3D reconstructions from bulky objects using sparse-angle data
Abstract »We have devised a setup for efficient 3D X-ray imaging of bulky objects, for example engines, large containers or massive valves. Accurate measurement of such targets requires a powerful source with mean photon energies well in the MeV scale. One problem with these measurements is that each 2D projection may take several minutes to complete. A whole scan consisting of hundreds of projections thus occupies the X-ray device for an entire day.
Our aim in this work is to demonstrate how we can limit the number of required projections without significantly compromising the image quality by using advanced reconstruction algorithms. By working with a smaller set of projections, not only do we limit the overall time needed for inspection, but also reduce the total exposure time of the detector, increasing the life-time of the X-ray device.
In this study, we have used a 6 MeV electron linear accelerator (LINAC) for taking 360 cone-beam projections of a target object in a circular geometry. The classical NDT S-band LINAC was equipped with a standard 1 mm tungsten target. It can work with 4.5 microsecond pulses and 300 Hz repetition rate, achieving 6 Gy/min dose rate at 1 m distance.
From the full data set with one-degree angular increment, we select a subset 60 projections as the sparse data set. We compare the quality of the 3D reconstructions obtained using (1) the full data set with a standard algorithm (FDK) and (2) the sparse data set with an advanced algorithm based on statistical inversion (PSIG). We demonstrate how the relevant details in the reconstruction are preserved to sufficient accuracy even with the limited data set, increasing throughput of the LINAC by a factor of six.