Registration of three dimensional image data to 2D-image-derived data

[0001] This application claims the benefit of U.S. Patent Application Ser. No. 60/650,616, filed Feb. 7, 2005, the entire disclosure of which is incorporated herein by reference.

[0002] This invention relates to registration of three-dimensional data to a reference coordinate system, and more particularly to registration of three dimensional image data with an X-ray imaging coordinate system.

[0003] Interventional medicine is the collection of medical procedures in which access to the site of treatment is made by navigation through one of the subject's blood vessels, body cavities or lumens. Interventional medicine technologies have been applied to manipulation of medical instruments which contact tissues during surgical navigation procedures, making these procedures more precise, repeatable and less dependent of the device manipulation skills of the physician. Some presently available interventional medical systems for directing the distal tip of a medical device from the proximal end of the medical device use computer-assisted navigation and a display means for providing a visual display of the medical device along with anatomical images obtained from a separate imaging apparatus. Such systems can provide a visual display of blood vessels and tissues, obtained from a Fluoroscopy (X-ray) imaging system for example, and can display a projection of the medical device being navigated to a target destination using a computer that controls the orientation of the distal tip of the medical device.

[0004] In some cases, it may be difficult for a physician to become oriented in a three dimensional setting using a display of a single-plane X-ray image projection. Enhancement or augmentation of the single-plane X-ray image may be required to aid the physician in visualizing the orientation of the medical device and three-dimensional tissue surfaces and objects in the body. A method is therefore desired for enhancing a display image of the medical device and anatomical surfaces to include three-dimensional images of surfaces and objects in the body. Likewise path information obtained from a three dimensional data set may be used to augment path information derived from two dimensional images for use in visualization, navigation and computer-controlled steering.

[0005] The present invention relates to a method for determining a transformation of a three-dimensional pre-operative image data set to obtain a registration of the three-dimensional image data with an X-ray image of a subject's body, in particular in the context of a remote navigation system. In one embodiment of the present invention, a method is provided for obtaining registration of a three-dimensional image data set of an anatomical vessel with corresponding two-dimensional image data of the vessel in an X-ray imaging system, where the method comprises identifying two points on an anatomical vessel on at least two X-ray image planes, identifying two similar points on the surface of the corresponding three-dimensional anatomical image data, determining the orientation direction of the vessel from the two identified image data surface points, determining the orientation direction of the vessel from the two points obtained from the at least two X-ray image planes, and calculating a transformation of the three-dimensional image data to obtain a best fit registration of the direction derived from the image surface points with the direction derived from the X-ray image data points.

[0006] Another embodiment of the present invention may further provide a method for automatically determining a centerline for the corresponding three-dimensional image data of the vessel using a software algorithm, determining a direction of the vessel from the two points obtained from the X-ray image planes, and calculating a transformation of the centerline direction derived from the three-dimensional image data to the direction derived from the X-ray image data points to obtain a registration of the three-dimensional image data with the two-dimensional image data of the X-ray imaging system.

[0007] Another embodiment of the present invention may provide a method for registration of three-dimensional anatomical image data with two-dimensional X-ray imaging system data to allow for overlay of pre-operative anatomical images onto the X-ray imaging planes.

[0008] Yet another embodiment of the present invention, may provide a method for registration of three dimensional anatomical image data with a three dimensional reconstruction that has been obtained from image processing of at least a pair of two dimensional images, or that has been obtained from user identification and marking of the anatomy on at least a pair of two dimensional images or the claims.

[0009] Further aspects of the present invention will become apparent from the detailed description provided hereinafter. It should be understood that the detailed description and specific examples, while indicating various embodiments and methods of the invention, are for illustration purposes only and are not intended to limit the scope of the invention.

[0010] FIG. 1 is a series of cross-sectional images 20 of a three-dimensional volumetric image data set in accordance with the principles of the present invention;

[0011] FIG. 2 is a rendering of a three-dimensional volumetric image of a vessel of a subject body; and

[0012] FIG. 3 is an illustration of a heart and corresponding pulmonary veins having points that may be used to obtain a suitable registration of a pre-operative three-dimensional image of a heart with an X-ray image in accordance with the principles of the present invention.

[0013] Corresponding reference numerals indicate corresponding parts throughout the several views of the drawings.

