Field of the Invention This invention pertains in general to the field of three-dimensional image analysis. More particularly, the invention relates to a method and device for extracting blood vessel networks from medical image data.

Background of the Invention

Medical image data can be difficult to interpret due to the complex anatomies captured in the data. Blood vessels in the body are part of complex three-dimensional networks of vessels and the superposition in 2-dimensional mappings of such networks in medical image data pose significant difficulties in applications where it is desirable to resolve the true three-dimensional anatomy, such as in advanced biometrics, or for extraction of physiological data, or for health monitoring.

Previous techniques for resolving the aforementioned anatomies lack sufficient accuracy in distinguishing the relevant structures from the irrelevant ones, and/or rely on inherently deficient techniques that are unable to reconstruct the three-dimensional reality.

Hence, an improved method and device would be advantageous and in particular allowing for resolving the true three-dimensional structures in medical image data.

Summary of the Invention Accordingly, embodiments of the present invention preferably seeks to mitigate, alleviate or eliminate one or more deficiencies, disadvantages or issues in the art, such as the above-identified, singly or in any combination by providing a device and a method according to the appended patent claims.

According to a first aspect of the invention a method comprising extracting a vessel network from medical image data is provided. The method comprises extracting linear features of said vessel network from the image data, the linear features comprising a 2D projection of said vessel network, and determining a verified tree model of the vessel network by validating the linear features against a physical criterion parameter related to the vessel network.

According to a second aspect of the invention a device adapted to extract a vessel network from medical image data is provided, the device comprising an image processing unit configured to extract linear features of said vessel network from the image data, the linear features comprising a 2D projection of said vessel network, and a second processing unit configured to calculate a verified tree model of the vessel network by validating the linear features against a physical criterion parameter related to the vessel network.

Further embodiments of the invention are defined in the dependent claims, wherein features for the second and subsequent aspects of the invention are as for the first aspect mutatis mutandis.

Embodiments of the invention provide for resolving the three-dimensional anatomies with high accuracy in distinguishing the relevant structures from the irrelevant ones.

Embodiments of the invention provide for reconstructing the three-dimensional reality of overlaid blood vessel networks in two-dimensional image data.

It should be emphasized that the term "comprises/comprising" when used in this specification is taken to specify the presence of stated features, integers, steps or components but does not preclude the presence or addition of one or more other features, integers, steps, components or groups thereof.

Brief Description of the Drawings

These and other aspects, features and advantages of which embodiments of the invention are capable of, will be apparent and elucidated from the following description of embodiments of the present invention, reference being made to the accompanying drawings, in which

Fig. 1 is an illustration of medical image data displaying vessel networks;

Fig. 2 is an illustration of the linear features of the vessel networks in the image data of

Fig. 1 ;

Fig. 3 is another illustration of the linear features of the vessel networks in the image data of Fig. 1 being part of the illustration of a verified tree model of the vessel network according to an embodiment of the invention;

Fig. 4 is a magnified portion of the linear features in Fig. 3;

Fig. 5 is another illustration of the linear features of the vessel networks of Fig. 4; Fig. 6 is another illustration of the linear features of the vessel networks of Fig. 3 being part of the illustration of a verified tree model of the vessel network according to an embodiment of the invention, including an illustration of spatially transforming a first blood vessel network in relation to a secondary vessel network;

Fig. 7 is a flow chart illustrating a method according to an embodiment of the invention; Fig. 8 is an illustration of a device according to an embodiment of the invention performing a method according to an embodiment of the invention as illustrated in Fig. 7;

Fig. 9 is an illustration of false features; left Fig.: false feature consisting of one bifurcation and one ending; and right Fig.: false feature consisting of two bifurcations due to a 'crossing' of veins on different layers;

Fig. 10a-d is an illustration of typical patterns of veins;

Fig. 11 is a schematic view of the feature extraction part of a previous biometric authentication method;

Fig. 12 is a schematic view of the feature extraction part of an authentication method according to an embodiment of the invention, i.e. determination of a verified tree model according to an embodiment of the invention, where the "3D modeling" unit in Fig. 12 may be comprised in a device as illustrated in Fig. 8 performing the method as illustrated in Fig. 7;

Figs. 13a-c is an illustration of a vein that passes two layers (a) may give rise to false bifurcation, marked with a hollow circle, in the 2D view (b);

Fig. 14 is an illustration of a tree structure; and

Fig. 15 is another illustration of the linear features of the vessel.

