Computer-implemented method for evaluating a three-dimensional angiography data set, evaluation system, computer program

Abstract

A computer-implemented method, an evaluation system, a computer program and an electronically readable memory medium for evaluating a three-dimensional angiography data set. Computer-implemented method for evaluating a three-dimensional angiography data set (1), in particular a computer tomography angiography data set, of a vascular tree, in particular a coronary artery tree, of a patient, wherein based on a comparison of angiography information (2) of the angiography data set (1) with reference information (3) describing at least one of a plurality of categories of anatomical variations, variation information (5) is determined which describes at least one anatomical variation category belonging to at least one of the plurality of anatomical variation categories which is relevant for an anatomical variation of the vascular tree.

Description

Technical Field

[0001] This invention relates to a computer-implemented method for evaluating a three-dimensional angiographic dataset of a patient's vascular tree, particularly a coronary artery tree, and the angiographic dataset, particularly a computed tomography (CT) angiography dataset. The invention further relates to an evaluation system, a computer program, and an electronically readable storage medium. Background Technology

[0002] Medical imaging is a commonly used tool for a variety of medical tasks, particularly for diagnosis and / or treatment planning and detection. In particular, angiography describes techniques that specifically utilize contrast agents to obtain angiographic datasets of a patient's blood vessels, specifically at least one vascular tree. Two main areas of application for angiography include coronary angiography and cerebral angiography, such as in the case of stroke.

[0003] Regarding coronary angiography, coronary artery disease (CAD) is a leading cause of death worldwide. CAD is associated with dynamic atherosclerotic plaques that narrow blood vessels. These plaques can, for example, develop due to inflammation or turbulence in blood flow. Recent studies have shown promising results for identifying culprit lesions using a non-invasive diagnostic procedure called coronary computed tomography angiography (CCTA), which may require invasive procedures and / or pharmacological treatment. In particular, deep learning methods have been proposed, for example, to evaluate coronary computed tomography angiography datasets using separate evaluation algorithms to detect lesions in the heart or coronary artery tree. Even, for example, cardiac region-based angiography datasets have been proposed to simulate hemodynamics to measure hemodynamic parameters such as fractional flow reserve (FFR).

[0004] Normally, the physiological function of the coronary arteries is to perfuse the myocardium, particularly the myocardial layer. The coronary tree comprises the three main branches of the coronary arteries that surround the entire heart: the right coronary artery (RCA), the left anterior descending artery (LAD), and the circumflex branch or circumflex branch (LCX) of the left coronary artery. Schemes for labeling segments within the coronary tree have been proposed by several associations and organizations.

[0005] While definitions of the anatomical structure of the main branches exist, several non-pathological variations of this structure can be present in patients. These variations include different branches acting as major contributors to perfusion in certain cardiac regions, the presence of additional branches between the LAD and LCX, a U-shaped loop at the beginning of the RCA (the so-called shepherd's crook), high output at the coronary ostium, and acute output from the LCX. Similar variations are also found in the vascular tree of the brain.

[0006] Information regarding the presence of such normal (i.e., non-pathological) anatomical variations can be beneficial for several diagnostic and / or planning tasks. Although these normal anatomical variations are non-pathological, they can be relevant to patient risk stratification and management decisions. Additionally, such anatomical variations can complicate minimally invasive interventions. Therefore, these anatomical variations should be included in the report. On the other hand, because anatomical variations can alter the course of blood flow, they can influence the results of assessment algorithms, particularly calculations related to deep learning-based methods and / or hemodynamic FFR.

[0007] Today, the existence of anatomical variants is qualitatively assessed by human readers of anatomical datasets. For individual organs, such as the heart, automated methods for detecting anatomical variants are known. For example, in R. Hannah et al.'s article, "A Hybrid Method for Automatic Anatomical Variant Detection and Segmentation," FIMH2011: Functional Imaging and Modeling of the Heart, pp. 333–340, to address organ variability in topology, they proposed creating a set of organ model variants and selecting the most suitable model variant for the patient at hand. Summary of the Invention

[0008] Therefore, the object of the present invention is to provide a robust method for automatically assessing the presence of anatomical variations in a vascular tree.

[0009] This objective is achieved by providing a computer-implemented method, evaluation system, computer program, and electronically readable storage medium according to the independent claims. Advantageous embodiments are described in the dependent claims.

[0010] In the initially described method, variant information is determined by comparing angiographic information from an angiography dataset with reference information describing at least one of several anatomical variant categories that are associated with anatomical variants of the vascular tree.

