Semisupervised Medical Image Segmentation through Prototype-Based Mutual Consistency Learning

Table 2

Quantitative comparison between our method and other semisupervised methods on CVC-ClinicDB under 10% and 20% labeled data.

Method	Labeled/unlabeled	DSC (%)	JI (%)	SE (%)	AC (%)	95HD (mm)	ASD (mm)
Fully supervised	490/0	84.16	76.00	85.23	96.83	29.76	8.84
Supervised-only	49/0	61.01	50.72	66.96	92.68	80.68	28.56
MT [5]	49/441	62.36	52.38	66.97	93.08	77.20	26.43
DAN [10]	49/441	64.11	53.49	69.05	93.01	73.83	23.84
EM [49]	49/441	62.44	51.47	67.92	92.70	81.96	28.08
UAMT [8]	49/441	64.10	52.83	70.52	93.05	72.67	26.07
ICT [50]	49/441	63.35	53.33	68.38	92.79	68.38	23.27
PMCL (ours)	49/441	69.00	58.50	74.86	93.55	66.68	23.08
Supervised-only	98/392	72.33	61.44	75.65	94.48	60.12	19.23
MT [5]	98/392	73.04	63.33	76.74	94.59	48.86	13.11
DAN [10]	98/392	74.10	63.59	76.98	94.87	55.19	17.55
EM [49]	98/392	73.94	63.98	75.93	94.77	55.52	14.83
UAMT [8]	98/392	72.67	62.52	78.84	94.59	53.36	15.16
ICT [50]	98/392	74.85	64.87	76.84	94.91	50.35	14.53
PMCL (ours)	98/392	75.64	65.61	81.09	94.76	48.15	14.58

The bold values suggest the best performance compared to other state-of-the-art methods.