Aortic rupture risk estimation

Abstract

A computer-implemented method, an apparatus, and a computer program product for estimating an aortic rupture risk are disclosed. The method comprises: obtaining imaging data comprising three-dimensional imaging data representing an aorta of a subject; obtaining additional data related to the subject, wherein the additional data comprises multi- dimensional stress data of the aorta of the subject; fitting an aortic structural model to the imaging data to determine a geometry of the subject's aorta, and processing the determined geometry and the multi-dimensional stress data by a first estimator configured to determine a first aortic rupture risk estimate; processing the imaging data by a second estimator configured to determine a second aortic rupture risk estimate, wherein the second estimator has been trained using training data comprising three-dimensional imaging data of a plurality of aortas; determining a combined aortic rupture risk estimate based on the first aortic rupture risk estimate and the second aortic rupture risk estimate; and outputting the combined aortic rupture risk estimate and including the indicative rupture direction of the aorta.

Description

[0001] AORTIC RUPTURE RISK ESTIMATION

[0002] Technical Field

[0003] The present solution generally relates to a computer-implemented method, an apparatus, and a computer program product for estimating an aortic rupture risk.

[0004] Background

[0005] Aortic aneurysms are a common incidental discovery in medical imaging studies of the chest. A risk of an aortic rupture is associated with all aortic aneurysms, but providing an accurate prediction of the risk is not possible with current techniques. Due to this, a large number of patients with aortic aneurysms attend follow-up imaging studies to monitor the progression of the aneurysm. This places a large burden on healthcare systems and their limited imaging resources.

[0006] Summary of the Invention

[0007] The scope of protection sought for various embodiments of the invention is set out by the independent claims. Various embodiments are disclosed in the dependent claims. The embodiments and features, if any, described in this specification that do not fall under the scope of the independent claims are to be interpreted as examples useful for understanding various embodiments of the invention.

[0008] A computer-implemented method for estimating an aortic rupture risk comprises: obtaining imaging data comprising three-dimensional imaging data representing an aorta of a subject; obtaining additional data related to the subject, wherein the additional data comprises multidimensional stress data of the aorta of the subject; fitting an aortic structural model to the imaging data to determine a geometry of the subject’s aorta, and processing the determined geometry and the multi-dimensional stress data by a first estimator configured to determine a first aortic rupture risk estimate and an indicative rupture direction of the aorta; processing the imaging data by a second estimator configured to determine a second aortic rupture risk estimate, wherein the second estimator has been trained using training data comprising three-dimensional imaging data of a plurality of aortas; determining a combined aortic rupture risk estimate based on the first aortic rupture risk estimate and the second aortic rupture risk estimate; and outputting the combined aortic rupture risk estimate including the indicative rupture direction of the aorta.

[0009] The method may further comprise determining the indicative rupture direction based on a direction of a largest stress of the multi-dimensional stress data.

[0010] The three-dimensional imaging data representing the aorta of the subject may comprise a set of two-dimensional images of the aorta of the subject.

[0011] The training data may comprise, for each aorta of the plurality of aortas, a set of two- dimensional images representing the aorta.

[0012] The imaging data may comprise one or more of: magnetic resonance imaging, MRI, data, computed tomography, CT, data, and ultrasound data.

[0013] The imaging data may further comprise data representing velocity of blood flow in the aorta, and the first estimator may be further configured to determine the first aortic rupture risk estimate also based on the velocity of blood flow in the aorta.

[0014] The method may further comprise obtaining further additional data related to the subject, and at least one of the first estimator and the second estimator may be configured to process the further additional data to determine the respective aortic rupture risk estimate.

[0015] The further additional data of the subject may comprise one or more of the following risk factors associated with aortic aneurysms e.g. systolic blood pressure, age, sex, body mass index, smoking history, diabetes, hypertension, hypercholesterolemia, genetic aortic diseases, structure of aortic valve (bicuspid or tricuspid) and family history (parents, siblings, offspring) related to aneurysms.

[0016] The method may further comprise processing, by the first estimator, the further additional data of the subject to determine the first aortic rupture risk estimate.

[0017] The method may further comprise processing, by the second estimator, the further additional data of the subject to determine the second aortic rupture risk estimate.