[0011] Therefore, this invention proposes determining the presence of variations based on comparison with reference information, wherein, in a preferred embodiment, this comparison is based on the structural features of the vascular tree. That is, the structure of the vascular tree can be analyzed in a first preparation step. Specifically, at least one structural piece of information about the vascular tree can be determined as angiographic information by at least one structural evaluation algorithm. This angiographic information is then compared with reference information, which also includes or relates to such structural features. Therefore, in this invention, it is preferred that the structure, particularly specific structural features such as the centerline and / or lumen of a vessel, be identified as carrying all the relevant necessary information to determine the presence of anatomical variations, as will be shown in detail below.

[0012] In this way, the presence of anatomical variations can be determined fully automatically, allowing for a simple and robust implementation. Beyond pathology and patient-specific anatomy, the detection and labeling of anatomical variations introduces new, universal semantic concepts, which can be used for computer-aided detection (CADe). In particular, physicians treating patients are provided with additional clinically relevant information that can be considered in diagnosis and / or, especially, for example, in treatment planning for minimally invasive interventions.

[0013] The general method of the present invention is applicable to any vascular tree and any imaging modality; however, the preferred application area is computed tomography angiography results relating to coronary artery trees or vascular trees in the brain.

[0014] In a generally applicable advantageous embodiment of the invention, in the first step, a comparison indicating the general presence of anatomical variants is performed, wherein if no anatomical variant is detected in the comparison, the anatomical variant category associated with the absence of the anatomical variant is determined as variant information; otherwise, in the second step, belonging to at least one other anatomical variant category is determined by at least one further comparison. That is, normal anatomical structures can be defined, for example, by selecting the option with the highest prevalence in patients for each possible type of anatomical variant. Such normal anatomical structures have already been defined in the art and can form the basis for the first comparison. If the first comparison indicates no deviation pointing to the presence of anatomical variants different from the normal anatomical structure, the result can already be used to determine variant information, thus eliminating the need for further comparison. Therefore, computational effort and time can be saved in many cases. In embodiments, further comparisons regarding individual anatomical variant categories or groups of anatomical variant categories can be specifically implemented, wherein the comparison information of the first comparison in the first step is evaluated to select which anatomical variant categories are further compared. Of course, it is also possible to perform a comparison of structures that are further hierarchically layered in multiple second steps, for example, to perform a comparison for a group of anatomical variant categories, wherein a comparison of members in a group is performed only if the comparison of that group returns that there is at least one anatomical variant in the group.

[0015] The following describes in detail two specific and advantageous embodiments of the inventive concept that can be used cumulatively, i.e., different methods of the two specific methods can be used for different categories of anatomical variants and / or different comparative stratification levels, and / or both specific methods can be applied for at least one category of anatomical variant, for example to allow for rationality checks and / or further improvements in robustness.

[0016] In a first method particularly advantageously suited for the assessment of coronary artery trees, such as the Coronary Computed Tomography Angiography Dataset (CTTA Dataset), the centerlines of at least a portion of the vessels in the vascular tree are determined as structural information to form a centerline tree, wherein at least one set of rules for at least one category of anatomical variants is provided as reference information, each set of rules including at least one condition for at least one feature of the centerline tree, wherein in comparison, the condition of at least one set of rules in at least one set of rules is applied, and the anatomical variant category associated with the set of rules is determined when all conditions of the set of rules associated with the anatomical variant category are satisfied by the centerline tree.

[0017] In the course of this invention, it should be understood that, regarding vascular trees, particularly coronary artery trees, the centerline, while a simple and easily manipulated geometric construct, encompasses all information related to typical normal anatomical variant categories. Additionally, conditions can be formulated for the centerline of the centerline tree, indicating the presence of a particular anatomical variant when all conditions are met. In a preferred embodiment, additional anatomical features derived from an anatomical structure dataset can also be considered in the rule set; that is, in a preferred embodiment, at least one condition evaluates anatomical structure information describing at least one additional anatomical feature that is not part of the vascular tree, wherein this anatomical structure information is derived from an angiography dataset and / or an additional image dataset registered with the angiography dataset. For example, to determine which vessel primarily supplies a predefined region, the region can be segmented and the distance from the centerline to the region can be compared, inferring that the vessel closest to the centerline to the region is the vessel primarily supplying that region. In other examples, the relative positions with more additional anatomical features can be evaluated.

[0018] In a general example, at least one condition may be to compare at least one angle at a bifurcation (or multiple bifurcations) with a threshold angle, and / or at least one curvature in at least one vascular segment with a threshold curvature, and / or at least one number of multiple bifurcations in at least one vascular segment with a reference number, and / or the distance between centerline features, and / or at least one additional anatomical feature described by anatomical information derived from an angiography dataset and / or an additional image dataset, and / or to check the presence of at least one type of multiple bifurcation, particularly three bifurcations.

