Fig. 1. Non-convex regularized spherical deconvolution algorithm

Fig. 2. Fiber orientation distributions of synthetic data obtained by different methods. (a) Real fiber orientation of the synthetic data; (b) synthetic data with four different diffusion weighted gradient directions (noise free); (c) synthetic data with four different diffusion weighted gradient directions (signal-to-noise ratio is 17); (d) result of RL-SD; (e) result of RL-TV; (f) result of the proposed method

Fig. 3. Mean angular errors of different methods

Fig. 4. Fiber orientation distributions of phantom data obtained by different methods.(a) Real fiber structure; (b)-(d) results of RL-SD, RL-TV, and the proposed method on interested region A; (e)-(f) results of RL-TV and the proposed method on interested region B