Semisupervised Medical Image Segmentation through Prototype-Based Mutual Consistency Learning

Table 3

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

Method	Labeled/unlabeled	DSC (%)	JI (%)	SE (%)	AC (%)	95HD (mm)	ASD (mm)
Fully supervised	304/0	81.05	74.16	80.41	98.78	20.57	5.10
Supervised-only	30/0	36.92	28.26	38.31	94.68	150.09	76.87
MT [5]	30/274	42.56	33.74	43.85	93.99	156.51	77.51
DAN [10]	30/274	44.04	35.11	44.23	95.48	133.45	69.00
EM [49]	30/274	43.39	34.66	42.42	95.60	124.81	62.38
UAMT [8]	30/274	43.45	34.84	42.57	95.43	156.10	82.26
ICT [50]	30/274	44.77	36.70	45.63	94.76	135.35	69.11
PMCL (ours)	30/274	51.14	42.81	50.29	95.82	103.89	42.62
Supervised-only	60/244	62.09	53.85	58.55	97.39	97.72	44.42
MT [5]	60/244	65.65	56.87	67.79	97.42	99.50	48.61
DAN [10]	60/244	66.14	58.02	66.65	97.66	91.36	45.87
EM [49]	60/244	65.29	56.74	67.19	97.37	112.44	45.24
UAMT [8]	60/244	64.82	55.81	64.27	97.77	115.44	46.78
ICT [50]	60/244	65.96	57.74	65.66	97.71	93.75	47.81
PMCL (ours)	60/244	67.42	59.97	68.80	97.91	79.39	38.87

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