RTK/ImageQuality

This page summarizes the existing and the future solutions in RTK for improving image quality of cone-beam (CB) CT images. It is based on discussions at the RTK meeting on image quality.

X-ray source imperfections

• Geometric blurring can be corrected by the scatter glare correction detailed in the detector imperfections section.
• Exposure fluctuations from projection to projection are common. They can be corrected by rtk::I0EstimationProjectionFilter which automatically estimates a weighting constant per projection using an histogram analysis. This filter only works if there are pixels in each projections that measure x-rays that traversed air only (except maybe a few projections using a revursive least-square (RLS) algorithm). The filter does not have any parameter except the bitshift template value for the reduction of the histogram size. It is implemented for integer pixel types only.
• Focal spot motion cannot be corrected currently. It would require geometric calibration for each acquired projection using auto calibration.

Detector imperfections

• Variations of the flat field image and the dark field image is known (due to temperature changes or ghosting, see, e.g., [Siewerdsen and Jaffray, Med Phys, 1999]). Acquisition of these two images before and after the acquisition is the best solution when it is possible. There is no other solution in RTK except for the automatic detection of the constant I0 value (see fluctations of the source exposure).
• Lag corresponds to a short term effect of the detector. The PhD of[: the [ rtk::] filter implements correction implements

, see [Poludniowski et al, PMB, 2011] implemetend in [rtk::]

Beam hardening

• The acquired data may be linearized for a given material using rtk::LookupTableImageFilter as explained, e.g., in Fig. 1 of [Brooks and Di Chiro, PMB, 1976]. There are several solutions to compute the lookup table:
  • Compute it from the knowledge of the spectrum of the x-ray source and the detector response,
  • Measure the attenuation for several thicknesses of the material of interest,
  • Do a tomography of a homogeneous object (e.g., a cylinder) and find the lookup table using the rtk::WaterPrecorrectionImageFilter which implements [Kachelriess et al, Med Phys, 2006].
  • Estimate the p and q parameters of Equation 1 in [Ohnesorge et al, Eur Radiol, 1999] which comes down to a two-parameters beam-hardening correction and convert it to a lookup table.
• The algorithm of [Kyriakou et al, Med Phys, 2010] may easily be implemented from the existing code in RTK.

Statistical noise

• RTK has a fast 2D median filter for projection images for a few kernel dimensiosn, see rtk::MedianImageFilter. A GPU version of the median filter will be developed in Salzburg (Austria).
• Median filters do not preserve edges (see [Arias-Castro and Donoho, Annals of Statistics, 2009]. A multi-pass median filter is required which might be investigated in Louvain-La-Neuve (Belgium) in the future.
• The Savitzky–Golay filter is a promising solution that will be investigated in Louvain-La-Neuve (Belgium) in the future. This solution also provides derivatives of the image.

Truncated projection images

• The rtk::FFTRampImageFilter implements the heuristic solution of [Ohnesorge et al, Med Phys, 2000]. The parameter TruncationCorrection must be adjusted.
• Exact reconstruction based differentiated backprojection and inverse Hilbert filtering (see, e.g., [Noo et al, PMB, 2004]) is investigated in Lyon (France).