when loaded into a computer unit's main memory, cause the computer unit to execute the method steps explained above. The storage medium can be a hard drive, a floppy disk or CD, a solid-state storage device (SSD), a USB storage device, a network storage device (cloud storage), or another conventional data storage device.

[0056] Further embodiments and advantages emerge from the dependent patent claims and from the following description of embodiments with reference to figures.

[0057] The figures show: Fig. 1 imaging device (state of the art) Fig. 2 Graph representation (state of the art) Fig. 3 Flow diagram (state of the art) Fig. 4 nodes with three connected nodes Fig. 5 Short open ends Fig. 6 Large areas Fig. 7 Flow diagram

[0058] Fig. 1 schematically shows an imaging device suitable for implementing the invention. The imaging device comprises an imaging modality. The imaging modality is a C-arm X-ray system 1. The C-arm X-ray system 1 is shown merely as an example. It is only essential that the imaging modality used is suitable for generating angiography image data. Thus, for example, a computed tomography system or a magnetic resonance imaging system could also be used to implement the invention.

[0059] The imaging device further comprises a device for processing image data 3. The device for processing image data 3 receives image data from the imaging modality and processes it. The image data is processed by the computer unit 4. For this purpose, the computer unit 4 is connected to a working memory 2. Program instructions can be loaded into the working memory 2, which cause the computer unit 4 to execute the method steps encoded by the program instructions. Depending on the type of imaging modality, the processing of the image data can include preprocessing of image data, three-dimensional image reconstruction, denoising, and other post-processing steps. The only essential thing is that the processing of the angiographic image data includes the generation of a representation of blood vessels, for example, using a segmentation method or a DSA method.

[0060] Fig. Figure 2 schematically shows the processing of angiographic image data for the automatic segmentation of an AVM or an AVM nidus according to the prior art. The starting point is the line representation 9 of blood vessels of an AVM shown on the left of the figure. The lines are displayed in a very coarse image grid to facilitate the transfer to the graph representation 10 shown in the center, which uses the same coarse image grid. The line representation 9 shown can be obtained using any method known in the prior art.

[0061] The graph representation 10 shown in the center is generated from the line representation 9 shown on the left. The same coarse grid is used as in the line representation 9 shown on the left. In the graph representation 10 shown in the center, nodes 5 are represented as points and edges or connections 6 as lines. Nodes 5 can have one or more neighbors, i.e., one or more nodes 5 directly connected to them. Approximately in the center of the grid, a dashed-bordered area 7 with a large number of nodes 5 and connections 6 can be seen. Due to its arrangement and interconnection, area 7 appears to the observer to be the origin or starting point of the graph representation 10.

[0062] Shown on the right is a segmented graph representation 11 as a result of automatic segmentation of the AVM nidus. The automatic segmentation according to the state of the art first determines nodes 19 that are directly connected to more than two nodes 5. Such multiple directly connected nodes 19 are grouped with other multiple directly connected nodes 19 to which they are directly connected. In the center of the graph representation 10, an area 7 is shown with a dashed border, in which a plurality of such multiple directly connected nodes 19 are located. These nodes 19 are grouped into a single node 8. The node 20 adjacent to this area has only two direct connections 6 to other nodes 5. The doubly directly connected node 20 is not grouped with the multiple directly connected nodes 19 in this step.It initially remains as a doubly directly connected node and is only added to the AVM nidus in a later step, namely in step S7 explained below. Due to its arrangement and interconnection, node 8 appears to the observer as the origin or starting point of graph representation 10.

[0063] In principle, one or more such aggregated nodes can occur in a graph representation 10 of a blood vessel system. The node or aggregated node with the highest number of directly connected nodes, i.e., the node or aggregated node with the highest degree, is designated as the original node 8.

[0064] Fig. 3 shows the automatic AVM segmentation according to the prior art as a schematic flow diagram. In step S1, a 3D image data set is received from a device for processing image data 3. In step S2, a representation of a vascular system is determined in the 3D image data set, for example, using a segmentation method or a DSA method. In step S3, a line representation 9 of the vascular system is determined. In step S4, a graph representation 10 comprising nodes 5 and connections 6 of the line representation 9 is determined. In step S5, multiply connected nodes 19 of the graph representation 10 are combined into one or more combined nodes as described above. In step S6, the maximum number of direct connections to other nodes 5 that have nodes 5 or combined nodes 8 of the graph representation 10 is determined.Then, a node 5 or aggregate node 8 with the maximum number of directly connected nodes 5 is determined as the source node 8. In step S7, all nodes 5 connected to the source node 8 are determined that are either directly connected to two other nodes 5 or directly connected to more than a minimum number of nodes 5, where the minimum number may be higher than three and is preferably four.

