Antialiasing backprojection for helical MDCT.

Helical CTs are well known to suffer from aliasing artifacts because of their finite longitudinal sampling pitch. The artifact pattern is typically strong streaks from bone edges in clinical images. Especially in the case of multidetector row CT, the artifact resulting from longitudinal aliasing is often called a windmill artifact because the visible streaks form a windmill pattern when the object is of a particular shape. The scan must be performed using a very thin slice thickness, i.e., fine sampling in the longitudinal direction, with a longer scan time to mitigate this aliasing artifact. Some elaborate longitudinal interpolation methods to remediate longitudinal aliasing have been proposed, but they have not been successful in practice despite their theoretical importance. A periodic swing of the focal spot in the longitudinal direction, a so-called z-flying focal spot, was introduced recently to achieve finer sampling. Although it is a useful technique, some important deficiencies exist: It is sufficiently effective only near the isocenter and is difficult to apply to a scan using a thick slice thickness, even though longitudinal aliasing is more serious at the thicker scan. In this paper, the author addresses the nature of interlaced (or unequally spaced) sampling and derives a new principle of data treatment that can suppress the aliased spectra selectively. According to this principle, the common practice of image reconstruction, which backprojects data along the original sampling ray path, is never the best choice. The author proposes a new scheme of backprojection, which involves the longitudinal shift of projection data. A proper choice of longitudinal shift for backprojection provides effective and selective suppression of aliased spectra, with retention of the original frequency spectrum depending on the level of focus swing. With this shifted backprojection, the swing of focus can be made much smaller than for a conventional z-flying focal spot. The required amount of shift for backprojection is position dependent. Nevertheless, its implementation in the reconstruction process can be achieved simply by relocating the x-ray source and detector assembly from positions of actual scanning. Through simulation, the combination of shifted backprojection and the small swing of focus is evaluated. Results confirm that the artifact attributable to longitudinal aliasing is well suppressed in the entire field of view, whereas the penalty on the slice sensitivity profile (or longitudinal resolution) can be kept minimal. Moreover, this method solves other deficiencies of z-flying focus, such as inapplicability to scans with a thicker slice thickness.