Algorithm 1.

An iterative algorithm to solve the optimization problem in Equation (8).

Input:	Baseline MRI training data of S subjects and F dimensional feature: X ∈ RS×F T time points clinical scores of S subjects and C dimensional clincial score vector: 𝕐 = {Y(t) ∈ RS×C, t = 1, …, T} Parameters: regularization paramters and iteration times
Output:	Weight projection matrix: 𝕎 = {W(t) ∈ RF×C, t = 1, …, T}
	Set iteration r = 0 and initialize W(t) ∈ RF×C according to the linear model for each time point;
	Initialization: ${\hat{\text{W}}}^{(0)}=\left[{\text{W}}^{(1)},{\text{W}}^{(2)},\dots ,{\text{W}}^{(t)},\dots ,{\text{W}}^{(T)}\right]$
	Repeat
	for t = 1 to T
	Calculate Lf, Ls, $L_c^{(t)}$ and LD, according to the above definitions;
	Update ${\hat{\text{W}}}_r^{(t)}$ by solving the Sylvester problem in equation (13);
	End for
	${\hat{\text{W}}}_{r+1}^{(t)}=\left[{\text{W}}^{(1)},{\text{W}}^{(2)},\dots ,{\text{W}}^{(t)},\dots ,{\text{W}}^{(T)}\right]$;
	r = r + 1;
	until (r = 50 or $\hat{\text{W}}-{\text{W}}_{\mathbf{\text{2}}}<{10}^{–\mathbf{6}}$)
Return	W(t), (1 ≤ t ≤ T)