Description of embodiments Specific embodiments of the invention will now be described with reference to the accompanying drawings. This invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these

embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art. The terminology used in the detailed description of the embodiments illustrated in the accompanying drawings is not intended to be limiting of the invention. In the drawings, like numbers refer to like elements. The following description focuses on an embodiment of the present invention applicable to reconstructing the three-dimensional reality of overlaid blood vessel networks in medical image data. However, it will be appreciated that the invention is not limited to this application but is an example of the use of structural and physical properties to reconstruct the multi-dimensional reality of medical and other transport networks at arbitrary size through the 2D-view of limited visibility.

The topography of blood veins

Fig. 1 is a medical image 202 of a vein pattern in an organ. The vein pattern can be seen as networks 201 , 212, of blood vessels 213. A vessel network may in this context be construed as groups of vessels or veins that distribute a flow of blood in a tree-wise fashion e.g. towards body extremities.

The topography of blood veins is based on the circulation through the body bringing oxygen to the body extremities. They come from the arteries (the highway along the organs), while locally the veins spread out to the vessels where the oxygen is delivered. Therefore each vein comes from a single location at the artery and spreads in a tree-wise fashion. This looks like a binary tree, where the structure is repeatedly split in two (bifurcation) till the end. In other words, the only physically plausible nodes are 'bifurcations' and 'ends'. Bypasses can occur to guarantee access to extremities in the forward pass, but this does not change the overall nature of the tree- wise distribution.

The typical image of a body area shows some vein parts in a 2-dimensional mapping. Hence irregular or 'false' features will pop up because of crossing or sub-depth veins. From the experience of blood sampling, there are easy places to locate them (elbow and pulse) and difficult ones (hand). This has to do with the structure itself but also with the presence of water in the tissue that scatters the incoming light, which diffuses the visibility in the image. Furthermore, different details will be at different depth. Being slightly out of focus and having more tissue (to pass through) adds to the lack of definition in the image. Lastly, the size of the veins is usually influenced by health factors. Therefore a reliable model with reproducible measures is important. Vein diameter sets together with blood pressure the blood flow. Where flow can be measured in tact with the heart beat, an accurate estimation of vein diameters brings blood pressure. Doing this for a number of veins from the image raises the accuracy of the calculation.

The 2-dimensional image is used for person verification. The move to a 3-dimensional model removes the false features and brings identification and authentication in reach. It is possible to distinguish between: (a) verification, where it must be certified that a person is who he states to be, (b) identification, where it must be found out who the person is, and (c)

authentication, where it must be guaranteed that the presented person is one with sufficient permission. Measuring the blood flow dynamics extends the model for health identification. As the blood flow will considerably go down in the vessels, where the oxygen is delivered, little dynamics will be found in the final texture. In the following it is discussed how such 3-dimensional detail can be retrieved, for example generally described by:

^• Building a mathematical binary tree model for analysis of the vein structure;

^• and complete the mathematical binary tree model from analysis of the image quality while taking physical plausibility into account;

^• and reason from dynamic flow analysis to provide for metricity in the model. False features

Prior techniques come with the drawback that all features are classified as endings and bifurcations. Then a post-analysis shows that not all features are correctly accepted as some are not physically plausible as vein parts. Such are called 'false' or 'abnormal' and come in two categories, procedural and structural. An improper skeletonizing procedure of the original, noisy image brings false features in pairs such as two false bifurcations, two false endings and one false bifurcation with one false ending (Fig. 9 left). These are the procedural abnormal features. In Fig. 9 right we find mixing procedural with structural abnormality in the same figure. This type of false features is generated only when veins on different layers cross each others on the 2D view. In this case, it indicates overlapping of veins.

