Algorithm 1.

Weighted Combinations in one voxel

1:	Input: K multi-fascicle models with weights wk and the number N of fascicles in the output.
2:	Output: A multi-fascicle model: ${\sum }_{j=1}^N{f}_j^s{e}^{-b{\mathit{g}}^T{\mathit{D}}_j^s\mathit{g}}$
3:	for k in 1 to K do ▷ Construct the complete model
4:	for j in 1 to N do
5:	i ← (k − 1)N + j
6:	$f_i^c\leftarrow w_k{f}_j^k$
7:	${\mathit{D}}_i^c\leftarrow {\mathit{D}}_j^k$
8:	end for
9:	end for
10:	π ← Initialization ( ${\{D_i^c,f_i^c\}}_{j\le NK}$) ▷ Initialize clustering
11:	while π has not converged do
12:	for j in 1 to N do ▷ M-Step
13:	$log{\mathit{D}}_j^s\leftarrow \left({\sum }_{i:{\pi }_i=j}{f}_i^{c}log{D}_i^c\right)/\left({\sum }_{i:{\pi }_i=j}{f}_i^c\right)$
14:	$f_j^s\leftarrow {\sum }_{i:{\pi }_i=j}{f}_i^c$
15:	end for
16:	for i in 1 to KN do ▷ E-Step
17:	for j in 1 to N do
18:	$B_i(j)\leftarrow \text{Tr}\left({\mathit{D}}_i^{c-1}{\mathit{D}}_j^s\right)-log\left|{\mathit{D}}_i^{c-1}{\mathit{D}}_j^s\right|$
19:	end for
20:	πi ← arg minl Bi(l)
21:	end for
22:	end while