Tomosynthesis reconstruction from multi-beam X-ray sources

We investigate methods for reconstructing tomosynthesis data using arrays of microfabricated X-ray sources and area CCD detectors. Tomosynthesis is a 3D imaging technique for limited-angle tomography that uses multiple radiographic images taken from an X-ray source placed at several positions to estimate a 3D distribution of X-ray attenuation. In our implementation, the moving X-ray source is replaced with multiple carbon nanotube field-emission X-ray sources fabricated on a single wafer. The sources are individually addressable, so the motion of the single X-ray source in traditional tomosynthesis is replaced by sequential sampling of the individual sources. Reconstruction is performed using the ordered-subsets convex (OSC) algorithm with pre-computed geometry factors and an image shearing technique for efficiency. Because of the limited resolution of tomosynthesis along the primary direction of projection, reconstructions are computed on non-cubic voxels for relatively thick slabs. We demonstrate the capabilities of the reconstruction technique on simulated data and breast phantom data from a prototype device. Reconstructions for an 1 I-beam system with 1000x1000 pixels per projection onto a 800x800x20 grid require 55 minutes (10 iterations) on a 3.2GHz workstation with 2 GB memory. We conclude that this implementation of the OSC algorithm is effective for reconstructing tomosynthesis datasets.