JP2012221448A - Method and device for visualizing surface-like structures in volume data sets - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method and a device for visualizing surface-like structure in volume data sets.SOLUTION: The method includes the steps of: defining local coordinate systems LCSi at sample points P of a volume data set; transforming external parameters from a global coordinate system GCS into the local coordinate systems LCSi; calculating gradient vector components Gai, Gbi, Gci within the local coordinate systems LCSi of the sample points Pi; and using the gradient vector components Gai, Gbi, Gci for calculating a surface normal N at a given position Q of the volume data set. The external parameters from the global coordinate system GCS at the given position Q is calculated by using the transformed external parameters of the local coordinate systems LCSi of the sample points Pi, and shading or illumination at the given position Q is then performed by using the calculated surface normal N at the given position Q and the calculated external parameters at the given position Q.

Description

  The present invention relates to a method and apparatus for visualizing surface-like structures in a volume data set such as volume rendering and in particular for illuminating such surface-like structures for real-time rendering of sample data by ultrasound data acquisition. . The present invention specifically relates to 3D medical imaging for volume rendering of 2D images from 3D data volumes. Furthermore, the invention relates to a computer program product capable of implementing the method and apparatus described in independent claims 1 and 11.

  CT or MR imaging as well as 3D ultrasound imaging generally generates 3D data formatted in a coordinate system that is not a Cartesian coordinate system. Such a non-Cartesian grid is, for example, an acoustic grid of an ultrasound probe that acquires an image of the human heart, for example. Such data representing the volume is typically scan converted along a Cartesian coordinate system before volume rendering. When displaying a scanned object, it is necessary to display the surface of the object, and such a surface must be displayed "as natural as possible". One way to do that is to use a virtual external light source and incorporate such external light source contributions into the volume rendering integration. Such 3D shading is used to render such 3D datasets, and application programming interfaces such as OpenGL and DirectX are used for 3D shading.

  For illumination models such as the Blinn-Phong shading model, surface normals must usually be defined to evaluate the shading model at a selected location. The Blinn-Phong shading model is an improved version of the Phong reflection model and is performed for each point on the surface to be displayed. For this and other shading models, it is common to define a light vector and a surface normal at a specific location on such a surface to calculate diffuse reflection, and another vector (line of sight) to calculate specular reflection. Vector or camera vector) must be defined.

  A general method for approximating a surface normal at a specific position is to use a finite differencing scheme such as a forward difference method, a backward difference method, or a center difference method. In the case of volume data, such surface normals are calculated using all three dimensional differences (spatial derivatives), also called spatial gradient vectors:

式１

Formula 1

  The gradient of the scalar field is the direction of greatest change, so the surface normal of the isosurface at a fixed position is equal to the normalized gradient as

式２

Formula 2

  An inexpensive and quick way to calculate the slope of the sample data volume is to calculate the center difference, and a neighboring grid value is used. This is illustrated in FIG. 1 of the present application.

FIG. 1 shows the tip A of an acoustic probe that emits a beam 1 such as an ultrasonic beam to obtain a two-dimensional cross-sectional image of a scanned object such as a human heart. In order to calculate the gradient G at the selected sample point P (here only two-dimensional calculation is shown so that the invention can be better understood), two gradient components G _a and G _b must be calculated. . Note that a and b are axes of a local (finite) coordinate system.

G _a = (f (P _{+ x} ) −f (P _−x )) / 2
_{G b = (f (P -y} ) -f (P + y)) / 2
Formula 3

  Obviously, in the three-dimensional volume data set, the value of f for sampling the three-dimensional volume data must be obtained at six positions. Seven f values must be determined for each shadow location (eg, each voxel in the three-dimensional voxel space), including the center sample f (P) used for the shading operation.

In the prior art, it is known to speed up such calculations by pre-calculating gradient vectors at each discrete location. The difference is calculated at each discrete voxel position in the three-dimensional Cartesian coordinate system and stored as _a four-element vector (G _a , G _b , G _c , f) along with the original scalar value. In terms of performance, the number of references when shading a surface in such a three-dimensional voxel space is reduced. Shading is done by using only gradient vector components and scalar values, and their positions can be illuminated by using pre-calculated 4-element vectors. However, the main drawback is an increase in data size. Volume data increases at least four times, which can be a problem on machines with small amounts of memory.

