LCC-Net: A Lightweight Cross-Consistency Network for Semisupervised Cardiac MR Image Segmentation

Table 1

Comparative study of the proposed LCC-Net on the ACDC dataset. We randomly selected 2 subjects as the labeled data and the remaining 78 subjects as the unlabeled data. The models are tested on 20 unseen subjects. denotes the perturbations in the feature space, denotes the noise perturbation in the feature space, and denotes the dropout perturbation in the feature space.

Method	Labeled data	DICE (%)				Hausdorff (mm)
		LV	RV	MYO	Mean	LV	RV	MYO	Mean
U-Net (upbound)	80 subjects	93.2	85.8	88.9	89.3	2.2	4.8	2.8	3.3
U-Net (baseline)	2 subjects	76.1	24.7	69.1	56.6	9.2	17.0	11.4	12.5
L-Unet		69.9	36.7	60.2	55.6	10.7	19.2	12.7	14.2
LCC-Net w/o		78.2	52.0	69.6	66.6	7.8	16.6	9.8	11.4
LCC-Net w/o		78.0	54.8	70.7	67.8	5.8	11.3	7.8	8.3
LCC-Net w/o		80.6	53.6	73.0	69.0	5.6	11.9	6.9	8.1
LCC-Net		82.0	58.1	73.0	71.0	4.3	13.9	6.3	8.2