Signal-to-Noise Ratio Improvement in Dynamic Contrast-enhanced CT and MR Imaging with Automated Principal Component Analysis Filtering

Abstract

To develop a new automated filtering technique and to evaluate its ability to compensate for the known low contrast-to-noise ratio (CNR) in dynamic contrast material-enhanced (DCE) magnetic resonance (MR) and computed tomographic (CT) data, without substantial loss of information. Clinical data acquisition for this study was approved by the institutional review board. Principal component analysis (PCA) was combined with the fraction of residual information (FRI) criterion to optimize the balance between noise reduction efficiency and information conservation. The PCA FRI filter was evaluated in 15 DCE MR imaging data sets and 15 DCE CT data sets by two radiologists who performed visual analysis and quantitative assessment of noise reduction after filtering. Visual evaluation revealed a substantial noise reduction while conserving information in 90% of MR imaging cases and 87% of CT cases for image analysis and in 93% of MR imaging cases and 90% of CT cases for signal analysis. Efficient denoising enabled improvement in structure characterization in 60% of MR imaging cases and 77% of CT cases. After filtering, CNR was improved by 2.06 ± 0.89 for MR imaging (P < .01) and by 5.72 ± 4.82 for CT (P < .01). This PCA FRI filter demonstrates noise reduction efficiency and information conservation for both DCE MR data and DCE CT data. FRI analysis enabled automated optimization of the parameters for the PCA filter and provided an optional visual control of residual information losses. The robust and fast PCA FRI filter may improve qualitative or quantitative analysis of DCE imaging in a clinical context. http://radiology.rsna.org/lookup/suppl/doi:10.1148/radiol.10100231/-/DC1.