Noise reduction for low-dose single-slice helical CT sinograms

Abstract

Helical computed tomography (HCT) has several advantages over conventional step-and-shoot CT for imaging a relatively large object, especially for dynamic studies. However, HCT may increase X-ray exposure significantly. This work aims to reduce the radiation by lowering X-ray tube current (mA) and filtering low-mA (or dose) sinogram noise of HCT. The noise reduction method is based on three observations on HCT: (1) the axial sampling of HCT projections is nearly continuous as detection system rotates; (2) the noise distribution in sinogram space is nearly a Gaussian after system calibration (including logarithmic transform); and (3) the relationship between the calibrated data mean and variance can be expressed as an exponential functional across the field-of-view. Based on the second and third observations, a penalized weighted least-squares (PWLS) solution is an optimal choice, where the weight is given by the mean-variance relationship. The first observation encourages the use of Karhunen-Loève (KL) transform along the axial direction because of the associated correlation. In the KL domain, the eigenvalue of each principal component and the derived data variance provide the signal-to-noise ratio (SNR) information, resulting in a SNR-adaptive noise reduction. The KL-PWLS noise-reduction method was implemented analytically for efficient restoration of large volume HCT sinograms. Simulation studies showed a noticeable improvement, in terms of image quality and defect detectability, of the proposed noise-reduction method over the Ordered-Subsets Expectation-Maximization reconstruction and the conventional low-pass noise filtering with optimal cutoff frequency and/or other filter parameters.