(19)
(11)EP 2 048 617 B1

(12)EUROPEAN PATENT SPECIFICATION

(45)Mention of the grant of the patent:
13.05.2020 Bulletin 2020/20

(21)Application number: 08166295.9

(22)Date of filing:  10.10.2008
(51)Int. Cl.: 
G06T 7/30  (2017.01)
G06T 15/06  (2011.01)

(54)

Method, system and software product for providing efficient registration of volumetric images

Verfahren, System und Softwareprodukt zur Bereitstellung einer effizienten Registrierung von volumetrischen Bildern

Procédé, système et produit logiciel pour la fourniture de l'enregistrement suffisant d'images volumétriques


(84)Designated Contracting States:
AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MT NL NO PL PT RO SE SI SK TR

(30)Priority: 12.10.2007 US 871298

(43)Date of publication of application:
15.04.2009 Bulletin 2009/16

(73)Proprietor: Kofax International Switzerland Sàrl
1209 Genève (CH)

(72)Inventors:
  • Dekel, Doron
    Toronto Ontario M3H 4X1 (CA)
  • Chandrashekara, Raghavendra
    Toronto Ontario M4Y 1V5 (CA)

(74)Representative: Murgitroyd & Company 
Murgitroyd House 165-169 Scotland Street
Glasgow G5 8PL
Glasgow G5 8PL (GB)


(56)References cited: : 
US-A1- 2005 111 720
US-A1- 2005 249 398
  
  • FRED S. AZAR, KIJOON LEE, REGINE CHOE, ALPER CORLU, SOREN D. KONECKY, AND ARJUN G. YODH: "Joint analysis of non-concurrent magnetic resonance imaging and diffuse optical tomography of breast cancer", PROC. SPIE:OPTICAL TOMOGRAPHY AND SPECTROSCOPY OF TISSUE VII, vol. 6434, 21 January 2007 (2007-01-21), pages 1-10, XP040234451,
  • ALI KHAMENE ET AL: "A Novel Projection Based Approach for Medical Image Registration", 1 January 2006 (2006-01-01), BIOMEDICAL IMAGE REGISTRATION LECTURE NOTES IN COMPUTER SCIENCE;;LNCS, SPRINGER, BERLIN, DE, PAGE(S) 247 - 256, XP019035659, ISBN: 978-3-540-35648-6 * the whole document *
  • WALLIS J W ET AL: "Use of volume-rendered images in registration of nuclear medicine studies", NUCLEAR SCIENCE SYMPOSIUM AND MEDICAL IMAGING CONFERENCE, 1994., 1994 IEEE CONFERENCE RECORD NORFOLK, VA, USA 30 OCT.-5 NOV. 1994, NEW YORK, NY, USA,IEEE, US, vol. 3, 30 October 1994 (1994-10-30), pages 1429-1432, XP010150347, DOI: DOI:10.1109/NSSMIC.1994.474615 ISBN: 978-0-7803-2544-9
  • H.-M. CHAN AND A.C.S. CHUNG: "Efficient 3D-3D Vascular Registration Based on Multiple Orthogonal 2D Projections", LECTURE NOTES IN COMPUTER SCIENCE:BIOMEDICAL IMAGE REGISTRATION, vol. 2717, 23 June 2003 (2003-06-23), pages 301-310, XP002617029,
  • RUECKERT * D ET AL: "Nonrigid Registration Using Free-Form Deformations: Application to Breast MR Images", IEEE TRANSACTIONS ON MEDICAL IMAGING, IEEE SERVICE CENTER, PISCATAWAY, NJ, US, vol. 18, no. 8, 1 August 1999 (1999-08-01) , XP011035886, ISSN: 0278-0062
  
Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art. 99(1) European Patent Convention).


Description

FIELD



[0001] The present invention relates generally to computer-generated images and, more particularly, to a method for mapping locations in one 3-dimensional image to homologous locations in another 3-dimensional image.

BACKGROUND



[0002] Registration between volumetric images (data volumes) can involve mapping locations in 3-dimensional (3D) images to homologous locations in other 3D images. The capability of automatically performing elastic (or non-rigid) registration between volumetric images has many potential benefits, especially in medical applications where 3D images of patients are routinely acquired using CT, MRI, PET, SPECT and ultrasound scanners. Known methods for automatic 3D registration are described in United States Patent No., 6,909,794 and WO2006118548 and references thereof and in the book "Handbook of Biomedical Image Analysis, volume 3: Registration Models" by Suri et al. (2005) and references thereof, as well as in FRED S. AZAR, KIJOON LEE, REGINE CHOE, ALPER CORLU, SOREN D. KONECKY, AND ARJUN G. YODH: "Joint analysis of non-concurrent magnetic resonance imaging and diffuse optical tomography of breast cancer", PROC. SPIE:OPTICAL TOMOGRAPHY AND SPECTROSCOPY OF TISSUE VII, vol. 6434, 21 January 2007 (2007-01-21), pages 1-10, XP040234451.

[0003] Many robust 3D registration methods are iterative: one or more initial guesses are generated and then refined by a directed search for mapping parameters that maximize a similarity metric. Typically a large amount of computation needs to be performed per iteration to derive new mapping vectors, resample at least one of the 3D images on a curved grid and compute a global similarity function. The number of initial starting points (guesses) and optimization iterations required to ensure success increases exponentially when the degree of similarity between the contents of the image volumes is low under a rigid mapping, for example when images from a patient are registered to an annotated atlas created using images of one or more other patients. Elastic 3D registration, therefore, has had only limited usefulness in clinical practice to date.

SUMMARY



[0004] It is an object of the present invention to provide a method, a system and a computer program product for registering 3D medical image data. The object can be achieved by the features of the independent claims. Further embodiments are characterized by the dependent claims.

[0005] A method for registering 3D image data according to the appended claims is disclosed.

[0006] According to another aspect of the present invention, a system for registering 3D image data is provided comprising: a processor; and, a memory for storing a first 3D image, a second 3D image, and means for configuring the processor to perform the steps of the method according to the appended claims.

[0007] According to another aspect of the present invention, a computer program product is provided for use on a computer system to register 3D image data, the computer program product comprising: a recording medium; and, means recorded on the medium for instructing the computer system to perform the steps of the method according to the appended claim.

BRIEF DESCRIPTION OF THE DRAWINGS



[0008] Embodiments are described below in further detail, by way of example, with reference to the accompanying drawings, in which

Figure 1 is a flowchart of a method for registering 3D image data in accordance with an embodiment of the present invention;

Figure 2 is a sectional representation illustrating a rendering transform in accordance with an embodiment of the present invention ;

Figure 3 shows 2D images representing the surface parameters generated from a first and second 3D image in accordance with the rendering transform described in Figure 2;

Figure 4 is a flowchart of a method for registering two 2D projections using a single affine transform to initialize the elastic registration in accordance with an embodiment of the present invention;

Figure 5 illustrates a method for generating multiple candidate affine transforms in accordance with an embodiment of the present invention.

Figure 6 is a graphical representation of a method for generating a plurality of candidate affine transforms for two separate regions of a first 2D projection in accordance with an embodiment of the present invention.

Figure 7 is a flowchart of a method for registering two 2D projections using two affine transforms to initialize the elastic registration in accordance with an embodiment of the present invention;

Figure 8 illustrates the four main views of a single anatomical 3D image;

Figures 9 and 10 illustrate a method by which a region of interest can be isolated using contours and/or marked positions in accordance with an embodiment of the invention;

Figures 11 and 12 illustrate a method by which the results of the registration of one or more 2D projection pairs can be used to initialize a 3D elastic registration; and

Figure 13 is a block diagram of a computer system for implementing an embodiment of the invention.


DETAILED DESCRIPTION



[0009] The described embodiments of the present invention relate to methods for registering 3D image data using 2D projections of the 3D images. In at least one embodiment the 3D images are computed tomography (CT) scans of two different patients where the first 3D image is a subset of the anatomy scanned in the second 3D image. The second 3D image may be a full head-to-toe scan and can be annotated with locations, curves and regions of special interest, forming a reference atlas, while the first 3D image may be of an unknown region in the body. By registering the first 3D image to the second 3D image, the locations, curves, and regions from the second 3D image can be copied to their homologous locations in the first 3D image, enabling the automation of many operations that would otherwise require extensive user interaction, such as organ segmentation and documentation.

[0010] Reference is now made to Figure 1, in which a method 100 for registering 3D image data in accordance with an embodiment of the present invention is illustrated. In steps 102 and 104 a first 3D image 106 and a second 3D image 108 are received. In step 110 a rendering transform controlled by a number of rendering parameters is selected to apply to the 3D images 106 and 108. The values of the rendering parameters are selected based on the contents of the first and second 3D images 106 and 108 and the purpose of the registration. In steps 112 and 114 the rendering transform determined in step 110 is applied to the 3D images 106 and 108 to produce 2D projection images 116 and 118. In step 120 the two 2D projection images 116 and 118 are registered to produce an elastic registration 122 that maps locations in the second 2D projection 118 to homologous locations in the first 2D projection 116. Various of the steps of method 100 will be described in further detail below.

[0011] In step 110 of method 100 described above, a rendering transform is determined. Rendering can be described as the process of computing a 2D image from a 3D image. The 2D image may be computed by projection from a specified viewpoint or viewing direction. One common rendering technique is ray casting. In this technique, for each pixel in the 2D image a virtual ray is transmitted from a projection plane towards the 3D image data. The ray is sampled at regular intervals throughout the 3D image and the pixel value corresponding to the ray is computed from the samples according to a rendering ray transform. Useful rendering ray transforms include selecting the sample with the highest value, and computing a simulated light interaction with samples using the local data gradient. Techniques for rendering of 3D data are well known in the art and are described in many publications, for example in the book "Introduction to Volume Rendering" by Lichtenbelt et al. (1998).

