Pierre Croisille

Automatic registration of MR first-pass myocardial perfusion images

Magnetic resonance perfusion imaging has become a technique of choice in the evaluation of patients with suspected coronary artery disease (CAD). In order to improve the quantification of perfusion parameters (such as signal intensity amplitude and upslope), an automatic registration technique is proposed. The results are compared to manually registered perfusion sequences. Perfusion maps computed from original and registered data sets are also compared. Automatic registration can be efficiently used as a post-processing technique to improve further qualitative and quantitative evaluation strategies.

A strategy to quantitatively evaluate MRI/PET cardiac rigid registration methods using a Monte Carlo simulator

The goal of this work is to present a strategy to validate cardiac MRI/PET registration methods. The strategy relies on a MRI/PET image reference data set including a computer generated PET data set of the thorax and its structures. This data set was produced using a Monte Carlo simulator from segmented T1-weighted MRI thorax data. From the reference data set as a gold standard, test transformations are randomly generated and used to quantify registration accuracy. The validation approach has been applied to our own rigid registration method with three different similarity measures: Correlation Ratio, Correlation Coefficient and Mutual Information. In this study, we observed that the Correlation Ratio gave better results both for thorax and heart image registration.

An Adaptive Denoising Method Dedicated to Cardiac MR-DTI

Diffusion tensor magnetic resonance imaging (MRDTI) is noise sensitive, and the noise can seriously affect the subsequent characteristic parameter calculations. This paper proposes an adaptive denoising method based on a sparse representation for cardiac diffusion weighted images in MR-DTI. The method first generates a dictionary from the cardiac diffusion weighted images and then a dictionary training algorithm is applied to adapt the dictionary so that it better fits the features of the observed image. The denoising is achieved by gradually approximating the underlying image using the atoms selected from the generated dictionary. The results on both simulated images and real DT-MRI images from ex-vivo and invivo human hearts show that the proposed denoising method performs well in preserving image fine features and contrast. Keywords-image denoising; cardiac MR-DTI; sparse representation; adaptive dictionary