Model-Based Image Segmentation

Abstract

A method and system are presented for mapping boundary detection features of at least one source triangulation mesh of known topology to a target triangulation mesh of arbitrary topology. A region of interest in a volumetric image associated with each triangle of the target triangulation mesh is provided to a feature mapping network. The feature mapping network assigns a feature selection vector to each triangle of the target triangulation mesh. The associated region of interest and the assigned feature selection vector for each triangle of the target triangulation mesh are provided to a boundary detection network. A predicted boundary based on features of the associated region of interest selected by the assigned feature selection vector is obtained from the boundary detection network.

Description

[Technical Field]

[0001] The present invention relates to the field of model-based image segmentation, and in particular to model-based image segmentation using triangulated surface meshes. [Background technology]

[0002] Model-based image segmentation is used in a variety of applications to automatically segment objects from images. For example, model-based image segmentation techniques are used in medical image processing to segment organs or other body parts from volumetric medical images.

[0003] Model-based segmentation techniques using triangulated surface meshes have proven to be fast, robust, and accurate. In these techniques, a shape prior is encoded into the surface mesh, and the mesh is adapted to the image. The shape prior means that objects in the image can be segmented even if some parts of the object's boundary cannot be detected, and the image can be segmented quickly because only image data close to the mesh surface is processed to adapt the mesh to the image. Summary of the Invention [Problem to be solved by the invention]

[0004] Conventional model-based segmentation techniques use image gradients and other features to detect the boundaries of objects in an image. featureRecently, deep learning-based techniques have been developed to improve boundary detection. Brosch, T. et al., 2018 (“Deep Learning-Based Boundary Detection for Model-Based Segmentation with Application to MR Prostate Segmentation”, MICCAI (2018), pp. 512-522) describe a neural network that is trained to predict the boundary of each triangle in a mesh. The boundary of each triangle in a mesh is calculated using the features specific to that triangle. shape The detection is based on a combination of

[0005] However, these techniques require specific shape Since the mesh is trained to detect boundaries by assigning a fixed topology, i.e., a fixed number of vertices and triangles, it requires a surface mesh with a fixed topology. This means that the number of triangles in the new modified topology is shape Because ,is not defined, it prevents any changes to the topology, such as adding triangles to the mesh, removing triangles from the mesh, or otherwise rearranging or refining the topology.,Thus, such model-based segmentation techniques cannot be used to segment images where the shape of a body part differs significantly from the shape prior, for example, due to anatomical abnormalities such as extra vertebrae or artificial abnormalities such as screws from spinal fixation.,Some body parts, such as the rectum, have such large shape flexibility that it is not possible to encode the shape prior to reliably contour them.

[0006] Therefore, there is a need for a model-based segmentation technique that can use triangulated surface meshes of arbitrary topology. [Means for solving the problem]

[0007] The invention is defined by the claims.

[0008] According to an embodiment of the present invention, boundary detection of at least one source triangulated mesh of known topology is performed. shape A computer-implemented method is provided for predicting boundaries of objects within a region of interest, comprising mapping a triangulated mesh of arbitrary topology onto a target triangulated mesh of arbitrary topology.

[0009] This method is shape providing a mapping network with regions of interest in the volumetric image associated with each triangle of a target triangulated mesh; shape Using a mapping network, shape assigning a selection vector to each triangle of the target triangulation mesh; and for each triangle of the target triangulation mesh, determining an associated region of interest and the assigned shape providing the selection vector to a boundary detection network; and for each triangle of the target triangulated mesh, shape of the relevant regions of interest selected by the selection vector shape and obtaining a predicted boundary from the boundary detection network based on

[0010] The proposed concept is that when determining the boundaries of a triangle, triangles of meshes with different topologies that have similar local image environments are subjected to the same algorithm applied to them. shape This is based on the proposal that we should have a triangle-specific selection vector for each triangle, which requires a fixed mesh topology. shape Rather than a boundary detection network using a selection vector, the proposed embodiment is a second network shape Use a mapping network to assign a color to each triangle of a mesh of arbitrary topology based on the appearance of the region of the image surrounding the triangle. shape Assign a selection vector.