[0012] In one embodiment the rendering transform is determined so that the 2D projections (i.e. 116 and 118 of Figure 1) contain a preselected structure. The structure may be an anatomical structure vessels or other types of tissues or organs, such as skin or muscles. Since bones have similar shapes and spatial arrangements in different individuals, and their HU values are always higher than surrounding fat and muscle tissue, they are good "anchors" for inter-patient registration.

[0013] Reference is now made to Figure 2 wherein a rendering transform in accordance with an embodiment of the present invention is applied to a trans-axial 2D slice image 201 of a 3D image. A typical 3D image generated by a CT or magnetic resonance imaging (MRI) scanner is comprised of a stack of parallel 2D slice images, oriented perpendicular to the long axis of the patient body (trans-axial slices), with the front of the patient body or head facing one of the sides of the image. Although the rendering algorithm must be applied to every 2D slice image, for ease of explanation the application of a rendering transform to a 3D image will be explained in relation to a single 2D slice image 201.

[0014] In the exemplary embodiment shown in Figure 2 a projection plane 202, represented by an intersection line through the slice, is aligned with an edge of the 2D slice image 201. In one embodiment the projection plane 202 is selected to be parallel to one of the side faces of the 3D image such that when the 3D image is a CT or MRI scan of the patient, the projection plane 202 corresponds with a front, back, left side or right side view of the patient. Where the 3D images are generated by a medical scanner, such as a CT scanner, the particular view (front, back, left side or right side) desired can easily be located using the DICOM descriptor information. In cases where the descriptor indicates a patient orientation which is not aligned with the coordinate axes of the 3D image, trans-axial slice image 201 may represent a plane that is at an oblique angle to the coordinate axes of the 3D image.

[0015] Rays 204 are cast into each 2D slice image 201 at every pixel spacing increment 203 in the direction perpendicular to the projection plane 202. As noted above, each ray represents a pixel 214 in the 2D projection. The increment 203 may be selected such that the registration quality is maximized and the processing time minimized. In one embodiment the increment 203 is set to 1 mm.

[0016] Each ray 204 is sampled at regular intervals throughout the 2D slice image 201 and the sample value is compared to a threshold value to determine if the sample point corresponds to a particular structure. For example, since bone typically has a HU value of 150 or greater, if the particular structure being sought is bone then the threshold value would be 150. In the exemplary embodiment shown in Figure 2, the 2D slice image 201 contains a bony structure 206 having higher HU values than its surroundings. Accordingly, any rays 204 that intersect with the bony structure 206 will meet or exceed a 150 HU threshold.

[0017] At the point 211 a ray 204 meets or exceeds the threshold whereupon a plurality of surface parameters are calculated. Where the structure being sought is an anatomical structure such as bone, the surface parameters describe the features of the anatomical surface of the structure in a small 3D region surrounding point 211. If the ray reaches the end 210 of the 2D slice image 201 before the sample value meets or exceeds the threshold then one of the surface value parameters is assigned a special value to denote that fact. In one embodiment the surface parameters include the following four parameters: Nx, Ny, Nz and D where Nx, Ny, and Nz form a surface normal unit vector 207 derived from a 3D Zucker-Hummel operator centered around 211, and D represents the distance 208 between the projection plane 202 and the surface location 211. Other surface parameters representing curvature features, such as principal curvature radii and shape index ("COSMOS, Representation Scheme for 3D Free-Form Objects", Dorai et al, IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 19, NO. 10, OCTOBER 1997) may be successfully used as well, although they are more computationally expensive and more sensitive to noise. A pixel in which the threshold was not crossed is known as a background pixel and is identified by setting the D parameter to a negative value.

[0018] The plurality of surface parameters generated for each pixel of the 2D projection are stored with the 2D projection. The D surface parameter of each pixel can then be used later to map each non-background pixel back to its corresponding 3D location.

[0019] If the 2D projections are expected to contain noise and small narrow regions, such as minor blood vessels, that are often non-homologous across patients, it is advantageous to remove them from the 2D projections by performing a morphological opening of the non-background pixels. A morphological opening can generally be described as a process of removing small objects from an image and is implemented by erosion followed by dilation of the image.

[0020] Reference is now made to Figure 3 wherein 2D images created from the surface parameters generated in accordance with Figure 2 are illustrated. The top images 302, 304, 306 and 308 represent the surface parameters D, Nx, Ny, and Nz respectively generated from a second 3D image 108 of a full body in accordance with the rendering transform described in Figure 2. Similarly, the bottom images 310, 312, 314 and 316 represent the surface parameters D, Nx, Ny and Nz respectively generated from a first 3D image 106 of a partial body. In these images the surface parameters have been scaled and shifted to allow them to be displayed such that the minimum value is shown in black and the maximum in white. Background pixels are mapped to black. A morphological opening was performed on the images to remove noise and blood vessels that do not overlay bone.

[0021] After the 2D projections 116 and 118 are generated through the rendering of the first and second 3D images 106 and 108 in steps 112 and 114, the 2D projections 116 and 118 are registered in step 120 of method 100. Registration includes determining at least one affine transform and using this at least one affine transform to initialize an iterative optimization of an elastic registration. In one embodiment the elastic registration is described by a free-form deformation. An affine transform can be broadly described as a mapping between two coordinate spaces. An implementation of an affine transform comprises a number of parameters that control the mapping. In one embodiment an affine 2D transform is controlled by the following seven parameters: TX, TY, SX, SY, RX, RY and θ. TX and TY represent the translation in the lateral (x) and longitudinal (y) directions respectively. SX and SY represent the lateral and longitudinal scaling. RX and RY represent the location of a rotation centre point and θ represents the angle of rotation.

[0022] Reference is now made to Figure 4 wherein a method 400 for elastically registering two 2D projections using one affine transform is illustrated. In the first step 402 multiple candidate affine transforms are generated and evaluated. A candidate affine transform is an estimate of the correct mapping of locations from the second 2D projection 118 to analogous locations in the first 2D projection 116.

[0023] One measure of the quality of the affine transform is its corresponding similarity metric value. The similarity metric value quantifies the degree of similarity between values of pixels of the first 2D projection and the corresponding values of pixels of the second 2D projection as determined by the affine transform. Only pixel locations where at least one of the two pixel locations is not a background location are considered in computing the similarity metric. Various functions for evaluating the total similarity of the pixels across the images can be used. Common examples of such functions include cross-correlation, mutual information, mean-square difference and ratio image. In one embodiment the sum of the squared difference in one or more of the surface parameters (i.e. Nx, Ny, Nz) is used to evaluate the similarity.

[0024] In step 404 the N candidate affine transforms generated in step 402 with the highest similarity values are selected and iteratively optimized according to known iterative affine registration optimization methods. N is any integer equal to or greater than one. In one embodiment N is equal to 5.

[0025] In step 406 the optimized affine transform generated in the previous step 404 with the highest similarity value is used to initialize the parameters of the elastic registration. A number of known elastic registration approaches may be used. In one embodiment a multi-resolution free-form deformation registration is used, defined by an array of displacement vectors specified on a deformation grid of control points. The multi-resolution free-form deformation registration is described in the paper "Nonrigid registration using free-form deformations: Application to breast MR images", D. Rucker et al., in IEEE Transaction on Medical Imaging, 18(1):712-721, August 1999. In the initialization step 406, the optimized affine transform generated in previous step 404 is used to compute the displacement vector at each grid location.

[0026] In step 408 the elastic registration is iteratively optimized to further increase the similarity metric. In one embodiment the elastic registration is optimized using gradient descent with a distortion penalty term incorporated into the similarity measure to prevent excessive warp or folding of the deformation grid.

[0027] Since an affine transform comprises multiple parameters, each of which can accept a wide range of values, the number of candidate affine transforms to evaluate in step 402 in order to ensure success in subsequent steps can be very large. Accordingly, it would be advantageous to develop a method of generating a limited number of candidate affine transforms that includes candidate affine transforms that are likely to produce a good elastic registration and excludes candidate affine transforms that are unlikely to produce a good elastic registration.

[0028] Reference is now made to Figure 5 wherein a method 500 of generating a limited number of candidate affine transforms in accordance with an embodiment of the present invention is illustrated. Where the first and second 3D images 106 are images of a human body only a limited number of values for the affine transform parameter θ need to be evaluated since the orientation of the longitudinal axis of the bodies shown in the first and second 3D images and corresponding 2D projections is made similar by the rendering process. However, a full range of longitudinal translations (TY) needs to be evaluated.

[0029] To generate candidate affine transforms for a full range of longitudinal translations (TY) the longitudinal axis of the second 2D projection is divided into a plurality of increments 525.

[0030] At least one candidate affine transform is generated for each increment 525 of the longitudinal axis. The longitudinal translation parameter (TY) of the affine transform is set to the increment value and the remaining affine parameters are generated by comparing a region of the first 2D projection to a region of the second 2D projection and adjusting the affine transform parameters to maximize the similarity value of these two regions. The regions of the 2D projections selected for the comparison are based on the particular longitudinal axis increment being evaluated.

[0031] In one embodiment the process of generating candidate affine transforms further comprises aligning the longitudinal axis of the first 2D projection 116 and the longitudinal axis of the second 2D projection 118. The first 2D projection 116 is then shifted along the longitudinal axis of the second 2D projection increment by increment such that for each increment a region of the first 2D projection 116 overlaps with a region of the second 2D projection 118. It is the overlapping regions of the two 2D projections at a specific increment that are compared to determine the affine transform parameters for that specific increment.

[0032] This embodiment can be alternatively described in the following manner. The first and second 2D projections 116 and 118 each have a top edge and a bottom edge. The bottom edge of the first 2D projection 116 is aligned with the bottom edge of the second 2D projection 118 such that a region of the first 2D projection 116 overlaps with a region of the second 2D projection 118. The first 2D projection 116 is then shifted along the longitudinal axis of the 2D projection one increment at a time. For each increment there are different overlapping regions of the first 2D projection and the second 2D projection. It is the overlapping regions of the first and second 2D projections at each increment that are compared to determine the affine transform parameters for that specific increment.

