Table 5.

Evaluation of the temporal loss regularization using different weighing factors $w_{n-1}$, $w_{n-2}$. Results are shown for the architecture Five-Path-4D with respect to predicting the motion $\Delta s$ between the template and the last volume of a sequence. Errors are given in mm

$w_{n-1}$	$w_{n-2}$	MAE $\Delta s_x$	MAE $\Delta s_x$	MAE $\Delta s_x$	rMAE	aCC (%)
0	0	$0.16\pm 0.18$	$0.13\pm 0.15$	$0.10\pm 0.09$	$0.13\pm 0.14$	98.58
1	0	$0.15\pm 0.22$	$0.12\pm 0.13$	$0.11\pm 0.10$	$0.13\pm 0.15$	98.15
0.75	0	$0.14\pm 0.13$	$0.11\pm 0.10$	$0.13\pm 0.10$	$0.14\pm 0.11$	98.90
0.5	0	$0.10\pm 0.09$	$0.14\pm 0.11$	$0.10\pm 0.08$	$0.12\pm 0.10$	99.02
0.25	0	$0.11\pm 0.11$	$0.14\pm 0.13$	$0.11\pm 0.09$	$0.12\pm 0.11$	98.92
1	1	$0.11\pm 0.10$	$0.19\pm 0.17$	$0.10\pm 0.09$	$0.14\pm 0.12$	98.71
0.75	0.75	$\mathbf{0.09}\pm \mathbf{0.09}$	$\mathbf{0.11}\pm \mathbf{0.10}$	$\mathbf{0.10}\pm \mathbf{0.08}$	$\mathbf{0.10}\pm \mathbf{0.09}$	$\mathbf{99.06}$
0.75	0.5	$0.12\pm 0.10$	$0.10\pm 0.11$	$0.10\pm 0.08$	$0.11\pm 0.10$	99.03

Bold highlights the best performing method/results