[0011] The ability to predict the boundaries of meshes of arbitrary topology means that the shape of the mesh can be adjusted to better delineate structures in the image. Another advantage that may be provided by the proposed embodiments is the ability to predict the boundaries of the boundary detection network and shape Once mapping networks are trained, no training data may be required to use these networks for new topologies.

[0012] In some embodiments, the method may further comprise obtaining a target triangulated mesh of any topology by adding triangles to or removing triangles from a triangulated mesh of known topology. In this way, boundaries may be predicted for a mesh having triangles added to or removed from it, for example to locally refine the mesh to improve the delineation of body parts in volumetric medical images, or to add structure to the mesh to contain anomalies in the shape of a body part.

[0013] In another embodiment, the method may further comprise obtaining a target triangulated mesh of arbitrary topology by obtaining a segmentation of the volumetric image using a voxel-by-voxel segmentation technique, and generating the target triangulated mesh based on the obtained segmentation. In this way, boundaries for structures for which there is no existing triangulated surface mesh can be obtained. This allows for the delineation of body parts with large geometric flexibility, such as the rectum.

[0014] The region of interest associated with each triangle can be positioned according to the normal of that triangle. shape The mapping network will assign the same number of triangles to triangles whose associated regions of interest look similar when the triangles are positioned at the same angle. shape A selection vector can be assigned.

[0015] shape The mapping network can be trained using a first training algorithm configured to receive a training input and an array of known outputs, where the training input includes a region of interest associated with triangles of a mesh of an arbitrary topology, and the known output includes a known boundary of the region of interest.

[0016] In some embodiments, the first training algorithm further comprises: for each region of interest associated with a triangle of a mesh of any topology: shape A step of assigning a selection vector and, for each triangle of a mesh of any topology, assigning an associated region of interest and the assigned shape providing the selection vector to a boundary detection network; obtaining a predicted boundary from the boundary detection network for each triangle of a mesh of any topology; and based on the predicted boundary from the boundary detection network and the known boundary, shape and training the weights of the mapping network. shape The mapping network is used to predict boundaries within a region of interest. shape Select the best region of interest and shape It can be trained to learn the correspondence between the selection vectors.

[0017] In some embodiments, the method further comprises: shape Providing a mapping network with a set of standardized coordinates for each region of interest.

[0018] In some embodiments, the method further comprises: shape There may be the step of providing a mapping network with the relative position of each triangle of the target triangulated mesh.

[0019] The boundary detection network can be trained using a second training algorithm configured to receive a training input and an array of known outputs, where the training input includes training images having a mesh of known topology and the known outputs include known boundaries, and the training algorithm trains the boundary detection network to detect boundaries in triangles of the mesh of known topology.

[0020] According to another aspect of the invention, at least one source triangulation mesh of known topology is shape to a target triangulated mesh of arbitrary topology; and segmenting an object from the volumetric image using the target triangulated mesh.

[0021] Such proposed methods can enable segmentation of objects that cannot be segmented by existing model-based segmentation techniques. For example, model-based segmentation techniques that use fixed-topology triangulated surface meshes are not suitable for segmenting anatomical structures with abnormalities and body parts that may have large variations in shape. shape By mapping ,to a mesh of arbitrary topology, these structures can be segmented.

[0022] According to another aspect of the invention there is provided a computer program comprising code means for carrying out the above method when the program is run on a processing system.

[0023] According to another aspect of the present invention, a region of interest in a volumetric image associated with each triangle of a target triangulated mesh is determined by: shape providing it to a mapping network; shape For each triangle of the target triangulated mesh using the mapping network shape assigning a selection vector and, for each triangle of the target triangulated mesh, assigning an associated region of interest and the assigned shape providing the selection vector to a boundary detection network; and for each triangle of the target triangulated mesh, shape of the relevant regions of interest selected by the selection vector shape and adapted to obtain a predicted boundary from the boundary detection network based on:

[0024] In some embodiments, the processing system is further adapted to position the region of interest associated with each triangle according to the normal of that triangle.

[0025] In some embodiments, the processing system calculates, for each triangle of the target triangulated mesh, at least one of a set of normalized coordinates of an associated region of interest and a relative position of the triangle by: shape It is further adapted to provide to a mapping network.

