Cone beam volume CT image artifacts caused by defective cells in x-ray flat panel imagers and the artifact removal using a wavelet-analysis-based algorithm.

The application of x-ray flat panel imagers (FPIs) in cone beam volume CT (CBVCT) has attracted increasing attention. However, due to a deficient semiconductor array manufacturing process, defective cells unavoidably exist in x-ray FPIs. These defective cells cause their corresponding image pixels in a projection image to behave abnormally in signal gray level, and result in severe streak and ring artifacts in a CBVCT image reconstructed from the projection images. Since a three-dimensional (3-D) back-projection is involved in CBVCT, the formation of the streak and ring artifacts is different from that in the two-dimensional (2-D) fan beam CT. In this paper, a geometric analysis of the abnormality propagation in the 3D back-projection is presented, and the morphology of the streak and ring artifacts caused by the abnormality propagation is investigated through both computer simulation and phantom studies. In order to calibrate those artifacts, a 2D wavelet-analysis-based statistical approach to correct the abnormal pixels is proposed. The approach consists of three steps: (1) the location-invariant defective cells in an x-ray FPI are recognized by applying 2-D wavelet analysis on flat-field images, and a comprehensive defective cell template is acquired; (2) based upon the template, the abnormal signal gray level of the projection image pixels corresponding to the location-invariant defective cells is replaced with the interpolation of that of their normal neighbor pixels; (3) that corresponding to the isolated location-variant defective cells are corrected using a narrow-windowed median filter. The CBVCT images of a CT low-contrast phantom are employed to evaluate this proposed approach, showing that the streak and ring artifacts can be reliably eliminated. The novelty and merit of the approach are the incorporation of the wavelet analysis whose intrinsic multi-resolution analysis and localizability make the recognition algorithm robust under variable x-ray exposure levels between 30% and 70% of the dynamic range of an x-ray FPI.