Shape model of the fetal face between 24 and 34 gestational weeks from segmented 3D US

This fetal face dataset is constituted of:

- 184 3D surfaces of segmented fetal faces from 3D ultrasound acquired between 24 and 34 gestational weeks. Meshes have been scaled to adjust for gestational age (GA): data_age_rescaled.zip

- a spreadsheet reporting individual GA and diagnosed facial abnormality categories: Patient DB 2023.csv

- a shape model, or more precisely the parameters of the Deformetrica's deterministic atlas estimated for the above data: shape model.zip

Complete description of the data and methodology is presented in [1] and [2]. The main objective of this study is to study the normal changes and the most common abnormality of the fetal face morphology at this stage to assist perinatal clinical follow-up.

We summarise here the main information.

Participants were prospectively recruited from two clinical centres. Exclusion criteria were based on the absence of suitable image or follow-up data. Gestational age and a short categorisation of the diagnosed pathological facial abnormalities are listed in the attached spreadsheet, it includes, in particular: fetal growth restriction, facial cleft, chromosomal abnormality (T18, T21), etc.

Segmentation results were obtained by manual refinement of semi-automatic atlas-based approach. Meshes were then exported, scaled by GA, and rigidly aligned. Surfaces are available as vtk PolyData files in `data_age_rescaled.zip`.

Scaling, at gestational age t (in days), is given by the following growth model, estimated by least-square regression on our data and standardized to scale 1 at 29 weeks (203 days) of GA:

1/s(t) = 1 + 4.0077e-3 (t-203) - 3.2520e-5 (t-203)**2.

This model has been validated against standard available references and is useful to reduce the morphological variability to ease the shape analysis. It can easily be reverted using the provided GA in the spreadsheet.

Shape model files (`shape model.zip`) are respectively: the template (average) shape, the control points and momenta parametrizing the individual deformations from this template, the hyper-parameter of the model including the deformation kernel width kwd=7.0mm. The Deformetrica's framework is openly available there https://gitlab.com/icm-institute/aramislab/deformetrica.

Ethics

The study has been approved by local research ethic committees in both centres (see [1,2]) and data acquired following patient informed consent.

This dataset only contains non-identifiable data. In particular, the images themselves, that potentially contain extra information (compared to the final reconstruction) that may lead to the identification of the participants, are not included.

References

1. Sivera, R., Clark, A.E., Dall’Asta, A. et al. Fetal face shape analysis from prenatal 3D ultrasound images. Sci Rep 14, 4411 (2024). https://doi.org/10.1038/s41598-023-50386-9

2. Clark, A.E. Prenatal facial and brain analysis from 3D ultrasound. PhD thesis, Imperial College London (2024).