US-12622653-B2 - Method and apparatus for reconstructing CT images

US12622653B2US 12622653 B2US12622653 B2US 12622653B2US-12622653-B2

Abstract

A method for reconstructing CT images, comprises: providing CT recording data; reconstructing overlapping partial images; establishing displacement vectors for registering overlap regions of the partial images; interpolating a displacement vector field for each partial image from associated sets of the displacement vectors of the two side regions; creating an output image dataset based on the CT recording data and the displacement vector fields; and outputting the output image dataset.

Inventors

Andre RITTER
Thomas Allmendinger

Assignees

Siemens Healthineers Ag

Dates

Publication Date: 20260512
Application Date: 20230130
Priority Date: 20220131

Claims (20)

1 . A method for reconstructing Computed Tomography (CT) images, the method comprising: providing CT recording data including recordings of a plurality of overlapping partial image volumes; reconstructing a working image dataset from the CT recording data, the working image dataset including a plurality of partial images, and each respective partial image among the plurality of partial images having an overlap region with at least one corresponding partial image from among the plurality of partial images such that the plurality of partial images includes a plurality of overlap regions; establishing subsets of displacement vectors associated with two opposing side regions of each of the plurality of partial images; interpolating a corresponding displacement vector field for each respective partial image among the plurality of partial images from the subsets of the displacement vectors associated with the two opposing side regions of the respective partial image to obtain a plurality of displacement vector fields, two sides of the corresponding displacement vector field corresponding to the subsets of the displacement vectors, the interpolating including interpolating the subsets of the displacement vectors between the two sides of the corresponding displacement vector field based on a transfer function from a first subset of the displacement vectors to a second subset of the displacement vectors, and the subsets of the displacement vectors including the first subset and the second subset; creating an output image dataset based on the CT recording data and the plurality of displacement vector fields; and outputting the output image dataset.
2 . The method as claimed in claim 1 , wherein the establishing comprises: establishing a third subset of the displacement vectors for registering a first overlap region between a first partial image and a second partial image, the establishing the third subset of the displacement vectors including determining a two-dimensional boundary slice in the first overlap region, the first overlap region being among the plurality of overlap regions, and the plurality of partial images including the first partial image and the second partial image.
3 . The method as claimed in claim 2 , wherein the two-dimensional boundary slice is a plane perpendicular to a CT axis.
4 . The method as claimed in claim 3 , wherein the two-dimensional boundary slice is at a central position within boundaries of the first overlap region.
5 . The method as claimed in claim 2 , wherein the determining of the two-dimensional boundary slice comprises: assuming a set of displacement vectors with a value; and determining a position of the two-dimensional boundary slice along a CT axis such that quadratic differences of image values are at a minimum.
6 . The method as claimed in claim 1 , further comprising: defining a regular 2D grid of sampling points in a first overlap region of a first partial image among the plurality of partial images, the first overlap region being among the plurality of overlap regions; and associating a respective displacement vector with each sampling point among the regular 2D grid of the sampling points to obtain a third subset of the displacement vectors.
7 . The method as claimed in claim 6 , wherein the respective displacement vector points to a location in the first partial image from a corresponding sampling point among the sampling points; and the regular 2D grid of the sampling points are at least one of equidistant or form a boundary slice.
8 . The method as claimed in claim 7 , wherein a side region of the first partial image is defined based on the sampling points and a set of displacement vectors is associated with the side region of the first partial image, the set of displacement vectors including the third subset of the displacement vectors.
9 . The method as claimed in claim 1 , wherein the displacement vectors include a first displacement vector associated with a first sampling point of a first partial image and a second displacement vector associated with a corresponding sampling point of a second partial image, the first sampling point and the corresponding sampling point being in a first overlap region of the first partial image and the second partial image, the plurality of partial images including the first partial image and the second partial image, and the plurality of overlap regions including the first overlap region.
10 . The method as claimed in claim 9 , wherein the second displacement vector represents an inverse vector of the first displacement vector.
11 . The method as claimed in claim 9 , further comprising: determining the first displacement vector and the second displacement vector based on a first measure used for a similarity of image regions of the first partial image and the second partial image, the first measure being based on a square of differences of image values; and determining, via an iterative method, a minimizing vector based on current image values and image value gradients of the first partial image and the second partial image, the determining the minimizing vector being performed stepwise based on a step length at respectively displaced sampling points, and the minimizing vector being based on the first measure.
12 . The method as claimed in claim 1 , wherein the establishing includes setting values of a third subset of the displacement vectors for first image regions of the plurality of partial images to a first value, the first image regions having an image value below a limit value, and the limit value being above an air CT value.
13 . The method as claimed in claim 12 , wherein the limit value is between −1000 HU and −800 HU.
14 . The method as claimed in claim 1 , wherein the interpolating comprises: determining spacings of a relevant image region from two opposing side regions of a first partial image among the plurality of partial images, the relevant image region being between the two opposing side regions of the first partial image; determining a weighted relationship of the spacings from the two opposing side regions of the first partial image, a closer side region among the two opposing side regions of the first partial image having a greater weight in the weighted relationship; calculating a first displacement vector based on a ratio and a vector addition of a second displacement vector in a first side region and a third displacement vector in a second side region, the second displacement vector being among the first subset of the displacement vectors, the third displacement vector being among the second subset of the displacement vectors, and the first side region and the second side region being among the two opposing side regions of the first partial image; determining a minimizing vector based on current image values and image value gradients of two partial images among the plurality of partial images, the two partial images including the first partial image, the two partial images having a first overlap region in the first side region or the second side region, the minimizing vector being based on a first measure, the determining the minimizing vector being performed stepwise based on a step length at respectively displaced sampling points, and the first overlap region being among the plurality of overlap regions; and adding the minimizing vector to the first displacement vector.
15 . The method as claimed in claim 1 , further comprising: performing a smoothing of a third subset of the displacement vectors in a first overlap region or a first displacement vector field between two opposing side regions of a first partial image among the plurality of partial images, the first overlap region being among the plurality of overlap regions, the first displacement vector field being among the plurality of displacement vector fields; and at least one of performing a smoothing convolution or a low pass filtration to establish the third subset of the displacement vectors in the first overlap region, or performing a convolution with a Gaussian window in the first overlap region before interpolating the first displacement vector field.
16 . The method as claimed in claim 15 , wherein the first overlap region is a boundary slice; or the third subset of the displacement vectors in the first overlap region are established using a convolution with a Gaussian window.
17 . The method as claimed in claim 1 , wherein the creating the output image dataset includes reconstructing the plurality of partial images, displacing image regions of the plurality of partial images according to the plurality of displacement vector fields, and assembling the plurality of partial images based on the displacing; or the creating the output image dataset includes reconstructing the plurality of partial images based on the plurality of displacement vector fields using a back projection, a region of the output image dataset being formed based on information from first partial images among the plurality of partial images by weighting corresponding contributions of the first partial images.
18 . The method as claimed in claim 17 , wherein the output image dataset has a greater image resolution than the working image dataset.
19 . The method as claimed in claim 1 , wherein at least one of the output image dataset or the plurality of partial images are reconstructed via a different weighting of temporal portions of the CT recording data, the weighting being based on weighting values determined for a plurality of image points of the output image dataset or the plurality of partial images.
20 . The method as claimed in claim 19 , wherein the plurality of image points form a 3D image data grid; and a weight is associated with a number of displacement vectors.