[0033] In another embodiment a first margin 510 and a second margin 520 are defined to allow the first 2D projection 116 to extend outside the top and bottom edges of the second 2D projection 118 during shifting. This is necessary in cases where not all of the contents of the first 2D projection 116 can be found in the second 2D projection 118. For example, if the second 2D projection 118 contains a skeleton that goes from shoulders to feet and the first 2D projection 116 contains a skeleton that goes from head to mid-chest, the method should be able to compare the bottom part of the first 2D projection 116 with the top part of the second 2D projection 118. Accordingly, if the second 2D projection 118 is known to cover the full body then it is not likely that the entire contents of the first 2D projection 116 will not be found so the first and second margins 510 and 520 can be small. Otherwise the first and second margins need to be large (e.g. half the height of the first 2D projection 116).

[0034] In one embodiment the remaining affine parameters are generated in the following manner. The average lateral (x) co-ordinate of the non-background pixels in the overlapping region of the first 2D projection 116 and the average lateral (x) co-ordinate of the non-background pixels in the overlapping region of the second 2D projection 118 are computed and the lateral translation parameter (TX) is selected to align the two averaged lateral (x) co-ordinates. The lateral rotation center parameter (RX) and the longitudinal center parameter (RY) are set to the center 530 of the overlapping region of the second 2D projection 118. A small number of values around one for the lateral scaling parameter (SX) and the longitudinal scaling parameter (SY) are evaluated. A small number of values around zero, eg, {-2, 0, 2} degrees, are evaluated for θ. The affine transform parameters resulting in the highest similarity value for the given TX are saved for further processing.

[0035] In one embodiment each increment is equal to a single projection pixel. To reduce the time and processing power required to compute the similarity metrics for the candidate affine transforms, the resolution of the first and second 2D projections 116 and 118 may be reduced by averaging every n x n pixel block where n is equal to any integer greater than one. In one embodiment where the projection pixel size is 1mm, n is equal to 4, resulting in each reduced pixel representing a 4x4mm region.

[0036] Reference is now made to Figures 6 and 7 wherein a method for registering two 2D projections using two global affine transforms is illustrated. Where the first 2D projection 116 exceeds a certain size it may be necessary to obtain more than one affine transform to produce a good global registration. In one embodiment multiple affine transforms are required where the height of the first 2D projection 116 exceeds 70 cm. In step 702 the first 2D projection 116 is divided into a first region 600 and a second region 601 such that the first and second regions 600 and 601 are not co-extensive. In the exemplary embodiment shown in Figure 6, the first region 600 is the upper portion of the first 2D projection 116 and the second region is the lower portion of the first 2D projection 116. As shown in Figure 6, the first and second regions 600 and 601 may be overlapping.

[0037] In step 704 a number of candidate affine transforms are generated for mapping locations in the second 2D projection 118 to locations in the first region 600 of the first 2D projection 116. In one embodiment the candidate affine transforms are generated in accordance with the method described in relation to Figure 5.

[0038] In step 706 the top N candidate affine transforms generated in step 704 for the first region 600 of the first 2D projection 116 are iteratively optimized and the candidate affine transform with the highest similarity value is selected and designated the first region affine transform 650.

[0039] In step 708 a sub-set of the increments used in step 704 to generate the candidate affine transforms for the first region is created for generating candidate affine transforms for the second region 601 of the first 2D projection 116. The sub-set of increments is based on the first region affine transform 650 selected in the previous step 706. The first step is to calculate the longitudinal distance 620 between the centre 607 of the first region 600 of the first 2D projection 116 and the centre 604 of the second region 601 of the first 2D projection 116. The second step is to map the centre 607 of the first region 600 of the first 2D projection 116 back to location 610 of the second 2D projection 118 using the inverse of the first region affine transform 650 selected in the previous step 706. The sub-set of increments is then deemed to include those increments of the longitudinal axis of the second 2D projection 118 that fall between location 610 and the distance 620 between the centers 607 and 604 plus a tolerance range 630. The tolerance range may be proportional to the distance 620 between the centers 607 and 604. In one embodiment the tolerance is between 0.8 and 1.3 of the distance 620 between the centers 607 and 604.

[0040] In step 710 candidate affine transforms are generated for the second region 601 of the first 2D projection 118 for each increment in the sub-set of increments generated in the previous step 708. In one embodiment the candidate affine transforms for the second region 601 of the first 2D projection 116 are generated in accordance with the method described in relation to Figure 5.

[0041] In step 712 the top N candidate affine transforms generated in the previous step 710 are iteratively optimized and then the optimized candidate affine transform with the highest similarity value is selected and designated the second region affine transform 651.

[0042] In step 714 the elastic registration is initialized using the first and second affine transforms 650 and 651 selected in steps 706 and 714. In one embodiment the initialization involves the following steps. First, using the inverse of the first and second region affine transforms 650 and 651, the first region 600 centre 607 and the second region 601 centre 604 of the first 2D projection 116 are mapped back to locations 610 and 611 in the second 2D projection 118. Displacement vectors for grid locations in the second 2D projection that fall in the region 660 above location 610 are mapped using the first region affine transform 650. Displacement vectors for grid locations in the second 2D projection that fall in the region 662 below location 611 are mapped using the second region affine transform 651. Displacement vectors for grid locations that fall in the region 661 between locations 610 and 611 are mapped using a proportionally weighted blend of the first region affine transform 650 and the second region affine transform 651.

[0043] Once fully initialized, the elastic registration is iteratively optimized in step 716 using known methods. Some of these optimization methods were previously mentioned in relation to method 400.

[0044] The number of locations mapped between the second 3D image 108 and the first 3D image 106 can be increased by generating multiple 2D projections for each 3D image. For example, one 2D projection may be generated for each of the main views (front, back, left and right). Other views that could be generated include the four intermediate diagonals (front-left, front-right, back-left and back-right).

[0045] Reference is now made to Figure 8 wherein front, back, left and right views of a first and second 3D volume are illustrated. 2D images 802, 804, 806 and 808 correspond to a 3D volume of a whole body. 2D image 802 represents the front view, 2D image 804 represents the back view, 2D image 806 represents the left side view and 2D image 808 represents the right side view. 2D images 810, 812, 814 and 816 correspond to a 3D volume of a portion of a body and represent the front, back, left side and right side views respectively.

[0046] To generate the plurality of 2D projections (i.e. 2D projections 802, 804, 806 and 808) rendering transforms are determined for each view then applied to the 3D images. An elastic registration, initialized by one or more affine transforms, is then generated for each view.

[0047] As stated above, each affine transform comprises a plurality of parameters. In one embodiment the affine transform parameters include seven parameters: a lateral translation parameter (TX), a longitudinal translation parameter (TY), a lateral scaling parameter (SX), a longitudinal scaling parameter (SY), a lateral rotation centre parameter (RX,) a longitudinal rotation center parameter (RY) and rotation angle parameter (θ). Experimentation has indicated that a number of relationships exist between the affine transform parameters of different views. First, the longitudinal translation parameter (TY) and the longitudinal scaling parameter (SY) will be similar for all views. Second, the lateral scaling parameter (SX) and longitudinal rotational center parameter (RY) will be similar for opposite views. Examples of opposite views include front and back, and left and right. Third, the rotation angle parameter (θ) of one view will be very similar to the negation of the rotation angle parameter (θ) of the opposite view. Fourth, the lateral rotation centre parameter (RX) will be very similar to the width of the 2D projection minus the lateral rotation center parameter (RX) of the opposite view.

[0048] The affine transforms corresponding to the multiple views can be iteratively optimized by taking advantage of these similarity relationships between affine parameters. The following pseudo-code illustrates one way in which this iterative optimization can be implemented to direct the adjustment of parameters of different views during optimization iterations towards convergence on values consistent with these relationships: Generate an affine transform for each view; Loop N times Replace the TY, SY of each view by TY, SY averaged over all the views; Replace the SX, RY of each view by SX, RY averaged over the given view and its opposite; Replace the θ parameter of each view by its average with -θ of its opposite view; Replace the RX parameter of each view by its average with (image A width - RX) of its opposite view; Optimize the affine transform of each view separately; End loop;

[0049] N is any integer greater than 1. In one embodiment N is equal to 5. In a further embodiment the averages calculated in each iteration are weighted averages based on the similarity value of the affine transform such that affine transforms with a lower similarity value contribute less to the average.

[0050] Once the affine transformations have been optimized using the above process an elastic registration is initialized and iteratively optimized for each view separately.

[0051] Generating and registering multiple 2D projections from each 3D image also allows a region of interest to be isolated. For example, a clinician may be interested in isolating and analyzing a region of the anatomy, for example a single bone, a kidney or the liver.

[0052] Reference is made to Figure 9 and 10 wherein a method of isolating a region of interest using contours and/or marked positions is illustrated. In Figure 9 a first 2D projection 901 representing the front view of the second 3D image 108 and a second 2D projection 911 representing the left side view of the second 3D image 108 are generated. Similarly, a first 2D projection 902 representing the front view of the first 3D image 106 and a second 2D projection 912 representing the left side view of the first 3D image 106 are also generated.

[0053] The first and second 2D projections 901 and 911 of the second 3D image 108 include projection contours 904 and 914, which represent the region of the liver. These contours are drawn manually on the first and second 2D projections 901 and 911, based on anatomical knowledge of the location of the liver relative to the skeleton. The contour information is stored together with the 2D projection data corresponding to each 2D projection.

[0054] The first 2D projection 901 of the second 3D image 108 and the first 2D projection 902 of the first 3D image 106, and the second 2D projection 911 of the second 3D image 108 and the second 2D projection 912 of the first 3D image 106 are registered in accordance with the methods described herein. The contours 904 and 914 are then mapped from the first and second 2D projections 901 and 911 of the second 3D image 108 to the first and second 2D projections 902 and 912 of the first 3D image 106 resulting in contours 908 and 918 respectively. It should be clear to one skilled in the art that the mapped contours 908 and 918 do not need to be contained fully in the 2D projections 902 and 912.