  In the case of non-Cartesian data, the numerical calculation of the spatial derivative (gradient) and the resulting surface normal becomes more complicated, and the illumination calculation is based on the global Cartesian coordinate system, light source, observer position (camera position), etc. Is given in such a Cartesian coordinate system. Some shading models such as Blinn-Phong and Phong all use a Cartesian coordinate system for shading calculations.

For example, in the case of non-Cartesian data such as ultrasonic data given by so-called acoustic coordinates, a model such as a Blin-Phong model or a Phong reflection model cannot be used as it is, and for this reason, at a predetermined sample position as described above. Gradient vectors are used and any of the difference methods described above are used. This is because the coordinate system is not a Cartesian coordinate system. This is also shown in FIG. A local coordinate system with axes a and b does not provide the correct gradient vector component in direction a because the values f (P _{+ x} ) and f (P _−x ) are not in such a space. Therefore, in order to calculate the gradient vector G, the three-dimensional data set is first resampled into data in the Cartesian coordinate system to calculate the gradient component, whereby the gradient component is provided in the global Cartesian coordinate system.

  Another approach is shown in US Pat. No. 6,057,056, in which the gradient is first determined from the data in the acoustic domain and then converted to Cartesian coordinates or the display screen domain. A transformation matrix is specified for each position in the sampled three-dimensional acoustic data. The solution described above requires a sampling step into an expensive Cartesian geometry.

  Accordingly, it is an object of the present invention to provide a method and apparatus that visualizes surface-like structures in a volume data set that are not provided in a Cartesian coordinate system, thereby allowing the use of known illumination or shading models. Specifically, the present invention provides a fast gradient calculation and surface illumination solution for volume rendering in acoustic grids using real-time image processing units (GPUs).

  The present invention solves these objects by the features of independent claims 1 and 11. Preferred advantages are defined and claimed in each corresponding dependent claim.

  The main advantage of the present invention is that the data does not have to be resampled into Cartesian coordinate arrangements, and the present invention also makes the calculations easier and faster, thereby making known shading and lighting models It is possible to use.

The method of the present invention comprises:
a) defining a local coordinate system (LCS _i ) at the sample points (P) of the volume data set;
b) calculating gradient vector components (Gai, G _bi , G _ci ) in the local coordinate system (LCS _i ) of the sample points (P _i );
c) converting the external parameters into the local coordinate system (LCS _i ) at a sample point or a reference grid point selected from the global coordinate system (OCS);
d) calculating a surface normal (N) at a predetermined position (Q) of the volume data set using the gradient vector components (G _ai , G _bi , G _ci );
e) external from the global coordinate system (GCS) at the predetermined position (Q) by using the transformed external parameters of the local coordinate system (LCS _i ) at the selected sample points or reference grid points. Calculating a parameter.

  The external parameters preferably include position coordinates and / or vectors that define the position and / or direction, respectively, that determines the visualization (volume rendering) of the data set. The external parameter may be, for example, the position of one or several light sources, the observer position, the light vector used for rendering, and / or the observer vector / direction.

  For example, to visualize surface structures from various observer directions, it is desirable to be able to change these external parameters in real time, so a fast algorithm as provided by the present invention is important.

  According to one embodiment, the volume data set was acquired by medical imaging techniques such as ultrasound imaging, computed tomography, magnetic resonance imaging, positron emission spectroscopy and the like. Sample points are points that make up the data set.

  The local coordinate system is preferably a coordinate system that is aligned with the coordinate system (eg, acoustic coordinate system) of the volume data set at each sample point. Accordingly, the local coordinate system is slightly different for each sample point, and the local coordinate system preferably changes smoothly between the sample points.

  The global coordinate system is at the volume rendering stage, such as the position and orientation of the volume data set, as well as the position of one or several light sources, the position of the observer, the light vector used for rendering, and / or the direction of the observer. A coordinate system in which all necessary external parameters are defined in the same coordinate system. This is generally a Cartesian coordinate system, but is not a requirement.

