Table 2.

The performance of the proposed FCNN (model III) against two expert observers (references A and B) compared to a number of recently published work.

Method, year	$N_{\mathrm{Pat}}$	$N_{\mathrm{Img}}$	$N_{\text{test}}$	MAD (mm)	DSC (%)	SR (%)	PR (%)	Exec. time (s)
Proposed algorithm (reference A)	92	92	23	$2.3\pm 0.6$	$83\pm 6$	$87\pm 8$	$80\pm 9$	$1.1\pm 0.4$
Proposed algorithm (reference B)	92	92	23	$2.4\pm 0.8$	$83\pm 6$	$90\pm 7$	$77\pm 8$	$1.1\pm 0.4$
Kazemifar et al.,3 2018	85	85	$\sim 25$	—	$88\pm 12$	87	92	$\sim 1$
Shahedi et al.,7 2018	70	70	10	$1.9\pm 0.5$	$88\pm 2$	$94\pm 3$	$82\pm 4$	$22\pm 2$
Ma et al.,6 2017	92	92	92	—	84	—	—	—
Ma et al.,4 2016	15	15	15	—	$85\pm 3$	$83\pm 1$	—	—
Shi et al.10, 2016	24	330	258	$1.3\pm 0.8$	$92\pm 4$	$90\pm 5$	—	—
Shi et al.,9 2015	24	330	258	$1.1\pm 0.6$	$95\pm 3$	$93\pm 4$	—	—
Skalski et al.,13 2015	27	27	15	—	$81\pm 5$	—	—	—
Shao et al.,8 2015	70	70	70	$1.9\pm 0.2$	$88\pm 2$	84	86	—