Adaptive Filtering of Dynamic Contrast Enhanced Magnetic Resonance Images

conference paper

English

Dynamic contrast enhanced T1-weighted magnetic resonance imaging (DCE-MRI) is a powerful tool for cancer diagnosis, monitoring of treatment effects and evaluation of anticancer drugs. To get information about perfusion and microcirculation in a tissue, four-dimensional (3 space coordinates and time) dataset must be processed, i.e. converted to represent concentration of contrast agent and fitted with proper model. Both these steps are sensitive to noise presented in the data. Noise in the DCE-MRI is not negligible because high temporal and spatial samplings are required simultaneously. If signal to noise ratio (SNR) is too high, curve fitting of dilution curves with more sophisticated models is imprecise or not possible, hence noise suppression is important. Usual techniques for noise suppression as averaging in time or spatial domains remove high frequencies, which causes serious changes in the shapes of the dilution curves or blurring respectively. Median filtering is not useful because noise is approximately Gaussian. To improve SNR and preserve high frequencies, an adaptive filtering technique must be used. There is a lot of redundancy in DCE-MRI data because dilution curves for nearby voxels lying in same tissue type (e.g. in muscle, fat, bone) are the same except noise. Using this fact we are proposing to create a mean curve from neighboring curves with the same trend to suppress noise. To compute similarity of the curves, some measure over residuum (i.e. reference curve minus nearby curve) must be selected. Suitable ones are mean, variance, L2-norm of the residuum or any statistical test checking hypothesis that residuum is noise with Gaussian (or other) distribution.

DCE-MRI, denoising, adaptive filtering

BARTOŠ, M.; JIŘÍK, R.; KEUNEN, O.; TAXT, T.

2011

17. 1. 2011

University of Bergen

Bergen

978-82-993786-6-6

Abstract Book

88

88

1