  By calculating the gradient vector component of the sample point in the local coordinate system, it is not necessary to convert the sample point (and the corresponding gradient vector component) to a global Cartesian coordinate system. In practice, external parameters such as the position of the light source or camera or observer are converted from the global coordinate system to the local coordinate system of each sample point or selected sample point. Such a transformation of external parameters is possible by defining such a local coordinate system at the position to be shaded.

  A possible local coordinate system is a tangent vector space, which is used for local bump mapping calculations, for example. Here, the surface normal perturbation of the rendered object is referenced in the texture map at each sample point and applied before the illumination calculation is performed. Thereby, a richer and more detailed curved surface expression can be obtained (see, for example, Phong shading). Normal mapping and parallax mapping are the most commonly used bump mapping techniques. In this case, a tangent vector space is often used to define a local coordinate system.

  To create a surface tangent vector space, three vertical axes (ie, T, B, N) must be calculated. T (tangent vector) is parallel to the increasing direction of S or T on the parametric curved surface. N (normal vector) is perpendicular to the local surface. B (normal line) is perpendicular to both N and T and is on the surface like T. This is a “moving coordinate system” that moves with the sample points attached to the surface and shaded.

  In the case of acoustic coordinates, the three orthogonal axes are preferably aligned with the three directions of azimuth, elevation and range. For example, N corresponds to a range, T corresponds to elevation, and S corresponds to azimuth. The local coordinate system at each sample point is preferably aligned with the acoustic coordinates as well.

  The present invention uses a technique in which gradient vector components are calculated at each sample point in the local coordinate system of such sample point. The sample points are preferably lattice points of an acoustic grating, which is a special coordinate system, for example because at least three-dimensional having orthogonal axes (eg range, elevation and azimuth) in the direction of the acoustic grating. This is because a coordinate system ultrasound image is scanned and then a local coordinate system is defined with such sample points. The calculation of the gradient vector components in the local coordinate system is preferably, for example, calculating the spatial derivatives of all dimensions by using the discrete values of each adjacent sample point and determining the finite difference of the sample points. Can be done.

  If a predetermined location (sample points or voxels between such sample points) of the volume data set is to be illuminated, the surface normal at such predetermined location is calculated by using the gradient vector component of the sample points. be able to. The predetermined position is the position where it is desired to calculate the light reflection to create a volume rendered image. If the visualization or volume rendering process is performed by a ray casting method, the predetermined position is on such a ray, for example, and if a slice based method is used, the predetermined position is on such a slice. It becomes the pixel position. Most if not all of the predetermined positions are usually not the same as the sample points, and neither the selected sample points nor the points on the reference grid are the same.

  According to a preferred embodiment of the present invention, the surface normal at a given position first interpolates the non-normalized gradient vector components of adjacent sample points and then normalizes the interpolated gradient vector at that position. Is calculated by

  The next time the surface normal at a given position must be calculated, the preferred embodiment of the present invention can interpolate, for example, the denormalized gradient vector components of eight adjacent sample points, It is adjacent to such a predetermined position in such a three-dimensional (voxel) space. Thereby, the calculation is reduced to sample points only, but all other (voxel) positions are calculated by interpolating the gradient vector components of each adjacent sample point.

  According to an important aspect of the present invention, instead of interpolating the normalized gradient vector or normal vector of these sample points to obtain the corresponding gradient vector at a given location, the unnormalized gradient vector component is Interpolation is performed later, and normalization is performed at such a predetermined position.

  According to one embodiment, such interpolation between the gradient vector components of adjacent sample points is performed without considering the fact that the gradient vector components exist in different local coordinate systems. In other words, the interpolation is performed as if the gradient vector components from all contributing sample points are in the same local coordinate system. This approximation can be made because there is no sharp change in the geometric shape of the coordinate system, at least in acoustic coordinates, ie the orientation of the local coordinate system of adjacent sample points is only slightly different. This situation is different when sample points that are far from each other must be compared, otherwise the inaccuracy resulting from this procedure does not compromise the results.