[0018] The method may further comprise computing, by the first estimator, principal wall stresses of walls of the subject’s aorta based on the determined geometry; and determining, by the first estimator, the first aortic rupture risk estimate based on the computed principal wall stresses including the indicative rupture direction. The method may further comprise comparing, by the first estimator, the computed wall stresses to one or more predetermined stress thresholds to determine the first aortic rupture risk estimate.

[0019] The method may further comprise determining a wall thickness of the subject’s aorta based on the imaging data, and determining the first aortic rupture risk estimate also based on the determined aortic wall thickness.

[0020] The imaging data may represent the ascending aorta and / or the abdominal aorta of the subject.

[0021] The subject may be a human subject.

[0022] The method may further comprise determining, by the second estimator, the second aortic rupture risk estimate based on the first aortic rupture risk estimate, and the training data used to train the second estimator may further comprise a plurality of first aortic rupture risk estimates that have been determined by fitting the aortic structural model to the three- dimensional imaging data of the plurality of aortas to determine geometries of the of the plurality of aortas, and by processing the determined geometries.

[0023] The first estimator may be configured to determine the first aortic rupture risk estimate independently of the second estimator, and the second estimator may be configured to determine the second aortic rupture risk estimate independently of the first estimator.

[0024] The first aortic rupture risk estimate and / or the second aortic rupture risk estimate may comprise an aortic rupture probability.

[0025] The first aortic rupture risk estimate and / or the second aortic rupture risk estimate may comprise a critical diameter at which the subject’s aorta is expected to rupture.

[0026] The first aortic rupture risk estimate and / or the second aortic rupture risk estimate may comprise one or more rupture locations at which the subject’s aorta is expected to rupture.

[0027] The second estimator may comprise a convolutional neural network.

[0028] An apparatus is configured to perform the above-described method.

[0029] The apparatus may comprise at least one processor, at least one memory including computer program code, the at least one memory and the computer program code configured to, with the at least one processor, cause the apparatus to perform the method. A computer program product comprises computer program code configured to, when executed by at least one processor, cause an apparatus to perform the above-described method.

[0030] Brief Description of the Drawings

[0031] FIG. 1 is a flow chart illustrating a method for estimating an aortic rupture risk;

[0032] FIG. 2 is a block diagram illustrating embodiments of an apparatus for estimating the aortic rupture risk; and

[0033] FIG. 3 depicts an architecture of an embodiment of a computer program for estimating the aortic rupture risk.

[0034] Detailed Description of the Invention

[0035] The invention relates to a computer-implemented method, an apparatus, and a computer program product for estimating an aortic rupture risk of a subject. The invention provides an efficient solution for estimating the risk of aortic rupture, reducing the need to carry out resource-intensive follow-up studies for aortic aneurysm patients. As an alternative to, and / or in addition to estimating the risk of aortic rupture, the invention is suitable for estimating the risk of aortic dissecation (i.e. a tear in the inner layer of the aorta). Aortic dissecation and aortic rupture are closely related conditions, as dissecation can quickly lead to the complete rupture of the aorta.

[0036] The method is illustrated in the form of a flow chart in FIG. 1 , and embodiments of the apparatus are shown in FIG. 2. The apparatus 200 is configured to perform the method of FIG. 1 or any of its embodiments. The apparatus 200 may comprise means for performing the method of FIG. 1 or any of its embodiments. The means may comprise at least one processor 10, at least one memory 20 including computer program code 22, the at least one memory 20 and the computer program code 22 configured to, with the at least one processor 10, cause the apparatus 200 to perform the method of FIG. 1 or any of its embodiments. The at least one processor 10 may include e.g. a central processing unit (CPU) and / or a graphics processing unit (GPU). The at least one memory 20 may include e.g. random access memory (RAM) and / or non-volatile memory. The apparatus 200 may be but need not be dedicated hardware. The apparatus 10 may be a virtual machine. The method of FIG. 1 , described in more detail below, may be executed as a containerized application using operating system (OS) -level virtualization.

[0037] The invention may be implemented in the form of a computer program (product) or a computer-readable medium 30 comprising computer program code configured to, when executed by the at least one processor 10, cause the apparatus 200 to perform the method of FIG. 1 or any one of its embodiments. The computer-readable medium 30 may be a non- transitory computer-readable medium. An embodiment of the software implementation of the invention is shown in FIG. 3.