The procedural false features, mainly characterized to be topographically closer than normal to each other, can be eliminated using a distance criterion, or by back annotating in the original image, as they are ghosts. On the other hand, the structural abnormal feature is not a ghost and can be clearly identified in the original image. Their existence comes to pass when a number of normal features are (partly) overlapping in the current view. Therefore, the view on a feature comprises of either a singular item or a composition of parts in the 3D reality. For instance, crossing veins are not plausible to be a fork plus merge, but rather two veins at different depth, c.f. Fig. 10a-d showing a merge (10a), i.e. skin to body transport, a fork/split/bifurcation (10b), i.e. body to skin transport, a beginning/ending (10c), and crossing veins at different depth (1 Od). In other words, structural false features result from mapping the 3D reality on the 2D image. Veins running on different depths are seen as 'crossing' each other in the 2D view. In a 2D presentation, a crossing of branches that belong to different vein trees will lead to abnormal features in the detection procedure (Fig. 13a-c). Branch structure

It should be noted that the conventional approach in previous techniques is to eliminate all the abnormal features in order to achieve accurate feature detection (Fig. 11). From the original image, the features are detected and the false ones are blanked out, leaving only 2D recognizable parts. On the contrary, embodiments of the present invention may comprise reconstructing a 3-dimensional reality, e.g. by method 100, and device 900 which may comprise a

3D modeling unit (Figure 12), where a verified tree model is determined of the 3D blood vessel network populated by only normal (or verified) features, starting from the 2-dimensional view. Consequently, next to the overall image with recognizable features, a tree list is generated to ease further manipulation. The binary tree is a common search mechanism in computer algorithms. It is used to facilitate a search going from the root (the trunk of the tree) to the end terms (the leaves of the tree). Going from the root, the search is made by repeatedly deciding to go left or right in every split. The search on a tree is notorious, as without precautions it will take an exponential amount 5 of time. Therefore the algorithm may be equipped with a branch-cutting mechanism, which

decides whether a branch is predicted to be meaningful. As this mathematical model will be used for searching through the problem space to resolve the inconsistencies in the physical tree that represents the blood vein structure, such information will link between the two trees and remember whether false features are still unresolved in that branch.

o Where the mathematical notion of a binary tree is planar, i.e. can be depicted on a sheet of paper without crossings, the vein tree is a 3-dimensional physical reality and therefore not necessarily planar. In fact, the problem exists because the vein tree is hardly ever planar. Therefore, in this procedure a planar copy may gradually be made of the vein tree for matching and measuring, while rigorously noting the correspondence of these two trees. In this way, quick 5 searching may be assured on the abstract mathematics and validation of the real physics. The analysis of vein structure phenomena may be pursued with the only purpose of being able to make the mathematical model for algorithmic manipulation.

The branch-cutting prediction procedure may benefit from bifurcations being detected a priori (Fig. 5). The mathematical model and the vein structure are identical in the beginning of the o procedure, then gradually the abnormal features will be resolved and the mathematical model adjusted. Once the bifurcations, both real and false ones (also described as real crossings and false crossings below), are detected, it is easy to traverse the mathematical tree from root to leaves while the annotation to the original vein image may help further analysis.

The notable difference between the mathematical binary tree and blood veins is in the 5 2D structure. The mathematical tree can be depicted in 2D without crossings; an understandable tree picture will therefore have no crossings. In contrast, the blood vein structure is in 3D and therefore crossings will appear in a 2D presentation. These crossings may be utilized for finding the 3D reality from the 2D presentation according to an embodiment of the invention.

Consequently the nodal validation is performed on the presence of abnormal structure in a

0 branch. For instance, when the algorithm reaches point C in Figure 5, it discovers a 4-way

branching that result from overlapping two trees (also referred to as vessel network 201 , 212); the branching is not included in the structure of the upper tree but in the structure of the lower tree. Note that a 4-way crossing gives raise to a bifurcation-bifurcation false feature as depicted in Fig. 9 right. The algorithm marks the branches in point C as belonging to two different layers, and thus 5 builds a 3D model in that point. Image sharpness.