  According to another aspect of the invention, the light source (and other extrinsic parameters) are converted to a local coordinate system at several points in the three-dimensional data grid. These points may be selected sample points, or may be points of a reference grid that are defined in the acoustic grid but may not coincide with the sample points of the acoustic grid. According to one embodiment of the present invention, the conversion / transformation of external parameters to a local coordinate system is very computationally intensive and is not performed at every sample point, rather than several selected data sets. It is performed only at points (eg, every third to 50th sample point in each dimension) or points of a predefined reference grid, in which case the reference grid has a lower resolution (eg, 5 to 5) than the resolution of the sample points. 1/30).

  The conversion of the external parameter sample points or the points on the reference grid to the local coordinate system is performed, for example, by calculating a conversion matrix between the global coordinate system and the respective local coordinate system, and the external coordinates at the sample points or reference points. This is done by mapping to a local coordinate system.

  Next, according to a preferred embodiment of the present invention, the external parameters from the global coordinate system at the predetermined position are converted into selected sample points or reference grid points, preferably adjacent selected sample points or reference grids. Can also be calculated by using the transformed external parameters of the local coordinate system of neighboring points. This eliminates the need to calculate the transformation of the external parameters into the local coordinate system at each predetermined position, and only requires such calculations for selected sample points in the three-dimensional (acoustic) data grid. Time is further reduced. As mentioned above, these sample points may be grid points of a predefined reference grid as described in US Patent Application No. 2005/0253841 A1. Thus, according to one embodiment, the conversion from a Cartesian (global coordinate system) coordinate arrangement to an acoustic (local coordinate system) coordinate arrangement is not performed for all sample points, but at a coarser level.

  According to one embodiment, the extrinsic parameters at a given location are then interpolated from the nearest sample point or reference grid point where the extrinsic parameters have been converted to the local coordinate system. This is an approximation, but again, since there is no sharp change in geometry, interpolation is used to obtain a sufficiently smooth result.

  According to another preferred aspect of the invention, the shading of the predetermined position is performed by a conventional illumination model, and the surface normal at the predetermined position is used together with the calculated external parameters from the global coordinate system at the predetermined position. Is done.

According to the present invention, a volume data set derived from sampled raw ultrasound data can be used, and the gradient vector component in the local coordinate system is calculated from each sample point of the raw ultrasound data. And stored in a new data set containing the four-element vectors (G _a , G _b , G _c , F) along with the sample point scalar values. For any given location, these scalar vector components and such scalar values of adjacent sample points are then used to calculate the surface normal at the given location. Thereafter, the transformed external parameters of the local coordinate system of the sample points are used to calculate the external parameters at a predetermined position, which in turn uses the calculated external parameters and the calculated surface normal. Is illuminated by a conventional illumination model. Since related elements such as surface normal, light vector, viewer vector, semi-normal and reflection vector are in the local coordinate system of each sample point and are interpolated at each predetermined position, the conventional Blinn -A Phong lighting model can be used.

  According to the present invention, the sampled raw ultrasound data is first sent to a graphics processing unit (GPU), where it is then calculated using a shader, and from the global coordinate system, an external parameter of The conversion to the local coordinate system is performed, for example, by a graphic processing unit (GPU) of the ultrasonic computer system. Next, the gradient vector components in the local coordinate system are calculated by using center differences, Sobel operators or other estimators.

The present invention is also an apparatus for visualizing surface-like structures in a volume data set, comprising:
a) defining a local coordinate system at sample points of the volume data set; and b) a first arithmetic processing unit (PU1) for converting external parameters from a global coordinate system to the local coordinate system;
c) a second operation for calculating a gradient vector component of the sample point in the local coordinate system, and d) calculating a surface normal at a predetermined position of the volume data set using the gradient vector component. The present invention relates to a device for visualization with a processing device (PU2).

  Both processing units (PU1, PU2) are preferably graphic processing units that assist the CPU in minimizing the processing time required for lighting or shading. A graphics processing unit (GPU) displays a predetermined position illuminated by a conventional illumination model by using calculated external parameters and calculated surface normals, typically in the same Cartesian coordinate system as the display screen domain. In this case, the graphics processing apparatus generally transforms the external parameters from the global coordinate system at a predetermined position by interpolating the transformed external parameters of the local coordinate system of the adjacent sample points at such local grid points. Such points are described in the acoustic grid points (or points between these grid points) or in US application No. 2005/0253841 A1. Such a reference grid point.