[0038] The method of FIG. 1 comprises obtaining 102 imaging data that preferably comprises three-dimensional imaging data. The (three-dimensional) imaging data represents the aorta of a subject, or at least a part of the subject’s aorta, such the ascending aorta and / or the abdominal aorta. The subject is generally a human subject, but the invention may also be applied to other animals that have a sufficiently similar aortic anatomy as humans, such as other amniotes. The method of FIG. 1 further comprises obtaining additional data related to the subject. The additional data comprises multi-dimensional stress data. The multidimensional data stress data may be obtained by computing wall stresses of the walls of the subject’s aorta based on geometry of the subject’s aorta. In an example, the multidimensional stress data may be obtained by fitting an aortic structural model to the imaging data.

[0039] The apparatus 200 of FIG. 2 has several options for obtaining the imaging data. In a first option, the imaging data is stored in the memory 20 of the apparatus 200, e.g. in the database 24, and the apparatus is configured to obtain the data by reading it from the memory 20. In a second option, the data is stored in a computer-readable medium 30, and the apparatus is configured to obtain the data by reading it from the computer-readable medium 30. In a third option, the apparatus uses a communication interface 40 to obtain the data from a wired or wireless network. The apparatus may communicate via the network with a cloud server 50 configured to store the imaging data, and / or the apparatus may communicate with the imaging device(s) 52 used to measure the imaging data, directly or via the cloud server 50. In the embodiment of FIG. 3, the imaging data is received via a software interface 300, which may be e.g. an application programming interface (API).

[0040] Preferred types of imaging data include magnetic resonance imaging (MRI) data, computed tomography (CT) data, and ultrasound data. CT is most commonly used for follow-up studies of patients at risk of aortic rupture, but MRI could provide more accurate estimates of the risk. Various file formats exist for providing the imaging data; TIFF (Tagged Image File Format) is one example of a suitable file format. In general, three-dimensional images comprising a plurality of (e.g. 64) two-dimensional slices are suitable. The two-dimensional images may be parallel and / or form a stack of that forms a three-dimensional image. However, a set of two-dimensional images (e.g. 3 images) that do not form such a stack may also be used. For example, a subset of images that need not all be consecutive may be selected from a stack as the set of two-dimensional images. The two-dimensional images may or may not all be parallel with one another. Cardiac-synchronized imaging data is preferred to reduce motion artifacts.

[0041] To improve the accuracy of the estimate, data representing velocity of blood flow in the aorta may be included in the imaging data. For example, 4D MRI data may be used.

[0042] The obtained imaging data may comprise multiple different types of imaging data. For example, both (4D) MRI data and CT data may be obtained. Other combinations of two or more of the above-mentioned imaging data types (and other imaging data types) are suitable options as well.

[0043] In addition to the imaging data, additional data related to the subject may be obtained. For example, the additional data may be read from the memory 20 of the apparatus 200, obtained from the cloud server 50 via the communication interface 40, or received from a user via a user interface 60 of the apparatus. The additional data may include e.g. a blood pressure of the subject, an age of the subject, whether the subject is a smoker, the subject’s sex, and / or a body mass index of the subject. These additional data are processed by the clinical data module 304 that handles data validation and unit conversions.

[0044] Referring again to FIG. 1 , the method further comprises fitting an aortic structural model to the imaging data. This is performed by the geometry module 302 of FIG. 3. The result of the fitting is a determined geometry of the subject’s aorta. One way of performing the fitting is to extract edge points of the aorta from the two-dimensional image slices by applying edge detection techniques known in the field of image processing. The extracted edge points are then converted to a parametric surface, such as a three-dimensional polygon (e.g. triangle) model or another kind of a surface model. The geometry of the subject’s aorta can be used for computing wall stresses.

[0045] In this context, the geometry of the aorta refers to the three-dimensional shape and size of the aorta, expressed e.g. as an inner mesh that represents the inner surface of the aorta, and as an outer mesh that represents the outer surface of the aorta, the meshes referring e.g. to structural builds of a 3D model consisting of polygons. The thickness of the walls of the aorta at different locations can also be a part of the geometry; the wall thickness may be determined e.g. from distance(s) between the inner and outer meshes. The determined geometry is then processed 106 (see FIG. 1 ) by a first estimator 308 (see FIG. 3). The first estimator 308 is configured to determine a first aortic rupture risk estimate for the subject.

