Figure 7.

Ground truth comparisons. (a) Two hyperplane cuts of the multidimensional objective function ${\mathcal{L}}_j\left({\lambda }_t,{\lambda }_{t^{\prime }}\right)$, with respect to λ= (t_h, s_min), with s_min held constant and two values of the arguments, $t_h^{\prime }=t_h+1$ (red curve) and $t_h^{\prime }=t_h+2$ (green curve); values of the constants in ${\mathcal{L}}_j$ were α = 0.001 and β = 1.0. (b) Various iterations of the optimization algorithm, showing the values of the area difference between ground truth and the calculated borders. The area difference is obtained by summing the difference of areas for all image slices. The plot shows that despite the initial λ, the algorithm converges towards a constant non-zero area difference. (c) Visual inspection of borders obtained with an optimal solution λ* compared with the ground truth, superposed on the corresponding germinal center image.