  US Patent Application No. 2005/0253841 A1 discloses the deformation of the reference grid (coordinate system) as a function of the cutting plane defined in the three-dimensional data set. US Patent Application No. 2005/0253841 A1 proposes to use the intersection of the proxy geometry representing the scan volume and the cutting plane, the vertex processor of the graphics processing unit (GPU) then The grid is deformed and Cartesian coordinates and texture coordinates are determined for each grid point of the reference grid. However, in the present invention, the grid points need not be converted to Cartesian coordinates.

  The invention also relates to a computer program product comprising program code stored on a computer readable medium, wherein the computer program product performs the method described above when such computer program product is executed on a computer.

  Preferred embodiments of the present invention will be described in further detail by referring to the accompanying drawings.

It is a figure which shows the general acoustic lattice which has a local coordinate system in the sample point. From the global coordinate system is a diagram schematically showing the calculation of the local light vector L _Q in the conversion and the predetermined position Q of the external parameters to the local coordinate system. FIG. 6 illustrates an implementation of the present invention for real-time applications by calculating surface normals by interpolation of gradient vector components.

  FIG. 1 shows a conventional ultrasound (acoustic) grating with vertex A, beam 1 and row 2, where beam 1 and row 2 represent a two-dimensional representation of an acoustic coordinate system (eg, range and azimuth, respectively). To better understand the present invention, the third dimension (eg, elevation of a spherical coordinate system) is not shown in FIG.

As described above, the local gradient vector G is defined by the local gradient vector components G _a and G _b of the sample point P, and these components are calculated by the difference method as described above.

FIG. 2 schematically shows the conversion of the external parameters such as the light source LS from the global coordinate system GCS to the local coordinate system (that is, the first local coordinate system LCS ₁ and the second local coordinate system LCS ₂ ). ₁ is defined by the first sample point P ₁ and the second sample point P ₂ in the acoustic lattice as shown in FIG.

Each sample point P _i has a local gradient vector G _i having local gradient vector components G _ai and G _bi . The first sample point P ₁ has a first gradient vector component G _a1 and a second gradient vector component G _b1 , and the axis b ₁ of the _first local coordinate system LCS 1 is parallel to the beam 1. On the other hand, the second local coordinate system axis a ₁ is a row 2 tangent vector and is perpendicular to the _first coordinate system axis b ₁ .

The first local gradient vector G ₁ is defined by using adjacent sample points (sample point P ₂ , other sample points, etc.). At this time, a known difference method such as the central difference method (Equation 1), discrete convolution filtering such as a Sobel filter, or continuous convolution filtering such as a cubic B-spline or a derivative thereof is used (adjacent 4 × 4 × 4). ).

Figure 2 also shows the global coordinate system GCS having axes x and y, a light vector L _G given in such a global coordinate system GCS. Instead of re-sampling the sample points P ₁ and P ₂ to the global coordinate system GCS, the light source LS is converted into the local coordinate systems LCS ₁ and LCS ₂ , whereby the first local light vector L _L1 and the second local light A vector L _L2 is obtained. According to one embodiment, extrinsic parameters such as light sources are not converted to the local coordinate system of all sample points, only the local coordinate system of the coarser grid of sample points or the local coordinate system of the points of the reference grid Is converted to

FIG. 2 also shows a predetermined position Q, and the local light vector L _{Q at} such a predetermined position _Q is not necessarily calculated by converting the light source LS to the local coordinate system LCS _Q at the predetermined position Q. , Such a light vector L _Q may be a first local light vector L _L1 and a second local light vector L _{L2 of} sample points P ₁ and P ₂ , and further other sample points as required. Is preferably calculated by using the local light vector of. Therefore, the external parameters are converted to the local coordinate system LCS _Q at the predetermined position Q by interpolating the already converted external parameters from the selected adjacent sample points of the acoustic grid or the reference grid.

