Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T11/00—2D [Two Dimensional] image generation
  • G06T11/003—Reconstruction from projections, e.g. tomography
  • G06T11/006—Inverse problem, transformation from projection-space into object-space, e.g. transform methods, back-projection, algebraic methods
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T11/00—2D [Two Dimensional] image generation
  • G06T11/003—Reconstruction from projections, e.g. tomography
  • G06T11/008—Specific post-processing after tomographic reconstruction, e.g. voxelisation, metal artifact correction
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2211/00—Image generation
  • G06T2211/40—Computed tomography
  • G06T2211/421—Filtered back projection [FBP]

Definitions

• the invention relates to a method and a device for image reconstruction of a room volume from acquired projections.
• CTs computer tomography systems
• special multiprocessor systems are generally used in order to achieve acceptable reconstruction times of a few seconds per reconstructed slice, whereby in conventional CT devices a slice is taken layer by layer, while newer CT devices multi-cell detectors are able to scan up to four slices at the same time.
• a back-projection algorithm is generally used for the image reconstruction from projections acquired here.
• the gray value information contained in the projections is uniformly distributed along a beam from the projection pixel to the radiation source and weighted with a geometric distance factor.
• such rear projections are used in connection with a filtered projection, in which the projection data are first weighted depending on the location and subjected to filtering and then back-projected into a volume data set initialized with zero.
• the re-production can also include iterative methods, in which a projection and a back-projection operation are included in each step.
• the volume data can be read out from a memory for the reconstructed spatial volume image and, depending on the application, visualized immediately two-dimensionally in the form of sections or even three-dimensionally.
• This task is solved on the one hand by a method for the image reconstruction of a room volume from acquired projections, in which, during a reconstruction step, each acquired projection is fed once to data processing from a memory for the acquired projections and the intensity of a voxel of the reconstructed room volume image during the reconstruction step each for the voxel relevant projection is updated.
• a method for the image reconstruction of a room volume from acquired projections in which, during a reconstruction step, each acquired projection is fed once to data processing from a memory for the acquired projections and the intensity of a voxel of the reconstructed room volume image during the reconstruction step each for the voxel relevant projection is updated.
• this object is achieved by a device for image reconstruction of a room volume from acquired projections with a memory for the acquired projections and a memory for the reconstructed room volume image, in which these memories are linked together by data processing and which is characterized in that the data processing at least comprises two processing pipelines, each with at least one memory area of a projection cache for a projection or a region of one Projection and, on the other hand, are connected to at least one memory area of a voxel buffer, the voxel buffer being linked to the memory for the reconstructed spatial volume image and the projection cache to the memory for the acquired projections.
• a plurality of projections or regions of these projections are thus preferably processed in parallel, so that they only have to be loaded once, with all the relevant voxels being processed in accordance with the reconstruction specifications, in particular with such an arrangement for a projection or region of a projection which has been loaded once.
• a sub-cube from the voxel space can be processed in a sub-step of the reconstruction, such a sub-cube being stored in the voxel buffer for reasons of effectiveness and thus speed.
• these intensity data are available for evaluation in the buffer and it is not necessary to data from the actual memory for the reconstructed volume image, which is usually very time-consuming because of its size.
• a separate algorithm or a special hardware structure can be used to fill the buffer.
• the entire spatial volume image in the form of suitably selected subcubes can also advantageously increase the processing speed independently of the rest of the process. This is particularly the case when the corresponding voxel buffer is designed as a shift register.
• the projection information available in this region can be made available in a comparatively small memory, in particular a cache, a processing time being reduced by a cache, regardless of the rest of the inventive method or the other features of the described device, since Such a cache has significantly shorter access times for a computer unit connected to it.
• the corresponding region can be determined, for example, by projecting the respective voxels onto the projection plane and using the respective covered area.
• This selection of voxels is preferably a previously described sub-cube, as a result of which the advantages of the previously described approaches add up. It goes without saying that the subcubes do not necessarily have to have a cube shape; rather, any amount of voxels, preferably any simply connected amount of voxels, can be used if this reduces the number of projections or regions that are required in succession for a sequence of certain reconstruction steps can.
• voxel buffer is also advantageous independently of a projection cache or a parallel computer structure, since the access times of the computing unit can be shortened by such a buffer, since a large main memory, as is required for the recording of all the reconstructed image information, is essential has slower access speeds.
• a second voxel buffer can be provided, which can be connected to the processing pipeline as an alternative to the first voxel buffer and can exchange data with the memory for the reconstructed spatial volume image independently of the first voxel buffer.
• One of the two voxel buffers can exchange data with the memory for the reconstructed spatial volume image, while the other voxel buffer is used for the computing operations. When the calculation is finished, these connections can be exchanged with a simple switch. In this way, dead times are avoided by the data exchange between the voxel buffer and the memory for the reconstructed volume image.
• the invention proposes a device for image reconstruction of a room volume from acquired projections with a memory for the acquired projections and a memory for the reconstructed room volume image, which are linked to one another via data processing, in which the memory bandwidth is below the processing power of the overall system lies.
• a device aligned in this way is able to work faster than the memory enables and therefore requires that the hardware is used optimally.
• Memory bandwidth and processing power are preferably compared in voxels / second, although other criteria are also possible here, which allow a comparison between the performance of the memory for the reconstructed spatial volume image and for the acquired projections with the processing performance.
• spatial volume image encompasses any representation in which the information contained in the projections is determined three-dimensionally and stored. In particular, this can be an intensity distribution in a voxel space. The same applies to the "projections”.
• Figure 1 is a schematic representation of an X-ray system.
• FIG. 2 shows the X-ray system according to FIG. 2 in section
• FIG. 5 shows a schematic representation of the computer structure according to FIG. 3 with the link between volume and projection;
• FIGS. 3 to 5 shows a process sequence with the computer architecture according to FIGS. 3 to 5;
• 9 shows several projection planes for a sub-cube; 10 shows a possible arrangement of projections and image space to be reconstructed;