[0019] It should be noted at this point that segmentation labeling algorithms can typically be applied to centerline trees. Such segmentation labeling algorithms are known from existing technologies and can be applied to locate segments requested by conditions for comparison. As an example, see A. Fischer et al., “Deep Learning Based Automated Coronary Labeling For Structured Reporting Of Coronary CT Angiography In Accordance With SCCT Guidelines,” Journal of Cardiovascular Computed Tomography 14.3 (2020), pp. 21–22. Regarding centerline extraction algorithms, see Y. Zheng et al., “Robust and Accurate Coronary Artery Centerline Extraction in CTA by Combining Model-Driven and Data-Driven Approaches,” International Conference on Computational Medical Imaging and Computer-Aided Interventions, pp. 74–81, Nagoya.

[0020] In a second specific method, which can be particularly advantageously applied to vascular trees in the brain, at least one anatomical atlas dataset associated with at least one anatomical variant category is provided as reference information, wherein the anatomical atlas dataset is compared with at least one angiography dataset or a comparative dataset derived from the angiography dataset as angiography information. In other words, atlases of possible anatomical variants of the vascular tree can be registered to angiography information. In embodiments, the atlas version associated with at least one anatomical variant category having the minimum distance to actual anatomical structure information should represent the anatomical variant at hand. For example, the anatomical atlas dataset can be statistically derived from multiple baseline datasets of other patients showing the corresponding anatomical variant / anatomical variant configuration.

[0021] Specifically, a dataset of aberrant anatomical atlases related to the absence of anatomical variations (i.e., without (related) deviations from predefined normal anatomical structures) can be used, which can be statistically derived from multiple baseline datasets of patients showing aberrant anatomical variations. This dataset of anatomical atlases related to normal anatomical structures is particularly advantageous relative to the two-step comparison configuration described above; that is, in an advantageous embodiment,

[0022] - In the first step, if a comparison indicating the general presence of anatomical variations is performed, wherein if no anatomical variation is detected in the comparison, the anatomical variation category associated with the absence of anatomical variation is determined as variation information; otherwise, in the second step, at least one other anatomical variation category is determined by at least one further comparison.

[0023] - In the second step, a comparison is performed with at least one of the non-variant anatomical atlas datasets restricted to a subregion of the vascular tree to determine whether the angiography dataset belongs to an associated anatomical variant category, wherein the subregion is associated with at least one anatomical variant category.

[0024] Therefore, the dataset of aberrant anatomical atlases can be used in the first step to generally check for deviations from predefined normal anatomical structures. If no (relevant) deviations are found, i.e., the patient has normal anatomical structures, further comparison in the second step is unnecessary. However, if a deviation is found, further comparison continues in the second step. Here, since comparisons can be performed for sub-regions within the vascular tree, the fact that anatomical variants of the corresponding anatomical variant categories are located can be utilized. For example, if an anatomical variant category associated with a specific line of blood vessel is found in a segment, the comparison regarding that anatomical variant category can be limited to finding a sub-region of that segment. If no other anatomical variant category is located in that sub-region, the deviation indicates that the line of that segment is actually different from the standard.

[0025] Note that in this embodiment, but also generally, at least one similarity and / or correlation measure can be calculated and used in the comparison. For example, when comparing with a dataset of anatomy without variants, a threshold for the similarity measure can be defined, exceeding which indicates that a normal anatomical structure can be assumed. Of course, in this embodiment, multiple similarity and / or correlation measures can be used, and if a deviation from a predefined normal anatomical structure is determined in the first step, it is already permissible to predict that an anatomical variant may exist, thereby, for example, allowing for the selection of associated anatomical variant categories for further comparison, and the exclusion of other anatomical variant categories. Furthermore, depending on the actual value of at least one similarity and / or correlation measure, the examination of anatomical variant categories in the second step can be prioritized.

[0026] In a particularly preferred embodiment of the second method, structural information is also used as the basis for comparison. Preferably, the lumen of the vessels in the vascular tree is derived as structural information from the angiography dataset and used as the comparison dataset. That is, the lumen of the vascular system is extracted using lumen extraction algorithms known in the art. For example, since contrast agents have been used, the lumen can be determined during segmentation. Then, at least one anatomical atlas dataset can be registered to this segment. In this way, also for the second method, the data to be compared can be limited to practically relevant information.

[0027] Regarding the lumen detection algorithm, refer to F. Lugauer et al., “Precise lumen segmentation in computed tomography angiography”, MICCAI International Symposium on Medical Computer Vision, pp. 137-147, Springer (2014).

[0028] In a preferred embodiment, variant information can be used for automated report generation and / or to determine the suitability of downstream evaluation algorithms, particularly deep learning-based artificial intelligence evaluation algorithms. Thus, the methods described herein allow for automated report generation, for example, informing medical personnel that a particular anatomical variant may complicate a planned medical procedure. Furthermore, and particularly preferably, variant information can act as a “gatekeeper” for other evaluation algorithms, particularly deep learning-based methods that may have problems if they do not perform well on the input data in their training data. For example, if the evaluation algorithm is trained only for a specific anatomical variant configuration, problems may occur if a new anatomical variant configuration is applied as input. In this case, user information about the evaluation algorithm can be output, for example, indicating lower reliability of the results and / or prompting for human evaluation, particularly by reading the data. This “gatekeeper” function increases clinicians’ confidence in using the relevant deep learning-based methods. Examples of such evaluation algorithms include deep learning-based FFR calculation and lesion detection algorithms.