[0014] In one embodiment of the present invention, a method is provided for determining a transformation of a three-dimensional pre-operative image data set to obtain a registration of the three-dimensional image data with an X-ray imaging system. As shown in FIG. 1, the preferred embodiment of the present invention relates to a method for processing a three dimensional pre-operative image data set to perform a transformation that will obtain a “registration” of a pre-operative image 20, such as a vessel, to an X-ray image system. The method comprises the step of initially processing the three-dimensional visual image data. This image processing can be a means of visualizing three-dimensional structures by a series of cross-sectional images. MRI, CT and volumetric ultrasound are examples of some common methods of obtaining three-dimensional volumetric or anatomical data of an image of a heart, brain, or other area of a human body. In minimally invasive medical procedures such as cardiac catheterizations, it is often important to be able to register this data to imaging data that is acquired during the procedure. X-ray images are the preferred and commonly used imaging modality in the cardiac CathLab. Thus, the image data obtained during the procedure is usually two dimensional X-ray imaging data.

[0015] It is possible to use 2D data taken from different X-ray C-arm angulations to reconstruct structures such as arterial vessels. One method of doing this is to mark a set of points at corresponding locations in two images taken at two different C-arm orientations. This can be reconstructed as a set of 3D points by a remote navigation system computer. Furthermore, spline curves can be constructed to fit these points to represent the centerline of the vessel or to construct a vessel tree in three dimensions. While this may require the user to click on several points to perform the reconstruction, a 3D reconstruction from at least two contrast agent-enhanced 2D images could also be performed through image processing means in order to extract a 3D centerline vessel or vessel tree, as is done in some commercially available image processing systems. In the following, we shall refer to this type of 3D data reconstruction, whether based on user marking or on image processing, as 2D-image-derived data.

[0016] A registration between the 3D anatomical data and the 2D-image-derived data can be performed using one or more of the methods described in the following. Such a registration is needed by any remote navigation system that uses the 3D anatomical data as information for assisting device navigation.

[0017] In one embodiment, the user marks pairs of points near a branch on each of the vessels that meet at a branch of the vessel or network of vessels, on the at least 2X-ray images. This pair of points in effect defines the take-off orientation of that vessel at the branch. Similar points are picked on the surface of corresponding vessels in the 3D anatomical data, and the software finds corresponding points on the vessel centerline, as described in the algorithm description. At least three vessels, not all in the same plane, need to be marked in this manner to find a suitable transformation matrix that effects a registration between the two sets of data using the marked pairs of points. The registration could be done as landmark registration by directly associating corresponding points in a best-fit manner (using the standard Procrustes algorithm, for instance), or by finding the best-fit orientation that minimizes total orientation error of the picked vessel take-offs with the cost function approach given in the algorithm description.

[0018] In a second embodiment, the software could reconstruct the 3D vessel centerline and display it as a 3D model, possibly as a tubular surface, in a 3D window on a Graphical User Interface. The user could select pairs of points as described above, directly on the 3D model, and these could be used for the registration. In this case the 2D images are not directly used to perform the registration with the 3D anatomical data; instead the 3D reconstruction derived from the 2D images is.

[0019] In a third embodiment, the user marks at least 4 non-coplanar branch points in the 2D-image-derived data (which could either be directly the 2D images, or a 3D reconstruction derived from 2D images, as described above), and corresponding branch points in the 3D anatomical data. The software then performs a landmark registration between the data sets using the branch points as landmark points, using a standard method such as the Procrustes algorithm.

[0020] In a fourth embodiment, the user could mark a set of branch points and at least one direction corresponding to a vessel take-off orientation at a branch, and the software performs a best-fit registration based on finding a rigid transformation that minimizes a weighted sum of cost functions based on distances and orientations respectively.

[0021] The method of the various embodiments initially calls for the user to mark on at least two X-ray image planes two points on an anatomical vessel that provide a linear indication of the local direction of the vessel, which two anatomical vessel points have coordinates y₁ and y₂. The method then analyzes a three-dimensional image such as shown in FIG. 1 having a series of cross-sectional images 20 of a volume that was regularly sampled at some constant interval to create an image pixel data set that is utilized by image modeling software. The spaces 22 between each section are filled in so that the pixel data take on additional dimension and become volume pixels, or voxels. Because the voxels have the capacity of obscuring the view of other voxels in other sections, the voxel opacity value is preferably altered by changing the intensity of the voxel. Voxels with intensity values of zero are treated as completely transparent, and higher voxel values are more opaque.