[0026] According to another aspect of the present invention, there is provided a model-based image segmentation system, the system having the aforementioned processing system further configured to segment an object from a volumetric image using the target triangulation mesh, and a user interface configured to receive and display the segmented image of the object from the processing system.

[0027] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.

[0028] For a better understanding of the present invention and to show more clearly how the same may be carried into effect, reference will now be made, by way of example only, to the accompanying drawings in which: [Brief explanation of the drawings]

[0029] [Figure 1] Illustration of an object segmented by two meshes of different topologies. [Figure 2] Illustration of the triangular region of interest of the mesh in Figure 1, positioned according to the triangle normals. [Figure 3] 1 is a flow diagram of a method for mapping the shape of at least one source triangulated mesh of known topology to a target triangulated mesh of arbitrary topology, according to one embodiment of the present invention. [Figure 4] 1 is a flow diagram of a method for training a shape mapping network, according to one embodiment of the present invention. [Figure 5] 1 is a flow diagram of a method for model-based image segmentation, according to one embodiment of the present invention. [Figure 6] 1 is a diagram of a processing system according to one embodiment of the present invention. [Figure 7] FIG. 1 is a diagram of a model-based image segmentation system according to one embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0030] The present invention will now be described with reference to the drawings.

[0031] It should be understood that the detailed description and specific examples, while indicating exemplary embodiments of the system and method, are intended for purposes of illustration only and are not intended to limit the scope of the invention. These and other features, aspects, and advantages of the system and method of the present invention will become better understood from the following description, appended claims, and accompanying drawings. It should be understood that the drawings are merely schematic and are not drawn to scale. It should also be understood that the same reference numerals are used throughout the drawings to indicate the same or similar parts.

[0032] According to the proposed concept, boundary detection of at least one source triangulation mesh of known topology is performed. shapeAn approach is provided for mapping a region of interest in a volumetric image associated with each triangle of the target triangulation mesh to a target triangulation mesh of arbitrary topology. shape provided to the mapping network. shape The mapping network shape A selection vector is assigned to each triangle of the target triangulation mesh. The assigned vector is used for each triangle of the associated region of interest and the target triangulation mesh. shape The selection vector is provided to the boundary detection network. shape of the relevant regions of interest selected by the selection vector shape A predicted boundary based on is obtained from the boundary detection network.

[0033] Exemplary embodiments can be used, for example, in model-based image segmentation systems, such as medical imaging analysis systems. shape Selection vectors can be used for triangles of meshes with different topologies that have similar local environments, and the neural network can then compare the local environments of the triangles. shape This may be based at least in part on the realization that the algorithm may be trained to learn correspondences between selected vectors.

[0034] 1 shows an object 110 segmented by two meshes of different topologies. In the top image, the object 110 is segmented by a first mesh 120. A region of interest 130 associated with a triangle 125 of the first mesh 120 encompasses the triangle 125 and a portion of the boundary between the object 110 and an adjacent structure 140. In the bottom image, the object 110 is segmented by a second mesh 150, which has more surface triangles than the mesh 120 and therefore a different topology. A second region of interest 160 associated with a triangle 155 of the second mesh 150 encompasses the triangle 155 and a portion of the boundary between the object 110 and a second adjacent structure 170.

[0035] 2 shows regions of interest 130 and 160 positioned according to the normals of triangles 125 and 155, respectively. As can be seen from FIG. 2, triangles 125 and 155 have similar associated regions of interest when positioned in the same direction. Therefore, the same boundary detection shape can be used to predict the boundaries in both triangles 125 and 155. However, current boundary detection networks The unique shape of a triangle This means that a boundary detection network trained using mesh 120 cannot be used to detect the boundaries of triangles in mesh 150.

[0036] The present invention performs boundary detection of triangles 125 of mesh 120. shape to triangles 155 of mesh 150, we recognize that the boundary detection network trained on mesh 120 can be used to detect boundaries in triangles 155.

[0037] FIG. 3 illustrates a method for generating a triangulated mesh of at least one source triangulated mesh of known topology according to one embodiment of the present invention. shape 3 shows a computer-implemented method 300 for mapping N to a target triangulation mesh of arbitrary topology. At least one source triangulation mesh can be an average mesh generated from N meshes of the same known topology, each of the N meshes being a (ground truth) segmentation of N patient images.

