Session SA1.2

Denoising of Cardiac Diffusion Weighted Magnetic Resonance Images

LJ Bao, YM Zhu, WY Liu*, M Robini,
ZB Pu, I Magnin

Harbin Institute of Technology
Harbin, China

Diffusion tensor magnetic resonance imaging (DT-MRI) attracts much research interest recently for its unique three-dimensional fiber reconstruction ability. It is well known that myocardial fiber orientation is altered in various cardiac diseases and therefore detailed information about 3D fiber structures of myocardium can provide important cues to the diagnosis of heart disease. However cardiac DW (diffusion weighted) imaging is noise sensitive, and the noise can induce numerous systematic errors in subsequent parameter calculations. So improving the signal-to-noise ratio (SNR) is crucial for practical utility of DT-MRI in human hearts, and noise removal techniques constitute the most efficient way without additional acquisitions.
This paper proposes a sparse representation-based method for denoising cardiac DW images. The method first generates a dictionary of multiple bases according to the features of the observed image. A segmentation algorithm based on nonstationary degree detector is then introduced to make the selection of atoms in the dictionary more adaptive to the image’s features. Unlike classical approaches, sparse representation-based denoising (SPDN) is achieved by gradually approximating the underlying image using the atoms selected from the generated dictionary where each atom is corresponding to a desirable image structure.The segmentation employs the concept of nonstationarity degree (NSD) in view of its interesting properties for dealing with noisy data.
The results on both simulated signals and real cardiac DW images show that the proposed denoising method performs significantly better than conventional denoising techniques (compared with partial-differential-equation filter and wavelet algorithm in the experiments) by preserving image contrast and fine structures. It can effectively reduce the noise in cardiac DW images and improve the calculation accuracy of diffusion tensors as well as the principal eigenvector field of the heart. That would allow a more precise and robust fiber tracking of the myocardium.

(Abstract Control Number: 267)