Model-based Iterative Reconstruction for Flat-Panel Cone-Beam CT with Focal Spot Blur, Detector Blur, and Correlated Noise

TL;DR

This paper introduces a model-based iterative reconstruction method for flat-panel cone-beam CT that explicitly accounts for system blurs and noise correlations, resulting in improved image resolution and noise tradeoff.

Contribution

It develops a comprehensive forward model including focal spot and detector blur, and a reconstruction framework that incorporates noise correlation for enhanced image quality.

Findings

Increased resolution by 42% with correlated noise modeling.

Superior noise-resolution tradeoff compared to traditional methods.

Performance validated on simulated and test-bench data.

Abstract

While model-based reconstruction methods have been successfully applied to flat-panel cone-beam CT (FP-CBCT) systems, typical implementations ignore both spatial correlations in the projection data as well as system blurs due to the detector and focal spot in the x-ray source. In this work, we develop a forward model for flat-panel-based systems that includes blur and noise correlation associated with finite focal spot size and an indirect detector (e.g., scintillator). This forward model is used to develop a staged reconstruction framework where projection data are deconvolved and log-transformed, followed by a generalized least-squares reconstruction that utilizes a non-diagonal statistical weighting to account for the correlation that arises from the acquisition and data processing chain. We investigate the performance of this novel reconstruction approach in both simulated data and in CBCT test-bench data. In comparison to traditional filtered backprojection and model-based methods that ignore noise correlation, the proposed approach yields a superior noise-resolution tradeoff. For example, for a system with 0.34 mm FWHM scintillator blur and 0.70 FWHM focal spot blur, using the a correlated noise model instead of an uncorrelated noise model increased resolution by 42% (with variance matched at 6.9 x 10^-8 mm^-2 ). While this advantage holds across a wide range of systems with differing blur characteristics, the improvements are greatest for systems where source blur is larger than detector blur.