Title of Invention  METHOD AND APPARATUS FOR RECONSTRUCTING A THREEDIMENSIONAL IMAGE VOLUME FROMTWODIMENSIONAL PROJECTION IMAGES 

Abstract  Method and apparatus for reconstructing a threedimensional image volume from twodimensional projection images The invention relates to a method and an apparatus for reconstructing a threedimensional image volume (12) from twodimensional projection images (20) of a subject which have been taken from different projection directions by rotating the recording system (3, 4) around the subject (5), wherein the grayscale values of the voxels of the image volume (12) are calculated by back projection of the projection images (20). The invention is characterized in that prior to back projection at least one projection image (20) is modified in such a way that it corresponds to a projection image (24) taken with a virtual detector (22) whose axes are aligned parallel to the rotational axis (18) of the recording system. 
Full Text  2005P12002 US 1 Description Method and apparatus for reconstructing a threedimensional image volume from twodimensional projection images The invention relates to a method and an apparatus for reconstructing a threedimensional image volume from twodimensional projection images, as well as a computer program product and a digital storage medium with a program code for carrying out the method. In particular the invention pertains to a method or an apparatus in which the projection images are recorded from different projection directions by rotating a recording system around a subject, the voxels of the image volume being calculated by back projection of the projection images. In radiology the need often exists to reconstruct a threedimensional (3D) image from twodimensional (2D) images for diagnosis, therapy planning and during interventional procedures. The problem arises with socalled Carm xray systems, for example, in which the xray tube and detector are mounted on a Carm which is freely maneuverable around the patient. It is often desirable to reconstruct a threedimensional image volume from twodimensional xray images or projection images thus obtained. However, in this case the projection geometry is more complex than with a computer tomograph, as the latter xrays the subject under examination section by section with a fan beam, whereas the beam is coneshaped in the case of a carm machine. This statement is not true of multislice CT. For example, for 64row scanners an algorithm for conebeam projection geometry is also required. Moreover, Carm systems are often mechanically unstable and do not move exactly on a circular orbit around the subject, but produce e.g. slight vibrations. For reconstructing structures with low xray contrast (e.g. soft parts in medical data) a large number of projection 2005P12002 US 2 images are required, which greatly increases the computational cost/complexity and may therefore negatively impact the usability of the system. For example, during an interventional catheter treatment, the reconstruction must if possible be complete within a few seconds or a few minutes at the most. A method for such a reconstruction is disclosed in the article "Practical conebeam algorithm" by L.A. Feldkamp, L.C. Davis and J.W. Kress, Journal of the Optical Society of America 1, 612619 (1984). An implementation and adaptation of this system for use in Carm systems is described in "Enhanced 3Dreconstruction algorithm for Carm systems suitable for interventional procedures" by K. Wiesent, K. Barth, M. Navab, P. Durlak, T. Brunner, 0. Schiitz and W. Seissl, IEEE Transactions on Medical Imaging, Vol. 19, No.5 (2000). This method is based on back projection, i.e. the grayscale values of the projection images are summed for each voxel in the image volume at the corresponding image points. For each projection image, the image volume to be reconstructed is run through voxel by voxel, the projection of the voxel is computed and the grayscale value of the corresponding pixel of the projection image is added to the voxel value. This method requires that the projection geometry of the recording system has been separately determined in advance for each recording position by means of a calibrating method, allowing for any deformations and dynamic movements of the recording system, which means that in general the main axes of the volume to be reconstructed do not have a uniform orientation in the projection images. In general it must therefore be assumed for each projection that the spacings of adjacent voxels in the projection are not uniform in any direction. For the general projection geometry occurring in Carm systems, at least three additions and two divisions must be carried out for each voxel, see e.g. page 395 of the abovementioned article by K. Wiesent et al. in IEEE 2005P12002 US 3 Transactions on Medical