Table III.

Performance of all submissions for the LVQuan 2018 challenge on the training dataset with 145 subjects under the five-fold cross validation (CV) protocols. For each task, only the average performance is shown here and the best result is highlighted in boldface. Average MAE is shown for areas, dimensions and RWTs, and error rate is shown for cardiac phase. All the methods achieved performance better or close to the state-of-the-art DMTRL.

Methods	Area (mm2)	Dimension (mm)	RWT (mm)	Phase (%)
SG-based methods
ResUNet	62.3	0.79	0.68	6.72
ESUPNet	62	1.14	0.96	8
UNetMF⋆	141.7	1.77	1.39	-
MMED-S	120	1.25	1.03	7.8
SegNetRFf†	-	-	-	10
CNTCVX	176	2.23	1.75	10.3
DR-based methods
LDAMT	156	2.03	1.38	8.1
MMED-R	158	2.08	1.51	9.4
HQNet*	197	2.57	1.51	9.8
CNN3DST	190	2.29	1.42	3.85
FNN2D3D	188	2.42	1.42	8.76
Combined methods
MMED-SR	142	1.94	1.42	8.6
EnCNNU‡	124	2.27	1.62	13.7
DenseUMT	173	2.44	1.37	7.8
DMTRL	180	2.51	1.39	8.2

^⋆45 subjects in the training set were used for test;

^†only Pearson correlation coefficients were report on the training set;

^*7-fold CV was used;

^‡3-fold CV was used. For MMED, the results for all three modes were reported.