[0046] In the embodiment of FIG. 3, the first estimator 308 computes wall stresses of the walls of the subject’s aorta using the geometry of the aorta determined by the geometry module 302. The wall stresses may be multidimensional stress data, whereby values of the wall stresses can be defined for dimensions of the subject’s aorta. In example, the multidimensional stress data is three-dimensional data in a three-dimensional coordinate system, whereby the multidimensional stress data comprises components of the wall stresses for each coordinate axis, e.g. X-, Y- and X-axis. The wall stresses may be computed analytically or numerically, e.g. with finite element (FE) methods. As an alternative to FE methods, the first estimator 308 or the geometry module 302 may segment the aorta (i.e. its determined geometry) to a plurality of tubular segments (e.g. 100 tubular segments), and the first estimator 302 may compute one or more wall stresses for each of the tubular segments, providing surprisingly accurate estimation results. This may be performed as an analytical computation.

[0047] The first estimator 308 compares the computed wall stresses to one or more predetermined stress thresholds. Each predetermined stress threshold specifies a stress at which the tissue will rupture at a specific location or area of the aorta. The first estimator compares the computed wall stresses, for example maximum or maxima of the computed wall stresses, to the predetermined stress thresholds, and determines the first aortic rupture estimate accordingly. For example, if a computed wall stress at a specific location of the aorta is greater than or equal to a predetermined stress threshold of the location or the area containing the location, the first aortic rupture estimate will indicate a high risk of rupture at the location. The computed wall stresses may comprise principal wall stresses incorporated with the directional indication (Hamrock, Schmid, Jabobsson. Fundamentals of Machine Elements, 2nd Edition, pp. 54...60). The directional indication gives information about the direction of the rupture of the aortic wall.

[0048] In an example, the principal wall stresses are further calculated using the three-dimensional 3x3 stress tensor S total from equation 1. This includes the formation of the characteristic equation S total - A,I, where X represents the eigenvalues and I is 3x3 identity matrix. The determinant of the characteristic equation is solved, i.e. det{ Sy total - X I } = X3+ It X2+ 12X3+ 13= 0, and the solving of this third order equation gives a solution as Xi = { Xi, X2, X3} where Xi , X2and X3 are the three principal wall stresses. The directions of the principal wall stresses are the directions where only three principal wall stresses and no structural shear stresses are existing and therefore can be written n(Sjj total - A, I) = 0, where n is unknown matrix variable consisting three direction vectors n-i, n2 and ns for each principal wall stresses. When any one solved three principal wall stresses are substituted for X in the previous equation, it reduces to only two independent linear equations. By taking account that n-i2+ n22+ ns2= 1 (rule of the direction cosines) and ns = x n2 (all the direction vectors are perpendicular with each other), we have three equations for solving ni, n2 and ns and thus the directions are known. Now, the fracture (or the rupture) advances in isotropic or near-isotropic material perpendicular against the direction of the largest principal wall stress which gives the direction of the indicative rupture direction.

[0049] The first estimator may also take the thickness of the walls of the subject’s aorta into account. The first estimator 308 may obtain the thickness from the geometry of the aorta as described above. Alternatively or additionally, the geometry module 302 or the first estimator 308 may fit the imaging data to a population model (i.e. a known model of the aorta that is based on measurements and / or imaging data of a plurality of individual aortas) to obtain an estimate of the thickness of the walls of the aorta. Thickness values may be defined for multiple locations or areas of the subject’s aorta, as used in equations 1 and 2 described further below.

[0050] The predetermined stress thresholds may be obtained from biomechanical measurements of aorta samples. A suitable measurement protocol is presented in section 2.4, and example values of suitable tissue rupture thresholds (failure stresses) are presented in Figure 3C, of Kiema M, Sarin JK, Kauhanen SP, Torniainen J, Matikka H, Luoto ES, Jaakkola P, Saari P, Liimatainen T, Vanninen R, Yla-Herttuala S, Hedman M, Laakkonen JP: Wall Shear Stress Predicts Media Degeneration and Biomechanical Changes in Thoracic Aorta, Front Physiol. 2022 Jul 7;13:934941 , doi: 10.3389 / fphys.2022.934941 . Preferably, the average values of the healthy subjects shown in Figure 3C of the above-mentioned document are used as the threshold values for the locations specified (inner and outer curve).

[0051] When the imaging data comprises data representing the velocity of blood flow in the aorta (e.g. 4D MRI data), the first estimator 308 can be further configured to determine the first aortic rupture estimate also based on the velocity of the blood flow in the aorta. The velocity information can be used e.g. in computing the wall stresses of the subject’s aorta.

