Fetal face shape analysis from prenatal 3D ultrasound images

Abstract

3D ultrasound imaging of fetal faces has been predominantly confined to qualitative assessment. Many genetic conditions evade diagnosis and identification could assist with parental counselling, pregnancy management and neonatal care planning. We describe a methodology to build a shape model of the third trimester fetal face from 3D ultrasound and show how it can objectively describe morphological features and gestational-age related changes of normal fetal faces. 135 fetal face 3D ultrasound volumes (117 appropriately grown, 18 growth-restricted) of 24-34 weeks gestation were included. A 3D surface model of each face was obtained using a semi-automatic segmentation workflow. Size normalisation and rescaling was performed using a growth model giving the average size at every gestation. The model demonstrated a similar growth rate to standard head circumference reference charts. A landmark-free morphometry model was estimated to characterize shape differences using non-linear deformations of an idealized template face. Advancing gestation is associated with widening/fullness of the cheeks, contraction of the chin and deepening of the eyes. Fetal growth restriction is associated with a smaller average facial size but no morphological differences. This model may eventually be used as a reference to assist in the prenatal diagnosis of congenital anomalies with characteristic facial dysmorphisms.

Figure 1

Size measurements of the fetal face. (a) face and head sizes with respect to age before rescaling: our regression model (black), reference model for head circumference from Kiserud et al. (red). The quadratic growth curves are estimated using quantile regression (medians are represented by dashed lines, 5th and 95th percentiles by dotted lines). (b) face size measurements after rescaling. Regression models are shown for AGA cases (blue), FGR cases (orange) and both combined (black).

Figure 2

(a, b) Mean template of the fetal face, the template is warped by subject-specific deformations to match each individual morphology. (c) Average reconstruction error at each point, measured as the distance between the reconstructions and the original models. Small errors are yellow/orange and larger errors purple. d) Shape variability measured using the standard deviation of the mesh node positions over all of the reconstruction. Purple denotes the areas with the highest variability and yellow/orange the most stable areas.

Figure 3

PCA projection of the first two modes of the SSM model. The first mode is represented horizontally, the second vertically. Each grey dot represents one subject. The faces illustrate the model at −/+ 2 standard deviations.

Figure 4

The effect of age on fetal facial morphology. (a, b) Model of the average morphology at (a) 24 weeks and (b) 34 weeks of gestational age. (c) Illustration of the local z-values assessing the effect of gestational age on morphology. The colour bar is normalised to show levels corresponding to p=0.5 (purple), p=0.1 (orange) and p=0.05 (yellow). (d) divergence of the deformation vector field (momenta).

Figure 5

Differences associated with FGR: (a) average control face, (b) average FGR face (normal diagnosis only), (c) group comparison z-value assessing for the effect of FGR on the morphology, the colour-bar ticks are normalized to show values corresponding to p=0.5 (purple), p=0.1 (orange), and p=0.05 (light yellow) for the t² statistic. d) divergence of the deformation vector field (momenta).

Figure 6

Two cases with normal diagnosis: case A top row and B (the most “abnormal normal”) bottom row. From left to right: original mesh, model reconstruction, deformation z-score, orthogonal z-score.

Figure 7

Two 3D US volumes collected from the same participant during the same appointment. The volume on the right was selected as the facial border were clearer and there was no fetal limb in contact with the fetal face as seen in the left image.

Figure 8

Flowchart of recruitment, 3D US volume acquisition, segmentation quality and final numbers for analysis.

Figure 9

Overview of the atlas-based segmentation algorithm steps: (a, b) 3D US volume (c) 4 manual landmarks (eyes, nose, midpoint of the lips) (e) monogenic signal map (e) segmentation mask f) 3D surface mesh.

Figure 10

Schematic representation of the statistical shape model: template shape to individual cases/deformations.