[0055] The mapped contours 908 and 918 may further be used to generate a contour extrusion intersection sub-volume which encloses the region of interest (i.e. the region of the liver). The sub-volume is defined by the intersection in 3D of the mapped contours 908 and 918 extruded in the direction of the rendering projection. The sub-volume generation is illustrated in Figure 10 using a 2D slice 930 of the first 3D image 106. The region enclosed by the mapped contours 908 and 918 intersect the 2D slice 930 along lines 940 and 950 respectively. Line segments 940 and 950 are then extruded in the projection direction. Region 1010 is the intersection of the extrusions and includes a region 1000 representing the region of interest (i.e. the region of the liver). Once the region of interest is isolated and separated by this method from other regions of similar intensity and size, such as when the liver region is separate from most of the pancreas, the stomach and the spleen regions, the region of interest can subsequently be more accurately segmented using thresholding and then applying mathematical morphology operators to the sub-volume, such as opening, region growing and selection of the largest component.

[0056] The 2D projections 901 and 911 generated from the second 3D image 108 may contain, in addition to the contours 904 and 914 or in place of the contours 904 and 914, manually marked positions 903 and 913. These manually marked positions are positions that are likely to be located within a region of interest (e.g., the liver). The marked positions 903 and 913 in the first and second 2D projections 901 and 911 of the second 3D image 108 are mapped through the registration process described above to positions 942 and 952 in the first and second 2D projections 902 and 912 respectively. The intersection point 1020 of the projections of positions 942 and 952 in 3D may be used as a seed for segmentation algorithms using region growing. Segmentation in the context of digital imaging is the process of partitioning a digital image into multiple regions and is typically used to locate objects and boundaries. Segmentation using region growing involves starting a region with a single pixel. Adjacent pixels are recursively examined and added to the region if they are sufficiently similar to the region. Segmentation using seeded region growing is described in further detail in R. K. Justice et al., "Medical image segmentation using 3D seeded region growing", Proc. SPIE Vol. 3034, p. 900-910, Medical Imaging 1997: Image Processing, Kenneth M. Hanson; Ed., April 1997. Where the projections of positions 942 and 952 do not directly intersect in 3D as shown in Figure 10, the intersection point 1020 may be selected as the mid-point of the shortest line segment between the projection lines.

[0057] Depending on the purpose of the registration, the contour extrusion intersection sub-volume may undergo a secondary registration with an atlas of the region of interest (e.g., liver) to produce a more accurate definition of the region of interest (e.g., liver). The registration may be performed directly in 3D, as is known in the art, or may be performed in accordance with the registration methods described herein. The computational savings of performing registration in accordance with the described embodiments over a direct 3D registration are reduced for this secondary registration since the region being registered is relatively small.

[0058] The contour/marked position method of isolating a region of interest is particular useful when a clinician wants to isolate a soft tissue region of the anatomy (e.g., liver) when the rendering transform is selected so that the 2D projections contain bone structures.

[0059] Elastic registration of one or more 2D projections generated in accordance with any of described embodiments is further used to initialize registration of the corresponding 3D images. Reference is now made to Figures 11 and 12 wherein a method of using the results of at least one 2D projection registration to initialize a 3D elastic registration is illustrated. If during the rendering process the depth, d, to a specified threshold value (i.e. HU) is stored for each pixel in the 2D projection then each of the pixels in the 2D projection can be mapped to a 3D location defined by coordinates (x, y, z). For example, point p' in a 2D projection 118 derived from a second 3D image 108 can, using the distance value d1, be mapped to 3D location p defined by co-ordinates (px, py, pz). Similarly the homologous point q' in a 2D projection 116 derived from a first 3D image 106 can, using the distance value d2, be mapped to 3D location q defined by co-ordinates (qx, qy, qz). The 3D locations generated from the 2D projection 118 form a set P. Similarly, the 3D locations generated from 2D projection 116 form a set Q. The relationship between P and Q is derived from the registration of 2D projection 118 and 2D projection 116 and is used to initialize the 3D registration.

[0060] The initialization may involve the following steps. First, a 3D transformation, T_global, is generated to best match P and Q. The transformation comprises a global affine transform and a local free-form deformation. The global affine transform is an affine transform which accounts for the bulk difference between the point sets P and Q. The global affine transform is calculated by minimizing the mean squared value of the distance between the mapped positions of points in set P with the position of points in set Q. The residual non-affine component is estimated by T_local. T_local is similarly calculated by minimizing the mean squared value of the residual distances between the mapped positions of points in set P with the points in set Q. Once the transform has been initialized, it can be provided as input to the 3D volumetric registration algorithm.

[0061] Algorithms for computing T_global and T_local are described in B.K. P. Horn, "Closed-form solution of absolute orientation using unit quaternions", Journal of the Optical Society of America, 4:629-642, April 1987 and S. Lee et al., "Scattered data interpolation with multilevel B-splines", IEEE Transactions of Visualization and Computer Graphics, 3(3): 228-244, July-September 1997 respectively.

[0062] Reference is now made to Figure 13 wherein a block diagram of a computer system 1300 suitable for implementing an embodiment of the present invention is illustrated. Specifically, the computer system 1300 comprises a processor 1310 connected to memory 1312. The memory 1213 may be random access memory (RAM). The processor 1310 is also connected to an input/output controller 1308, which controls access to a keyboard 1302, mouse 1304 and monitor 1306.

[0063] In accordance with an aspect of the invention, the processor 1310 is configured to implement an embodiment of the invention. In this case, a first and second 3D image 106 and 108 would be stored in memory 1312. Then, the processor 1310 would perform steps on the first and second 3D images 106 and 108, analogous to the steps described in relation to Figure 1, to register data of the first 3D image 106 with data of the second 3D image 108. In accordance with another aspect of the present invention, a computer program product is provided comprising a recording medium on which is recorded a program for suitably configuring the processor 1310 to perform these steps.

[0064] The embodiments described herein have been shown and described by way of a number of examples.


Claims

1. A method for registering 3D medical image data, the method comprising:

a) receiving (102, 104) a first 3D medical image (106) and a second 3D medical image (108);

b) determining (110) at least one rendering transform for applying to each of the first 3D medical image (106) and the second 3D medical image (108), wherein determining (110) comprises:

i) selecting a first structure in the first 3D medical image (106) and a second structure in the second 3D medical image (108), the second structure being homologous to the first structure, wherein the first structure and the second structure are anatomical structures; and

ii) determining (110) the at least one rendering transform such that at least one 2D projection (116) corresponding to the first 3D medical image comprises the first structure, and at least one 2D projection (118) corresponding to the second 3D medical image comprises the second structure and is similarly oriented to the at least one 2D projection (116) corresponding to the first 3D medical image;

c) applying (112, 114) the at least one rendering transform to the first 3D medical image (106) to provide at least one 2D projection (116) corresponding to the first 3D medical image and the second 3D medical image (108) to provide at least one 2D projection (118) corresponding to the second 3D medical image;

characterized in that the method further comprises:
d) determining (120) at least one elastic registration (122) for mapping one or more locations in the at least one 2D projection (118) corresponding to the second 3D medical image to homologous one or more locations in the at least one 2D projection (116) corresponding to the first 3D medical image, wherein determining (120) each of the at least one elastic registration comprises:

i.) generating a plurality of candidate affine mappings wherein each candidate affine mapping in the plurality of candidate affine mappings defines a mapping of the one or more locations in the at least one 2D projection (118) corresponding to the second 3D medical image to the homologous one or more locations in the at least one 2D projection (116) corresponding to the first 3D medical image;

ii.) determining a plurality of similarity values, wherein for each candidate affine mapping in the plurality of candidate affine mappings the plurality of similarity values includes an associated similarity value corresponding with the quality of the registration under that candidate affine mapping;

iii.) selecting from the plurality of candidate affine mappings a candidate affine mapping with a highest associated similarity value; and

iv.) initializing the elastic registration based on the candidate affine mapping with the highest associated similarity value;

the method further comprising
for each point (p') in the second structure of a 2D projection (118) corresponding to the second 3D medical image, storing 3D location information for mapping that point back to a corresponding point (p) in the second 3D medical image, thus forming a first 3D point set (P);
for each point (q') in the first structure of a 2D projection (116) corresponding to the first 3D medical image, which is homologous to a point (p') in the second structure of the 2D projection (118) corresponding to the second 3D medical image, storing 3D location information for mapping that point back to a corresponding point (q) in the first 3D medical image, thus forming a second 3D point set (Q);
and
initializing a 3D elastic registration based on the relationship between the first 3D point set (P) and the second 3D point set (Q), as derived from the elastic registration of the 2D projection (118) corresponding to the second 3D medical image and the 2D projection (116) corresponding to the first 3D medical image.
 
2. The method as defined in claim 1 wherein the first structure and the second structure are anatomical structures consisting of at least one of bones, organs, and blood vessels.
 
3. The method as defined in claim 1 wherein
b) comprises determining a first rendering transform and a second rendering transform of the at least one rendering transform such that the first rendering transform provides a 2D projection (116) corresponding to the first 3D medical image and a 2D projection (118) corresponding to the second 3D medical image from a first viewing direction and the second rendering transform provides a 2D projection (116) corresponding to the first 3D medical image and a 2D projection (118) corresponding to the second 3D medical image from a second viewing direction, wherein the first viewing direction is a different viewing direction than the second viewing direction.
 
4. The method as defined in claim 1 wherein
the one or more locations in the at least one 2D projection (118) corresponding to the second 3D medical image comprise a contour (904, 914) surrounding an area of interest.
 
