Computer-implemented method and apparatus for comparing images

Abstract

The invention relates to a computer-implemented method for comparing images (B, B1), comprising the following steps:-receiving images of the same subject recorded at different times; -determining structures (S) in the image and generating directed acyclic graphs on the basis of the structures (S) in the image, where each graph has a specified point (P, P1); registering the image (G1) of the at least one second image (B1) to the image (G) of the first image (B); -determining a correspondence between the points (P, P1) of the registered graph on the basis of the spatial proximity of the points (P, P1) in combination with the link structure of the graph; and registering at least a region of the image specified by the corresponding point (P, P1) from the registered map, and outputting at least the registered region of the image. The invention further relates to a device, to a control device, and to a medical system.

Description

Technical Field

[0001] The invention relates to a computer-implemented method and device for comparing images, a control device for controlling a medical system, in particular a diagnostic system or an image recording system, and a medical system. Background Art

[0002] In computed tomography angiography (CT angiography or CTA for short), CT recordings of a patient's blood vessels, such as the coronary arteries of the heart, are acquired and subsequently evaluated, typically over a period of time. In this process, not only are 3D regions typically mapped based on individual 3D points, but also compact regions are important. Such regions include, for example, lesions along the course of the vessel or areas with stents.

[0003] For specific applications, such as multi-phase CTA, correspondences between points on a segmented vascular tree are typically found from one phase to another. This can be achieved with the aid of a registration algorithm that operates on the underlying image or on a segmented anatomical structure, such as a vascular centerline tree. Registration can be either rigid or flexible.

[0004] However, known automatic registration methods based on feature points are very prone to errors and sometimes give incorrect courses of blood vessels, which is very disadvantageous in subsequent diagnosis by humans. Summary of the Invention

[0005] The object of the present invention is to provide a method and a device for comparing images, a control device for controlling a medical system, and a medical system, by means of which the above-mentioned disadvantages are avoided.

[0006] This object is achieved by a computer-implemented method according to an embodiment of the invention, an apparatus according to an embodiment of the invention, a control device according to an embodiment of the invention, and a medical-technical system according to an embodiment of the invention.

[0007] The computer-implemented method according to the present invention is used for (in particular automatically) comparing images. The method comprises the following steps:

[0008] - receiving images with the same theme (Motiv) recorded at different times,

[0009] - determining a structure in the image, and generating a directed acyclic graph based on the structure in the image, wherein each graph has a specified point,

[0010] - registering at least one map of the second image to the map of the first image,

[0011] -Determine the correspondence between points of the registered graph based on the spatial proximity of the points combined with the link structure of the graph,

[0012] - registering at least a region of the image specified by the corresponding points according to the registered map and outputting at least the registered region of the image,

[0013] - Optional: Compare regions of the image specified by corresponding points.

[0014] This method approach enables fully automated matching, which links 3D time points, particularly from 3D CCTA scans, via a suitable intermediate geometric representation. The resulting representation can, for example, encode the coronary artery topology, particularly continuously branching vessels, as a graph-theoretic tree. This enables compact association of regions, which in turn can be used to provide intuitive side-by-side views of these regions from initial and subsequent examinations, such as a curved view of a focused lesion.

[0015] The method is typically performed using medical images of blood vessels, in particular images of the coronary arteries of the heart. While the method can be applied in two dimensions, it is preferred that the images be three-dimensional, such as CT images. The method is particularly advantageous for photon counting CT (PCCT). The method is preferably used for automated image comparisons, in particular for medical follow-up examinations, as it allows for highly effective automatic detection of changes in blood vessels over time.

[0016] First, an image is received. In order to better understand the following steps, it should be assumed that this is a 3D CT image. The image may be previously recorded and / or available in an image database.

[0017] The images should show the same subject, i.e. the same vascular structure. By "same subject" it is meant that it can be the same subject (i.e. the corresponding vascular structure of the same patient), but can also be the same type of subject, i.e. the corresponding vascular structure in different patients. The method is not only advantageous when showing changes in the vascular structure of a patient, but is also advantageous for comparing the corresponding vascular structures of multiple patients. For a better understanding, it can be imagined that the coronary arteries of a patient are observed over a time course (across multiple examinations). A possible time course is the different cardiac phases during one cardiac cycle during a single examination.