[0052] When additional data, such as blood pressure data, is obtained, the first estimator 308 may further determine the first aortic rupture estimate based on the additional data. The additional data may be multidimensional stress data of the aorta of the subject. To take the blood pressure data into account, the first estimator 308 may use equation 1 : where Sy total is the total aortic wall stress expressed as a stress tensor, Sypreis the stress tensor based on blood pressure and Sy _exti is the stress tensor based on external loads and / or displacements caused by the heart or other elements of the human body. Calculation of Sy pre is done based on the imaging data; the calculation may be performed using various simplified or numerical methods. An example method is presented in Lang, H., A., Toroidal elastic stress for pressurized elbows and pipe bends, International Journal of Pressure Vessels and Piping, Vol. 15, No. 4, 1984, Pages 291-305. Calculation of the Sy _exti is also done based on the imaging data; the calculation may be performed using various simplified or numerical methods. An example method is presented in Lang, H., A., In-plane bending of a curved pipe or toroidal tube acted on by end couples, International Journal of Pressure Vessels and Piping, Vol. 15, No. 1 , 1984, Pages 27-35; especially equation (3).

[0053] In a more specific example, the first estimator 308 determines the first aortic rupture estimate using equation 2: where Sy is the wall stress expressed as a stress tensor, p; is the total blood pressure, Rj is the radius of the aorta, obtained e.g. from the determined geometry of the aorta, in a particular section (e.g. a tubular section) i of the aorta, f(R)j is the aortic wall thickness (obtained e.g. from the determined geometry of the aorta) as a function of the radius of the aorta, Gy is a geometric correction coefficient, By is a bending stiffness, by is a displacement or torsion (i.e. translation and / or rotation) of the aorta caused by movement of the heart, and I is a moment of inertia of the aortic cross-section, which may be computed from a cross-section of the wall of the aorta, including the dimensions of the cross-section.

[0054] The imaging data is also processed 108 (see FIG. 1 ) by a second estimator 310 (see FIG. 3) configured to determine a second aortic rupture risk estimate. The second estimator is an artificial intelligence estimator and it has been trained using training data to determine the second aortic rupture estimate. A target of the training may be to obtain a sensitivity or specificity of 86 % to 90 %, for example. This may be achieved with a training data set comprising labeled images of approximately 200 to 2000 different subjects. The training data comprises imaging data that is similar to that obtained via the interface 300; three- dimensional imaging data of aortas of different (human) subjects.

[0055] The aortas of the training data may include both healthy aortas and aortas that have ruptured. The training data has been captured prior to rupture of the aortas (that have eventually ruptured) to allow the second estimator to learn to predict the risk of rupture before it occurs.

[0056] The training data may be labeled with truth values. For example, the truth values can take two values (rupture or no rupture) depending on the outcome of a specific aorta depicted in a training image.

[0057] Alternatively, more than two labels (e.g. low risk, medium risk, high risk) may be used as the truth values. In addition or as an alternative to using the outcome (rupture or no rupture) of each aorta, a medical professional may set the labels to provide a more fine-grained estimate of the risk of rupture for each aorta in the training data. This is especially useful when little or no training data is available for aortas that have eventually ruptured. For example, a medical professional may label each aorta of the training data on a scale from 1 (healthy) to 5 ((almost) certain to rupture or already ruptured), the risk of rupture increasing with the value of the label.

[0058] The second estimator 310 is not limited to a specific implementation, but a neural network, and specifically a convolutional neural network is preferred. A pre-trained neural network, such as ResNet-50, may be configured to act as the second estimator 310. However, other types of artificial intelligence and / or machine learning classifiers and estimators may also be used.

[0059] When additional data about the subject is obtained via the interface 300, the second estimator may further process the additional data to obtain the second aortic rupture estimate. Useful data that improve the performance of the second estimator include the age of the subject, the sex of the subject, whether the subject is a smoker, and the body mass index of the subject, for example. To be able to process such data, the second estimator has been trained with corresponding additional data of the subjects whose aortas are depicted in the training data.

[0060] The first estimator 308 and the second estimator 310 may, but need not process the same imaging data. For example, the first estimator may process 4D MRI data representing the subject’s aorta and obtain the benefits of the included flow information, and the second estimator may process CT data representing the same subject’s aorta for a more computationally efficient estimation. However, as 4D MRI data is not always available, the first estimator may process (3D) MRI data and the second estimator may process CT data. As mentioned, both estimators may additionally or alternatively process the same data, such as CT data, which is generally better available than any kind of MRI data.

