Table 5.

Detailed information on the growth model, quality enhancement and visualization studies.

Topic	Author	Year	Type of method	Brain structure	Gestational age	Description of method	Learning strategy (if applicable)	US machine, US Probe, 2D/3D	Number of subjects	Outcome
Abnormality detection	Zhou et al.	2021	DL	Brain	17–32	A CNN for classification was combined with the Java fuzzy cognitive maps algorithm to filter the found features before final classification.	Data-augmentation: none Annotations: diagnosed based on the pathological results of the fetus Strategy: not mentioned	Voluson E8§, ?, 2D	?	Accuracy: Week 17–19: 0.64
Growth model	Bihoun et al.	2020	N	BPD, HC	16–36	Comparison of the resulting growth curve based on Salomon equation and the Intergrowth 21-st growth curves was performed, for a population from rural Burkina Faso.		FFsonic UF-4100†, 3.5–5.0 MHz TAb, 2D	276	Error = - 0.01 mm for HC
Growth model	Burgos-Artizzu et al.	2021	DL	TT	16–42	A CNN, pre-trained to detect key brain structures, was trained to estimate the gestational age from the brain image. Within the architecture of the CNN, regular convolutions were replaced by a series of slightly altered coordinated convolution layers, which incorporated image resolution into the computation.	Data-augmentation: none Annotations: GA was determined by CRL measurements on first-trimester ultrasound Strategy: no cross-validation, external test set	Voluson E6, S8 and S10§, and Aloka¶, 3–7 MHz Tab, 2D	598	Error: 14.2 days
Growth model	Namburete et al.	2014	ML	Silvian fissure	18–27	A Regression Forest was trained on image features extracted from the Silvian fissure to predict the GA of the given image.	Data-augmentation: none Annotations: combination of first day last menstural period (LMP) and first trimester US measurements Strategy: 12-fold cross-validation, external test set	HD9‡, 2–5 Mhz, 3D	32	Error: left hemisphere = 6.11 days right hemisphere = 6.66 days
Growth model	Wyburd et al.	2021	DL	Sylvian fissure, parieto-occipital fissure, calcarine sulcus	19–30	The 3D VGG-Net and 3D ResNet architectures were compared to predict the GA from the different structures. Furthermore, attention maps for GA prediction were studied for the different structures.	Data-augmentation: none Annotations: combination of first day last menstural period (LMP) and first trimester US measurements Strategy: 12-fold cross-validation, external test set	?, ?, 3D	811	Error: Sylvian fissure: 3.4 days Parieto-occipital fissure: 4.9 days Calcarine sulcus: 5.0 days
Quality enhancement	Perez-Gonzalez et al.	2020	DL	Brain	14–27	Several partially occluded ultrasound images of the same object were merged using a pipeline of CNNs. Two CNNs were used to segment the fetal skull, one was used to register the fetal brain to a common reference space, and the final CNN was used to merge different acquisitions together by learning how to weigh their influence on the resulting image.	Data-augmentation: none Annotations: expert obstetrician Strategy: cross-validation, no external test set	?, 8–20 MHz, 3D	18	Increase image sharpness: 34.9%
Visualization	Pooh et al.	2016	N	Brain	8–31	HDlive§ software was used to visualize the cerebral vascular structure.		Voluson E10§, 6–12 MHz TVa, 3D
Visualization	Tutschek et al.	2009	N	Atrium, LV, corpus callosum, CER, cerebellar vermis, CM, CP, CSP, falx cerebri, frontal horns, interhemispheric fissure, occipital horns, thalami, temporal horns	Late first trimester to mid-trimester	4Dview§ software.		Voluson 730 or E8§, ?, 3D	22
Visualization	Tutschek et al.	2009	N	Atrium, LV, corpus callosum, CER, cerebellar vermis, CM, CP, CSP, falx cerebri, frontal horns, interhemispheric fissure, occipital horns, thalami, temporal horns	Late first trimester to mid-trimester	4Dview§ software.		Voluson 730 or E8§, ?, 3D	22

Legend to brain structures: BPD = biparietal diameter, CER = cerebellum, CM = cistera magna, CP = choroid plexus, CSP = cavum septi pelllucidi, LV = lateral ventricles, TT = trans-thalamic plane. Legend to description of method: CNN = convolutional neural network, VGG-net, ResNet = widely used network architectures; a brief explanation can be found in Supplementary Material 4. 2D = two dimensional, 3D = three-dimensional, a ∗ indicates longitudinal data. ^†Fukunda Denshi, ^‡Philips, Bothell, WA 98021, USA, ^§GE Medical Systems, Zipf, Austria, ^¶Aloka Co, Ltd, Tokyo, Japan.