• FIG. 11 shows the selection of a voxel slice, a voxel row selected in the voxel slice or a voxel ice selected in the voxel row;
• FIG. 13 shows a computer structure according to the prior art with a plurality of parallel memories for the intensity information.
• a person 1 is irradiated by means of a radiation source 2.
• projections 3 can be recorded with a corresponding detector, which ultimately reflect the interaction of the corresponding beam cone 4 with the body of the irradiated person.
• the radiation source 2 and the corresponding detector are arranged so as to be rotatable about the person, so that different projection directions can be recorded.
• other egg devices in which a rear projection is necessary can also be used.
• objects can also be examined accordingly.
• the ascertained projections 5 are stored in a corresponding memory 6 for the acquired projections. From these, a room volume image is to be determined, which is stored in a memory 7 for the reconstructed room volume image (see FIGS. 1 to 5).
• the two memories 6 and 7 are linked to one another via data processing 8, the data processing 8 comprising a projection cache 9 in the exemplary embodiments shown in FIGS. 3 and 4.
• Data from the projection memory 6 can be stored in this projection cache 9 as required.
• the projection cache 9 has individual memory segments 10, in each of which transmitted pixel data of a projection plane are stored.
• a hardware or processing pipeline 11 is provided for each storage unit 10 and is assigned to a cell 12 of a voxel intermediate storage 13 designed as a shift register.
• the hardware pipeline 11 reads the necessary projection pixels from the projection memory 9 for the volume element stored in the memory 12. It then calculates the contribution for the volume element which is added to the previous contribution of the volume element.
• the intensities along the rays “source-projection pixels” are distributed uniformly over the relevant voxels in the present embodiment, whereby — depending on the specific embodiment — a function can also be provided which takes into account a geometric attenuation. For each voxel, it must be calculated on the basis of each projection on which point the center of the voxel is mapped.
• the intensity of the point in the projection plane 5 is generally calculated by bilinear interpolation of the neighboring pixel intensities. The determined value is then multiplied by the inverse square of the “source-voxel” distance and added to the previous contribution in the voxel.
• a second voxel intermediate storage 14 is provided in the embodiment according to FIG Hardware pipelines 11 is connected.
• the parallel processors of the hardware pipelines 11 are optimally used.
• the voxel buffer 13 is emptied and reloaded accordingly, so that it is then available for further calculations, and the calculations for the data in the voxel memory 14 are completed.
• the voxel memory 13 is then connected again to the hardware pipelines 11, while the data exchange with the memory 7 takes place for the voxel memory 14.
• region is understood to mean a relatively small amount of, preferably simply contiguous, pixels, the size of the amount being selected such that this amount of pixels can be loaded into the projection cache 9 without further ado.
• any voxels or projections 5 or regions 15 can be loaded into the projection cache 9 and in the voxel buffer 13 or 14.
• the regions 15 read into the projection cache 9 are preferably correlated with one another.
• the correlation can be selected such that a sub-cube 17, the number of voxels, is selected from the image space 16 to be reconstructed preferably corresponds to the number of storage elements 12 in the voxel intermediate storage 13 or 14.
• a sub-cube 17 with four voxels with an edge length is shown as an example in FIG. 7 and can be stored in a voxel buffer 13 with 64 memory elements.
• All the regions 15 which contain the relevant image data corresponding to the respective projection direction are then loaded into the projection cache 9, as is shown by way of example in FIGS. 5 and 9.
• the respective calculations can be carried out, all voxels of the sub-cube 17 for the respective regions 15 loaded in the projection cache 10 being calculated in parallel by the shift register.
• All projection directions are ideally stored in the projection cache 9. However, a cost-benefit assessment can be carried out after only part of the required projections 5 have been stored in the projection cache 9 and a corresponding data exchange has been provided in the meantime.
• the process sequence thus carried out is shown in FIG. 6.
• the reconstruction process is essentially based on two steps: first the data is filtered, then the back projection is carried out.
• the x-ray system has rotated about an axis which is parallel to the normal to one of the side surfaces of the volume cube or the image space 16 to be reconstructed (cf. FIG. 10).
• voxel disks 17A of thickness N can then be viewed separately parallel to this side surface and reconstructed from corresponding projection lines 5A (numbered as an example in FIG. 10) of the projections 5.
• voxel cubes 17 of size NxNxN can be read from these slices, this preferably being carried out iteratively according to voxel lines 17B.
• the contribution of a stack of NxNxN projections (projection block) onto each of the voxels of this read cube is calculated and added to the current values of the voxels.
• the projections 5 can then be used in each case in their rows 5A relevant for a voxel disk 17A for calculating the individual voxels.
• each pipeline 11 preferably contains enough memory for this data. Due to this suitable arrangement of projection to volume, it can be assumed that there are several adjacent lines in the projection image. In order for the method to work even more efficiently, the individual lines are preferably reloaded during the calculation (2-way memory) in order to avoid access conflicts, or several memory banks are used.
• the size of the memory in this embodiment variant is preferably approximately voxel cube width * side length of the projection image (typically 512 or 1024) values (typically 16 bits).
• 8 multipliers are preferably used, ie 512 multipliers in total.
• the calculation process looks as follows:
• cubes and rows of several projections can be kept loaded at the same time and can be The mechanism of each voxel of the cube can be matched with each of the loaded projections. This has the advantage that the calculated slice is finished after it has been run through all the projections and almost all the lines used in the projections no longer have to be loaded.
• FIGS. 12 and 13 Exemplary embodiments of the prior art are shown in FIGS. 12 and 13, each of which has parallel structures 108 and 208, however, either the projection planes 105 are stored in a plurality of memories 106 or the voxel spaces 207 are stored in parallel memories, which means that the Costs increase significantly.
• the present invention can be implemented, for example, in a C-arm angiography system or on a linear acceleration in conjunction with an electronic portal imagine device (EPID).
• the reconstruction can be carried out, for example, by a filtered back projection or by imperative methods.
• the projection data are first weighted depending on the location and subjected to filtering. If the filtering is implemented in the frequency domain using well-optimized software for Fourier transformation, the step can be regarded as relatively non-critical.
• For the back projection of the filtered profiles these and a volume data set initialized with zero are first loaded into the memory of the corresponding system card.
• the rear projection takes place as described above.
• the interactive process for image reconstruction Each step includes a projection and a back projection operation. Special methods and architectures are already known for realizing an efficient projection operation (ray tracing).
• the above-described method and the above-described device can be used for a voxel-based return position.