Imaging. In the algorithm specified there, the voxels of the image volume are sampled in three nested loops in the x, y and zdirection. In the innermost loop, the relevant zcoordinate is then multiplied by the projection matrix and the homogeneous pixel coordinates (r, s, t) thereby obtained are normalized so that the position of the corresponding pixel in the image coordinates is finally yielded as u = r/t and v = s/t. Finally the pixel closest to the position (u, v) must be determined and accumulated to the grayscale value of the voxel at the position (x, y, z). As these operations must be performed in the innermost loop, this results in considerable computational cost/complexity. In addition, random memory access to the projection data is required which negatively impacts the necessary memory transfer between main memory and cache. Pared down to its essential algorithmic processing steps, the known backprojection algorithm can be described as follows: For each slice of the 3D volume (with uniform zvalue) Determine the projection of the vertex of the slice For each row of the slice (with uniform yvalue) Determine the projection of the row start point as a function of the slice vertex For each voxel of the row (xdirection) Determine the projection of the voxel as function of the start point of the row (3 additions and 2 divisions) Accumulate the grayscale value of the image point to the voxel value (1 addition) End End End. 2005P12002 US 4 The object of the invention is to provide a rapid reconstruction method and corresponding apparatus which can be used even with mechanically unstable recording systems. This object is achieved by the invention as claimed in claim 1 and the apparatus as claimed in claim 9, as well as with the computer program product as claimed in claim 7 and the digital storage medium as claimed in claim 8. Advantageous embodiments of the invention are set forth in the subclaims. As claimed in claim 1 the problem is solved by modifying at least one projection image prior to back projection in such a way that it corresponds to a projection image taken by a detector whose vertical columns are aligned parallel to the axis of rotation of the recording system. The advantage of this modification of one or more projection images is that the projection geometry is simplified such that the projections of all the voxels lying on a straight line parallel to the rotational axis of the recording system are equidistant. If these voxels are sampled consecutively, the pixel corresponding to them on the modified projection image can be determined from the previous point by simple addition of the spacing, thereby considerably reducing the computational cost/complexity. Because of the lower nesting depth of the program loops required for that purpose, the additional arithmetic operations required for modifying the projection images are irrelevant compared to the savings witnin tne bac^L projection. Klso with. the method according to the invention, the actual projection geometries of the recording system are advantageously determined in advance for each recording position by means of a calibrating process. It has already been possible to confirm the reproducibility of these projection geometries for a number of equipments. 2005P12002 US 5 The modified projection image is preferably obtained from the original projection image by application of a projective transformation (= homography). Such transformations with which the points of a plane are mapped to another plane are known in the prior art and are used e.g. for correcting the screen projections for projectors. Such a transformation corresponds to the image changes occurring if the screen is inclined. The projection matrix associated with the relevant projection image is preferably modified likewise by multiplication by a homography. According to a preferred embodiment, the area of the virtual detector can cover either the entire projection region of the recording system or only the maximum reconstructable volume. The maximum reconstructable volume is generally taken to mean the volume which is covered by all the projection images. By matching the modified projection image to this volume, reconstruction can therefore be more efficiently organized. Advantageously, the zdirection of the subject's coordinate system runs parallel to the rotational axis of the recording system. In this case the zdirection for the back position is preferably sampled in the innermost loop and the vertical image coordinates of the projection image are computed by the addition of an increment that is constant for the x and ycoordinates of this loop. This uses the property of the modified projection geometry that the projections of the voxels along the zdirection are equidistant in the modified projection image and can therefore be