[0061] In the embodiment of FIG. 3, the first estimator 308 and the second estimator 310 are configured to exchange data. The second estimator 310 may be configured to determine the second aortic rupture risk estimate based on the first aortic rupture risk estimate determined by the first estimator 308. In this case, the training data used to train the second estimator further comprises a plurality of first aortic rupture risk estimates that have been determined by the first estimator 308 or another corresponding estimator used for training purposes, i.e. by fitting the aortic structural model to the three-dimensional imaging data of the plurality of aortas to determine geometries of the of the plurality of aortas, and by processing the determined geometries.

[0062] Additionally, or as an alternative to the above, intermediate data used by the first estimator, such as the wall thickness of the aorta, may be input to the second estimator 310. The second estimator processes e.g. the wall thickness of the aorta to determine the second aortic rupture risk estimate. In this case, wall thickness data of the aortas of the training data is included in the training data used to train the second estimator 310.

[0063] Alternatively, the first and / or second estimators may operate independently or even in complete isolation from each other. The first estimator 308 may be configured to determine the first aortic rupture risk estimate independently of the second estimator, such that the first estimator 308 does not take any inputs from the second estimator 310. Alternatively or additionally, the second estimator 310 may configured to determine the second aortic rupture risk estimate independently of the first estimator 308, such that the second estimator 310 does not take any inputs from the first estimator 308.

[0064] The method of FIG. 1 further comprises determining 110 a combined aortic rupture risk estimate based on the first aortic rupture risk estimate and the second aortic rupture risk estimate. In the embodiment of FIG. 3, this is performed by the results data postprocessing unit 312. As the combined aortic rupture risk estimate comprises the information of both the first and the second aortic rupture risk estimates, a better, more reliable estimate of the subject’s risk for their aorta to rupture can be delivered. The first aortic rupture risk estimate and the second aortic rupture risk estimate may be combined simply by concatenating the risk estimates, or when the risk estimates are of the same type (see below), by computing an average and a range of variation of the first aortic rupture risk estimate and the second aortic rupture risk estimate. For example, arithmetic mean, geometric mean, and / or a weighted mean may be used. A confidence interval provided by the first estimator 308 may be included in the combined aortic rupture risk estimate.

[0065] The first estimator 308 and the second estimator 310 are modules of a modular digital twin 306. The modules of the modular digital twin 306 perform different analyses of the subject’s risk for an aortic rupture. The combination of the two different analyses performed by the first estimator 308 and the second estimator 310 improves the accuracy of the overall aortic rupture risk estimate. The steps performed by the first estimator 308 and the second estimator 310 may be performed computationally in parallel or sequentially. Parallel processing has the advantage of reduced execution time, but in some cases, such as when the first estimator 308 provides the first aortic rupture risk estimate to be processed by the second estimator 310, sequential processing is used.

[0066] The first and second estimators can express the determined aortic rupture risk estimates in several ways. Three different options for determining and expressing the risk of rupture are presented below. The first and second aortic rupture estimates may, but need not use the same way of expressing the risk. Each of the first and second aortic rupture risk estimates may include one or more of the following options.

[0067] One option is for the estimator to determine the risk as an aortic rupture probability, expressed e.g. as a percentage value ranging from 0 to 100, or as one of a plurality of predetermined categories (e.g. low risk, medium risk, high risk). The latter corresponds to one of the options for labeling the truth data used to train the second estimator 310, and is especially useful when the second estimator 310 is configured to determine the second aortic rupture risk estimate based on the first aortic rupture risk estimate.

[0068] A second option is to determine the risk as a critical diameter at which the subject’s aorta is expected to rupture. The critical diameter is preferably an inner diameter of the aorta. In relation to this option, a difference between the current (inner) diameter of the subject’s aorta and the critical (inner) diameter may be computed and provided with or as the risk estimate. A smaller difference correlates with an increased risk. When performed by the first estimator 308, the critical inner diameter may be obtained using equation 3: where Smax is the maximum failure stress of the aortic wall, a is the critical, local inner radius on the aorta (critical inner diameter = 2*critical inner radius), t is the thickness of the aortic wall, and R the local center line radius of the ascending aortic arc, and <|) is rotating angle around the local center line radius. The skilled person can obtain the critical diameter by solving equation 3 for the radius R using analytical and / or numerical methods.

