Patent Application
20080181355
The present invention relates to a method and arrangement in X-ray imaging, in particular three-dimensional imaging, and more especially tomosynthesis.
Tomosynthesis is used to create a three-dimensional image volume of a person's body part, e.g. her breast, or an object, using X-rays. Currently, tomosynthesis breast imaging is available only for research purposes, but an increasing number of market analysts believe that tomosynthesis breast imaging will become more widely used than conventional two-dimensional mammography.
Tomosynthesis is essentially a limited form of Computed Tomography (CT). Normally, several projection images, e.g. 5 or 30, are acquired from slightly different angles, using a modified X-ray system. Each projection image is essentially a conventional 2-dimensional digital X-ray image of the examined object. The projection images are then combined using special purpose software for reconstruction of a 3-dimensional image volume, which is a 3-dimensional array of voxels, wherein each voxel is essentially a value corresponding to X-ray attenuation in one location of the real world. The image volume may also be regarded as a stack of layers, wherein each layer is a 2-dimensional image. The stack of images can be displayed in a sequence.
In breast imaging, efficient workflow is very important, in particular for specialized screening mammography clinics where healthy patients are examined on a regular basis. Speed requirements and cost control have driven many mammography clinics to introduce a workflow that resembles an assembly line at a factory. Most often, patients leave the clinics before a radiologist looks at the images. Thus, the demands and requirements for breast imaging systems are different compared to other applications of medical X-ray imaging.
The reconstruction time is a problem in breast imaging, due to high demands for speed, and computationally expensive reconstruction algorithms. The long processing time arises partly due to truncation of the object at the border of projection images. Truncation cause difficulties when using the same methods as for mainstream CT, such as filtered back-projection and direct Fourier methods. Another problem is that it is desired to produce an image volume with many more voxels than there are pixels in the stack of projection images, which calls for regularization when using iterative methods. Recent research indicates that image quality can be improved using extremely computationally expensive regularization, such as the TV-norm regularization (e.g. Sidky, Kao, Pan 2006). Therefore, it is expected that reconstruction time or computational cost will remain a challenge for many years to come.
The long reconstruction time is a problem in normal workflow, wherein the operator of the X-ray apparatus looks at the acquired image and determines whether or not it can be used for diagnosis. In case of failure, a new image must be taken before the patient leaves the examination. The main risk of failure is bad positioning, wherein an important part of the breast is not visible. Other risks for reconstruction are apparatus failure, large dirt particles, silicon implants and metal pieces such as piercing.
According to prior art, projection images may be previewed. Unfortunately, each projection image tends be characterized by a substantially worse image quality than a reconstructed volume, since it contains the combination of X-ray quantum noise and disturbing super-imposed tissue. Normally, each projection image is very noisy, since each projection image is acquired using fraction of the total dose.
Image quality is important for preview in non-screening breast imaging, wherein patients have been called back to further study something that was seen in earlier images. The operator of the X-ray apparatus shall look in the image to determine whether or not that something is visible and well depicted in the image. If not, the operator may acquire another image with a different positioning.
The object of the present invention is to provide an effective aid for quickly determining whether or not a tomosynthesis image acquisition was successful. The aid is quick enough to fit in the workflow of virtually any modern mammography clinic. The aid is also inexpensive to manufacture.
For these reasons, an arrangement for a quick three-dimensional volume preview is presented.
The present invention effectively reduces the risk of a patient needs to be recalled due to bad image quality.
In contrast to prior art, wherein the projection images are displayed as a preview, the present invention
The present invention produces and outputs two image volumes from the same set of projection images, a preview image volume and a high quality diagnostic image volume. According to the present invention, a preview image volume is reconstructed using projection images at a lower resolution. There are two different pipes of processing for preview image volume and the diagnostic image volume.