determined from the previous point by simple addition of the spacing. For the accumulation oi the grayscale value oi the image point of a projection image to the grayscale value of the voxel, the most adjacent pixel ("nearest neighbor") of the projection image is preferably used. Alternatively the grayscale value can also be computed by interpolation from a plurality of 2005P12002 US 6 nearest neighbors. However, the advantage of using the nearest neighbor is that, by means of this method, the memory accesses to the projection image can be optimally adapted for caching within the computing unit. Whereas in the prior art random access to a plurality of rows of the projection image had to be enabled within the inner loop, with the invention this can be limited to one image row. During backprojection the modified projection image is therefore preferably stored in a data memory which is organized in such a way that vertically adjacent pixels follow one another. If the linear storage of the projection image is oriented in this way, access in the innermost loop to an image row or column, and therefore to a very restricted memory area, is limited and optimized for caching. Instead, if the linear storage of the projection image is organized on a rowbyrow basis, the virtual detector can also be rotated through 90° by appropriate modification of the homography. If, as described in the abovementioned article by L.A. Feldkamp, weighting of the distance between voxel and projection center is to take place orthogonally to the detector within the backprojection, in contrast to the prior art, the weighting factor required for this purpose can be computed outside the inner loop, which also considerably reduces the number of necessary operations for this enhanced algorithm. The invention will now be explained in greater detail with reference to the accompanying drawings in which: Fig. 1 schematically illustrates a Carm system with which the invention can be implemented; Fig. 2 shows the projection geometry according to the prior art ; 2005P12002 US 7 Fig. 3 schematically illustrates a modification of a projection image; and Fig. 4 schematically illustrates the projection geometry after the modification. Fig. 1 schematically illustrates a Carm xray system 1 having a Carm 2 on which an xray tube 3 and an xray detector 4 are mounted opposite one another. The angular position of the Carm can be rotated (angled) around a table 6 with a patient 5 supported thereon in order to obtain xray images from different projection directions. The resulting 2D images are forwarded to a control and computing unit 7 comprising at least a data memory 8 and a computing module 9 incorporating a processor or the like. The recorded projection images and the if necessary reconstructed image volumes can be viewed on a monitor 11. Fig. 2 shows the projection geometry for such a Carm system prior to modification of the projection image. The figure shows the detector plane 10 with, projected thereon, a representation of the volume 12 to be reconstructed which constitutes the image volume to be reconstructed. The figure therefore shows the positions of the voxels in the image dataset 12, projected onto the detector plane 10, and corresponds to the actually measured projection image 20. The main axes of the image volume to be reconstructed are denoted by x, y and z, whereas the main axes of the projection image are marked u and v. Since the detector, as is frequently the case with a Carm system, is not aligned exactly to the rotational axis, the zaxis in the image volume is not parallel to the edge of the detector and therefore to the vertical image direction v. 2005P12002 US 8 For the back position, the image dataset 12 is run through voxel by voxel and the image coordinates u, v of the projection are computed in each case, the current reconstruction slice being labeled 14. If, for example, the xdirection is sampled along the arrow marked 16 in the innermost loop, it becomes apparent that adjacent voxels are not projected equidistantly, but with different spacings Dul, Au2 and Du3. In other words the projection of each voxel must be recomputed from its three coordinates x, y and z and cannot be derived e.g. by addition of a constant spacing Du or Dv from the projection of an adjacent point. Fig. 3 shows an example of the step in which a projection image 2 0 is modified in such a way that it corresponds to a projection image 24 taken by a virtual detector 22 whose columns are aligned parallel to the rotational axis 18 of the recording system. The xray focal point is shown at 17. The subject (not shown) that is projected onto the actual detector plane 20 is in the cone of the beam. To make the projective transformation clear, the projection image 2 0 is shown as a grid. When this projection image is modified, an