System for uncollimated digital radiography

摘要

The inversion algorithm based on the maximum entropy method (MEM) removes unwanted effects in high energy imaging resulting from an uncollimated source interacting with a finitely thick scintillator. The algorithm takes as input the image from the thick scintillator (TS) and the radiography setup geometry. The algorithm then outputs a restored image which appears as if taken with an infinitesimally thin scintillator (ITS). Inversion is accomplished by numerically generating a probabilistic model relating the ITS image to the TS image and then inverting this model on the TS image through MEM. This reconstruction technique can reduce the exposure time or the required source intensity without undesirable object blurring on the image by allowing the use of both thicker scintillators with higher efficiencies and closer source-to-detector distances to maximize incident radiation flux. The technique is applicable in radiographic applications including fast neutron, high-energy gamma and x-ray radiography using thick scintillators.

主权项

1. A digital radiographic imaging system, comprising:
an uncollimated radiation source adapted to produce a substantially conical profile radiation beam directed at an object/space to be imaged; a radiation converter adapted to be positioned adjacent the object/space opposite the radiation source and to convert incident radiation into a different form of radiation representing an observed image of the object/space; a digital image recorder operably connected to the radiation converter to digitally record the observed image; and a computerized system adapted to reconstruct the observed image as a deblurred image that substantially compensates for the conical profile of the radiation beam, by inverting a predetermined conical blurring model (that is based on the geometries of the radiation source and radiation converter, and represented by the probability transfer matrix) on the digitally recorded observed image using the maximum entropy method (MEM), wherein the conical blurring model is used as the linear operator of interest and a smoothed background image pre-recorded by the digital image recorder is used as a Bayesian prior.