The tree-like vein structure has a depth and consequently a vein outside the lens focus will be less sharp and a deeper laying vein will show with more scattering of the reflected light. A physically plausible model of the 3D reality needs to have continuity in such characteristics. Conventionally, applying methods such as Unsharp Mask easily enhances the sharpness, or more precisely the acutance, of the captured image. In our invention, however, the change of sharpness on the same branch of a tree is a clear indicator for the vein changing depth. Note that a vein can be not clearly visible. This does not mean that the vein is not there. Looking at the pixel values, there can be indication of some structure. If such structure connects two neighboring ends, there is reason to assume a deeper vein to be present.

Such information is lost if multiple images of the same scenery are taken with varying intensity of the reflected light, as usually done in 3D capturing such as for plenoptic devices (Adelson, E.H. and Wang, J.Y.A. Single lens stereo with plenoptic camera. IEEE Transactions on PAMI, Vol. 14, No. 2, pp. 99 - 106, 1992). For instance, a layered multiple image capturing will lead to false endings in Fig. 13a. Figs. 13a-c illustrates that a vein passing two layers (a) may give rise to false bifurcation, marked with a hollow circle, in the 2D view (b). A traditional 3D handling removes the false bifurcation but introduces a number of false endings, marked with solid circles, at least one per layer (c). Embodiments of the invention comprise focusing on the change in sharpness between layer 1 and layer 2 and hence reconstruct the 3D scenery as actually existing in reality.

Embodiments of the invention may comprise the steps of;

Step 1. Pre-process the image using line thickness and pixel intensity;

Step 2. Separate the image into independent vein structures that are well contained in the picture discarding spillover parts that are largely outside the viewed area;

Step 3. Validate the identified trees for physical plausibility. If not successful, retry step

2.

Bandwidth Analysis.

In a typical image there may be multiple independent trees (Fig. 14). Each root is treelike connected to its own endings and none of these trees can prematurely abort. Therefore the goal of removing the crossings in the image is to establish these trees. Each tree has a number of qualities that can be used. One quality is the flux, the amount of blood that can be passing through. Blood will be pulsed through the system and consequently it cannot be accelerated. Blood flow may be decisive for making the decision on which multiple tree structure is viable for the example in Fig. 14. In Fig. 14 the captured structure in the left can be interpreted in different ways, each leading to unique constellation of multiple independent trees. Resolving the choice may be achieved by looking at the blood flow capacity. The flow will not be available in a static image. But the underlying process is the physical potential. Blood has to flow as it otherwise will clod; in contrast, it will not fly as otherwise the vessels cannot pass their attritions. It may be noticed that the blood flow from vessels to an artery may be hard to analytically describe but the inverse may be easier to solve. In most cases, such an analysis may not be a primary factor. In general, the tree will detail locally and the flow analysis may give more a health indicator than a reconstruction directive. Embodiments of the invention provide for verifying a 3D structure of a blood vessel network from the flood of blood in the vessels.

Spatial transformation.

Once the different vein trees are distinguished, the abnormal crossing features are eliminated by the principle of 'cutting and folding'. This notion is known from graph planarization, where an arbitrary graph is locally unfolded to remove (line) crossings. Later, it returned to map graphs onto a network with geometrically fixed resources, for example to optimize the use of AND- and OR-gates of Boolean equations on the available gates in a Programmable Logic Array. In our situation, we want to create a mathematical crossover-free model from the physical reality to navigate in a simple way while the observations can be 1-on-1 interpreted in the physical space. For example, the upper tree or vessel network 201 in Fig. 6, is flipped vertically in comparison to Fig. 3 to eliminate the crossing with the lower tree, or secondary vessel network 212. This will change the positions of bifurcations and endings in the upper tree, which exemplifies that the notational tree is only structurally but not geometrically showing the physical reality. For structures with higher degree of overlapping between vein trees the unfolding may result in a frame that is larger than the original image frame.

A complication may be that the graph is undirected. Therefore the difference between the root and the ends are not marked separately. They are placed at the edge of the picture where 'the pencil is put on paper' (Fig. 15). Fig. 15 is an illustration of linear features when only part of an undirected graph is visible in the frame, root and leaves are not distinguishable.