[0065] Fig. Figure 4 schematically illustrates method steps of the inventive computer-implemented method for automatically segmenting an AVM or AVM nidus. The inventive segmentation can be based on the result of a segmentation according to the prior art. It is irrelevant whether the starting point was determined using the previously described method or another method known in the prior art. The only essential requirement is that a segmented graph representation 11 of the AVM nidus with a segmented source node 8 is available.

[0066] According to the invention, in step S8, isolated branching nodes 12 are identified in the segmented graph representation 21, which are defined by two conditions. First, they have three nodes directly connected to them. Second, they have a neighboring branching node 13, which in turn is directly connected to at least four nodes. Such a neighboring branching node 13 can be directly connected to the isolated branching node 12. It can also be indirectly connected to the isolated branching node 12 via one or more further nodes, which in turn are each directly connected to exactly two nodes. In other words, the isolated branching node 12 can be indirectly connected to the branching node 13 via one or more non-branched nodes.

[0067] If one or more isolated branching nodes 12 are identified according to the criteria described above, these are combined with the original node 8.

[0068] If, as described above, one or more isolated branch nodes 12 have been combined with the source node 8, the method is then executed again, beginning with step S7.

[0069] Fig. Figure 5 schematically illustrates further method steps of a computer-implemented method for automatically segmenting an AVM or AVM nidus according to an advantageous development of the invention. After step S8 is completed, according to the development, in step S9, 22 regions are identified as open ends 15 in the graph representation if they fulfill two conditions. First, they comprise an end node 14 that is directly connected to only one other node. Second, they comprise either no node or exclusively those nodes that are themselves directly connected to only two nodes.

[0070] If one or more open ends 15 are identified as described above, the nodes 14 and possibly further nodes of these open ends 15 are combined with the original node 8.

[0071] Fig. 6 schematically explains further method steps of a computer-implemented method for automatically segmenting an AVM or AVM nidus according to an advantageous development of the invention. According to the development, according to step S10 in the graph representation 23 connected regions are identified as large components 16 if they satisfy the following two conditions. First, the number of nodes 17 of the large component 16 that are directly connected to exactly three other nodes 5 of the large component 16 must be less than the number of nidus connections 18, increased by two, via which nodes 5 of the large component 16 are connected to the AVM nidus. Second, the nodes 5 of the large component 16 may only be directly connected to the original node 8 or to other nodes 5 of the large component 16.

[0072] If one or more large connected areas 16 are identified as described above, they are added to the source node 8 and combined with it.

[0073] Fig. Figure 7 shows the automatic AVM segmentation according to the invention as a schematic flow diagram. The method according to the invention starts with a graph representation 10 of the AVM nidus. In the previously described step S7, connected nodes 5 are determined and combined with the source node 8. In step S8, branch nodes 12 connected to the source node 8 are searched for. If such branch nodes 12 are found, they are combined with the source node 8, and the method is repeated starting from step S7.

[0074] If no such branching nodes 12 are found, the method continues with step S9. In step S9, short open ends 15 are determined and, if necessary, combined with the source node 8.

[0075] Next, in step S10, large connected regions 16 connected to the source node 8 are searched for and, if necessary, combined with the source node 8. Regarding step S10, it is only important that it is performed after the preceding step S8. However, step S10 does not necessarily have to be performed after step S9. The order of execution of steps S10 and S9 can be chosen arbitrarily.

[0076] In order to be able to display or use the automatically determined AVM nidus segmentation in the original underlying 3D image dataset, an assignment of those voxels of the 3D image dataset that correspond to the nodes 5 determined in the graph representation 10 and the original node 8 is carried out in step S11. The determined nodes from the graph representation 10 are assigned to the line representation 9 in the sense of a mapping. Through this assignment, the nodes 5, 8 are mapped from the logical coordinate system of the graph representation 10 into the spatial coordinate system of the line representation 9 and thus assigned to the corresponding spatial coordinates. For each node 5, 8 determined in the graph representation 10, the nearest spatial point in the line representation 9 is determined.