5. The method as defined in claim 4 wherein the at least one 2D projection (116) corresponding to the first 3D medical image comprises at least two 2D projections (116) corresponding to the first 3D medical image and the method further comprises generating a contour extrusion intersection sub-volume (1010) using the contour (908, 918) of the at least two 2D projections (116) corresponding to the first 3D medical image.
 
6. The method as defined in claim 5 wherein

the first structure and the second structure comprise skeletal structure;

the contour (904, 914) in each 2D projection of the at least one 2D projection (118) corresponding to the second 3D medical image is selected to surround specific soft tissue based on a location of the specific soft tissue relative to the skeletal structure of that 2D projection (118) such that the homologous contour (908, 918) for each 2D projection (116) of the at least one 2D projection corresponding to the first 3D medical image surrounds the specific soft tissue of a first patient.


 
7. The method as defined in claim 3 wherein

the first structure comprises a first anatomical surface;

the second structure comprises a second anatomical surface;

the method further comprises, for each point of a first plurality of points in the first structure of the at least one 2D projection (116) corresponding to the first 3D medical image, deriving a first surface parameter from the first anatomical surface of the first structure;

the method further comprises, for each point of a second plurality of points in the second structure of the at least one 2D projection (118) corresponding to the second 3D medical image, deriving a second surface parameter from the second anatomical surface of the second structure; and
d) comprises, for each point in the first plurality of points and for each point in the second plurality of points determining (120) the at least one elastic registration (122) based on the first surface parameter and the second surface parameter.


 
8. The method as defined in claim 7 wherein

the method further comprises, for each point of the first plurality of points deriving a plurality of first surface parameters including the first surface parameter from the first anatomical surface of the first structure;

the method further comprises, for each point of the second plurality of points deriving a plurality of second surface parameters including the second surface parameter, from the second anatomical surface of the second structure; and
d) comprises, for each point in the first plurality of points and for each point in the second plurality of points determining (120) the at least one elastic registration (122) based on the plurality of first surface parameters and the plurality of second surface parameters.


 
9. The method of claim 2 wherein
the at least one 2D projection (116) corresponding to the first 3D medical image has a first longitudinal axis and a first lateral axis;
the at least one 2D projection (118) corresponding to the second 3D medical image has a second longitudinal axis and a second lateral axis where the second longitudinal axis is divided into a plurality of longitudinal increments; and
d) comprises for each longitudinal increment (525) in the plurality of longitudinal increments, generating a candidate affine mapping defined by a plurality of mapping parameters for mapping locations of the at least one 2D projection (118) corresponding to the second 3D medical image to homologous locations in the at least one 2D projection (116) corresponding to the first 3D medical image where generating the candidate affine mapping comprises

aligning the first longitudinal axis with the second longitudinal axis;

shifting the at least one 2D projection (116) corresponding to the first 3D medical image along the second longitudinal axis to the longitudinal increment (525) such that a first overlap region of the at least one 2D projection (116) corresponding to the first 3D medical image overlaps with a second overlap region of the at least one 2D projection (118) corresponding to the second 3D medical image;

selecting the plurality of mapping parameters to maximize an associated similarity value where the associated similarity value is a measure of how similar the first overlap region is to the second overlap region under that candidate affine mapping; and

determining (120) the at least one elastic registration (122) based on the candidate affine mapping with the highest associated similarity value.


 
10. The method of claim 2 wherein
the at least one 2D projection (116) corresponding to the first 3D medical image has a first longitudinal axis and a first lateral axis;
the at least one 2D projection (118) corresponding to the second 3D medical image has a second longitudinal axis and a second lateral axis where the second longitudinal axis is divided into a plurality of longitudinal increments; and
d) comprises, for each longitudinal increment (525) in the plurality of longitudinal increments,

v) generating a candidate affine mapping defined by a plurality of mapping parameters for mapping locations in the at least one 2D projection (118) corresponding to the second 3D medical image to homologous locations in the at least one 2D projection (116) corresponding to the first 3D medical image where generating the candidate affine mapping comprises

vi) defining a first region along the first longitudinal axis of the at least one 2D projection (116) corresponding to the first 3D medical image;

vii) based on the longitudinal increment (525), defining a second region along the second longitudinal axis of the at least one 2D projection (118) corresponding to the second 3D medical image;

viii) selecting the plurality of mapping parameters to maximize an associated similarity value where the maximized associated similarity value is a measure of how similar the first region is to the second region under that candidate affine mapping;
after v), vi), vii) and viii), for each longitudinal increment (525) in the plurality of longitudinal increments, determining (120) the at least one elastic registration (122) based on the candidate affine mapping with a highest associated similarity value of a plurality of maximized similarity values, wherein the plurality of maximized similarity values comprises the highest similarity value for each longitudinal increment in the plurality of longitudinal increments.


 
11. The method of claim 10 wherein the plurality of mapping parameters include a longitudinal scaling parameter for adjusting scaling of the at least one 2D projection (116) corresponding to the first 3D medical image along the second longitudinal axis and a lateral scaling parameter for adjusting scaling of the at least one 2D projection (116) corresponding to the first 3D medical image along the second lateral axis.
 
12. The method of claim 2 wherein d) comprises
determining whether the at least one 2D projection (116) corresponding to the first 3D medical image exceeds a specified threshold size and where the at least one 2D projection (116) corresponding to the first 3D medical image exceeds the specified threshold size, dividing the at least one 2D projection (116) corresponding to the first 3D medical image into a first region (600) and a second region (601) where the first region (600) and the second region (601) are not co-extensive;
determining a first elastic registration for mapping locations in the at least one 2D projection (118) corresponding to the second 3D medical image to homologous locations in the first region (600) of the at least one 2D projection (116) corresponding to the first 3D medical image; and
determining a second elastic registration for mapping locations in the at least one 2D projection (118) corresponding to the second 3D medical image to homologous locations in the second region (601) of the at least one 2D projection (116) corresponding to the first 3D medical image where the second elastic registration is based on the first elastic registration.
 
13. The method of claim 10 wherein
the at least one 2D projection (116) corresponding to the first 3D medical image has a first resolution defined by a first number of points in the at least one 2D projection (116) corresponding to the first 3D medical image;
the at least one 2D projection (118) corresponding to the second 3D medical image has a second resolution defined by a second number of points in the at least one 2D projection (118) corresponding to the second 3D medical image;
the method further comprises reducing the first resolution of the at least one 2D projection (116) corresponding to the first 3D medical image by averaging each nxn block of points prior to determining (120) the at least one elastic registration (122) wherein n is any integer greater than one; and
the method further comprises reducing the second resolution of the at least one 2D projection (118) corresponding to the second 3D medical image by averaging each nxn block of points prior to determining (120) the at least one elastic registration (122) wherein n is any integer greater than one.
 
14. The method as defined in claim 1 wherein the at least one 2D projection corresponding to the first 3D medical image comprises a first and a second 2D projections and the at least one 2D projection corresponding to the second 3D medical image comprises a first and a second 2D projections, and

b) comprises determining a first and a second rendering transforms;

c) comprises applying the first and second rendering transforms to the first 3D medical image to provide the first and second 2D projections and the second 3D medical image to provide the first and second 2D projections, respectively; and

d) comprises determining a first and a second elastic registrations.


 
15. A system for registering 3D medical image data comprising:

a processor (1310);

a memory (1312) for storing a first 3D medical image (106), a second 3D medical image (108), and means for configuring the processor to perform the steps of any of claims 1 to 14.


 
16. A computer program product for use on a computer system (1300) to register 3D medical image data, the computer program product comprising:

a recording medium;

means recorded on the medium for instructing the computer system (1300) to perform the steps of any of claims 1 to 14.


 
17. The method as defined in claim 14 wherein
each of the first and second 2D projections corresponding to the second 3D medical image comprises a marked position within an area of interest;
in d), the first and second elastic registrations are operable to map the marked positions of the first and second 2D projections corresponding to the second 3D medical image to homologous locations in the first and second 2D projections corresponding to the first 3D medical image;
wherein the method further comprises:

g) generating a seed point from the intersection in 3D of the homologous locations in the first and second 2D projections corresponding to the first 3D medical image; and

h) using the seed point as a starting point for a segmentation algorithm based on region growing.


 


Ansprüche

1. Ein Verfahren zum Registrieren medizinischer 3D-Bilddaten, wobei das Verfahren Folgendes beinhaltet:

a) Empfangen (102, 104) eines ersten medizinischen 3D-Bilds (106) und eines zweiten medizinischen 3D-Bilds (108);

b) Bestimmen (110) mindestens einer Rendering-Transformation zur Anwendung auf sowohl das erste medizinische 3D-Bild (106) als auch das zweite medizinische 3D-Bild (108), wobei das Bestimmen (110) Folgendes beinhaltet:

i) Auswählen einer ersten Struktur im ersten medizinischen 3D-Bild (106) und einer zweiten Struktur im zweiten medizinischen 3D-Bild (108), wobei die zweite Struktur zur ersten Struktur homolog ist, wobei die erste Struktur und die zweite Struktur anatomische Strukturen sind; und

ii) Bestimmen (110) der mindestens einen Rendering-Transformation so, dass mindestens eine 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, die erste Struktur beinhaltet und mindestens eine 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, die zweite Struktur beinhaltet und zur mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, ähnlich orientiert ist;

c) Anwenden (112, 114) der mindestens einen Rendering-Transformation auf das erste medizinische 3D-Bild (106), um mindestens eine 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, bereitzustellen, und das zweite medizinische 3D-Bild (108), um mindestens eine 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, bereitzustellen;

dadurch gekennzeichnet, dass das Verfahren ferner Folgendes beinhaltet:
d) Bestimmen (120) mindestens einer elastischen Registrierung (122) zum Abbilden einer oder mehrerer Stellen in der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, auf eine oder mehrere homologe Stellen in der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, wobei das Bestimmen (120) jeder der mindestens einen elastischen Registrierung Folgendes beinhaltet:

i.) Generieren einer Vielzahl von in Frage kommenden affinen Abbildungen, wobei jede in Frage kommende affine Abbildung in der Vielzahl von in Frage kommenden affinen Abbildungen eine Abbildung der einen oder der mehreren Stellen in der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, auf die eine oder die mehreren homologen Stellen in der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, definiert;

ii.) Bestimmen einer Vielzahl von Ähnlichkeitswerten, wobei für jede in Frage kommende affine Abbildung in der Vielzahl von in Frage kommenden affinen Abbildungen die Vielzahl von Ähnlichkeitswerten einen zugeordneten Ähnlichkeitswert, der mit der Qualität der Registrierung bei dieser in Frage kommenden affinen Abbildung korrespondiert, umfasst;

iii.) Auswählen einer in Frage kommenden affinen Abbildung mit einem höchsten zugeordneten Ähnlichkeitswert aus der Vielzahl von in Frage kommenden affinen Abbildungen; und

iv.) Einleiten der elastischen Registrierung basierend auf der in Frage kommenden affinen Abbildung mit dem höchsten zugeordneten Ähnlichkeitswert;

wobei das Verfahren ferner Folgendes beinhaltet:

für jeden Punkt (p') in der zweiten Struktur einer 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, Speichern von 3D-Stelleninformationen zum Zurückabbilden dieses Punkts auf einen korrespondierenden Punkt (p) im zweiten medizinischen 3D-Bild, wodurch mithin eine erste 3D-Punktmenge (P) gebildet wird;

für jeden Punkt (q') in der ersten Struktur einer 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, der zu einem Punkt (p') in der zweiten Struktur der 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert,

homolog ist, Speichern von 3D-Stelleninformationen zum Zurückabbilden dieses Punkts auf einen korrespondierenden Punkt (q) im ersten medizinischen 3D-Bild, wodurch mithin eine zweite 3D-Punktmenge (Q) gebildet wird; und

Einleiten einer elastischen 3D-Registrierung basierend auf der Beziehung zwischen der ersten 3D-Punktmenge (P) und der zweiten 3D-Punktmenge (Q), wie von der elastischen Registrierung der 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, und der 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, abgeleitet.


 
2. Verfahren gemäß Anspruch 1, wobei die erste Struktur und die zweite Struktur anatomische Strukturen, die aus mindestens einem von Knochen, Organen und Blutgefäßen bestehen, sind.
 
3. Verfahren gemäß Anspruch 1, wobei
b) Bestimmen einer ersten Rendering-Transformation und einer zweiten Rendering-Transformation der mindestens einen Rendering-Transformation so, dass die erste Rendering-Transformation eine 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, und eine 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, aus einer ersten Blickrichtung bereitstellt und die zweite Rendering-Transformation eine 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, und eine 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, aus einer zweiten Blickrichtung bereitstellt, wobei die erste Blickrichtung eine andere Blickrichtung als die zweite Blickrichtung ist, beinhaltet.
 
4. Verfahren gemäß Anspruch 1, wobei
die eine oder die mehreren Stellen in der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, eine Kontur (904, 914), die einen Bereich von Interesse umgibt, beinhaltet/beinhalten.
 
5. Verfahren gemäß Anspruch 4,
wobei die mindestens eine 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, mindestens zwei 2D-Projektionen (116), die mit dem ersten medizinischen 3D-Bild korrespondieren, beinhaltet und das Verfahren ferner Generieren eines Konturextrusionsschnittpunkt-Subvolumens (1010) unter Nutzung der Kontur (908, 918) der mindestens zwei 2D-Projektionen (116), die mit dem ersten medizinischen 3D-Bild korrespondieren, beinhaltet.
 
6. Verfahren gemäß Anspruch 5, wobei
die erste Struktur und die zweite Struktur skelettale Strukturen beinhalten;
die Kontur (904, 914) in jeder 2D-Projektion der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, so, dass sie ein betreffendes Weichgewebe umgibt, basierend auf einer Stelle des betreffenden Weichgewebes relativ zur skelettalen Struktur dieser 2D-Projektion (118) ausgewählt wird, sodass die homologe Kontur (908, 918) für jede 2D-Projektion (116) der mindestens einen 2D-Projektion, die mit dem ersten medizinischen 3D-Bild korrespondiert, das betreffende Weichgewebe eines ersten Patienten umgibt.
 
7. Verfahren gemäß Anspruch 3, wobei
die erste Struktur eine erste anatomische Oberfläche beinhaltet;
die zweite Struktur eine zweite anatomische Oberfläche beinhaltet;
das Verfahren ferner für jeden Punkt einer ersten Vielzahl von Punkten in der ersten Struktur der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, Ableiten eines ersten Oberflächenparameters von der ersten anatomischen Oberfläche der ersten Struktur beinhaltet;
das Verfahren ferner für jeden Punkt einer zweiten Vielzahl von Punkten in der zweiten Struktur der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, Ableiten eines zweiten Oberflächenparameters von der zweiten anatomischen Oberfläche der zweiten Struktur beinhaltet; und
d) für jeden Punkt in der ersten Vielzahl von Punkten und für jeden Punkt in der zweiten Vielzahl von Punkten Bestimmen (120) der mindestens einen elastischen Registrierung (122) basierend auf dem ersten Oberflächenparameter und dem zweiten Oberflächenparameter beinhaltet.
 
8. Verfahren gemäß Anspruch 7, wobei
das Verfahren ferner für jeden Punkt der ersten Vielzahl von Punkten Ableiten einer Vielzahl erster Oberflächenparameter, die den ersten Oberflächenparameter umfassen, von der ersten anatomischen Oberfläche der ersten Struktur beinhaltet;
das Verfahren ferner für jeden Punkt der zweiten Vielzahl von Punkten Ableiten einer Vielzahl zweiter Oberflächenparameter, die den zweiten Oberflächenparameter umfassen, von der zweiten anatomischen Oberfläche der zweiten Struktur beinhaltet; und
d) für jeden Punkt in der ersten Vielzahl von Punkten und für jeden Punkt in der zweiten Vielzahl von Punkten Bestimmen (120) der mindestens einen elastischen Registrierung (122) basierend auf der Vielzahl erster Oberflächenparameter und der Vielzahl zweiter Oberflächenparameter beinhaltet.
 
9. Verfahren gemäß Anspruch 2, wobei
die mindestens eine 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, eine erste Längsachse und eine erste Querachse aufweist;
die mindestens eine 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, eine zweite Längsachse und eine zweite Querachse aufweist, wobei die zweite Längsachse in eine Vielzahl von Längsinkrementen aufgeteilt ist; und
d) für jedes Längsinkrement (525) in der Vielzahl von Längsinkrementen Generieren einer in Frage kommenden affinen Abbildung, die durch eine Vielzahl von Abbildungsparametern zum Abbilden von Stellen der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, auf homologe Stellen in der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, definiert ist, beinhaltet, wobei das Generieren der in Frage kommenden affinen Abbildung Folgendes beinhaltet:
Ausrichten der ersten Längsachse an der zweiten Längsachse;
Verschieben der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, entlang der zweiten Längsachse entsprechend dem Längsinkrement (525), sodass eine erste Überlappungsregion der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, mit einer zweiten Überlappungsregion der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, überlappt;
Auswählen der Vielzahl von Abbildungsparametern, um einen zugeordneten Ähnlichkeitswert zu maximieren, wobei der zugeordnete Ähnlichkeitswert ein Maß dafür ist, wie ähnlich die erste Überlappungsregion der zweiten Überlappungsregion bei dieser in Frage kommenden affinen Abbildung ist; und
Bestimmen (120) der mindestens einen elastischen Registrierung (122) basierend auf der in Frage kommenden affinen Abbildung mit dem höchsten zugeordneten Ähnlichkeitswert.
 
10. Verfahren gemäß Anspruch 2, wobei
die mindestens eine 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, eine erste Längsachse und eine erste Querachse aufweist;
die mindestens eine 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, eine zweite Längsachse und eine zweite Querachse aufweist, wobei die zweite Längsachse in eine Vielzahl von Längsinkrementen aufgeteilt ist; und
d) für jedes Längsinkrement (525) in der Vielzahl von Längsinkrementen Folgendes beinhaltet:

v) Generieren einer in Frage kommenden affinen Abbildung, die durch eine Vielzahl von Abbildungsparametern zum Abbilden von Stellen in der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, auf homologe Stellen in der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, definiert ist, wobei das Generieren der in Frage kommenden affinen Abbildung Folgendes beinhaltet:

vi) Definieren einer ersten Region entlang der ersten Längsachse der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert;

vii) basierend auf dem Längsinkrement (525) Definieren einer zweiten Region entlang der zweiten Längsachse der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert;

viii) Auswählen der Vielzahl von Abbildungsparametern, um einen zugeordneten Ähnlichkeitswert zu maximieren, wobei der maximierte zugeordnete Ähnlichkeitswert ein Maß dafür ist, wie ähnlich die erste Region der zweiten Region bei dieser in Frage kommenden affinen Abbildung ist;

nach v), vi), vii) und viii) für jedes Längsinkrement (525) in der Vielzahl von Längsinkrementen Bestimmen (120) der mindestens einen elastischen Registrierung (122) basierend auf der in Frage kommenden affinen Abbildung mit einem höchsten zugeordneten Ähnlichkeitswert einer Vielzahl maximierter Ähnlichkeitswerte, wobei die Vielzahl maximierter Ähnlichkeitswerte den höchsten Ähnlichkeitswert für jedes Längsinkrement in der Vielzahl von Längsinkrementen beinhaltet.
 
11. Verfahren gemäß Anspruch 10, wobei die Vielzahl von Abbildungsparametern einen Längsskalierungsparameter zum Anpassen einer Skalierung der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, entlang der zweiten Längsachse und einen Querskalierungsparameter zum Anpassen der Skalierung der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, entlang der zweiten Querachse umfasst.
 