[0022] A Gaussian blurring operation may be applied to the initial segmented three-dimensional volumetric data to provide a weighted mask or filter in the processing of the image intensity of the image voxels. Let I(l, m, n) be the image intensity at the voxel indexed by (l, m, n). We can then set: $\begin{array}{c}{K}_i=\frac{K^{\prime }}{\sum _{j=2}^2{K}_j^{\prime }},\\ \mathrm{where}{K}_{-2}^{\prime }=e^{\frac{-4}{2{\sigma }^2}},\\ K_{-1}^{\prime }=e^{\frac{-1}{2{\sigma }^2}},\\ K_0^{\prime }=1,K_1^{\prime }\\ =K_{-1}^{\prime },\\ K_2^{\prime }=K_{-2}^{\prime }\mathrm{and}\end{array}$ $I_1\left(l,m,n\right)=\sum _{i=-2}^2{K}_i{I}_1\left(l+i,m,n\right),\mathrm{followed}\mathrm{by}$ $I_2\left(l,m,n\right)=\sum _{i=-2}^2{K}_i{I}_1\left(l,m+i,n\right),\mathrm{and}\mathrm{then}$ $I_3\left(l,m,n,\right)=\sum _{i=-2}^2{K}_i{I}_1\left(l,m,n+i\right)$

[0023] The processing of voxel intensity enables rendering of a blurred image of the three-dimensional image volume. An example of a three-dimensional rendering is depicted in FIG. 2, which shows a section of a vascular vessel 24 having curved surfaces. The method comprises selecting and converting a surface point on the vessel 24 to a point in the center of the vessel by the use of a suitable algorithm. Many imaging software programs utilized in processing three-dimensional volumetric images allow the user to “snap to surface” for selecting a desired surface point on a rendered three-dimensional image. For surface point 26, the tangents to the curves of the surface at the point all lie on a tangent plane, and a local gradient direction for the three-dimensional image voxel indices at point 26 may be determined by: $\begin{array}{cc}{\mathrm{grad}}_x{I}_3\left(l,m,n\right)=1/2\left[I_3\left(l+1,m,n\right)-I_3\left(l-1,m,n\right)\right]{\mathrm{grad}}_y{I}_3\left(l,m,n\right)=1/2\left[I_3\left(l,m+1,n\right)-I_3\left(l,m-1,n\right)\right]{\mathrm{grad}}_z{I}_3\left(l,m,n\right)=1/2\left[I_3\left(l,m,n+1\right)-I_3\left(l,m,n-1\right)\right],\mathrm{which}\mathrm{leads}\mathrm{to}G=\left[{\left({\mathrm{grad}}_x{I}_3\right)}^2+{\left({\mathrm{grad}}_y{I}_3\right)}^2+{\left({\mathrm{grad}}_z{I}_3\right)}^2\right]\stackrel{->}{n}=\frac{1}{\sqrt{G}}\left({\mathrm{grad}}_x{I}_3,{\mathrm{grad}}_y{I}_3,{\mathrm{grad}}_z{I}_3\right)& \left(1\right)\end{array}$

[0024] Equation (1) above is a unit vector {right arrow over (n)} indicating the local gradient direction. Starting at point p₁ (26), the algorithm picks successive voxels in the direction of {right arrow over (n)} corresponding to the local gradient direction. If the line of successive voxels intersects a predetermined consecutive number of zero-value voxels (transparent), then the algorithm goes back to point p₁ and proceeds to pick successive voxels in the direction of −{right arrow over (n)}. In the preferred embodiment the predetermined consecutive number N of zero values is about 10 to 15 voxels. The algorithm proceeds in either direction of travel from point p₁ until a gradient value of at least a predetermined fraction of the gradient magnitude at point p₁ is encountered, where such point is also a local maximum and at least a predetermined distance from point p₁. For example, the gradient may be about 0.75 the gradient magnitude G_ijk at point p₁, and a minimum distance of about 5 millimeters from p₁ corresponding to a typical pulmonary vein. This point p₁′ is representative of the diametrically opposite side of the vessel or pulmonary vein from p₁ on the chosen image section. The center of the vessel x₁ can then be defined by: $\begin{array}{cc}{\stackrel{->}{\chi }}_1=\frac{1}{2}{\left({\stackrel{->}{\chi }}_{P1}+\stackrel{->}{{\chi }_{P^{\prime }1}}\right)}_{{}^{}}& \left(2\right)\end{array}$