It may be reasonable to assume that the wider lines on the picture edge represent incoming veins, but pictures are arbitrarily limited in view so by accident also vessels can appear on the edge. Even though these vessels will be in less wide lines, the assumption of what represents an incoming vein and what is an ending vessel needs to be ascertained. Therefore, the tree assumption may be a minimal requirement for the 3D model, but the flow analysis may make the ultimate decision. Another aspect is bypasses which make mazes into the tree, similar to reconvergent fanout in Boolean logic. The mazes can be found by a technique called levelizing, which is usually applied to establish the computational order of gate evaluation. In the presence of mazes, the levelizing will come to a premature stop. A solution is to have a cost value that will point out the most likely signal line to cut so that the overall levelizing process is achieved by a minimal set of cuts. The procedure may be used for the tree representation, and the presence of the related maze (also called cluster) may be notated so that improvements can still be made when the bandwidth information becomes available afterwards. Extracting linear features of the vessel network from the image data and determining a verified tree model of the 3D vessel network

Fig. 2 shows the image data in Fig. 1 after binarization and skeletonization, i.e. to get the lines-thinned image. Further, isolated pixel removal has been applied to remove unwanted isolated pixels. Linear features 203, representing the blood vessels in the vessel networks 201 , 212, is thereby extracted from the image data 202. The linear features are a 2D projection of a vessel network 201 , 212, which in reality extend in three dimensions in the imaged organ.

According to an embodiment of the invention a method 100 of extracting a vessel network 201 , 212, from medical image data 202 is provided, see Fig. 7. The method 100 comprises extracting 101 linear features 203 of the vessel network from said image data 202, and determining 102 a verified tree model 204 of the vessel network 201 , 212, by validating the linear features 203 against a physical criterion parameter related to the vessel network 201 , 212. By validating the linear features 203 against criteria that must be fulfilled due to physical requirements or physical plausibility in the organ or the vessel network 201 , 212, a verified tree model 204 of the actual three-dimensional structure of the vessel network 201 , 212, can be determined. Linear features 203 that does not fulfill or comply with the set physical criterion parameter may then be discarded from belonging to the current vessel network 201 , or 212, that is verified. A vessel network 201 can thereby be verified and extracted from a second vessel network 212 in the 2D image data 202, see Fig. 3, that would otherwise appear as belonging to the same network of vessels due to overlapping of vessels, i.e. linear features 203, as seen in Fig. 2. The vessel network 201 may accordingly be distinguished from the second vessel network

212. Hence, physical plausibility is taken into account in order to resolve the actual three- dimensional structure.

Fig. 4 is an enlarged section of the upper left portion in Fig. 3. The 2D projection of the vessel network 201 , 212, comprises endings 206 and crossings 207, 210, of linear features 203. Each of the crossings 207, 210, comprises overlapping of linear features 203 in a common point 208, and branches 209, 211 , of linear features 203 extending from the common point 208.

The physical criterion parameter may comprise a pre-determined spatial criterion for the linear features 203, and validating the linear features 203 may comprise validating 103 the crossings 207, 210, against this pre-determined spatial criterion. The vessel network 201 , 212, must comply with certain physical parameters, such as how the included vessels are arranged spatially, in relation to each other, and/or how they are connected, as required by the nature of the imaged organ, and further comply with aspects of physical plausibility in spatial arrangement in such organ. Therefore, by setting a spatial criterion the verified tree model 204 may be determined by excluding linear features 203 that are not fulfilling the criterion, or does not comply with the aforementioned physical plausibility, from belonging to the current vessel network 201 , or 212, that is verified.

The pre-determined spatial criterion may be determined as a number of branches 209, 211 , in each of the crossings 207, 210. The vessels in a vessel network 201 , 212, have the function to transport oxygen in the organs, and assume generally a tree-wise distribution for this function. Validating the linear features 203 may thus comprise distinguishing 104 false crossings 207 from real crossings 210 in the vessel network based on the number of branches in each of the crossings 207, 210. Fig. 4 illustrates a false crossing 207 in a particular vessel network 201 , i.e. the branches 209 does not belong to the vessel network 201 that is being verified, but belongs instead to second vessel network 212. In this example, the crossing 207 has four branches from the common point 208, or intersection of the linear features, i.e. branches 209 extending upwards and downwards in Fig. 4, and branches 211 extending left and right in Fig. 4, from point 208, respectively. As the blood vessels split repeatedly into two in a tree-wise fashion, crossing vessels are not plausible to be a split (i.e. bifurcation) plus a merge. Thus three branches are more plausible than four branches in a crossing. Hence, a crossing having more than three branches may be determined 106 as a false crossing 207, and/or a crossing having less or equal to three branches may be determined 107 as a real crossing 210. Fig. 4 illustrates vessel network 201 having three branches extending from the crossing 210, which thus is determined as a real crossing 210, thereby complying with the actual three dimensional vessel structure in the vessel network 201.

Determining the verified tree model may thus comprise including 105 the real crossings 210 in the verified tree model 204 of the particular vessel network 201.

False crossings 207 comprises false branches 209 of linear features 203 that intersect with the particular vessel network 201 being verified in this embodiment. Thus, determining the verified tree model 204 may comprise excluding 108 the false branches 209 from the verified tree model 204 when the false crossing 207 is identified, and further, determining 109 the excluded false branches 209 as secondary linear features belonging to one or more secondary vessel networks 212. This provides therefore for eliminating linear features from the 2D projection of the vessel network 201 that are part of another overlapping secondary vessel network 212.

False branches 209 may be excluded from a verified tree model 204 of a particular network 201 , or 212, by analyzing other physical criterion parameters of the vessel network 201 , 212, such as the flow of blood in the vessels, relative thickness of the vessels, or the relative depth of the vessels in the tissue, as described further below.

Hence, extraction 110 of the vessel network 201 from a secondary vessel network 212 by determining the verified tree model 204 is possible, and the problems from projecting the three dimensional blood vessel structure in 2D are overcome.

Further, as illustrated in Fig. 6, vessel network 201 may be spatially transformed 111 in relation to the secondary vessel network 212. Flipping the vessel network 201 in relation to the secondary vessel network 212 removes the false crossings 207 and the vessel networks 201 , 212 can be treated separately more easily in subsequent analysis.

The physical criterion parameter may relate to a flow of blood in the vessel network 201 , 212. As the flow of blood must take a certain path in the vessels, analysis and determination of such flow can be used to exclude or include linear features 203 from a particular vessel network 201 , or 212, when determining the verified tree model 204 of that particular vessel network. For example, when a false crossing 207 has been identified, the flow of blood may be determined in each of the branches 209, 211 , extending therefrom, and if the flow in a pair of branches 211 , e.g. extending left and right from common point 208 in Fig. 4, is similar, the branches 211 may be determined as belonging to the same vessel network 201 , whereas branches 209 in Fig. 4 are determined as false branches 209 not belonging to network 201. E.g. starting from branch 211 far right in Fig.4 and moving left towards point 208 where branches 209 cross vertically, the branches 209 may be excluded from belonging to vessel network 201 because the flow of blood in branches 209 would most likely be different than in branches 211 extending to the left and right of common point 208.

Therefore, validating the linear features 203 may comprise distinguishing 112 vessel network 201 from the second vessel network 212 based on the flow of blood.

The flow of blood may be determined 113 by analyzing a temporal change in color, contrast, or brightness, or in any combination thereof in the image data of the blood vessels. The image data may thus include a sequence of images in time, a video sequence, for resolving such changes in time. By observing changes in aforementioned parameters the continuity of the flow may be traced in the various branches 209, 211 , of the vessel network 201 , 212, for determining branches that are in fluid communication and thereby connected to the same vessel network as the particular vessel network currently verified. Further, the temporal change in the thickness or diameter of vessels 213 in the vessel network 201 , 212, may be analyzed and determined as an indicator for the blood flow. This is due to the pulsating action of the vessels as the blood is pumped through with the heart beat. The pulsating action increases the diameter of the blood vessels temporarily.

Hence, distinguishing the vessel network 201 from the second 212 vessel network may comprise determining 114 at least two branches, 209 or 211 , corresponding to said vessels, having a substantial equal flow of blood as belonging to the vessel network 201 , or the second vessel network 212, being verified in the tree model 204. E.g. connected branches 209, or 211 , having synchronized increase in vessel diameter over time may be determined as belonging to the same vessel network, 209 or 211.