[0038] The method 300 begins at step 320, where a region of interest in the volumetric image associated with each triangle of the target triangulated mesh is determined. shape provided to the mapping network.

[0039] In step 340, shapeThe mapping network assigns a region of interest (ROI) to each triangle of the target triangulation mesh based on the associated ROI. shape Assign a selection vector.

[0040] In step 360, the associated region of interest and the assigned triangles for each triangle of the target triangulated mesh are shape The selection vector is provided to a boundary detection network.

[0041] In step 380, the assigned shape of the relevant regions of interest selected by the selection vector shape Based on this, the predicted boundary of each triangle of the target triangulated mesh is obtained from the boundary detection network.

[0042] The target triangulation mesh can be obtained by modifying a triangulation mesh of known topology or by generating a new mesh from scratch. For example, triangles can be added to or removed from existing shape priors encoded in the MBS model to locally refine the mesh or add additional structure to the mesh. Alternatively, an initial (coarse) segmentation technique, such as a voxel-wise segmentation technique, can be used to obtain an initial segmentation of the volumetric image, and the target triangulation mesh can be obtained by generating a mesh based on the initial segmentation, for example, by using marching cubes or some other mesh generation technique. Suitable initial (coarse) segmentation techniques will be apparent to those skilled in the art and may include the use of deep learning networks.

[0043] The volumetric image may be, for example, a volumetric medical image, such as a computed tomography (CT) image, a magnetic resonance (MR) image, a nuclear medicine image such as a positron emission tomography (PET) image or a single photon emission computed tomography (SPECT) image, or a volumetric ultrasound image.

[0044] Each triangle of the target triangulation mesh is associated with a region of interest in the volumetric image. Each region of interest includes its associated triangle, the portion of the object that the target mesh outlines, and a portion of the object's periphery. Each region of interest can be positioned so that the associated triangles are at the same angle. For example, each region of interest may be positioned according to the normal of its associated triangle.

[0045] Each region of interest is shape Each region of interest can be accompanied by additional information, for example, if the objects to be contoured are pre-registered to a common coordinate system, such as atlas-based registration for volumetric medical images. shape The mapping network can be provided with a standardized set of coordinates for each region of interest. shape The mapping network can be provided with the relative positions of the triangles of the target triangulated mesh associated with each region of interest, for example, the spherical coordinates of each triangle, or the distance vector from each triangle to the centroid of the mesh.

[0046] shape The mapping network is shape An artificial neural network is trained to assign a selection vector. An artificial neural network is a type of machine learning algorithm, i.e., a self-training algorithm that processes input data to generate or predict output data. shapeThe input data of the mapping network includes regions of interest associated with triangles of a mesh of arbitrary topology, and the output data is the assigned shape Contains the selection vector.

[0047] The structure of an artificial neural network (or simply neural network) is inspired by the human brain. A neural network consists of layers, each containing multiple neurons. Each neuron contains a mathematical operation. In particular, each neuron can contain different weighted combinations of one type of transformation (e.g., the same type of transformation, such as sigmoid, but with different weightings). In the process of processing input data, each neuron's mathematical operation is performed on the input data to produce a numerical output, and the output of each layer in the neural network is fed sequentially to the next layer. The final layer provides the output.

[0048] Methods for training neural networks are well known. Typically, such methods involve obtaining a training dataset containing training input data entries and corresponding training output data entries. An initialized machine learning algorithm is applied to each input data entry to generate a predicted output data entry. The error between the predicted output data entry and the corresponding training output data entry is used to modify the machine learning algorithm. This process can be repeated until the error converges and the predicted output data entry is sufficiently similar to the training output data entry (e.g., ±1%). For example, the weighting of each neuron's mathematical operation can be modified until the error converges. This is commonly known as a supervised learning technique. Known methods for modifying neural networks include gradient descent, backpropagation algorithms, etc.

[0049] shapeThe mapping network can be trained using training input data entries corresponding to regions of interest associated with triangles of a mesh of any topology, and training output data entries corresponding to known boundaries of the regions of interest. Thus, the training data set can include a set of training images with adapted meshes of any topology and annotated to depict boundaries of objects in the images.

[0050] FIG. 4 illustrates a method for manufacturing a semiconductor device according to an embodiment of the present invention. shape 4 shows a computer-implemented method 400 for training a mapping network.

