TABLE III.

Results of experiments on cortical plate segmentation. We compare three different transfer learning approaches with our proposed method of multi-task learning with heterogeneous data. We ran paired t-tests between the four results (three transfer learning trials and multi-task learning trial), separately for each of the three datasets. Statistically better results (p = 0.01), were marked with bold type.

Training/fine-tuning data	Test data	DSC	HD95 (mm)	ASSD (mm)	ECE	MCE
Train on CP- younger fetus	CP- younger fetus	0.90 ± 0.03	0.82 ± 0.03	0.21 ± 0.03	0.07 ± 0.02	0.20 ± 0.04
↳ Fine-tune on CP- older fetus	CP- older fetus	0.80 ± 0.05	1.05 ± 0.22	0.40 ± 0.12	0.13 ± 0.06	0.31 ± 0.07
↳ Fine-tune on CP- newborn	CP- newborn	0.92 ± 0.07	0.80 ± 0.03	0.19 ± 0.02	0.07 ± 0.02	0.19 ± 0.02
Train on CP- older fetus	CP- older fetus	0.82 ± 0.04	1.10 ± 0.21	0.37 ± 0.08	0.10 ± 0.04	0.22 ± 0.10
↳ Fine-tune on CP- younger fetus	CP- younger fetus	0.90 ± 0.04	0.88 ± 0.04	0.22 ± 0.04	0.10 ± 0.03	0.17 ± 0.06
↳ Fine-tune on CP- newborn	CP- newborn	0.92 ± 0.03	0.89 ± 0.04	0.18 ± 0.02	0.07 ± 0.01	0.17 ± 0.03
Train on CP- newborn	CP- newborn	0.92 ± 0.03	0.84 ± 0.01	0.18 ± 0.03	0.06 ± 0.03	0.12 ± 0.04
↳ Fine-tune on CP- younger fetus	CP- younger fetus	0.90 ± 0.03	0.85 ± 0.03	0.22 ± 0.04	0.08 ± 0.02	0.19 ± 0.03
↳ Fine-tune on CP- older fetus	CP- older fetus	0.81 ± 0.05	0.96 ± 0.18	0.34 ± 0.14	0.16 ± 0.05	0.34 ± 0.05
Train a single model for all datasets	CP- younger fetus	0.90 ± 0.02	0.81 ± 0.01	0.18±0.03	0.05±0.03	0.13±0.08
	CP- older fetus	0.85±0.03	0.90±0.16	0.30±0.05	0.04±0.02	0.09±0.05
	CP- newborn	0.92 ± 0.02	0.81 ± 0.02	0.16±0.02	0.03±0.01	0.07 ±0.02