[0025] The above procedure is repeated for a second point p₂ on the vessel, to obtain a second center point in the vessel, x₂. The method then determines the line from x₁ to x₂, which will be used as inputs to a registration algorithm. In a second embodiment of the present invention, the software algorithm is further capable of automatically fitting or constructing a centerline for the vessel, which would yield the same direction of the vessel as the above method for determining the line from x₁ to x₂. In either embodiment, the line from x₁ to x₂ provides a pulmonary vein direction, and is defined by: $\begin{array}{cc}{\stackrel{\rightharpoonup }{m}}_1=\frac{\left(\stackrel{->}{{\chi }_2}-\stackrel{->}{{\chi }_1}\right)}{\stackrel{->}{{\chi }_2}-\stackrel{->}{{\chi }_1}}& \left(3\right)\end{array}$

[0026] The two anatomical vessel points that the user marked on an X-ray image comprise endpoints having coordinates y₁ and y₂. The method then proceeds to determine a rigid transformation that takes the set {{right arrow over (n)}, {right arrow over (y)}} to the set {{right arrow over (m)}, {right arrow over (x)}} as closely as possible. First, a rotation matrix R that takes {right arrow over (n)} to {right arrow over (m)} is found by initially defining a cost function C: $\begin{array}{cc}{C}_0\equiv \sum _{i=1}^N{\left({\stackrel{\rightharpoonup }{m}}_i-R{\stackrel{\rightharpoonup }{n}}_i\right)}^T\left({\stackrel{\rightharpoonup }{m}}_i-R{\stackrel{\rightharpoonup }{n}}_i\right)=2\sum _{i=1}^N\left(--{\stackrel{\rightharpoonup }{m}}_i^{T}R{\stackrel{\rightharpoonup }{n}}_i\right)& \left(4\right)\end{array}$

[0027] The rotation matrix R can be found by minimization of the cost function C using a variety of standard computational methods known to those skilled in the art. The rotation matrix R can then be used to perform a rigid transformation of the three-dimensional image data points to the X-ray image coordinates, for providing a suitable registration of the three-dimensional vessel image with an X-ray image. The vessel registration method may be used, as an example, to mark points on the pulmonary veins in a three-dimensional pre-operative image of a heart 30 as shown in FIG. 3, and navigating a catheter tip to correspondingly mark the same pulmonary vein points in an X-ray image to obtain a registration of a pre-operative heart image.

[0028] This method of aligning a pre-operative 3D data set to an intra-operative data set reconstructed from 2D image data is especially useful for registration of vessel trees, where typically vessel take-off orientations at branch points are easily identified by a user. In this case, a good registration of the respective vessel trees, or portion thereof, can be determined as a best-fit reorientation based on minimization of C (as described above) followed by a simple (best-fit) translation to match up (at least one) corresponding branch point(s).

[0029] Alternatively a Procrustes-type method can be employed to find the best rigid transformation (a combination of a rotation and translation) that match up corresponding landmark points in the two datasets. This method is useful for chamber-type objects, possibly with vessel-type structures emanating from the chambers. The registration of a three-dimensional image 20 will allow, among other things, overlay of a visual representation of a pre-operative image object onto an X-ray image plane, which can serve as both a visual tool and an aid in surgical navigation. The method involves pre-processing the three-dimensional pre-operative image data set to obtain a set of at least two points on the three-dimensional vessel image that provide directional information.

[0030] The advantages of the above described embodiment and improvements should be readily apparent to one skilled in the art, as to enabling determining a transformation for use in obtaining registration of a three-dimensional image object with an X-ray image display. Additional design considerations may be incorporated without departing from the spirit and scope of the invention. Accordingly, it is not intended that the invention be limited by the particular embodiment or form described above, but by the appended claims.