Description

CROSS-REFERENCE TO RELATED APPLICATION(S) The present application claims priority under 35 U.S.C. § 119 to German Patent Application No. 10 2022 200 999.1, filed Jan. 31, 2022, the entire contents of which are incorporated herein by reference. FIELD One or more example embodiments of the present invention relate to a method and an apparatus for reconstructing CT images, in particular for a CT stack artifact correction method. BACKGROUND In CT (computed tomography) imagingin the region of the thorax, movements such as, for example, the heartbeat or changes in the lungs caused by breathing can lead to inconsistent recording data. CT images which are reconstructed from inconsistent recording data show these inconsistencies in the CT images as artifacts in the form of “blurred” structures or discontinuities such as jumps at transitions. These artifacts can be reduced in that, for example, by way of high rotation speeds or dual source techniques, the time span that is needed to cover a suitable rotation angle interval during a recording is lessened. If the typical length scales on which a change takes place within this time interval are small compared with the voxel spacings (ca. 1 mm) achieved later, then the artifacts are also small, but still present. For the reduction of these artifacts, in a known movement pattern, the recording is often performed with defined movement conditions. In order to achieve this, in, for example, cardiac imaging with CT, apart from the temporal resolution, above all the control of the recording and reconstruction via the analysis of an ECG signal obtained during the recording plays an important part. Normally, above all, the data which can be obtained or taken into account in the reconstruction during a defined rest position of the heart are recorded. In recordings of an extended region, there is a particular difficulty. The recording region of a CT detector often does not necessarily cover a desired recording region along the CT axis (e.g. that of the heart) completely. A complete coverage is then achieved, for example, by way of a continuous table feed during the recording (spiral CT) or by way of a table feed between a plurality of successive recordings with a static patient (sequence CT). The consequence thereof is that data from the same recording time point is not present for all the positions along the CT axis used in the recording. In particular, during the recording of the heart, it is thus possible that for two different positions, only data from different heart cycles is available. Or conversely, for the same position, data from a plurality of heart cycles is available. It can occur that recording data which originates from comparable regions of a plurality of heart cycles nevertheless has spatial differences, for example, by way of an overlaid breathing movement or other differences between the heart cycles which cannot necessarily be represented and recognized in the ECG. Thereby, despite a high temporal resolution, inconsistent data can arise, which can find expression therein that in the reconstructed CT image, overlaid structures from a plurality of heart cycles are recognizable. However, for example, discontinuities between the slices can also arise, since for instance, subregions are so greatly displaced during the patient movement that they cannot be captured by the detector region in any of the heart cycles. In special SOMATOM CT scanners, the possibility of a so-called True Stack reconstruction exists. Herein, each CT layer is reconstructed from only a single defined heart cycle, even if there were recording data from a plurality of heart cycles for one slice. Therein, coherent slice stacks (stacks) which can be associated with exactly one determined heart cycle (thus also a time point) are formed. A plurality of such stacks can then follow one another. In each stack, it can be assumed that artifacts evoked by movement only play a part if the time interval for the recording of this stack is too large, as described above. If there are inconsistencies between the stacks (e.g. due to breathing movement), then this can reveal itself in that discontinuities arise at the boundary between two stacks. These are distinguished by truncated structures or doubled structures along the CT axis. A further-developed method for solving the discontinuity problem at stack boundaries has been disclosed by Lebdev et al. in “Stack Transition Artifact Removal (STAR) for Cardiac CT” (Med. Phys. 46 (11), 4777-4791; November 2019). This is based on forming a displacement vector field via image registration, on the basis of which the discontinuities at the stack boundaries are lessened in that the image structures are deformed according to the vector fields thus obtained. For this purpose, the fact is made use of that the stacks usually have a greater coverage than the finally displayed coverage along the CT axis. By this mechanmism and/or means, an overlap region of the stacks is forme