Object-based Rician Noise Estimation for MR Images

Pierrick Coupé (1) , José V. Manjón (2) , Elias Gedamu (3) , Douglas Arnold (1) , Montserrat Robles (2) , Louis D. Collins (1)

1 MNI - McConnell Brain Imaging Centre
2 ITACA
3 Montreal Neurological Institute and Hospital

The MR images noise variance is an important measure used for many applications such as denoising, registration or image quality assessment. The real and the imaginary parts of the MR complex raw data are considered as corrupted by white additive Gaussian noises with the same variance. By taking the magnitude of the complex data, the noise is transformed into Rician noise. This noise is usually described by Rayleigh distribution in the background and approximated by Gaussian noise in the foreground when Signal Noise Ratio (SNR) is high enough. These descriptions of noise distribution has been used in the majority of noise estimation methods. Nevertheless, the Rayleigh model of the background can fail when ghosting artefact are presents (i.e. signal different to zeros), and the Gaussian approximation of foreground is no longer valid for low SNR images. To overcome these limitations an object-based estimation taking into account the Rician nature of the noise.

Domaines: Imagerie médicale; Ingénierie biomédicale

Pierrick Coupé, José V. Manjón, Elias Gedamu, Douglas Arnold, Montserrat Robles, et al.. Object-based Rician Noise Estimation for MR Images. Organization for Human Brain Mapping 2009 Annual Meeting, Jul 2009, United States. pp.s81, ⟨10.1016/S1053-8119(09)70571-3⟩. ⟨hal-00645453⟩