[0051] The method 400 begins at step 410, shape The mapping network receives regions of interest from a training dataset that are associated with triangles of a mesh of arbitrary topology. The regions of interest can be positioned so that their associated triangles are at the same angle. In step 420, for each region of interest: shape The mapping network is based on the triangles associated with the region of interest. shape Assign a selection vector.

[0052] In step 430, the associated region of interest and the associated shape The selection vector is provided to a boundary detection network.

[0053] In step 440, the predicted boundary for each triangle is obtained from the boundary detection network. shape of the relevant regions of interest selected by the selection vector shape Based on this, we predict the boundary for each triangle.

[0054] In step 450, based on the obtained predicted boundaries and known boundaries from annotations on the images that make up the training dataset, shapeThe weights of the mapping network are trained. This step involves calculating the distance between each predicted boundary and its corresponding known boundary, and then applying the calculated distances to the network. shape and training the weights of the mapping network.

[0055] These steps are: shape Assigned by the mapping network shape The selection vector can be used iteratively until the boundaries predicted by the boundary detection network are sufficiently similar to their corresponding known boundaries (eg, until they are within 1% of each other).

[0056] Returning to FIG. 3, the trained shape assigned to each triangle of the target triangulated mesh by the mapping network shape The selection vector, along with the associated region of interest for each triangle, is provided to a boundary detection network in step 360 .

[0057] The boundary detection network is a triangle-specific shape of the images selected by the selection vector shape A second artificial neural network is trained to predict boundaries in the volumetric image at triangles of a mesh of known topology based on

[0058] The boundary detection network can be trained using training input data entries corresponding to at least one training image having a mesh of known topology and training output data entries corresponding to known boundaries in the training image. Thus, the training dataset can include a set of training images having an adapted mesh of known topology and annotated to depict boundaries of objects in the images. The training images can be shapeThese can be the same images used to train the mapping network, but with a mesh of known topology rather than an arbitrary topology adapted to them. The boundary detection network can be trained using the method described in Brosch, T. et al., 2018.

[0059] In this way, the source mesh of a known topology can be thought of as the average mesh generated from all N meshes of the same known topology used to train the boundary detection network, and assigned to each triangle of the source mesh. shape is over all N meshes shape Thus, the source mesh of known topology is a representation of all meshes of known topology used to train the boundary detection network, and is assigned to each triangle of the source mesh. shape represents the best match of that triangle across all N meshes. shape is.

[0060] In step 380, a boundary detection network provides predicted boundaries for each triangle of a target triangulated mesh of any topology. The boundary detection network uses a set of algorithms specific to the triangles of a mesh of known topology that are used to train the boundary detection network. shape Using the selection vector to predict the boundaries of triangles in a target mesh of arbitrary topology shape You can select shape Mapping networks are shape Which of the selection vectors is used to predict the boundary at each triangle of the target triangulated mesh? shape This is because we have decided which is the best choice.

[0061] In other words, shape The mapping network is triangle-specific. shapeThe selection vector is mapped onto the target triangulation mesh, and the boundary detection network shape The selection vector is used to determine the boundary of each triangle in the target triangulated mesh. shape Next, the boundary detection network selects which of the selected regions of interest associated with each triangle of the target mesh should be used. shape Predict the boundary based on

[0062] FIG. 5 illustrates a computer-implemented method 500 for model-based image segmentation, according to one embodiment of the present invention.

[0063] The method begins at step 510 by computing a triangulated mesh of at least one source triangulated mesh of known topology. shape is mapped to a target triangulated mesh of any topology according to any of the methods described above.

[0064] In step 520, the target triangulation mesh is used to segment an object from the volumetric image. For example, if the volumetric image is a volumetric medical image, the target triangulation mesh may be used to segment an organ or some other body part from the volumetric medical image. Methods for using triangulation meshes to segment an object from a volumetric image are known and will be apparent to those skilled in the art.

[0065] 6 illustrates a processing system 600 according to one embodiment of the present invention. shape It comprises a mapping network 640 and a boundary detection network 660 .

[0066] For each triangle of the target triangulation mesh, an associated region of interest 620 in the volumetric image is shape provided to the mapping network 640; shape A selection vector 662 is assigned to each triangle based on its associated region of interest 620 .