Preferably, the full projection images are stored in a first memory buffer. The contents of this memory buffer will be used for reconstructing the diagnostic image volume, but that may wait. The primary focus is reconstruction of a preview image volume. Preferably, the projection images are sub-sampled, and the sub-sampled data is stored in a second buffer, which is used for reconstruction of the preview images. Preferably, the sub-sampling reduces the number of pixels by 4 or 16 times. The sub-sampled images are used for reconstructing a three-dimensional image volume, and the resolution in the image volume follows the resolution in the projection images. Preferably, the reconstruction of the preview image is performed in essentially the same way as for the diagnostic image volume, except for different parameters related to image quality and geometrical differences for the preview resolution. The primary parameter difference is that fewer layers are reconstructed, since the preview image volume shall have at least as elongated voxels as the diagnostic image volume. Thus, one octave of sub-sampling reduces the number of pixels by 2*2=4 times in the projection images, and the number of reconstructed voxels will be reduced by 2̂3=8 times. Thanks to one octave of sub-sampling, the reconstruction time will be at least 4 times, and probably close to 8 times faster. There are reasons to believe that two octaves of sub-sampling will provide enough image quality for preview, which speeds up reconstruction up to 64 times. Further speed is possible by computational approximations or sub-optimal parameters. Most tomosynthesis reconstruction algorithms are iterative (e.g. EM or Lange-Fessler 1995) and the result converges over a number of iterations. Preferably, the preview image is computed with at most half number of iterations compared to the diagnostic image volume, which doubles speed. In addition, it is possible to skip computationally expensive regularization, which simplifies computations and also tend to substantially decrease the number of iterations.
Different embodiments of the invention may use different methods of reducing the resolution of the projection images, such as iterators or similar mechanisms for reducing or omitting pixels. The preferred embodiment of the present invention computes averages of neighboring pixels values, whereby noise is reduced. It may also be possible to leave out entire projection images, but the preferred implementation uses all available projection images.
In the most preferred embodiment of the present invention, the projection images are acquired using a multi-slit scanner, wherein the x-ray source and detector rotate around a common rotation axis in order to simultaneously obtain a set of projection images. The images are reconstructed in a coordinate system, which is curved around the rotation axis, like polar coordinates. Thanks to the curved coordinate system, slabs of the image volume can be reconstructed with very little dependence on data pixels and voxels outside the thin slab. The low dependence enables massive parallelism when computing the diagnostic image volume. The coordinate system turns out to be even more advantageous when computing the preview image volume. After sub-sampling, the dependence is negligible, and it is possible to perform reconstruction by massive parallelism, wherein each cell of parallelism is based on only one pixel column from each projection image. After sub-sampling, the data set is small enough to fit into the CPU cache in a normal computer, which further speeds up the calculation. Each column can be processed as a vector with constant address shifts, whereby the preferred implementation is suitable vector operations and Single Instruction, Multiple Data (SIMD) instructions in computer hardware. The current trends of computer hardware are parallelism by threads, parallelism by vector operations and increasing speed difference of access to CPU cache compared to Random Access Memory (RAM). The preferred implementation is promising with respect to all major trends fits well into all of these trends in modern computer hardware.
The preferred embodiment of the present invention is inexpensive and provides a tentative image much faster than the diagnostic image volume, thus re-calls of patients may be avoided.
Another aspect of the present invention is a reconstruction of a second diagnostic image volume, based from the same projection image data. A second main diagnostic image volume may be computed whenever the diagnosis is unusually difficult, or there are special image artifacts, which require an extra powerful method for reconstruction.
The invention may also be implemented as computer program comprising procedures for executing the steps mentioned earlier.
The invention is not limited to the shown embodiments but can be varied in a number of ways without departing from the scope of the appended claims and the arrangement and the method can be implemented in various ways depending on application, functional units, needs and requirements etc. The reconstruction arrangement can of course be used in other x-ray applications.
The above mentioned and described embodiments are only given as examples and should not be limiting to the present invention. Other solutions, uses, objectives, and functions within the scope of the invention as claimed in the below described patent claims should be apparent for the person skilled in the art.