12. Verfahren gemäß Anspruch 2, wobei d) Folgendes beinhaltet:

Bestimmen, ob die mindestens eine 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, einen vorgegebenen Größenschwellenwert überschreitet, und, wenn die mindestens eine 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, den vorgegebenen Größenschwellenwert überschreitet, Aufteilen der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, in eine erste Region (600) und eine zweite Region (601), wobei die erste Region (600) und die zweite Region (601) nicht flächengleich sind;

Bestimmen einer ersten elastischen Registrierung zum Abbilden von Stellen in der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, auf homologe Stellen in der ersten Region (600) der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert; und

Bestimmen einer zweiten elastischen Registrierung zum Abbilden von Stellen in der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, auf homologe Stellen in der zweiten Region (601) der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, wobei die zweite elastische Registrierung auf der ersten elastischen Registrierung basiert.


 
13. Verfahren gemäß Anspruch 10, wobei
die mindestens eine 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, eine erste Auflösung, die durch eine erste Anzahl von Punkten in der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, definiert wird, aufweist;
die mindestens eine 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, eine zweite Auflösung, die durch eine zweite Anzahl von Punkten in der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, definiert wird, aufweist;
das Verfahren ferner Verringern der ersten Auflösung der mindestens einen 2D-Projektion (116), die mit dem ersten medizinischen 3D-Bild korrespondiert, durch eine Mittelwertbildung für jeden nxn-Punktblock vor dem Bestimmen (120) der mindestens einen elastischen Registrierung (122), wobei n eine beliebige ganze Zahl größer als eins ist, beinhaltet; und
das Verfahren ferner Verringern der zweiten Auflösung der mindestens einen 2D-Projektion (118), die mit dem zweiten medizinischen 3D-Bild korrespondiert, durch eine Mittelwertbildung für jeden nxn-Punktblock vor dem Bestimmen (120) der mindestens einen elastischen Registrierung (122), wobei n eine beliebige ganze Zahl größer als eins ist, beinhaltet.
 
14. Verfahren gemäß Anspruch 1, wobei die mindestens eine 2D-Projektion, die mit dem ersten medizinischen 3D-Bild korrespondiert, eine erste und eine zweite 2D-Projektion beinhaltet und die mindestens eine 2D-Projektion, die mit dem zweiten medizinischen 3D-Bild korrespondiert, eine erste und eine zweite 2D-Projektion beinhaltet und

b) Bestimmen einer ersten und einer zweiten Rendering-Transformation beinhaltet;

c) Anwenden der ersten und der zweiten Rendering-Transformation auf das erste medizinische 3D-Bild, um die erste und die zweite 2D-Projektion bereitzustellen, bzw. das zweite medizinische 3D-Bild, um die erste und die zweite 2D-Projektion bereitzustellen, beinhaltet; und

d) Bestimmen einer ersten und einer zweiten elastischen Registrierung beinhaltet.


 
15. Ein System zum Registrieren medizinischer 3D-Bilddaten, das Folgendes beinhaltet:

einen Prozessor (1310);

einen Speicher (1312) zum Speichern eines ersten medizinischen 3D-Bilds (106), eines zweiten medizinischen 3D-Bilds (108) und von Mitteln zum Konfigurieren des Prozessors zum Durchführen der Schritte gemäß einem der Ansprüche 1 bis 14.


 
16. Ein Computerprogrammprodukt zur Nutzung in einem Computersystem (1300) zum Registrieren medizinischer 3D-Bilddaten, wobei das Computerprogrammprodukt Folgendes beinhaltet:

ein Aufzeichnungsmedium;

in dem Medium aufgezeichnete Mittel zum Anweisen des Computersystems (1300) zum Durchführen der Schritte gemäß einem der Ansprüche 1 bis 14.


 
17. Verfahren gemäß Anspruch 14, wobei
sowohl die erste als auch die zweite 2D-Projektion, die mit dem zweiten medizinischen 3D-Bild korrespondieren, eine markierte Position innerhalb eines Bereichs von Interesse beinhalten;
die erste und die zweite elastische Registrierung in d) dazu dienen, die markierten Positionen der ersten und der zweiten 2D-Projektion, die mit dem zweiten medizinischen 3D-Bild korrespondieren, auf homologe Stellen in der ersten und der zweiten 2D-Projektion, die mit dem ersten medizinischen 3D-Bild korrespondieren, abzubilden;
wobei das Verfahren ferner Folgendes beinhaltet:

g) Generieren eines Saatpunkts aus dem 3D-Schnittpunkt der homologen Stellen in der ersten und der zweiten 2D-Projektion, die mit dem ersten medizinischen 3D-Bild korrespondieren; und

h) Nutzen des Saatpunkts als Ausgangspunkt für einen Segmentierungsalgorithmus basierend auf einem Regionenwachstum.


 


Revendications

1. Un procédé pour enregistrer des données d'images médicales 3D, le procédé comprenant :

a) la réception (102, 104) d'une première image médicale 3D (106) et d'une deuxième image médicale 3D (108) ;

b) la détermination (110) d'au moins une transformée de rendu pour appliquer à chaque image parmi la première image médicale 3D (106) et la deuxième image médicale 3D (108), la détermination (110) comprenant :

i) la sélection d'une première structure dans la première image médicale 3D (106) et d'une deuxième structure dans la deuxième image médicale 3D (108), la deuxième structure étant homologue à la première structure, la première structure et la deuxième structure étant des structures anatomiques ; et

ii) la détermination (110) de l'au moins une transformée de rendu de telle sorte qu'au moins une projection 2D (116) correspondant à la première image médicale 3D comprenne la première structure, et qu'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D comprenne la deuxième structure et soit orientée de façon similaire à l'au moins une projection 2D (116) correspondant à la première image médicale 3D;

c) l'application (112, 114) de l'au moins une transformée de rendu à la première image médicale 3D (106) pour fournir au moins une projection 2D (116) correspondant à la première image médicale 3D et à la deuxième image médicale 3D (108) pour fournir au moins une projection 2D (118) correspondant à la deuxième image médicale 3D ;

caractérisé en ce que le procédé comprend en outre :
d) la détermination (120) d'au moins un enregistrement élastique (122) pour mapper un ou plusieurs emplacements dans l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D sur un ou plusieurs emplacements homologues dans l'au moins une projection 2D (116) correspondant à la première image médicale 3D, la détermination (120) de chaque enregistrement parmi l'au moins un enregistrement élastique comprenant :

i.) la génération d'une pluralité de mappages affines candidats, chaque mappage affine candidat dans la pluralité de mappages affines candidats définissant un mappage du ou des emplacements dans l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D sur l'emplacement ou les emplacements homologues dans l'au moins une projection 2D (116) correspondant à la première image médicale 3D ;

ii.) la détermination d'une pluralité de valeurs de similarité, dans lequel pour chaque mappage affine candidat dans la pluralité de mappages affines candidats, la pluralité de valeurs de similarité inclut une valeur de similarité associée correspondant à la qualité de l'enregistrement sous ce mappage affine candidat ;

iii.) la sélection parmi la pluralité de mappages affines candidats d'un mappage affine candidat ayant une valeur de similarité associée la plus élevée ; et iv.) l'initialisation de l'enregistrement élastique sur la base du mappage affine candidat avec la valeur de similarité associée la plus élevée ;

le procédé comprenant en outre
pour chaque point (p') dans la deuxième structure d'une projection 2D (118) correspondant à la deuxième image médicale 3D, le stockage d'informations d'emplacement 3D pour mapper ce point en retour sur un point correspondant (p) dans la deuxième image médicale 3D, formant ainsi un premier ensemble de points 3D (P) ; pour chaque point (q') dans la première structure d'une projection 2D (116) correspondant à la première image médicale 3D, qui est homologue à un point (p') dans la deuxième structure de la projection 2D (118) correspondant à la deuxième image médicale 3D, le stockage d'informations d'emplacement 3D pour mapper ce point en retour sur un point correspondant (q) dans la première image médicale 3D, formant ainsi un deuxième ensemble de points 3D (Q) ; et
l'initialisation d'un enregistrement élastique 3D sur la base de la relation entre le premier ensemble de points 3D (P) et le deuxième ensemble de points 3D (Q), tel que dérivé de l'enregistrement élastique de la projection 2D (118) correspondant à la deuxième image médicale 3D et de la projection 2D (116) correspondant à la première image médicale 3D.
 
2. Le procédé tel que défini dans la revendication 1 dans lequel la première structure et la deuxième structure sont des structures anatomiques constituées d'au moins un élément parmi des os, des organes, et des vaisseaux sanguins.
 
3. Le procédé tel que défini dans la revendication 1 dans lequel
b) comprend la détermination d'une première transformée de rendu et d'une deuxième transformée de rendu de l'au moins une transformée de rendu de telle sorte que la première transformée de rendu fournisse une projection 2D (116) correspondant à la première image médicale 3D et une projection 2D (118) correspondant à la deuxième image médicale 3D à partir d'un premier sens de visualisation et que la deuxième transformée de rendu fournisse une projection 2D (116) correspondant à la première image médicale 3D et une projection 2D (118) correspondant à la deuxième image médicale 3D à partir d'un deuxième sens de visualisation, le premier sens de visualisation étant une direction de visualisation différente du deuxième sens de visualisation.
 
4. Le procédé tel que défini dans la revendication 1 dans lequel l'emplacement ou les emplacements dans l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D comprennent un contour (904, 914) entourant une zone d'intérêt.
 
5. Le procédé tel que défini dans la revendication 4
dans lequel l'au moins une projection 2D (116) correspondant à la première image médicale 3D comprend au moins deux projections 2D (116) correspondant à la première image médicale 3D et le procédé comprend en outre la génération d'un sous-volume d'intersection d'extrusion de contour (1010) à l'aide du contour (908, 918) des au moins deux projections 2D (116) correspondant à la première image médicale 3D.
 