Similarly, the normal vector of the predetermined position Q N is the gradient vector component G _a2 of the gradient vector components of the first sample point P ₁ G _a1, G _b1 and G _C1 and the second sample point P2 (not shown) , G _b2 and G _c2 , thus eliminating the need to calculate the vector required for the illumination model at each given location by transforming the coordinate system or extrinsic parameters, and the external at discrete sample points. It is only necessary to calculate the parameter conversion value and interpolate the non-normalized gradient vector component of each sample point P ₁ to obtain the surface normal N (and other vectors) at the predetermined position Q. In this case as well, the external parameters are converted to the local coordinate system LCS _Q at the predetermined position Q by interpolating the already converted external parameters from the selected adjacent sample points of the acoustic lattice or the reference lattice.

  In that way, the calculation cost can be reduced by calculating the light source LS only at the grid points of the acoustic grid or the reference grid only in the local space, and by using linear interpolation for intermediate points or pixels. it can. Thereby, most of the conversion calculation cost can be reduced. This is because the lattice is coarse.

  The present invention accepts that the value at each predetermined location (each voxel in the 3D Cartesian voxel space) is determined by interpolation, so that within a grid (acoustic grid, reference grid, or part or combination of these grids). The computational model can be reduced by transforming the external parameters at only some or all of the points. Since such an interpolation result is only used for lighting calculations, this is considered a reasonable approximation. The human eye is extremely sensitive to high frequency changes in lighting (eg, sharp edges and hard cuts are easily visible). However, since there is no sharp change in geometry, fairly smooth results can be obtained by using the interpolation technique of the present invention. The human eye is usually unaware of the difference in lighting results.

  FIG. 3 illustrates another preferred embodiment of the present invention for real-time applications such as life-form 3D ultrasound scanning. Gradient calculation is often an obstacle when real-time applications are required. Currently, the pre-process gradient is calculated by the CPU and later transferred to the GPU. According to the present invention, it is possible to execute the preliminary calculation by transferring the original raw data in the rendering stage. Modern hardware supports such rendering of individual 3D volume textures into individual slices.

Raw data as shown in stage I of FIG. 3 is first transferred to the GPU in its native format (eg, 8 bits per 11 samples). In the second stage II, each sample is repeated and the slope is calculated. Center differences, Sobel operators or other estimators can be used. Thereafter, the denormalized gradient components (G _a , G _b , G _c , F) are stored in a new three-dimensional volume as schematically shown in FIG. Here, it is preferable to use high precision (eg 16 or 32 bit float) to avoid any quantization artifacts.

  According to the present invention, it is advantageous not to normalize the precomputed gradients at this stage, but instead to store the unnormalized gradient vector components.

As shown in step III in FIG. 3, the surface normal N at the predetermined position Q, the first gradient G ₁ sample point P _1, the second gradient G ₂ sample points P _2, the third sample It is calculated by interpolating the gradient vector component of the gradient G ₄ of the point P ₃ of the gradient G _3, and a fourth sample point P _4. Each gradient has its own gradient vector component, all of which are used to define a gradient vector at a given location Q, and only normalizing this gradient vector yields a surface normal N. Can do.

The invention has the advantage that the method used is well adapted to the adoption of a reference grid. This is because the workload in the pixel shader stage is low and high performance real-time gradient rendering is obtained. Furthermore, all lighting calculations are performed in the local coordinate system. The gradient volume in the acoustic space can be calculated and stored in advance so that illumination is expedited.

Claims (14)