[0029] In a further preferred embodiment of the invention, at least one anatomical variant category is used in relation to image features in an angiography dataset caused by imaging artifacts, particularly stacking artifacts. Some image artifacts, particularly so-called stacking artifacts, have the appearance of anatomical variants in angiography datasets. Therefore, the strategy according to the invention can also be applied to artifact detection in angiography datasets, as imaging artifacts can share morphological manifestations similar to anatomical variants in angiography data, such as gaps and / or offsets in the depicted blood vessels. For example, if the angiography dataset is compiled from at least two image stacks, gaps or offsets can occur due to various reasons during the imaging process, resulting in, for example, separation of two parts of the blood vessel. This is called a stacking artifact and can still be detected using the invention. Other examples of imaging artifacts can be local blurring artifacts, for example, if blurred streaks in an image make the vessel wall less clear, as if the contrast agent has penetrated. Such artifacts can also be assigned to an anatomical variant category and can be detected using the invention. Detection of such imaging artifacts can lead to corresponding user information, allowing the user to distinguish between true anatomical variants and anatomical abnormalities caused by imaging artifacts.

[0030] As explained, at least one anatomical variant category can be associated with normal anatomical variants, i.e., non-pathological anatomical variants. As a specific example of anatomical variants in a vascular tree, at least one anatomical variant category can involve at least one dominant variant, and / or at least one regional supply variant, and / or at least one additional and / or fewer vascular variants, and / or at least one line variant and / or at least one ostial abnormality variant. For example, in the case of a coronary tree, different vessels can supply perfusion, where dominance is determined by the side of the vascular system supplying the posterior side of the heart. Right dominance is present in 70% of cases and is therefore normal, left dominance is observed in 10%, and co-dominance can even occur in 20%. Co-dominance refers to the posterior descending artery (PDA) and posterolateral branch appearing together from both the right and left sides of the vascular system. For example, the dominance in a patient can be detected by extracting the central line and performing rule-based analysis on the central line line. The dominant side should have a segment of the posterior descending artery (PDA), and in the case of co-dominance, both PDAs should be present.

[0031] Further anatomical variations involve the supply to different regions. For example, the supply to the inferior wall can be analyzed. For instance, if the PDA has an early origin and then flows along the diaphragmatic surface of the right ventricle (RV) towards the apex, the anatomical variation can be termed "premature origin of the PDA." In another variation, the LAD may encircle the apex and supply the apical portion of the inferior wall, referred to as "encircling the LAD." Finally, multiple branches can exist, in which case the PDA is very small and multiple branches from the terminal RCA, LCX (left circumflex artery), and obtuse marginal branches can supply the inferior wall. In another example, the atrioventricular node supply can be analyzed. Blood supply to the atrioventricular node occurs 90% of the time from the atrioventricular branch of the RCA, or 10% from the LCX. As a final example, the sinoatrial node supply can be analyzed, i.e., the main coronary artery supplies the sinoatrial node. This supply occurs 60% of the time from the RCA, although in some cases it can also be from the terminal RCA or LCX. Regarding this regional supply variant, in the first method, detection can be performed based on segmenting the supply region of interest as an additional anatomical feature, where the nearest centerline of the extracted centerline of the centerline tree can be identified, since the condition for a vessel with a primary supply region of interest can be that its centerline is closest to the supply region of interest.

[0032] In the general group of another anatomical variant, there may be an excess or deficiency of additional vessels, i.e., vessels may be absent. A prominent example in coronary artery cases is the so-called intermediate branch, where the left main coronary artery (LMCA) does not divide into two branches, LCX and LAD, but instead into three branches: LCX, LAD, and the additional intermediate branch. This can be detected by analyzing the central tree structure, i.e., if there are three bifurcations following the LMCA.

[0033] An example of a missing vessel (segmentation) is the so-called LMCA (left main coronary artery) atresia, where the LMCA is absent and the LAD and LCX are separated from the left aortic sinus (LSV) but adjacent ostia appear. This can be detected, for example, if multiple ostia for the left artery are found in a centerline tree or luminal comparison dataset.

[0034] Regarding line variations, examples include myocardial bridging. This anatomical variation can be detected with an incidence rate as high as 25%. In myocardial bridging, atypical lines of the coronary arteries are found, where the line descends within the myocardium during cardiac systole and causes compression of the vessel. Using a central line tree in the first approach, this can also be detected by segmenting the myocardium as an additional anatomical feature and providing corresponding anatomical information, where the condition checks whether the vessel's central line penetrates the segmented myocardial mask.

