Adaptive two-pass cone-beam artifact correction using a FOV-preserving two-source geometry: A simulation study
The processing steps of the second-pass correction method: (1) Acquisition , (2) reconstruction , (3) modeling , (4) second-pass acquisition, i.e., simulated projection , (5) second-pass reconstruction , (6) artifact separation by subtraction , and (7) subtraction of the artifacts from the original image . Shown are sagittal slices of the Forbild thorax phantom with static heart insert. L/W: 0/500 HU.
The CT acquisition geometry.
The reconstruction zones. The central 360° zone can be reconstructed completely even for short scan acquisition. The two neighboring zones above and below are the 180° zones, which can only be reconstructed completely in case of a full-scan acquisition. The outermost zones cannot be reconstructed because any position in those zones is illuminated over less than 180° even with a full scan acquisition.
The circular geometry. The central gray rectangle represents the object. The large hexagon is a central cross section of the 360° zone . The small hexagon is the cross section of the 360° zone of the second pass. The reduced second-pass cone angle is denoted by .
The stereo tube. The two vertically separated sources are switched on and off in alternating fashion at the acquisition frame rate while they synchronously rotate around the patient on circular arcs.
The stereo tube geometry. When the sources are collimated like this, any voxel within the cylindrical FOV is in one or both of the sources’ 360° regions. For clarity the cone and 360° region of source 2 are not drawn.
Left: The aperture weighting functions for the two sources used during backprojection. Right: Stereo reconstruction of the Forbild head phantom with heart insert. Contrary to the usual behavior of axial reconstruction, where the cone-beam artifacts become stronger toward the edges of the image, in stereo images the artifacts are strongest in the center.
Corrected image roughness as a function of . The left curve belongs to the example in Fig. 10; the right one to the example in Fig. 13.
Study showing sagittal slices of a phantom containing five ellipsoidal disks, stacked in axial direction, i.e., along , and immersed in water. The disks are 160 mm in diameter, 25 mm spaced apart, and 10 mm thick. The CT values of the disks correspond to 800 HU. Upper left: Phantom. Upper right: Uncorrected stereo reconstruction. Lower left: Corrected image. Lower right: Difference between corrected and uncorrected images. L/W: 0/500 HU. The central vertical lines mark the locations of the profile plots of Fig. 9.
Profiles along the vertical lines shown in Fig. 8.
Example of cone-beam artifact correction. Coronal (left column) and sagittal (right column) images of the Forbild thorax phantom with heart insert. Uncorrected (upper row) and corrected (lower row) reconstructions. L/W: 0/300 HU (left column) and 50/300 HU (right column). The lines labeled “” mark the location of line profiles shown in Fig. 15.
Example of cone-beam artifact correction. Transaxial (first and third rows), coronal (second and fourth rows, left column), and sagittal (second and fourth rows, right column) images of the Forbild thorax phantom with dynamic heart insert. Uncorrected (upper two rows) and corrected (lower two rows) reconstructions. L/W: 0/100 HU (transaxial), 0/300 HU (coronal), and 50/300 HU (sagittal).
Example of cone-beam artifact correction. Coronal (left column) and sagittal (right column) images of a medical head phantom. Uncorrected (upper row) and corrected (center row) reconstructions. L/W: 0/300 HU. The lower row shows the corresponding difference images. L/W: −70/170 HU. The boxes frame image regions in which noise figures are calculated (see Table IV).
Example of cone-beam artifact correction. Coronal (left column) and sagittal (right column) images of a medical thorax phantom. The heart is filled with contrast agent. Uncorrected (upper row) and corrected (center row) reconstructions. L/W: 0/400 HU. The lower row shows the corresponding difference images. L/W: 83/170 HU (left) and 48/170 HU (right). The boxes frame image regions in which noise figures are calculated (see Table IV).
CT value line scans taken from the image data corresponding to Fig. 10 at three random positions with high CT value gradients.
Relevant scanner geometry parameters used in this work.
Additional stereo tube parameters.
Reconstruction FOVs and voxel sizes for the presented examples.
Noise figures measured in the marked regions of Figs. 13 and 14.
Article metrics loading...
Full text loading...