[0067] Associated regions of interest 620 and assigned for each triangle of the target triangulated mesh shape The selection vector 662 is provided to a boundary detection network 660. The boundary detection network 660 detects the assigned shape of the associated region of interest 620 selected by the selection vector 662 shape 664 is used to determine the predicted boundary 680 for each triangle of the target triangulated mesh.

[0068] In some embodiments, the processing system 600 is further adapted to position the region of interest 620 associated with each triangle such that each triangle is at the same angle. For example, the processing system can be adapted to position the region of interest 620 associated with each triangle according to the normal of that triangle.

[0069] In some embodiments, the processing system 600 shape The processing system 600 may be further adapted to provide additional information to the mapping network 640. For example, the processing system 600 may generate a set of standardized coordinates for each region of interest 620. shape and / or to provide the relative position of each triangle of the target triangulated mesh to the mapping network 640. shape It may be adapted to provide to a mapping network 640 .

[0070] 7 illustrates a model-based segmentation system 700 according to one embodiment of the present invention, which comprises a processing system 600 further adapted to segment an object from a volumetric image using a target triangulation mesh to generate a segmented image 710, as described above, and a user interface 720.

[0071] User interface 720 is configured to receive segmented image 710 from processing system 600 and display segmented image 710. The user interface may further receive and display the original volumetric image and / or the target triangulated mesh from processing system 600.

[0072] It will be understood that the disclosed methods are computer-implemented methods, and therefore also the concept of a computer program comprising code means for implementing the described methods when said program is run on a processing system.

[0073] Those skilled in the art can readily develop a processor to perform any of the methods described herein. Accordingly, each step in the flowchart may represent a different action performed by a processor and may be performed by a respective module of the processor.

[0074] As described above, the system utilizes a processor to perform data processing. The processor may be implemented in a variety of ways using software and / or hardware to perform the various functions required. The processor typically employs one or more microprocessors that can be programmed using software (e.g., microcode) to perform the necessary functions. The processor may be implemented as a combination of dedicated hardware to perform some functions and one or more programmed microprocessors and associated circuitry to perform other functions.

[0075] Examples of circuitry that may be used in various embodiments of the present disclosure include, but are not limited to, conventional microprocessors, application specific integrated circuits (ASICs), and field programmable gate arrays (FPGAs).

[0076] In various implementations, the processor may be associated with one or more storage media, such as volatile and non-volatile computer memory, e.g., RAM, PROM, EPROM, and EEPROM. The storage media may be encoded with one or more programs that, when executed on the one or more processors and / or controllers, perform the necessary functions. The various storage media may be attached to the processor or controller, or may be transportable, such that one or more programs stored on the storage media can be loaded into the processor.

[0077] Variations to the disclosed embodiments can be understood and realized by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality. A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain means are recited in mutually different dependent claims does not indicate that a combination of these means cannot be used to advantage. A computer program can be stored / distributed on an appropriate medium, such as an optical storage medium or a solid-state medium, supplied together with or as part of other hardware, but can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems. When the term "adapted to" is used in the claims or the description, the term "adapted to" has the same meaning as the term "configured to." Any reference signs in the claims should not be construed as limiting the scope.

Claims

1. 1. A computer-implemented method for predicting boundaries of objects within a region of interest, the method being suitable for mapping shapes that detect the boundaries of triangles of at least one source triangulated mesh of known topology onto triangles of a target triangulated mesh of arbitrary topology, the method comprising: providing a region of interest in a volumetric image associated with each triangle of a target triangulation mesh to a shape mapping network, the region of interest comprising each triangle and a portion of a boundary of the object outlined by the target triangulation mesh, the shape mapping network being configured to assign, based on the region of interest, to each triangle of the target triangulation mesh a shape selection vector for selecting which shape should be used to determine a boundary in each triangle of the target triangulation mesh, thereby mapping shapes that detect the boundary of triangles of the at least one source triangulation mesh to triangles of the target triangulation mesh; assigning the shape selection vector to each triangle of the target triangulated mesh using the shape mapping network; for each triangle of the target triangulated mesh, providing the region of interest and the assigned shape selection vector to a boundary detection network, the boundary detection network being configured to detect a predicted boundary for each triangle of the target triangulated mesh based on the shape of the region of interest selected by the assigned shape selection vector; for each triangle of the target triangulated mesh, obtaining a predicted boundary from the boundary detection network based on the shape of the region of interest selected by the assigned shape selection vector; 1. A computer-implemented method comprising:

2. 2. The computer-implemented method of claim 1, further comprising the step of obtaining the target triangulated mesh of an arbitrary topology by adding triangles to or removing triangles from a triangulated mesh of a known topology.