A method of visualizing a surface-like structure in a volume data set, wherein the visualization is determined by a set of external parameters including position coordinates and / or orientations that define the visualization,
a) defining a local coordinate system (LCS _i ) at the sample points (P) of the volume data set;
b) calculating gradient vector components (G _ai , G _bi , G _ci ) in the local coordinate system (LCS _i ) of the sample points (P _i );
c) transforming the external parameters into the local coordinate system (LCS _i ) at a sample point or a reference grid point selected from the global coordinate system (GCS);
d) calculating a surface normal (N) at a predetermined position (Q) of the volume data set using the gradient vector components (G _ai , G _bi , G _ci );
e) external from the global coordinate system (GCS) at the predetermined position (Q) by using the transformed external parameters of the local coordinate system (LCS _i ) at the selected sample points or points of the reference grid Calculating the parameter. The surface normal (N) at the predetermined position (Q) first interpolates the denormalized gradient vector components (G _ai , G _bi , G _ci ) of adjacent sample points (P), and then The method according to claim 1, wherein the method is calculated by normalizing the interpolated gradient vector (G) at a particular position (Q). The method according to claim 1 or 2, wherein the sample points (P) are lattice points of an acoustic lattice and the local coordinate system (LCS _i ) has at least three orthogonal axes. The predetermined position (Q) is a position between sample points (P), and the surface normal (N) at the predetermined position (Q) is denormalized at at least eight adjacent sample points (P). The method according to claim 1, wherein the method is calculated by interpolating gradient vector components (G _ai , G _bi , G _ci ). The gradient vector components (G _ai , G _bi , G _{ci in the} local coordinate system (LCS _i ) of the sample points (P _i ) are preferably calculated in all dimensions by the finite difference method. ) Is calculated according to any one of claims 1 to 4.   The extrinsic parameters from the global coordinate system (GCS) at the predetermined position (Q) are calculated by interpolating between the transformed extrinsic parameters from the nearest selected sample point or the nearest point of the reference grid, The method according to claim 1. g) Conventionally determining the predetermined position (Q) using the surface normal (N) at the predetermined position (Q) and the calculated external parameters from the global coordinate system (GCS) at the predetermined position (Q). 7. A method according to any of claims 1 to 6, comprising the additional step of shading with a lighting model. The volume data set is derived from the sampled raw ultrasound data, and the gradient vector components (G _ai , G _bi , Gci) in the local coordinate system (LCS _i ) are transformed into the raw ultrasound data Calculated for each sample point (P _i ) and stored in a new data set (G _ai , G _bi , G _ci , F) along with the scalar value (F) of the sample point (P _i ),
For any given position (Q), the scalar value (F) of the sample point (P ₁ ) adjacent to the gradient vector component (G _ai , G _bi , G _ci ) is the surface method at the given position (Q). Used to calculate the line (N),
The transformed external parameter of the local coordinate system (LCS _i ) of the selected sample point or reference grid point is used to calculate an external parameter at the predetermined position (Q), and the predetermined position (Q 8) is illuminated by a conventional illumination model by using the calculated extrinsic parameters and the calculated surface normal (N). The sampled raw ultrasound data is first sent to a graphics processing unit (GPU) where it is calculated using a vertex shader and the external parameters from the global coordinate system (GCS) The method according to claim 8, wherein the conversion to a local coordinate system (LCS _i ) is performed by a processing unit of an ultrasonic computer system. The gradient vector components (G _ai , G _bi , G _ci ) in the local coordinate system (LCS _i ) are calculated using a central difference, Sobel operator or other estimator. The method of any one of these. A device for visualizing the surface-like structure of a volume dataset,
a) defining a local coordinate system (LCS _i ) at the sample points (P) of the volume data set; and b) converting external parameters from a global coordinate system (GCS) to the local coordinate system (LCS _i ). A first arithmetic processing unit (PU1);
c) calculating gradient vector components (G _ai , G _bi , G _ci ) in the local coordinate system (LCS _i ) of the sample points (P _i ), and
d) a second arithmetic processing unit (N) for calculating a surface normal (N) at a predetermined position (Q) of the volume data set using the gradient vector components (G _ai , G _bi , G _ci ); PU2).   The graphics processing unit (GPU) displays the predetermined position (Q), and the predetermined position (Q) uses the calculated external parameter and the calculated surface normal (N) to obtain a three-dimensional The apparatus according to claim 11, which is illuminated by a conventional illumination model in a Cartesian coordinate system during volume rendering.   The graphics processing unit (GPU) also converts external parameters from the global coordinate system (GCS) at the predetermined position (Q) to the converted external parameters of the closest selected sample point or the closest point of the reference grid. 13. An apparatus according to claim 11 or 12, wherein the calculation is performed by interpolating between. A computer program product comprising program code for performing the method according to any of claims 1 to 10 when stored on a computer readable medium and executed on a computer.

Images