6. Le procédé tel que défini dans la revendication 5 dans lequel la première structure et la deuxième structure comprennent une structure squelettique ; le contour (904, 914) dans chaque projection 2D de l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D est sélectionné pour entourer un tissu mou spécifique sur la base d'un emplacement du tissu mou spécifique par rapport à la structure squelettique de cette projection 2D (118) de telle sorte que le contour homologue (908, 918) pour chaque projection 2D (116) de l'au moins une projection 2D correspondant à la première image médicale 3D entoure le tissu mou spécifique d'un premier patient.
 
7. Le procédé tel que défini dans la revendication 3 dans lequel la première structure comprend une première surface anatomique ;
la deuxième structure comprend une deuxième surface anatomique ;
le procédé comprend en outre, pour chaque point d'une première pluralité de points dans la première structure de l'au moins une projection 2D (116) correspondant à la première image médicale 3D, la dérivation d'un premier paramètre de surface à partir de la première surface anatomique de la première structure ;
le procédé comprend en outre, pour chaque point d'une deuxième pluralité de points dans la deuxième structure de l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D, la dérivation d'un deuxième paramètre de surface à partir de la deuxième surface anatomique de la deuxième structure ; et
d) comprend, pour chaque point dans la première pluralité de points et pour chaque point dans la deuxième pluralité de points la détermination (120) de l'au moins un enregistrement élastique (122) sur la base du premier paramètre de surface et du deuxième paramètre de surface.
 
8. Le procédé tel que défini dans la revendication 7 dans lequel le procédé comprend en outre, pour chaque point de la première pluralité de points, la dérivation d'une pluralité de premiers paramètres de surface incluant le premier paramètre de surface à partir de la première surface anatomique de la première structure ;
le procédé comprend en outre, pour chaque point de la deuxième pluralité de points, la dérivation d'une pluralité de deuxièmes paramètres de surface incluant le deuxième paramètre de surface à partir de la deuxième surface anatomique de la deuxième structure ; et
d) comprend, pour chaque point dans la première pluralité de points et pour chaque point dans la deuxième pluralité de points la détermination (120) de l'au moins un enregistrement élastique (122) sur la base de la pluralité de premiers paramètres de surface et de la pluralité de deuxièmes paramètres de surface.
 
9. Le procédé de la revendication 2 dans lequel
l'au moins une projection 2D (116) correspondant à la première image médicale 3D a un premier axe longitudinal et un premier axe latéral ;
l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D a un deuxième axe longitudinal et un deuxième axe latéral où le deuxième axe longitudinal est divisé en une pluralité d'incréments longitudinaux ; et
d) comprend pour chaque incrément longitudinal (525) dans la pluralité d'incréments longitudinaux, la génération d'un mappage affine candidat défini par une pluralité de paramètres de mappage pour mapper des emplacements de l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D sur des emplacements homologues dans l'au moins une projection 2D (116) correspondant à la première image médicale 3D où la génération du mappage affine candidat comprend l'alignement du premier axe longitudinal avec le deuxième axe longitudinal ;
le décalage de l'au moins une projection 2D (116) correspondant à la première image médicale 3D le long du deuxième axe longitudinal à l'incrément longitudinal (525) de telle sorte qu'une première région de chevauchement de l'au moins une projection 2D (116) correspondant à la première image médicale 3D chevauche une deuxième région de chevauchement de l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D ;
la sélection de la pluralité de paramètres de mappage pour maximiser une valeur de similarité associée où la valeur de similarité associée est une mesure de la façon dont la première région de chevauchement est similaire à la deuxième région de chevauchement sous ce mappage affine candidat ; et
la détermination (120) de l'au moins un enregistrement élastique (122) sur la base du mappage affine candidat avec la valeur de similarité associée la plus élevée.
 
10. Le procédé de la revendication 2 dans lequel
l'au moins une projection 2D (116) correspondant à la première image médicale 3D a un premier axe longitudinal et un premier axe latéral ;
l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D a un deuxième axe longitudinal et un deuxième axe latéral où le deuxième axe longitudinal est divisé en une pluralité d'incréments longitudinaux ; et
d) comprend, pour chaque incrément longitudinal (525) dans la pluralité d'incréments longitudinaux,

v) la génération d'un mappage affine candidat défini par une pluralité de paramètres de mappage pour mapper des emplacements dans l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D sur des emplacements homologues dans l'au moins une projection 2D (116) correspondant à la première image médicale 3D où la génération du mappage affine candidat comprend

vi) la définition d'une première région le long du premier axe longitudinal de l'au moins une projection 2D (116) correspondant à la première image médicale 3D ;

vii) sur la base de l'incrément longitudinal (525), la définition d'une deuxième région le long du deuxième axe longitudinal de l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D ;

viii) la sélection de la pluralité de paramètres de mappage pour maximiser une valeur de similarité associée où la valeur de similarité associée maximisée est une mesure de la façon dont la première région est similaire à la deuxième région sous ce mappage affine candidat ;

après v), vi), vii) et viii), pour chaque incrément longitudinal (525) dans la pluralité d'incréments longitudinaux, la détermination (120) de l'au moins un enregistrement élastique (122) sur la base du mappage affine candidat avec une valeur de similarité associée la plus élevée d'une pluralité de valeurs de similarité maximisées, la pluralité de valeurs de similarité maximisées comprenant la valeur de similarité la plus élevée pour chaque incrément longitudinal dans la pluralité d'incréments longitudinaux.
 
11. Le procédé de la revendication 10 dans lequel la pluralité de paramètres de mappage inclut un paramètre de mise à l'échelle longitudinale pour ajuster la mise à l'échelle de l'au moins une projection 2D (116) correspondant à la première image médicale 3D le long du deuxième axe longitudinal et un paramètre de mise à l'échelle latérale pour ajuster la mise à l'échelle de l'au moins une projection 2D (116) correspondant à la première image médicale 3D le long du deuxième axe latéral.
 
12. Le procédé de la revendication 2 dans lequel d) comprend
la détermination du fait que l'au moins une projection 2D (116) correspondant à la première image médicale 3D dépasse une taille seuil spécifiée et où l'au moins une projection 2D (116) correspondant à la première image médicale 3D dépasse la taille seuil spécifiée, la division de l'au moins une projection 2D (116) correspondant à la première image médicale 3D en une première région (600) et une deuxième région (601) où la première région (600) et la deuxième région (601) ne sont pas coextensives ;
la détermination d'un premier enregistrement élastique pour mapper des emplacements dans l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D sur des emplacements homologues dans la première région (600) de l'au moins une projection 2D (116) correspondant à la première image médicale 3D ; et
la détermination d'un deuxième enregistrement élastique pour mapper des emplacements dans l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D sur des emplacements homologues dans la deuxième région (601) de l'au moins une projection 2D (116) correspondant à la première image médicale 3D où le deuxième enregistrement élastique est basé sur le premier enregistrement élastique.
 
13. Le procédé de la revendication 10 dans lequel
l'au moins une projection 2D (116) correspondant à la première image médicale 3D a une première résolution définie par un premier nombre de points dans l'au moins une projection 2D (116) correspondant à la première image médicale 3D ;
l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D a une deuxième résolution définie par un deuxième nombre de points dans l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D ;
le procédé comprend en outre la réduction de la première résolution de l'au moins une projection 2D (116) correspondant à la première image médicale 3D en calculant la moyenne de chaque bloc nxn de points avant de déterminer (120) l'au moins un enregistrement élastique (122), n étant n'importe quel nombre entier supérieur à un ; et le procédé comprend en outre la réduction de la deuxième résolution de l'au moins une projection 2D (118) correspondant à la deuxième image médicale 3D en calculant la moyenne de chaque bloc nxn de points avant de déterminer (120) l'au moins un enregistrement élastique (122), n étant n'importe quel nombre entier supérieur à un.
 
14. Le procédé tel que défini dans la revendication 1 dans lequel l'au moins une projection 2D correspondant à la première image médicale 3D comprend une première et une deuxième projection 2D et l'au moins une projection 2D correspondant à la deuxième image médicale 3D comprend une première et une deuxième projection 2D, et

b) comprend la détermination d'une première et d'une deuxième transformée de rendu ;

c) comprend l'application des première et deuxième transformées de rendu à la première image médicale 3D pour fournir les première et deuxième projections 2D et la deuxième image médicale 3D pour fournir les première et deuxième projections 2D, respectivement ; et

d) comprend la détermination d'un premier et d'un deuxième enregistrement élastique.


 
15. Un système pour enregistrer des données d'images médicales 3D comprenant :

un processeur (1310) ;

une mémoire (1312) pour stocker une première image médicale 3D (106), une deuxième image médicale 3D (108), et des moyens pour configurer le processeur afin de réaliser les étapes de n'importe lesquelles des revendications 1 à 14.


 
16. Un produit de programme informatique pour une utilisation sur un système informatique (1300) afin d'enregistrer des données d'images médicales 3D, le produit de programme informatique comprenant :

un support d'enregistrement ;

des moyens enregistrés sur le support pour donner instruction au système informatique (1300) de réaliser les étapes de n'importe lesquelles des revendications 1 à 14.


 
17. Le procédé tel que défini dans la revendication 14 dans lequel
chacune des première et deuxième projections 2D correspondant à la deuxième image médicale 3D comprend une position marquée à l'intérieur d'une zone d'intérêt ;
en d), les premier et deuxième enregistrements élastiques sont utilisables pour mapper les positions marquées des première et deuxième projections 2D correspondant à la deuxième image médicale 3D sur des emplacements homologues dans les première et deuxième projections 2D correspondant à la première image médicale 3D ;
le procédé comprenant en outre :

g) la génération d'un point-graine à partir de l'intersection en 3D des emplacements homologues dans les première et deuxième projections 2D correspondant à la première image médicale 3D ; et

h) l'utilisation du point-graine comme point de départ pour un algorithme de segmentation sur la base d'une croissance de région.


 




Drawing










































REFERENCES CITED IN THE DESCRIPTION



This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.

Patent documents cited in the description




Non-patent literature cited in the description