3. obtaining said target triangulated mesh of any topology; obtaining a segmentation of the volumetric image using a voxel-wise segmentation technique; generating the target triangulated mesh based on the obtained segmentation; The computer-implemented method of claim 1 further comprising:

4. The computer-implemented method of claim 1 , wherein the region of interest associated with each triangle is oriented according to the normal of the triangle.

5. 5. The computer-implemented method of claim 1, wherein the shape mapping network is trained using a first training algorithm configured to receive a training input and an array of known outputs, the training inputs including regions of interest associated with triangles of a mesh of arbitrary topology, and the known outputs having known boundaries of the regions of interest.

6. The first training algorithm further comprises: assigning a shape selection vector to each region of interest associated with a triangle of a mesh of any topology; for each triangle of a mesh of any topology, providing the region of interest and the assigned shape selection vector to the boundary detection network; obtaining a predicted boundary for each triangle of a mesh of any topology from said boundary detection network; training weights of the shape mapping network based on the predicted boundaries from the boundary detection network and the known boundaries; 6. The computer-implemented method of claim 5, configured to perform:

7. A computer-implemented method as described in any one of claims 1 to 6, further comprising the step of providing a set of coordinates in a common coordinate system for each region of interest to the shape mapping network.

8. The computer-implemented method of claim 1 , further comprising providing the relative position of each triangle of the target triangulated mesh to the shape mapping network.

9. 9. The computer-implemented method of claim 1, wherein the boundary detection network is trained using a second training algorithm configured to receive a training input and an array of known outputs, the training input including training images having a mesh of known topology, and the known outputs including known boundaries.

10. 1. A computer-implemented method for model-based image segmentation, comprising: A method according to any one of claims 1 to 9, comprising the steps of mapping the shape of at least one source triangulated mesh of known topology onto a target triangulated mesh of arbitrary topology; segmenting an object from the volumetric image using the target triangulated mesh; 1. A computer-implemented method comprising:

11. A computer program comprising code means for performing the computer implemented method of any one of claims 1 to 10.

12. 1. A processing system for predicting boundaries of objects in a region of interest, comprising: providing a region of interest in the volumetric image associated with each triangle of a target triangulated mesh to a shape mapping network, the region of interest comprising each triangle and a portion of a boundary of an object outlined by the target triangulated mesh, the shape mapping network being configured to assign, based on the region of interest, to each triangle of the target triangulated mesh a shape selection vector for selecting which shape should be used to determine a boundary in each triangle of the target triangulated mesh, thereby mapping shapes that detect the boundary of at least one triangle of the source triangulated mesh to triangles of the target triangulated mesh; assigning said shape selection vector to each triangle of said target triangulated mesh using said shape mapping network; for each triangle of the target triangulated mesh, providing the region of interest and the assigned shape selection vector to a boundary detection network, the boundary detection network being configured to detect a predicted boundary for each triangle of the target triangulated mesh based on the shape of the region of interest selected by the assigned shape selection vector; for each triangle of the target triangulated mesh, obtaining the predicted boundary from the boundary detection network based on the shape of the region of interest selected by the assigned shape selection vector; A processing system that runs

13. The processing system of claim 12 , further configured to perform the step of determining an orientation of the region of interest associated with each triangle according to a normal of that triangle.

14. 14. The processing system of claim 12 or 13, further configured to perform the step of providing, for each triangle of the target triangulated mesh, to the shape mapping network at least one of a set of coordinates in a common coordinate system for the region of interest associated with each triangle and the relative position of the triangle.

15. 1. A model-based image segmentation system, comprising:

15. The processing system of any one of claims 12 to 14, further configured to segment an object from the volumetric image using the target triangulated mesh; a user interface that receives and displays the segmented image of the object from the processing system; A system having:

Images