Vronique Brion1, Olivier Riff1, Irina Kezele1, Maxime Descoteaux2, Denis Le Bihan1, Jean-Franois Mangin1, Cyril Poupon1, Fabrice Poupon1
We adressed the problem of the correction of the Rician noise, corrupting diffusion-weighted images at high b-values, in real-time. We combined a Linear Minimum Mean Square Error Estimator (LMMSE) together with a Kalman framework in order to compute in real-time the noise-free diffusion data, as well as the diffusion maps stemming from any local high angular resolution diffusion (HARDI) or hybrid diffusion (HYDI) model. A feedback is retropropagated from the Kalman filter to the LMMSE, in order both to reinforce the influence of the local structure onto the noise correction, and to prevent smoothing effects. The technique vas validated on synthetic and real data acquired at low signal to noise ratio (SNR) to assess its efficiency and the full pipeline was tested on the computation of orientation distribution functions.