[0069] Alternatively, the first estimator may obtain the critical (inner) diameter using equation 4: where, similarly to equation 2, Sy is the wall stress expressed as a stress tensor, pi_Crt is the critical blood pressure (i.e. the highest observable blood pressure, typically the systolic blood pressure of the subject), Rj is the radius of the aorta in a particular section (e.g. a tubular section) i of the aorta to be solved, f(Rj)j is the aortic wall thickness (obtained e.g. from the determined geometry of the aorta) as a function of the radius of the aorta, Gy is a geometric correction coefficient, By is a bending stiffness, by is a displacement or torsion (i.e. translation and / or rotation) of the aorta caused by movement of the heart, and I is a moment of inertia of the aortic cross-section, which may be computed from a cross-section of the wall of the aorta, including the dimensions of the cross-section. By solving for Rj, the critical diameter may be obtained as 2*Rj. Iterative methods may be used as Rj appears also inside the equation.

[0070] When the second option is used by the second estimator 310, critical diameters of the aortas of the training data, determined e.g. by the first estimator 308, have been included in the training data used to train the second estimator 310.

[0071] A third option is to determine the risk as one or more rupture locations at which the subject’s aorta is expected to rupture. When performed by the first estimator 308, these locations are obtained from the wall stresses (i.e. tensions) of the subject’s aorta. The locations may be obtained by identifying locations of the aorta based on wall stress values. In an example, the wall stresses may comprise principal stress values and the locations may be identified based on a maximum of the principal stress values or maxima of the principal stress values. Accordingly, it should be noted that there may be a single global maximum principal stress value or a plurality of local maxima principal stress values The location(s) of principal stress values, e.g. the global maximum principal stress value or the plurality of local maxima principal stress values (including the global maximum), may be identified by the first estimator 308.

[0072] When the third option is used by the second estimator 310, one or more rupture locations of the aortas of the training data, determined e.g. by the first estimator 308, have been included in the training data used to train the second estimator 310.

[0073] The type of the combined aortic rupture risk estimate based on the first aortic rupture risk estimate and the second aortic rupture risk estimate. The combined aortic rupture risk estimate comprises the same types of estimates as the first aortic rupture risk estimate and the second aortic rupture risk estimate. When the first aortic rupture risk estimate and the second aortic rupture risk estimate both comprise an estimate of the same type, these may be combined to a single estimate of the same type e.g. by averaging as described above.

[0074] The method of FIG. 1 further comprises outputting 112 the combined aortic rupture estimate. The outputting may comprise writing the combined aortic rupture estimate to the memory 20 of the apparatus 200 (see FIG. 2). Alternatively, or additionally, the outputting may comprise transmitting the combined aortic rupture estimate to the cloud server 50 via the communication interface 40. Yet alternatively or additionally, the outputting may comprise outputting the combined aortic rupture estimate via the user interface 60 of the apparatus 200. The user interface 60 may comprise e.g., a display and / or a speaker configured to output the combined aortic rupture estimate. In the embodiment of FIG. 3, the interface 300 is configured to output the combined aortic rupture estimate.

[0075] The risk estimate may be used by medical professionals to evaluate the need for further follow-ups. Unnecessary follow-up studies can thus be avoided, and subjects in need of further studies or procedures can be identified more efficiently. The limited resources of healthcare systems can thus be targeted in a more efficient manner.

[0076] If desired, the different functions discussed herein may be performed in a different order and / or concurrently with other. Furthermore, if desired, one or more of the above-described functions, such as (steps performed by) either the first estimator or the second estimator, may be optional or may be combined. Although various aspects of the embodiments are set out in the independent claims, other aspects comprise other combinations of features from the described embodiments and / or the dependent claims with the features of the independent claims, and not solely the combinations explicitly set out in the claims. It is also noted herein that while the above describes example embodiments, these descriptions should not be viewed in a limiting sense. Rather, there are several variations and modifications, which may be made without departing from the scope of the present disclosure as defined in the appended claims.