[0018] After receiving the image, the structures in the image (i.e. essentially vessels, in particular blood vessels) are determined. Based on the determined structures, a directed acyclic graph (i.e. essentially a vascular tree) is then generated in the image. Directed graphs are known in the prior art and comprise a certain number of nodes and a certain number of edges, each of which connects pairs of nodes to one another. Here, an edge is a directed edge that can only be passed in one direction. A directed acyclic graph is, in turn, a directed graph that does not contain directed cycles. This is an important aspect of the present invention, because limiting the possible results to such a graph (in particular a tree structure) eliminates a range of potential errors. In the following, a tree structure can be envisioned as a graph for vessels.

[0019] Each graph has a plurality of specified points. The points are used to be able to compare graphs from different images with each other. In theory, nodes can be simply regarded as points, i.e. a point is always set at a branch of a structure. However, this is problematic in practice, since not every branch is always reliably identified as such in different images. It is therefore preferred to set points on the edges (if necessary, in addition to the identified nodes, the nodes preferably also represent points here). The spacing between the points is preferably less than 1 cm, particularly preferably less than 3 mm. Since most work is done with digital images having voxels, the minimum spacing between two points is preferably less than 100 voxels, in particular less than 50 voxels, or even less than 10 voxels (this also applies to the spacing in pixels in pixel-based images).

[0020] It is particularly preferred that the granularity of the points in one image, applied to one image, is also applied to the other images of the other images. Thus, the points in all images preferably have essentially the same granularity or essentially the same spacing. The term "essentially" here means "with a deviation of at most 30%, in particular at most 10%."

[0021] In a further step, the maps of two or more images are registered to one another. For this purpose, one of the images is preferably selected and the map of the other image is registered to the map of said image. This can be achieved by: registering the image (with Figure 1 The images are preferably registered with one another (either initially or before generating the image), but preferably only the images are registered with one another, which requires significantly less computational effort than a complete image registration. The basic principles of image registration are known in the art. The registration is preferably based on an elastic iterative algorithm with one or more regularization terms for the closest points, which penalize stretching / reducing the distances between successive tree points, or deformation. Preferably, the current vessel name or other attributes are additionally used as orientation for the registration.

[0022] It should be noted that after registration, the above points do not necessarily overlap each other. In practice, only in rare cases (despite the best registration) do two points accidentally end up at the same location. This is due to changes and movements of the vessels and errors in image recording.

[0023] Now, the points are still set to be connected to each other. Even if the points are not exactly in the same position, this is feasible in all cases. If, for example, points are set only at the branches of the graph, then even if the vessel moves, the points (in the ideal image) correspond to the corresponding branches of the vessel and are related to each other. In the particularly preferred case of observing the (especially additional) points on the edge, branches cannot be identified either. So, for example, it is possible to infer a correspondence when the points are close to each other. How this can be achieved accurately is described in more detail below. In essence, a "correspondence" means that the corresponding points are on the same edge or in the same node. Therefore, the correspondence is determined based on the spatial proximity of the points in combination with the link structure of the graph. The link structure corresponds to the topological structure of the links of the graph, that is, how the graph is distributed in its branches. Thereby, the special structure of the graph (for example, in the form of a tree) directly affects the correspondence.

[0024] Now, if a region of interest (ROI) is found in the image, where a point of the image is located, a point corresponding to the point can be found in another image.

[0025] The original images, or at least their regions designated by corresponding points (for comparison), are then registered with one another. To make structures, such as vessels, correspond or comparable, image registration is used. Regions are registered based on the registered image; that is, the relevant regions of the structure (e.g., vessel) are deformed to resemble the image at that location. Since only the structure is relevant for the subsequent comparison, how image regions outside of the structure are registered is largely irrelevant. Image registration is applied to the structure itself. The registered regions (or the entire image) are then output, for example, by storage or display, for subsequent comparison.

[0026] It should be noted that the registration already provides a basis for comparison, since regions that match each other can thus be compared at a glance. However, even if a visual comparison is possible, an automated comparison is preferred, which assists the assessor in his or her work.

[0027] Then, during the comparison, the region of one of the other images is determined by corresponding points to be compared with the region of the first image. Thus, if the ROI in one image is the region of a stent or a stenosis, the image will be checked to see which points are located there, the corresponding points in the other image will be determined, the corresponding points will be checked in the corresponding image, and the region in the other image will be identified as the region where the ROI is located.