[0077] General remark: In the present patent application documents, regardless of the chosen grammatical gender of a particular term, persons with male, female or other gender identity are to be included equally.

Claims

[1] Computer-implemented method for processing image data for automated segmentation of an arteriovenous malformation comprising the steps: - Obtaining a graph representation (10) of a vascular system comprising nodes (5) and connections (6), - S5) combining directly connected nodes (19) of the graph representation (10) which are directly connected to more than two nodes (5, 19) into a combined node (8), - S6) Determining a maximum number of direct connections to other nodes (5) that have nodes (5) or aggregated nodes (8) of the graph representation (10), and determining a node (5) or aggregated node (8) having the maximum number of directly connected nodes (5) as the originating node (8), - S7) Combining nodes (5) directly connected to the source node (8) that satisfy one of the following conditions with the source node (8): - the nodes (5) are directly connected to exactly two nodes (5), or - the nodes (5) are directly connected to at least four nodes (5), characterized by the subsequent steps: - S8) combining a branch node (12) connected to the source node (8) with the source node (8) if the branch node (12) satisfies both of the following conditions: - the branching node (12) is directly connected to exactly three nodes (5), and - the branching node (12) is directly or indirectly connected to a further branching node (13) which is directly connected to at least four nodes (5), wherein the indirect connection comprises one or more nodes (5) which are each directly connected to exactly two nodes (5), - Repeating the process starting with step S7) if in step S8) nodes are combined with the original node (8), - Providing the graph representation determined in step S8) for display by a display device and / or for executing further image processing steps. [2] Method according to claim 1, characterized by the further step following step S8): - S9) combining nodes (5, 14) of an end component (15) connected to the originating node (8) with the originating node (8), wherein an end component (15) comprises an end node (14) which is directly connected to exactly one further node (5) and additionally comprises no node or only those nodes (5) which are directly connected to exactly two further nodes (5). [3] Method according to one of the preceding claims, characterized by the further step following step S8): - S10) combining nodes (5) of a large component (16) connected to the originating node (8) with the originating node (8), wherein a large component (16) consists of interconnected nodes (5), and wherein the number of nodes (5) within the large component (16) that are directly connected to exactly three nodes (5) is less than the number of nidus connections (18) of the large component (16) to the originating node (8) plus two, and wherein the nodes (5) of the large component (16) are directly connected exclusively to the originating node (8) and to other nodes (5) of the large component (16). [4] Method according to one of the preceding claims, wherein, if a plurality of nodes (5) or combined nodes (8) has the maximum number of nodes (5) directly connected thereto, in step S6) several of this plurality of nodes (5) or combined nodes (8) are selected as source nodes (8), and one or more of the subsequent method steps are carried out separately for each of the selected source nodes (8). [5] Method according to one of the preceding claims, wherein, depending on a result of the previously carried out method steps, one or more nodes (5) or combined nodes (8) with a number of directly connected nodes (5) which is less than the maximum number are selected as source nodes (8), and one or more of the method steps starting with step S7) are carried out separately for each of the selected source nodes (8). [6] Method according to one of the preceding claims, characterized by the following steps preceding step S5): - S1) Receiving a 3D image data set, - S2) Determining a representation of a vascular system in the 3D image data set, - S3) Determining a line representation (9) of the vascular system based on the representation of the vascular system, - S4) Determining a graph representation (10) of the line representation comprising nodes (5) and connections (6) between the nodes (5). [7] Method according to claim 6, characterized by following step: - S11) Identifying those voxels of the 3D image data set that correspond to the nodes (5) and origin nodes (8) determined in the graph representation (10), wherein for each determined node (5) and origin node (8) a nearest point in the line representation (10) of the vascular system is determined. [8] Device for processing image data (3) comprising a computer unit (4) which is arranged to carry out a method according to one of the preceding claims. [9] Imaging device comprising an imaging modality and a device for processing image data (3) according to claim 8, wherein the imaging modality is designed to provide the image data (3). [10] Computer program product comprising program elements which, when loaded into a working memory (2) of a computer unit (4) of a device for processing image data (3), cause the computer unit (4) to carry out the steps of the method according to one of claims 1 to 7. [11] Computer-readable storage medium on which program elements are stored which are suitable for being read and executed by a computer unit (4) and which, when loaded into a working memory of a computer unit (4), cause the computer unit (4) to carry out the steps of the method according to one of claims 1 to 7.

Images