image shall be generated which corresponds as precisely as possible to the image which would have been taken by a virtual detector 22 oriented parallel to the rotational axis 18. The corresponding image homography can be determined from the known parameters of the projection geometry. In the example shown, the slightly distorted grid 24 would be produced by the transformation. Fig. 4 shows the projection geometry for a projection image 24 modified in this way. It again shows the detector plane 10, which in this case is the detector plane of the virtual detector 22. By means of the modification as shown in Fig. 3, the zaxis of the image volume to be reconstructed is now parallel to the vertical main axis v of the projection image. Because of this parallelism, the spacings of the projections of adjacent voxels are equidistant on a straight line 16 2005P12002 US 9 placed in the zdirection. For the straight line 16 the spacing are e.g. Dv. This equal spacing effect can be made clear using the intercept theorem which states if two rays originating from a point are intersected by two parallel lines, the segments on the parallel lines behave like the corresponding segments, measured from the beam focal point, on each beam. With reference to Fig. 4, the columns of the image volume 12 are therefore parallel lines which are intersected by xrays originating from the xray focal point. Provided the columns of the image volume are therefore aligned parallel to the detector plane, the projections of the voxels are equidistant. Advantageously the image volume as shown in Fig. 4 is sampled differently from in Fig. 2, i.e. with the zdirection in the innermost loop. Pared down to the essential algorithmic processing steps, the back projection can be represented as follows: For each slice of the 3D volume (now with uniform yvalue) Determine the projection of the slice vertex For each row of the slice (now with uniform xvalue) Determine the projection of the row start point as a function of the slice vertex Determine the increment in the projection image For each voxel of the row (zdirection) Determine the projection of the voxel as function of the start point of the row (1 addition) Accumulate the grayscale value of the image point to the voxel value (1 addition) End End End. The sequence of the outer and middle computing loop can also be transposed. By this means, in the inner loop the number of 2005P12002 US 10 necessary additions is reduced by 50% and the divisions eliminated altogether, thereby considerably reducing computational cost/complexity. The advantage of the faster back projection achieved by the invention in clinical practice is that when using the method during an interventional procedure the waiting time between acquisition of the projection data and the availability of the reconstructed volume is significantly reduced. This is particularly important when low contrasts within the recorded subject necessitate reconstruction from several hundred projection images. A clinical example in which both rapid diagnosis and the visibility of small contrasts and therefore the processing of a large number of projection images is relevant is the case of bleeding inside the brain which can occur during an interventional treatment. By speeding up back projection, further algorithmic enhancements of reconstruction quality, requiring e.g. the multiple use of back projection with intervening correction to the projection data, become feasible. 2005P12002 US 11 Claims 1. A method for reconstructing a threedimensional image volume (12) from twodimensional projection images (20) of a subject which have been taken from different projection directions by rotating the recording system (3, 4) around the subject (5), wherein the grayscale values of the voxels of the image volume (12) are calculated by back projection of the projection images (20), characterized in that prior to back projection at least one projection image (20) is modified in such a way that it corresponds to a projection image (24) taken with a virtual detector (22) whose axes are aligned parallel to the rotational axis (18) of the recording system. 2. The method according to claim 1, characterized in that the modified projection image (24) is obtained by applying a projective transformation (homography) from the original projection image (20). 3. The method according to one of the preceding claims, characterized in that the area of the virtual detector (22) covers either the entire projection region of the recording system or only the maximum reconstructable volume. 