[0028] The basic concept of the present invention is to segment the vascular tree along the centerline of the vessels in the 3D image in each image, i.e., at different recording times, then register the vascular trees with one another, and then establish correspondences between points on the vascular trees. After registration, a step for establishing correspondences is preferably performed. In particular, point proximity is used to establish correspondences while ensuring that the resulting linked dual-tree structure does not contain loops.

[0029] The device according to the invention is used for comparing images (in particular automatically). The device comprises the following components:

[0030] a data interface designed to receive images of the same subject recorded at different times,

[0031] a determination unit configured to determine a structure in the image and to generate a directed acyclic graph based on the structure in the image, wherein each graph has a specified point,

[0032] a registration unit designed to register at least one second image representation to the first image representation,

[0033] a correspondence unit designed to determine the correspondence between the points of the registered graphs based on the spatial proximity of the points in combination with the link structure of the graphs,

[0034] a comparison unit designed to register at least the regions of the image specified by the corresponding points according to the registered map and to output at least the registered regions of the image, and preferably also to compare the regions of the image specified by the corresponding points.

[0035] The functions of the components of the device have been described above. The device is preferably designed to perform the method according to the present invention. Regarding the "comparison unit," this can also be referred to as a "second registration unit" or a "calculation unit." These designations serve to remind you that only a specific registration of the image regions allows for a better comparison of the structures.

[0036] The control device according to the present invention is used for controlling a medical system, in particular a diagnostic system and / or an image recording system. The control device comprises the apparatus according to the present invention and / or is designed to carry out the method according to the present invention.

[0037] The medical-technical system according to the invention is preferably a diagnostic system or an image recording system and comprises a control device according to the invention.

[0038] The present invention can be implemented in particular in the form of a computer unit with suitable software. For this purpose, the computer unit can, for example, have one or more cooperating microprocessors or the like. In particular, the present invention can be implemented in the computer unit in the form of suitable software program parts. An implementation in accordance with software generally has the advantage that computer units already in use can also be retrofitted in a simple manner by means of a software or firmware update so as to operate in the manner according to the present invention. In this regard, the object is also achieved by a corresponding computer program product having a computer program that can be directly loaded into a memory device of a computer unit and that has program segments so that all steps of the method according to the present invention are carried out when the program is executed in the computer unit. In addition to the computer program, such a computer program product can optionally include additional components, such as documentation and / or additional components, as well as hardware components, such as a hardware key (dongle, etc.) for using the software.

[0039] A computer-readable medium, such as a memory stick, a hard disk or other removably or permanently installed data carrier, on which program segments of a computer program that can be read and executed by the computer unit are stored, can be transferred to a computer unit and / or stored at or in the computer unit.

[0040] Other particularly advantageous designs and improvements of the present invention are derived from the embodiments according to the present invention and the following description, wherein claims of one claim category can also be improved in a manner similar to the claims and description parts of another claim category, and in particular individual features of different embodiments or variants can also be combined to form new embodiments or variants.

[0041] According to a preferred embodiment of the present method, when determining the correspondence, the link structure of the graph is incorporated in at least two stages, which comprises at least the following steps:

[0042] a) determining a correspondence between the branches of the registered images, preferably wherein groups of points on corresponding branches are determined and a correspondence is established between the groups, and

[0043] b) Determining the correspondence between groups of points of mutually corresponding branches of the registered images, preferably wherein only the points of corresponding groups are considered for this purpose.

[0044] A preferred method of establishing a correspondence is to first link the branches to each other. A branch is a path from a remote point of a graph (e.g. a tree) toward its root. For this purpose, a distance measure between branches can be preferably calculated. Assume that x i and y j are the registered points of the two branches, then the distance can be expressed as:

[0045] Σ i，j min i (|x i -y j |)+Σ i，j min j (|x i -y i |).

[0046] The two sums of this formula are used for symmetry of the expression.Other information, such as the segment labels of the branches and their differences, can be included in this distance measure.

[0047] Preferably, incorporating the graph's link structure into the determination of correspondences includes checking whether the correspondences create loops in the resulting graph, rejecting correspondences that contain loops. Since graphs are by definition acyclic, correspondences that result in loops are likely incorrect. In this manner, basic knowledge of vascular structure enables straightforward error reduction.