Claims

Claims1 . A computer-implemented method for estimating an aortic rupture risk, the method comprising: obtaining imaging data comprising three-dimensional imaging data representing an aorta of a subject; obtaining additional data related to the subject, wherein the additional data comprises multi-dimensional stress data of the aorta of the subject; fitting an aortic structural model to the imaging data to determine a geometry of the subject’s aorta, and processing the determined geometry and the multi-dimensional stress data by a first estimator configured to determine a first aortic rupture risk estimate and an indicative rupture direction of the aorta; processing the imaging data by a second estimator configured to determine a second aortic rupture risk estimate, wherein the second estimator has been trained using training data comprising three-dimensional imaging data of a plurality of aortas; determining a combined aortic rupture risk estimate based on the first aortic rupture risk estimate and the second aortic rupture risk estimate; and outputting the combined aortic rupture risk estimate including the indicative rupture direction of the aorta.

2. The method of claim 1 , comprising: determining the indicative rupture direction based on a direction of a largest stress of the multi-dimensional stress data.

3. The method of claim 1 or 2, wherein the three-dimensional imaging data representing the aorta of the subject comprises a set of two-dimensional images of the aorta of the subject.

4. The method of any preceding claim, wherein the training data comprises, for each aorta of the plurality of aortas, a set of two-dimensional images representing the aorta.

5. The method of any preceding claim, wherein the imaging data comprises one or more of: magnetic resonance imaging, MRI, data, computed tomography, CT, data, and ultrasound data.

6. The method of any preceding claim, wherein the imaging data further comprises data representing velocity of blood flow in the aorta, and wherein the first estimator is furtherconfigured to determine the first aortic rupture risk estimate also based on the velocity of blood flow in the aorta.

7. The method of any preceding claim, further comprising obtaining further additional data related to the subject, wherein at least one of the first estimator and the second estimator is configured to process the further additional data to determine the respective aortic rupture risk estimate.

8. The method of claim 7, wherein the further additional data comprises one or more of the following: blood pressure, age, sex, and body mass index of the subject.

9. The method of claim 8, further comprising processing, by the first estimator, the blood pressure to determine the first aortic rupture risk estimate.

10. The method of claim 8 or 9, further comprising processing, by the second estimator, the age, sex, and / or body mass index to determine the second aortic rupture risk estimate.11 . The method of any preceding claim, further comprising: computing, by the first estimator, principal wall stresses of walls of the subject’s aorta based on the determined geometry; and determining, by the first estimator, the first aortic rupture risk estimate based on the computed principal wall stresses including the indicative rupture direction.

12. The method of claim 11 , further comprising comparing, by the first estimator, the computed wall stresses to one or more predetermined stress thresholds to determine the first aortic rupture risk estimate.

13. The method of any preceding claim, further comprising determining a wall thickness of the subject’s aorta based on the imaging data, and determining the first aortic rupture risk estimate also based on the determined aortic wall thickness.

14. The method of any preceding claim, wherein the imaging data represents the ascending aorta and / or the abdominal aorta of the subject.

15. The method of any preceding claim, wherein the subject is a human subject.

16. The method of any preceding claim, further comprising: determining, by the second estimator, the second aortic rupture risk estimate based on the first aortic rupture risk estimate, wherein the training data used to train the second estimator further comprises a plurality of first aortic rupture risk estimates that have been determined by fitting the aortic structural model to the three-dimensional imaging data of the plurality of aortas to determine geometries of the of the plurality of aortas, and by processing the determined geometries.

17. The method of any preceding claim 1-15, wherein the first estimator is configured to determine the first aortic rupture risk estimate independently of the second estimator, and the second estimator is configured to determine the second aortic rupture risk estimate independently of the first estimator.

18. The method of any preceding claim, wherein the first aortic rupture risk estimate and / or the second aortic rupture risk estimate comprises an aortic rupture probability.

19. The method of any preceding claim, wherein the first aortic rupture risk estimate and / or the second aortic rupture risk estimate comprises a critical diameter at which the subject’s aorta is expected to rupture.

20. The method of any preceding claim, wherein the first aortic rupture risk estimate and / or the second aortic rupture risk estimate comprises one or more rupture locations at which the subject’s aorta is expected to rupture.21 . The method of any preceding claim, wherein the second estimator comprises a convolutional neural network.

22. An apparatus configured to perform the method of any preceding claim 1-21 .

23. The apparatus of claim 22, comprising at least one processor, at least one memory including computer program code, the at least one memory and the computer program code configured to, with the at least one processor, cause the apparatus to perform the method of any preceding claim 1-21 .

24. A computer program product comprising computer program code configured to, when executed by at least one processor, cause an apparatus to perform the method of any preceding claim 1-21 .

Images