Table 1.

Results on the simulated in-house dataset. The first 3 rows are our proposed methods trained with different losses. The following 4 rows each show the effect of changing one component of the baseline (Ours (${\mathcal{L}}_{\mathit{Focal}}$)). The last two rows are the results obtained with FLEXCONN and MIMoSA based on the same dataset.

	DSC	PPV	TPR	LFPR	LTPR	VD
Ours (${\mathcal{L}}_{L_2}$)	0.861	0.919	0.810	0.104	0.603	0.119
Ours (${\mathcal{L}}_{\mathit{Dice}}$)	0.847	0.904	0.798	0.150	0.597	0.118
Ours (${\mathcal{L}}_{\mathit{Focal}}$)	0.865	0.850	0.880	0.209	0.636	0.045
Only 2.5D	0.859	0.866	0.853	0.278	0.620	0.028
Stacked 2D	0.828	0.801	0.858	0.584	0.640	0.088	Ablation Study
Smaller Patch	0.858	0.850	0.868	0.236	0.644	0.040
U-Net	0.835	0.803	0.871	0.597	0.694	0.087
FLEXCONN [12]	0.707	0.624	0.832	0.667	0.546	0.393
MIMoSA [15]	0.424	0.530	0.370	0.851	0.544	0.316