Landscapes

• Physics & Mathematics (AREA)
• Engineering & Computer Science (AREA)
• General Physics & Mathematics (AREA)
• Theoretical Computer Science (AREA)
• Algebra (AREA)
• Mathematical Analysis (AREA)
• Mathematical Optimization (AREA)
• Mathematical Physics (AREA)
• Pure & Applied Mathematics (AREA)
• Image Generation (AREA)
• Apparatus For Radiation Diagnosis (AREA)
• Image Processing (AREA)

Applications Claiming Priority (5)

Publications (1)

Family

ID=26008391

Family Applications (1)

Country Status (5)

Families Citing this family (12)

Family Cites Families (8)

• 2002
  • 2002-01-26 DE DE10290311T patent/DE10290311D2/de not_active Expired - Fee Related
  • 2002-01-26 WO PCT/DE2002/000266 patent/WO2002061686A1/de not_active Application Discontinuation
  • 2002-01-26 JP JP2002561778A patent/JP2004523037A/ja active Pending
  • 2002-01-26 EP EP02708166A patent/EP1356432A1/de not_active Withdrawn
  • 2002-01-26 US US10/470,494 patent/US20040114728A1/en not_active Abandoned
• 2004
  • 2004-12-21 US US11/018,476 patent/US20050151736A1/en not_active Abandoned

Non-Patent Citations (2)

Also Published As

Similar Documents

Legal Events