The physical criterion parameter may relate to the relative depth of blood vessels in the vessel network 201 , 212. A change of focal plane in the image data 202 and the corresponding change of sharpness of the blood vessels may be determined 115 as a relative depth indicator of the linear features 203. This provides a possibility to distinguish e.g. false branches 209 from a particular vessel network 201 , see Fig. 4, as the false branches 209 run at a different depth in the tissue than branches 211 part of the particular vessel network 201. Hence, the overlapping features in the 2D projection can be resolved to determine the true vessel network 201.

Accordingly, validating the linear features 203 may comprise distinguishing 116 the vessel network 201 from the second vessel network 212 based on the relative depth indicator. Extracting the vessel network 201 from a secondary vessel network 212 may comprise determining 117 at least two branches 209 or 211 of blood vessels having substantial equal image sharpness as belonging to the particular vessel network 201 or the second vessel network 212.

The physical criterion parameter may relate to the relative thickness or diameter of blood vessels in the vessel network 201 , 212. It is more physically plausible that branch 211 , extending left of common point 208 in Fig. 4, is part of the vessel network 201 if it has a substantially equal diameter as branch 211 extending right from common point 208, due to the continuous nature of the blood vessel geometry. Branches 209 may have a different diameter than branches 211 , and are thereby determined as belonging to the same secondary vessel network 212, different from the first vessel network 201. Extracting the vessel network 201 from a secondary vessel network 212 may thus comprise determining 118 at least two branches of blood vessels 211 , or 209, having substantial diameter as belonging to one of the first vessel network 201 and the second vessel network 212. This further improves the accuracy in determining a verified tree model 204 of the vessel network 201 , 212, in reconstructing the three dimensional reality of the aforementioned.

According to one embodiment of the invention a device 900 adapted to extract a vessel network 201 , 212, from medical image data 202 is disclosed, as illustrated in Fig. 8. The device 900 may comprise an image processing unit 901 configured to extract linear features 203 of the vessel network 201 , 212, from the image data 202. The linear features 203 comprises a 2D projection of the vessel network 201 , 212. The device may further comprise a second processing unit 902 configured to calculate a verified tree model 204 of the vessel network 201 , 212, by being adapted to validate the linear features 203 against a physical criterion parameter related to vessel network 201 , 212. The device 900 may perform all of the method steps 101 -117 in the method 100 as described above relating to the first aspect of the invention, as illustrated by the schematic flow chart in Fig. 8.

The method 100, and/or the device 900 may be used for biometric identification, and/or for extraction of physiological data, and/or for health monitoring. This originates from that the complex 3D reality of overlapping and intersecting networks of blood vessels can be resolved and remodeled from 2D medical image data according to the inventive aspects of the invention, thereby obtaining the actual vessel structure in a verified tree model 204. Identification of a person can therefore be made more accurate, by applying the inventive aspects of the invention to image data of blood vessels of the particular person that is to be identified. Safety is therefore also improved, both from an authorization perspective, and from e.g. a patient perspective in the medical health care where e.g. extraction of physiological data can be improved and more accurate, as previously confusing blood vessel structures can be discarded from the 2D image data that are not part of the verified tree structure 204 of the blood vessel network 201 , 212, that is being analyzed and e.g. being diagnosed for defects and/or diseases. Health monitoring can therefore also be improved. Analysis and ultimately diagnosis and treatment of patients can be undertaken more efficiently, thereby saving resources. E.g. the improvements in image analysis of patient data provided by the inventive aspects of the invention, allows for obtaining and extracting a verified blood vessels structures or networks with a minimal amount of image data, dispensing with the need for comprehensive and expensive multilayer imaging, such as MRI.

The present invention has been described above with reference to specific embodiments.

However, other embodiments than the above described are equally possible within the scope of the invention. The different features and steps of the invention may be combined in other combinations than those described. The scope of the invention is only limited by the appended patent claims. More generally, those skilled in the art will readily appreciate that all parameters and configurations described herein are meant to be exemplary and that the actual parameters and/or configurations will depend upon the specific application or applications for teachings of the present invention is/are used.