[0048] Preferably, the link structure incorporated into the graph when determining correspondences includes a checkpoint hierarchy, whereby correspondences with a hierarchy that does not conform to the rules of a directed graph are rejected. This also directly incorporates basic knowledge of the vascular structure into the reduction of errors. For example, if an undirected graph is obtained, the correspondence can be discarded as a defect, and a search for new correspondences can be conducted.

[0049] Preferably, incorporating the link structure of the graphs when determining the correspondence comprises checking whether a topological structure is generated by the correspondence which does not correspond to either of the two graphs and rejecting the correspondence in this case.

[0050] Preferably, incorporating the link structure of the graph when determining the correspondence comprises matching points on corresponding branches of the graph on a path from the proximal end toward the distal end and / or on a path from the distal end toward the proximal end. In the event of a discrepancy, the correspondence is preferably rejected.

[0051] In all these alternatives, which can be used individually or in combination, the basic knowledge of the graph or the strict definition of a graph as directed and acyclic reduces possible errors. This allows associations identified as faulty to be rejected and new, error-free associations to be searched for.

[0052] To determine the correspondences between points (thus representing points in each graph, but also groups of points, such as branches), a distance matrix is ​​preferably determined and branch matchings are derived therefrom. It is preferred that spacing measures between pairs of points in the corresponding graphs be calculated and the distance matrix be calculated based on a plurality of distance measures for the pairs of points. Alternatively or additionally, it is also preferred to derive the branch matchings using the Hungarian method (also known as the Kuhn-Munkels algorithm).

[0053] The distance matrix can be calculated based on a distance measure (between points or complete graphs) by evaluating all pairwise distances. Using the Hungarian algorithm, matching between branches of a graph can be efficiently calculated. After matching the branches, correspondences between points can then be established.

[0054] In particular, the branches of the first graph can be traversed in a specific order, for example, according to decreasing importance or length of the branches. Given a given first branch, the second branch is determined by branch consistency. Using the Hungarian algorithm, the inconsistent points of the two branches can then be further matched to one another. However, it should be noted that, for example, when transitioning from the far end to the near end in the first graph (tree), the order of consistency in the second graph also proceeds from the far end to the near end.

[0055] Furthermore, adding a correspondence between two points may cause a conflict with previously added points. In this case, the correspondence should not be added.

[0056] Other conflict resolution strategies can also be advantageously applied here, such as backtracking and randomizing the processing order in the branch. Once the points of the two branches are consistent, they are continued in the next branch.

[0057] This results in a series of correspondences that can be further refined. For example, only a subset of the points of the two trees can be matched, which may result in a sparse match (only a few points). This sparse match can be made dense by adding correspondences between other points. If you go from the far end to the near end on the branch, you can match the points on the two branches again between the first and second correspondences (i.e., between corresponding pairs of points), thus creating a denser correspondence. This can also be performed, for example, with the help of the Hungarian algorithm. Here, the order of the points should be preserved in both branches.

[0058] Therefore, according to a preferred embodiment of the method, after the first determination of the assignment between the points, a further assignment is performed between points for which no assignment could be established after the first determination and / or which were subsequently inserted into the map.

[0059] Due to the correspondence, the ROI of the first image and thus the ROI at a certain moment are automatically associated with one or more regions of the other image (i.e., another moment). The region in the other image corresponds to the ROI in the first image. For example, if the ROI is a coronary artery lesion, the properties of the lesion can be associated at other moments (images) and summarized in one overview. For example, each lesion has a maximum degree of stenosis. The trend of the maximum degree of stenosis over time is important for the assessment of the disease process. Therefore, given the correspondence and the centerline vascular tree point (i.e., the point of the graph), the following steps are preferred:

[0060] 1. For a plurality of points (in particular for each point) in the ROI of the first image: determining a corresponding point in at least one further image,

[0061] 2. Determine the values ​​of a plurality of parameters for the ROI from the points in the first image, for example determine the maximum degree of stenosis,

[0062] 3. extracting the same parameter or the value of the same parameter from corresponding points of at least one other image,

[0063] 4. The values ​​are correlated, in particular over the time course of the image recording, in order to display the evolving course of events.

[0064] In addition to the example of stenosis, the method can also be used, in particular, to monitor the progression of plaque accumulation, plaque composition or susceptibility to plaque and for assessments based on PCI (Percutaneous Coronary Interaction), such as FFR (Fraktionellen Flussreserve).

[0065] Furthermore, the corresponding segments (corresponding to different moments in time) of the points in the 3D volume of the image ROI can be visualized, for example, using curved surface reconstruction (CPR). This allows for the display of additional attributes, such as the start and stop markers of the lesion, for each moment in time. These additional attributes can be mapped from the corresponding points of the lesion at a first moment in time (the image being examined) to subsequent moments in time (another image being examined). Alternatively or additionally, other lesions at the corresponding points in this subset can also have zero or more lesion areas. These markers can also be visualized at the corresponding moment in time.

[0066] The automated processing can also be used to create an overview report on changes, such as lesions, and their progression. Changes can be categorized, for example, based on the magnitude of the progression between moments. Progress can be displayed visually, for example with an equal sign, an up arrow, or a down arrow, to indicate whether something remains the same, increases, or decreases.

[0067] This automatically acquired progress information can also be used when creating cardiology reports.

[0068] Preferably, the graph runs as a line in the structure. Here, the structure is preferably a vessel in the body, in particular a blood vessel, and the directed acyclic graph is a vessel tree in the vessel. Preferably, the structure runs on the center line of the vessel in the image.

[0069] Preferably, the registration of the graphs is based on an elastic, iterative algorithm for closest points with one or more regularization terms that penalize deformation, stretching, or reduction of the spacing between successive tree points. Preferably, annotations of the segmentation and / or other information about the structure are additionally used as orientations for the registration.

[0070] For comparing regions of images (i.e. at different times), a preferred embodiment of the method comprises the following steps:

[0071] - selecting a region of interest in the first image, preferably a region with a structural defect, in particular a region with a lesion, a stent, a stenosis, a plaque or a region of percutaneous coronary intervention,

[0072] - specifying a point in the map in the first image that is in said region,

[0073] - determining corresponding points in a plurality of further images, and respectively specifying areas predetermined by the corresponding points in the plurality of further images,

[0074] - comparing the designated region with a region of interest (ROI) in the first image, preferably wherein a parameter value of the region of interest is determined and a plurality of corresponding parameter values ​​are derived from a plurality of designated regions and said parameter values ​​are compared with one another,

[0075] Optionally: the recording time of the image is determined and the course of change is determined taking into account the recording time.

[0076] In this case, it is preferred that, for the visualization of the determination results, the points are visualized by means of surface reconstruction.

[0077] According to a preferred embodiment of the method, the correspondence between the points of the registered images is additionally used to associate image points of the images outside the structure with one another, so that corresponding regions are obtained outside the structure. Preferably, the corresponding regions are compared with one another.

[0078] A preferred device is characterized in that the assignment unit is designed to first determine an assignment between the branches of the registered images and subsequently determine an assignment between points of groups of branches of the registered images that correspond to one another.

[0079] When determining the correspondence between branches of the registered graphs, preferably a distance measure is calculated between pairs of corresponding branches in the graphs, and a distance matrix is ​​calculated based on a plurality of distance measures for a plurality of pairs, and the branch matching is derived therefrom.

[0080] When determining the correspondence between groups of points of corresponding branches of the matched graph, it is preferred to recalculate the distance measures between pairs of points of corresponding groups in the graph and calculate a distance matrix based on multiple distance measures of multiple pairs, and derive the branch matching from it.

[0081] KI-based methods (KI: "Künstliche Intelligenz" Artificial Intelligence) are preferably used for the method according to the present invention. Artificial Intelligence is based on the principles of machine learning and is usually implemented using learning algorithms that are trained accordingly. The term "machine learning" is often used for machine-based learning, which also includes the principle of "deep learning."

[0082] Preferably, the components of the present invention are available as "cloud services." These cloud services process data, particularly using artificial intelligence, but can also be based on traditional algorithms or evaluated by humans in the background. Typically, a cloud service (hereinafter also referred to as the "cloud") is an IT infrastructure that provides, for example, storage space, computing power, and / or application software via a network. Communication between the user and the cloud occurs via data interfaces and / or data transmission protocols. In the current scenario, it is particularly preferred that the cloud service provide both computing power and application software.

[0083] Within the scope of a preferred method, the data obtained within the scope of the present invention is provided to a cloud service via a network. The cloud service comprises a computing system that generally does not include a user's local computer. The method can be implemented using a combination of commands in the network. The data calculated in the cloud is then sent to the user's local computer via the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The present invention will be further described in detail below based on exemplary embodiments with reference to the accompanying drawings. In the various figures, identical components are provided with identical reference numerals. The drawings are generally not to scale. The drawings show:

[0085] Figure 1 A schematic diagram of a CT system is shown with one exemplary embodiment of a control device according to the present invention for carrying out the method.

[0086] Figure 2 An image showing a vessel with a vascular tree,

[0087] Figure 3 A block diagram showing the method flow,

[0088] Figure 4 An overview of the effects of the method is shown. DETAILED DESCRIPTION

[0089] Figure 1 One embodiment of a computed tomography system (CT system) 1 is shown having a radiation detector 4 and a radiation source 5. The radiation source 5 is designed to irradiate the radiation detector 4 with radiation. The illustrated CT system 1 includes a gantry 2 having a rotor 3. The rotor 3 includes an X-ray source 5 as the radiation source 5 and a radiation detector 4 designed to detect X-ray radiation.

[0090] The rotor 3 can rotate about an axis of rotation 8. A patient 6 is supported on an examination bed 7 and can be moved along the axis of rotation 8 through the gantry 2. The head of the patient 6 is placed on a support aid L. In order to control the imaging system 1 and / or generate image data sets based on the signals detected by the radiation detector 4, a computer unit 9 is provided, which is connected to the gantry 2 via a data line D.

[0091] Typically, (raw) X-ray image data sets of the examination object 6 are recorded from multiple angular directions at each radiation energy by means of the radiation detector 4, i.e., two or more raw data sets. Subsequently, a (final) image data set can be reconstructed based on the (raw) X-ray image data sets by means of mathematical methods including, for example, filtered back projection or iterative reconstruction methods.

[0092] The computer unit 9 serves as a control device 9 for controlling the CT system 1. An input device 10 and an output device 11 are connected to the computer unit 9. The input device 10 and the output device 11 can enable, for example, user interaction or display of generated image data sets B.

[0093] The control device 9 comprises a device 12 according to the invention for comparing the images B, B1 according to the method according to the invention (see Figure 3 The device 12 includes a data interface 13 , a determination unit 14 , a registration unit 15 , a correspondence unit 16 and a comparison unit 17 .

[0094] Data interface 13 is used to receive images B and B1 of the same subject, recorded at different times by the CT system. It should be noted that device 12 is also advantageous in diagnostic systems. However, the example of CT system 1 is shown here so that the recording of image B is also considered. For example, control unit 9 registers the identity of patient 6 being examined and automatically searches for image B1 from previously performed examinations of patient 6.

[0095] The determination unit 14 is used to determine structures in the images B, B1 and to generate directed acyclic graphs G, G1 based on the vessels shown in the images B, B1 , wherein each graph G, G1 has a specified point P, P1 .

[0096] The registration unit 15 is used to register at least one map G, G1 of the second image B1 to the map of the first image B.

[0097] The correspondence unit 16 is configured to determine a correspondence between the points P, P1 of the registered graphs G, G1 based on the spatial proximity of the points P, P1 in combination with the link structure of the graphs G, G1.

[0098] The comparison unit 17 is configured to register at least the region of images B and B1 designated by corresponding points P and P1 based on the registered graphs G and G1, and to output at least the registered region of images B and B1. The comparison unit 17 is also configured to compare the regions of images B and B1 designated by corresponding points P and P1.

[0099] Figure 2 Image B shows a vessel S with a vessel tree G. The vessel tree G is a directed acyclic graph G. It can be seen that the vessel tree G extends primarily in the middle of the vessel S. The progression of the diagram is easy to follow purely visually. However, if one assumes that each of the dashed lines represents a point, an error would occur in the area of ​​the intersections on the left: the intersections are perceived as close.

[0100] In the circles above and below the intersections, potential misinterpretations are shown which might be able to be identified as corresponding points (the question marks should indicate possibility and uncertainty).

[0101] Thus, while misinterpretations about intersections would be errors that could be avoided by labeling the branches of the graph, misinterpretations of the lower parts can only be eliminated by not allowing the graph to have cycles.

[0102] Figure 3 A block diagram showing the method flow, Figure 1 A method for comparing images is shown.

[0103] In step I, 3D images B, B1 having the same subject recorded at different times are received.

[0104] In step II, blood vessels S are determined as structures S in images B, B1 and directed acyclic graphs G, G1 are generated based on the structures in images B, B1. Each graph G, G1 has a specified point P, P1.

[0105] In step III, the graph G1 of the at least one second image B1 is registered to the graph G of the first image B. Here, the corresponding branches A of the graphs G, G1 are shown as a representative example.

[0106] In step V, the correspondence between the points P, P1 of the registered graphs G, G1 is determined based on the spatial proximity of the points P, P1 in combination with the logical structure of the graphs G, G1.

[0107] In step VI, at least the areas of images B and B1 specified by corresponding points P and P1 are aligned according to the aligned graphs G and G1, and at least the aligned areas of images B and B1 are output, and the areas of images B and B1 specified by corresponding points P and P1 are compared.

[0108] Figure 4 An overview of the effectiveness of the method is shown. Two images B and B1 are shown above, each showing the same vessel S. On the left, it can be seen that vessel S has a stenosis, while on the right, it does not. The two images B and B1 are now registered with one another, thereby also registering the two images G and G1. This is done to more clearly show the area of ​​the stenosis. In practice, it is simpler to directly register the two images G and G1 with one another.

[0109] The search for corresponding points P, P1 is shown below. The points P, P1 are relatively close to each other, and the closest points are respectively associated as corresponding to each other. Now, corresponding areas of the images B, B1 can be selected and compared (bottommost diagram).

[0110] Finally, it is pointed out once again that the invention described in detail above is only an embodiment, which can be modified in completely different ways by a person skilled in the art without departing from the scope of the invention. In addition, the use of the indefinite article "a" or "an" does not exclude that the features involved can also exist in multiple forms. Similarly, the term "unit" does not exclude that the components involved are composed of multiple co-acting subcomponents, which can also be distributed spatially if necessary. The term "plurality" should be interpreted as "at least one". Regardless of the grammatical gender of the specific term, people with male or female gender identity are included.

Claims

1. A computer-implemented method for comparing images (B, B1), the method comprising the steps of: - receiving images (B, B1) of the same subject recorded at different times, - determining a structure (S) in the image (B, B1), and generating directed acyclic graphs (G, G1) based on the structure (S) in the image (B, B1), wherein each graph (G, G1) has a specified point (P, P1), - registering the map (G1) of at least one second image (B1) onto the map (G) of the first image (B), - determining the correspondence between the points (P, P1) of the registered graph (G, G1) based on the spatial proximity of the points (P, P1) in combination with the link structure of the graph (G, G1), - registering at least the region of the image (B, B1) specified by the corresponding point (P, P1) according to the registered map (G, G1) and outputting at least the registered region of the image (B, B1), - Optionally: comparing areas of said images (B, B1) specified by corresponding points (P, P1).

2. The method according to claim 1 , wherein the link structure of the graph (G, G1) is incorporated in at least two stages when determining the correspondence, comprising at least the following steps: a) determining a correspondence between the branches (A) of the registered graphs (G, G1), preferably wherein corresponding groups of points (P, P1) on said branches (A) are determined and a correspondence between said groups is established, and b) Determining the correspondence between groups of points (P, P1) of mutually corresponding branches (A) of the registered graph (G, G1), preferably wherein for this purpose only corresponding points (P, P1) of the groups are considered.

3. The method according to claim 1 or 2, wherein introducing the link structure of the graph (G, G1) when determining the correspondence relationship comprises: - checking whether a cycle is generated in the resulting graph (G, G1) by said correspondence, wherein correspondences with cycles are rejected, and / or - checking the hierarchy of said points (P, P1), wherein correspondences having a hierarchy that does not conform to the rules of the directed graph (G, G1) are rejected, and / or - checking whether said correspondence results in a topology that does not correspond to either of the two graphs (G, G1) and rejecting the correspondence in this case, and / or - matching points (P, P1) on corresponding branches (A) of the graph (G, G1) on the path from the proximal end towards the distal end and / or on the path from the distal end towards the proximal end with each other and preferably rejecting the correspondence in case of inconsistency.

4. The method according to claim 1 , wherein, in order to determine the correspondence between the points (P, P1), a distance matrix is ​​determined and branch matching is derived therefrom, Preferably, a distance measure is calculated between pairs of points (P, P1) in the corresponding graph (G, G1), and the distance matrix is ​​calculated based on a plurality of distance measures of the pairs of points (P, P1), and / or Preferably, the branch matching is derived using the Hungarian method.

5. A method according to any one of the above claims, wherein, after the correspondence between the points (P, P1) is first determined, another correspondence is performed between the following points (P, P1), for which no correspondence could be established after the first determination and / or these points (P, P1) are subsequently inserted into the graph (G, G1).

6. A method according to any of the above claims, wherein the graph (G, G1) is preferably extended as a line in the structure (S), preferably wherein the structure (S) is a vessel (S) in a body, and the acyclic graph (G, G1) is a vascular tree in the vessel (S), and preferably extends on a centerline of the vessel (S) in the image (B, B1).

7. A method according to any of the preceding claims, wherein the mutual registration of the graphs (G, G1) is based on an elastic iterative algorithm for the closest points (P, P1) with one or more regularization terms, which penalize deformations, stretching / reductions of the spacing between consecutive tree points (G, G1), preferably wherein additionally annotations of the segmentation and / or other information about the structure (S) are used as orientations for the registration.

8. The method according to claim 1 , wherein regions of the images are compared with one another, wherein comparing regions of the images (B, B1) comprises the following steps: - selecting a region of interest in the first image (B), preferably a region having a defect of said structure (S), in particular a region having a lesion, a stent, a stenosis, a plaque or a region of percutaneous coronary intervention, - specifying a point (P) in said area in said map in said first image (B), - determining corresponding points (P, P1) in a plurality of further images (B, B1) and respectively specifying areas in the plurality of further images (B, B1) that are predetermined by the corresponding points (P, P1), - comparing the designated region with the region of interest in the first image (B), preferably wherein a parameter value of the region of interest is determined, and a plurality of corresponding parameter values ​​are derived from a plurality of designated regions and compared with each other, - optionally: determining the recording time of the image (B, B1) and determining the course of change taking into account the recording time, Preferably, for visualization of the determination result, the point (P, P1) is visualized by means of surface reconstruction.

9. The method according to claim 1 , wherein the correspondence between the points (P, P1) of the registered images (G, G1) is additionally used to associate image points of the images (B, B1) that are outside the structure (S) with one another, so that corresponding regions are obtained outside the structure (S). Preferably, the corresponding regions are compared with each other.

10. A device (12) for comparing images (B, B1), the device (12) comprising: - a data interface (13) designed to receive images (B, B1) of the same subject recorded at different times, a determination unit (14) designed to determine a structure (S) in the image (B, B1) and to generate a directed acyclic graph (G, G1) based on the structure (S) in the image (B, B1), wherein each graph (G, G1) has a specified point (P, P1), a registration unit (15) designed to register at least one image (G, G1) of a second image (B, B1) to the image (G, G1) of the first image (B, B1), a correspondence unit (16) designed to determine the correspondence between the points (P, P1) of the registered graphs (G, G1) based on the spatial proximity of the points (P, P1) in combination with the link structure of the graphs (G, G1), - a comparison unit (17), which is designed to align at least the region of the image (B, B1) specified by the corresponding point (P, P1) based on the aligned map (G, G1) and to output at least the aligned region of the image (B, B1), and is preferably also designed to compare the region of the image (B, B1) specified by the corresponding point (P, P1).

11. The device (12) according to claim 10, wherein the corresponding unit (16) is designed to: - first determining the correspondence between the branches (A) of the registered graphs (G, G1), preferably wherein a distance measure is calculated between pairs of groups of corresponding branches (A) in said graphs, and a distance matrix is ​​calculated based on a plurality of distance measures of a plurality of pairs and the branch matching is derived therefrom, - and subsequently determining the correspondence between groups of points (P, P1) of mutually corresponding branches (A) of the aligned graph (G, G1), preferably wherein spacing measures between pairs of points (P, P1) of corresponding groups are recalculated in the graph, and a distance matrix is ​​calculated based on multiple distance measures of multiple pairs and branch matching is derived therefrom.

12. A control device (9) for controlling a medical system, in particular a diagnostic system and / or an image recording system, comprising an apparatus (12) according to claim 10 or 11 and / or being designed to carry out a method according to any one of claims 1 to 9.

13. A medical technology system, preferably a diagnostic system or an image recording system, comprising a control device (9) according to claim 12.

14. A computer program product comprising instructions which, when executed by a computer, cause the computer to perform the steps of the method according to any one of claims 1 to 9. 15 . A computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to perform the steps of the method according to claim 1 .

Images