[0035] In another example of a line variation, the LCX can indicate an acute origin. This anatomical variation characterizes an angle of less than 45 degrees between the LMCA and LCX. This can also be detected in a simple way from the centerline tree, such as by simply assessing the angle at the bifurcation and checking if that angle is below or equal to a threshold of 45 degrees.

[0036] A well-known further anatomical variation is the shepherd's cane in the RCA, occurring in approximately 5% of cases. In this variation, the RCA has a normal origin but typically takes a curved and high trajectory immediately after its origin from the aorta. While this can be determined by analyzing curvature from a centerline tree (especially after applying a segmented labeling algorithm), additional or alternative centerline trajectories can also be analyzed relative to the aorta as further anatomical features. For example, the RCA may remain two millimeters close to the aorta while the distance to the heart (which can also be segmented) increases at the beginning of the RCA.

[0037] Ultimately, oral abnormalities are detected as anatomical variations. For example, in a known "high origin" anatomical variation, the mouth may be located 5 mm or more above the junction of the aorta and sinus duct. This can also be detected by centerline analysis, for example, a few millimeters before the RCA (Renus Carinae) adjacent to the aorta, close to the heart.

[0038] As a final example, vascular replication should be mentioned, such as the slit RCA. This anatomical variant can be defined as an RCA characterizing a slit PDA, where the inner sub-section of the RCA leads to the terminal portion of the PDA, which in turn leads to the inner free wall of the right ventricle. Other (posterior) bifurcations of the RCA remain within the atrioventricular groove and form the highest portion of the posterior descending artery. This is also known as a “double RCA.” It can be detected through central ventricle line analysis.

[0039] Regarding the second method, in embodiments where subregions are compared in the second comparison step, the examples discussed herein can be localized in a sense that deviations in the associated subregions are typical of the presence of corresponding anatomical variations. Similar anatomical variations are known regarding vascular trees in the brain.

[0040] The present invention further relates to an evaluation system for a three-dimensional angiography dataset for evaluating a patient's vascular tree, the dataset being particularly a computed tomography angiography dataset, the vascular tree being particularly a coronary artery tree, comprising:

[0041] - The first interface is used to receive angiography datasets.

[0042] - A determination unit for determining variant information based on a comparison of angiographic information from an angiography dataset with reference information describing at least one of at least one of several anatomical variant categories related to anatomical variants in a vascular tree, and...

[0043] - The second interface is used to provide variant information.

[0044] The determining unit can be implemented, in whole or in part, as a software module in the processor of a suitable control device or processing system. An advantage of embodiments primarily in software form is that the control device can be easily upgraded via software upgrades to operate in accordance with at least one embodiment of the invention, even if the control device and / or processing system are already in use.

[0045] The term "unit" can be replaced by the terms "circuit" or "module". The term "unit" may refer in part to or include processor hardware (shared, dedicated, or grouped) that executes code and memory hardware (shared, dedicated, or grouped) that stores the code executed by the processor hardware.

[0046] This unit can be implemented in a control unit and may include one or more interface circuits. Different components of the system can be implemented in a delocalized system, such as in a network. In some examples, the interface circuits may include wired or wireless interfaces that connect to a local area network (LAN), the Internet, a wide area network (WAN), or a combination thereof. The functionality of any given unit of this disclosure can be distributed among multiple modules connected via the interface circuits. For example, multiple modules can allow for load balancing. In a further example, a server (also referred to as a remote or cloud) module can perform some functions on behalf of a client module.

[0047] In other words, the evaluation system may include a computing device configured to perform the method according to the invention. All features and descriptions relating to the method according to the invention are applicable to the evaluation system according to the invention.

[0048] As a computing device, the evaluation system may include at least one processor and at least one memory device, where reference information may be stored. Functional units may be implemented as hardware and / or software. Furthermore, other functional units and / or sub-units may be added to implement steps of the preferred embodiments of the method according to the invention.

[0049] For example, the evaluation system can be integrated into the imaging system of a computed tomography (CT) device. However, in a preferred embodiment, the evaluation system can be included in the reading workstation, so that variant information can be provided to a person reading the angiography dataset, such as a radiologist. In embodiments, the evaluation system can, of course, also provide further evaluation information by using a corresponding evaluation algorithm.

[0050] The computer program according to the invention can be directly loaded into the computing device of the evaluation system, and the device includes programming means such that when the computer program is executed on the computing device of the evaluation system, the computing device can perform steps according to the method of the invention. The computer program can be stored on an electronically readable storage medium according to the invention, which therefore includes control information such that when the electronically readable storage medium is used in the computing device of the evaluation system, the steps according to the method of the invention can be executed. The electronically readable storage medium according to the invention can be a non-transitory medium, such as a CD-ROM.

[0051] The solutions according to the present invention are described and claimed below with respect to systems and methods. Elements, features, advantages, or alternative embodiments described herein may be assigned to other claimed objects, and vice versa. In other words, claims to the evaluation system may be modified by features described or claimed in the context of the method, and vice versa. The functional features of the method are implemented by the target units providing the system. Furthermore, elements, features, advantages, or alternative embodiments described in conjunction with specific exemplary embodiments may be assigned to the present invention in their most general terms. Attached Figure Description

[0052] Other objects and features of the invention will become apparent from the following detailed description taken in conjunction with the accompanying drawings. However, the drawings are merely schematic diagrams designed solely for illustrative purposes and do not limit the invention. The drawings show:

[0053] Figure 1 This is a general flowchart of a preferred embodiment of the method according to the present invention.

[0054] Figure 2 This is a schematic diagram showing the first anatomical variant.

[0055] Figure 3 This is a schematic diagram illustrating the second anatomical variant.

[0056] Figure 4 This is a schematic diagram showing the third anatomical variant.

[0057] Figure 5 This is a schematic diagram showing the fourth anatomical variant, and

[0058] Figure 6 This is the functional structure of the evaluation system according to the present invention. Detailed Implementation

[0059] Figure 1 This is a general flowchart of a preferred embodiment of the method according to the invention. In this method, an angiography dataset 1, in this case a computed tomography angiography dataset, is evaluated regarding the presence of anatomical variations relative to a predefined normal anatomical structure. Angiography dataset 1 illustrates a vascular tree in the imaging region of the patient. Hereinafter, the coronary artery tree is exemplary referred to as a vascular tree; however, in other embodiments, the vascular tree may also include blood vessels in the patient's brain.

[0060] In preparation step S1, structural information is determined from the angiography dataset 1, which will be used hereinafter as angiography information 2 for comparison with reference information 3 regarding at least one anatomical variant category. In a preferred embodiment, the centerline can be extracted in step S1 using a centerline extraction algorithm, such that a centerline tree structure and / or lumen extraction algorithm can be used to segment the lumen of the vessels in the vascular tree. Additionally, preferably, a segmentation labeling algorithm can be applied to assign centerlines and / or lumens to segments, particularly according to a labeling scheme. This application is useful if, in comparison 4, particularly under the conditions of a rule set, the centerlines of the defined segments are to be analyzed.

[0061] exist Figure 1 In this example, a two-step comparison 4 is illustrated. In the first comparison step S2, a comparison indicating the general presence of anatomical variations is performed. For example, this comparison can be performed by comparing the luminal dataset as angiographic information 2 with the dataset of aberrant anatomical images (showing predefined normal anatomical structures) as reference information 3. Additionally or alternatively, if all conditions indicating the absence of variations are met, the rule set of reference information 3 using those conditions can be applied. In any case, if it is determined as a result of step S2 that no variations exist in step S3, variation information 5 is determined in step S4 such that it indicates the absence of anatomical variations with respect to predefined normal anatomical structures.

[0062] However, if the first comparison in step S2 has indicated the presence of an anatomical variant, then in the second comparison step S5, at least one other anatomical variant category to which the anatomical variant falls is determined. In any case, variant information 5 indicates the presence of an anatomical variant of at least one anatomical variant category, or the absence of an anatomical variant relative to normal anatomical structures, i.e., no deviation associated with normal anatomical structures, whose formation indicates another anatomical variant category with no associated deviation from predefined normal anatomical structures.

[0063] In optional step S6, the variant information can be used for further anatomical decisions and / or treatment. Preferably, variant information 5 can be used to automatically generate a report that also includes information on the anatomical variant, which can be used, for example, regarding further treatment of the patient, particularly if a minimally invasive intervention is planned. More preferably, variant information 5 can be used to determine the suitability of downstream evaluation algorithms in the "gatekeeper" function. For example, if the evaluation function is a deep learning-based AI evaluation function that has been trained using training data that does not include a certain anatomical variant that has not yet been discovered, it can output warnings and / or prompt manual reading.

[0064] Note that some anatomical variant categories can also be associated with imaging artifacts, which appear to be anatomical variants but are actually caused during the imaging process. An example is stacking artifacts; if such an imaging artifact is detected in step S6, corresponding user information is output and / or corrective measures are triggered.

[0065] Regarding the comparison in step S4, two advantageous methods are conceivable, and these two methods can also be used in combination, for example, by using the anatomical atlas dataset associated with the non-anatomical variant in the first comparison in step S2 to include angiographic information 2 of the segmented lumen, while in the second comparison step S5, the conditions of the rule set are used to evaluate the centerline tree.