4. The method according to one of the preceding claims, characterized in that the zdirection of the subject's coordinate system runs parallel to the rotational axis (18) of the recording system, and that for each modified projection image (24) the voxels of the image volume (12) are sampled in three loops in the x, y and zdirection and the projection of the voxel onto the projection image (24) is computed in each case and the corresponding grayscale value of the projection image (24) is accumulated to the grayscale value of the voxel, 2005P12002 US 12 wherein the zdirection is sampled in the innermost loop and the vertical image coordinate (v) of the projection image is computed by addition of a constant increment (Av) for the xand ycoordinates of this loop. 5. The method according to one of the preceding claims, characterized in that the nearest voxel of the projection image is used for accumulation of the grayscale value of the projection image (24). 6. The method according to one of the preceding claims, characterized in that the modified projection image (24) is stored during back projection in a data memory which is organized in such a way that vertically adjacent voxels follow another. 7. The method according to one of claims 4 to 6, characterized in that, prior to accumulation to the grayscale value of the voxel, the grayscale value of the projection image (24) is multiplied by a weighting factor which depends on the distance between voxel and projection center (17) orthogonally to the virtual detector (20), the weighting factor not being computed in the innermost loop. 8. A computer program product with program code stored on a machine readable carrier for carrying out the method according to one of claims 1 to 7 when the program product is executed on a computer. 9. A digital storage medium, with electronically readable control signals which can interoperate with a programmable computer system in such a way that a method according to one of claims 1 to 7 can be implemented. 10. An apparatus for reconstructing a threedimensional image volume (12) from twodimensional projection images (20) of a subject, comprising 2005P12002 US 13 a recording system with a detector, rotatable about the subject, for recording projection images (20) of the subject from different directions, and a computing module (9) for computing the grayscale values of the voxels of the image volume (12) by back projection of the projection images (20), characterized in that a computing module is set up to modify at least one projection image (20) in such a way that it corresponds to a projection image (24) taken with a virtual detector (22) whose columns are aligned parallel to the rotational axis (18) of the recording system (3, 4). 11. The apparatus according to claim 10, characterized in that the computing module (5) is set up to carry out the method according to one of claims 1 to 6. Method and apparatus for reconstructing a threedimensional image volume from twodimensional projection images The invention relates to a method and an apparatus for reconstructing a threedimensional image volume (12) from twodimensional projection images (20) of a subject which have been taken from different projection directions by rotating the recording system (3, 4) around the subject (5), wherein the grayscale values of the voxels of the image volume (12) are calculated by back projection of the projection images (20). The invention is characterized in that prior to back projection at least one projection image (20) is modified in such a way that it corresponds to a projection image (24) taken with a virtual detector (22) whose axes are aligned parallel to the rotational axis (18) of the recording system. 

0938kol2006 correspondence others.pdf
0938kol2006 description[cmplete].pdf
0938kol2006 priority document.pdf
938KOL2006(06032014)ABSTRACT.pdf
938KOL2006(06032014)ANNEXURE TO FORM 3.pdf
938KOL2006(06032014)CLAIMS.pdf
938KOL2006(06032014)CORRESPONDENCE.pdf
938KOL2006(06032014)DESCRIPTION (COMPLETE).pdf
938KOL2006(06032014)DRAWINGS.pdf
938KOL2006(06032014)FORM1.pdf
938KOL2006(06032014)FORM2.pdf
938KOL2006(06032014)OTHERS.pdf
938KOL2006(13032012)CORRESPONDENCE.pdf
938KOL2006(13032012)OTHERS.pdf
938KOL2006(24072013)ABSTRACT.pdf
938KOL2006(24072013)ANNEXURE TO FORM 3.pdf
938KOL2006(24072013)CLAIMS.pdf
938KOL2006(24072013)CORRESPONDENCE.pdf
938KOL2006(24072013)DESCRIPTION (COMPLETE).pdf
938KOL2006(24072013)DRAWINGS.pdf
938KOL2006(24072013)FORM1.pdf
938KOL2006(24072013)FORM2.pdf
938KOL2006(24072013)OTHERS.pdf
938KOL2006(24072013)PETITION UNER RULE 137.pdf
938KOL2006CORRESPONDENCE 1.2.pdf
938KOL2006CORRESPONDENCE1.1.pdf
Patent Number  263930  

Indian Patent Application Number  938/KOL/2006  
PG Journal Number  49/2014  
Publication Date  05Dec2014  
Grant Date  27Nov2014  
Date of Filing  18Sep2006  
Name of Patentee  SIEMENS AKTIENGESELLSCHAFT  
Applicant Address  Wittelsbacherplatz 2,80333 Munchen Germany,  
Inventors:


PCT International Classification Number  G06T17/00  
PCT International Application Number  N/A  
PCT International Filing date  
PCT Conventions:
