Table 3.

Segmentation accuracy on the 2017 MICCIA ACDC dataset. Segmentation accuracy is reported as Dice coefficient in the ACDC challenge, but as IoU elsewhere in this work; therefore, both are reported here. (Note that Dice = 2 × IoU/(1 + IoU)). Results are reported for the Network B variant of the Ω-Net; for the results by Isensee et al. (2017) published in STACOM; and for the same group’s unpublished arxiv.org revision Isensee et al. (2018). Boldface formatting indicates the best performing model for each foreground class. MICCAI: Medical Image Computing and Computer Assisted Intervention Society; ACDC: Automated Cardiac Diagnosis Challenge.

Structure	LV Bloodpool	RV Bloodpool	LV Myocardium
Jaccard index (IoU)
Ω-Net	0.912	0.852	0.803
Isensee et al. (2018)	0.896	0.832	0.826
Isensee et al. (2017)	0.869	0.784	0.775
Dice coefficient
Ω-Net	0.954	0.920	0.891
Isensee et al. (2018)	0.945	0.908	0.905
Isensee et al. (2017)	0.930	0.879	0.873