[0066] Specifically regarding the coronary tree, the use of the rule set includes: if all conditions indicating the presence of at least one anatomical variant category are met, it proves particularly advantageous, at least in step S5, because all relevant information relating to the usual anatomical variants is included in the central line, connections, and their trajectories. Reference will now be made to... Figures 2 to 5 Some examples are discussed. Typically, the type of anatomical variant / anatomical variant category can be detected using a set of rules, each set of rules including at least one condition, which includes overt variants, regional supply variants, variants with additional and / or missing vessels, line variants, and oral abnormal variants.

[0067] Figures 2 to 4 An example of an anatomical variation in the line of blood vessels is illustrated schematically, chosen for simple visualization. Figure 2 This involves myocardial bridging. Here, the blood vessel causes its portion of the line to partially pass through the myocardium, indicated by the segmented region 6. As can be seen, the central line 7 partially passes through region 6, clearly indicating the presence of myocardial bridging.

[0068] It is usually noted that, Figure 1In step S1, anatomical information describing at least one additional anatomical feature (here, the myocardium) can be determined from angiography dataset 1 and / or at least one additional image dataset registered to angiography dataset 1. Of course, anatomical information relating to other anatomical features can also be determined, for example, if it is to determine which vessel supplies a region of interest, such as the inferior wall, atrioventricular node, and / or sinoatrial node. In this case, it can be examined which centerline lies closest to the segmented region of interest.

[0069] Figure 3 This involves the acute origin of the LCX from LMCA 9 to LCX 8. To determine the presence of this anatomical variation, the angle 10 between the corresponding centerline 11 and centerline 12 at bifurcation 13 is compared to a threshold angle of 45 degrees. If this angle is less than or equal to 45 degrees, an acute origin of the LCX is present.

[0070] Figure 4 A shepherd's cane 14 is shown attached to the RCA 15, which originates from the aorta 16. For example, this shepherd's cane 14 can be detected by analyzing the line of curvature, particularly, of the centerline 17 of the RCA 15, which originates from the aorta 16.

[0071] Figure 5 The so-called high starting point is shown. As can be seen, the orifice 18 of the coronary artery 19 from the aorta 16 is very high, such that the coronary artery 19 of the RCA 15 extends a few millimeters adjacent to the aorta 16, as can be clearly seen from the corresponding centerline 20.

[0072] Regarding the use of the anatomical atlas dataset (second method), it should be noted that, for example, the comparison can be performed as registering, for example, the anatomical structure information 2 of a segmented lumen with the corresponding anatomical atlas dataset of reference information 3, typically producing a distance as a measure of similarity and / or relevance. For example, multiple different anatomical atlas datasets can be registered to anatomical structure information 2, and at least one anatomical variant category associated with the closest anatomical atlas dataset can be assigned to angiography dataset 1. However, in a preferred embodiment, where different anatomical variants of the anatomical variant category can be located differently, for sub-regions associated with different anatomical variant categories, repeated comparisons with anatomical atlas datasets without associated anatomical variants can be performed, thereby locating discrepancies and inferring the presence of anatomical variants accordingly.

[0073] Figure 6The functional structure of the computing device 21 of the evaluation system 22 according to the present invention is shown. The evaluation system 22 includes a first interface 23 for receiving angiography dataset 1. Furthermore, reference information 2 can be stored in a memory device 24. According to step S1, the received angiography dataset 1 can be processed by the structure extraction unit 25 to determine the angiography information 2 in the form of structured information, particularly the centerline and / or lumen.

[0074] In determining unit 26, variant information 5 is determined by comparison 4, as described above, particularly in steps S2 to S5. In optional use unit 27, variant information 5 can be further used for automatic report generation, gatekeeper functions, and / or regarding detected imaging artifacts, as described with respect to step S6. Second interface 28 is an output interface, particularly regarding variant information 5, but also automatically generates reports, etc.

[0075] The evaluation system 22 can be integrated into the imaging device and / or the readout workstation. Furthermore, the evaluation system can certainly provide additional evaluation services, particularly regarding the application of evaluation algorithms.

[0076] Although the present invention has been described in detail with reference to preferred embodiments, the invention is not limited to the disclosed examples, and those skilled in the art can derive other variations from the examples without departing from the scope of the invention.

