Table 4.

Previous work on placenta segmentation in ultrasound with specifications about training and testing data, average performance measured by the Dice score and subjects included in the study. (CNN: convolutional neural network; RNN: recurrent neural network; cGAN: conditional generative adversarial network; CV: average performance obtained in a cross-validation strategy.)

Reference	Method	Dice	Training + Validation	Testing	GA	Subjects
Stevenson et al. (2015)	Random walker, semi-automatic	0.87	–	88	First trimester	3D US, singleton
Oguz et al. (2016)	Multi-atlas label fusion	0.83 ± 0.05	–	14	First trimester	3D US, only anterior
Yang et al. (2019)	Multi-object, 3D CNN +RNN	0.64	50 + 10	44	First trimester (10–14 weeks)	3D US, singleton and twin
Looney et al. (2018)	3D CNN	0.81 ± 0.15	1097 + 100	1196	First trimester (11–14 weeks)	3D US, singleton
Oguz et al. (2018)	2D CNN + 3D Multi-atlas label fusion	0.88 ± 0.05 (anterior) 0.85 ± 0.05 (posterior)	384 slices	73	First trimester	3D US, singleton, 28 anterior 19 posterior
Oguz et al. (2020)	Semi-automatic, Multi-atlas label fusion	0.82 ± 0.06	–	73	First trimester (11–14 weeks)	3D US, singleton, 28 anterior 19 posterior
Schwartz et al. (2022)	2D and 3D CNNs	0.88 ± 0.05	99	25	First trimester (11–14 weeks)	3D US, singleton
Looney et al. (2021)	Single- and Multi- object, 3D CNN	0.85 ± 0.05	1893 + 150	50	First trimester (11–14 weeks)	3D US, singleton
Hu et al. (2019)	2D CNN + shadow detection layer	0.92 ± 0.04	954 + 205	205	First, second and trimester (8–34 weeks)	2D US, singleton and twin
Torrents-Barrena et al. (2019a)	3D cGAN	0.75 ± 0.12	61	61 (CV)	Second and third trimester (15–38 weeks)	3D US, singleton and twin
Ours	3D Multi-task CNN	0.87 ± 0.I0 (anterior) 0.80 ± 0.13 (posterior)	1188 (292 with segm.)	292 (CV)	Second and third trimester (19–33 weeks)	3D US, singleton