Claims

1. A computer-implemented method for evaluating a three-dimensional angiography data set (1) of a vascular tree of a patient, the three-dimensional angiography data set (1) comprising a computed tomography angiography data set, the vascular tree comprising a coronary artery tree, characterized in that, Based on the comparison of angiography information (2) from the angiography dataset (1) with reference information (3) describing at least one category of anatomical variant categories, variant information (5) describing at least one anatomical variant category among multiple anatomical variant categories related to the anatomical variants of the vascular tree is determined. The at least one structural piece of information in the vascular tree is determined as angiographic information by at least one structural evaluation algorithm (2); and The angiography information (2) is compared with reference information (3), which also includes or refers to the structural information. In the first step, a comparison indicating the general presence of the anatomical variant is performed, wherein if no anatomical variant is detected in the comparison, the anatomical variant category associated with the absence of the anatomical variant is determined as the variant information; otherwise, in the second step, the anatomical variant category is determined by at least one other anatomical variant category through at least one further comparison, wherein at least one anatomical atlas dataset associated with at least one anatomical variant category is provided as reference information (3). The anatomical atlas dataset is compared with the at least one angiography dataset (1) or a comparison dataset derived from the angiography dataset as angiography information (2), and at least one similarity metric and / or correlation metric is calculated and used in the comparison, and / or the lumen of the blood vessels in the vascular tree is derived from the angiography dataset (1) as structural information and used as the comparison dataset.

2. The method of claim 1, wherein, The centerlines (7, 11, 12, 17, 20) of at least a portion of the vascular tree are determined as structural information to form a centerline tree, wherein at least one set of rules for at least one category of the anatomical variant categories is provided as reference information (3), each set of rules including at least one condition for at least one feature of the centerline tree, wherein in the comparison, the condition of at least one set of rules in the at least one set of rules is applied, and the anatomical variant category associated with the set of rules is determined when all the conditions of the set of rules associated with the anatomical variant category are satisfied by the centerline tree.

3. The method of claim 2, wherein, At least one conditional assessment describes the anatomical information of at least one additional anatomical feature (6), which is not part of the vascular tree, wherein the anatomical information is derived from the angiography dataset (1) and / or an additional image dataset registered with the angiography dataset (1).

4. The method of claim 1, wherein, An atypical anatomy atlas dataset associated with atypical anatomy was used, which was statistically derived from multiple baseline datasets of patients showing atypical anatomy.

5. The method according to claim 4, characterized in that: - In the first step, if the general comparison indicating the presence of anatomical variants is performed, wherein if no anatomical variant is detected in the comparison, then the anatomical variant category associated with the absence of anatomical variants is determined as the variant information; otherwise, in the second step, belonging to at least one other anatomical variant category is determined through at least one further comparison. - In the second step, a comparison is performed with at least one of the aberrant anatomy atlas datasets restricted to a subregion of the vascular tree to determine whether the angiography dataset (1) belongs to the associated anatomical variant category, wherein the subregion is associated with at least one anatomical variant category.

6. The method according to any one of claims 1-5, characterized by, The variant information (5) is used for automatic report generation and / or to determine the suitability of downstream evaluation algorithms, including deep learning-based artificial intelligence evaluation algorithms.

7. The method according to any one of claims 1-5, characterized by, At least one anatomical variant category associated with image features in the angiography dataset (1) is used due to imaging artifacts, including stacking artifacts.

8. The method according to any one of claims 1-5, characterized by, At least one anatomical variant category is associated with a normal anatomical variant, including at least one dominant variant and / or at least one regional supply variant and / or at least one additional and / or less vascular variant and / or at least one line variant and / or at least one oral abnormality variant.

9. An evaluation system (22) for evaluating a three-dimensional angiography dataset (1) of a patient's vascular tree, the three-dimensional angiography dataset (1) comprising a computed tomography angiography dataset, the vascular tree comprising a coronary artery tree, comprising: - First interface (23) for receiving the angiography dataset (1). - Determining unit (26) for determining variant information (5) based on a comparison of angiographic information (2) of the angiography dataset (1) with reference information (3) describing at least one of a plurality of anatomical variant categories related to the anatomical variants of the vascular tree, and - The second interface (28) is used to provide the variant information. The at least one structural piece of information in the vascular tree is determined as angiographic information by at least one structural evaluation algorithm (2); and The angiography information (2) is compared with reference information (3), which also includes or refers to the structural information. In the first step, a comparison indicating the general presence of the anatomical variant is performed, wherein if no anatomical variant is detected in the comparison, the anatomical variant category associated with the absence of the anatomical variant is determined as the variant information; otherwise, in the second step, the anatomical variant category is determined by at least one other anatomical variant category through at least one further comparison, wherein at least one anatomical atlas dataset associated with at least one anatomical variant category is provided as reference information (3). The anatomical atlas dataset is compared with the at least one angiography dataset (1) or a comparison dataset derived from the angiography dataset as angiography information (2), and at least one similarity metric and / or correlation metric is calculated and used in the comparison, and / or the lumen of the blood vessels in the vascular tree is derived from the angiography dataset (1) as structural information and used as the comparison dataset.

10. A computer program, when executed on a computing device (21) of an evaluation system (22), wherein the computer program performs the steps of the method according to any one of claims 1 to 8.

11. An electronically readable storage medium on which the computer program of claim 10 is stored.

Images