Prenatal fetal face analysis from 3D ultrasound
Author(s)
Clark, Anna
Type
Thesis
Abstract
Background: Objective quantification of fetal face morphology from 3D ultrasound scans could improve in-utero identification of rare conditions with characteristic craniofacial abnormalities.
Aims: To develop and test an automatic algorithm for segmentation of the fetal face from 3D ultrasound volumes, to build a statistical modelling algorithm to describe normal fetal face variability between 24-and 34-weeks’ gestation and characterise structural abnormalities and/or dysmorphic facial features.
Methods: 3D fetal face US volumes were prospectively acquired between 24- 34-weeks’ gestation. An atlas-based semi-automatic segmentation algorithm was developed, tested, and a shape model of the fetal face built. Size normalisation and rescaling was performed using a growth model giving the average size at every gestation. The model’s ability to objectively describe ultrasound normal fetal faces and characterise dysmorphic features was tested in cases of trisomy 21, trisomy 18, facial cleft and skeletal dysplasia and in individual examples of genetic conditions.
Results: 176 fetal face 3D ultrasound volumes (117 appropriately-grown, 18 growth-restricted, 41 abnormal) were included. The segmentation algorithm successfully segmented the fetal face, reducing the need for manual refinements. The shape model objectively described normal fetal face variability, and the use of shape scores introduced a methodology to extract more detailed information regarding facial morphology in known structural abnormalities and genetic syndromes. Global facial assessment only identified trisomy 18 as different to ultrasound normal cases. However, the use of regional scores evaluating specific facial zones, highlighted distinct abnormalities in different facial areas depending on the underlying diagnosis when compared to the ultrasound normal cohort.
Conclusion: Fetal facial morphology can be objectively characterised from 3D ultrasound analysis and in this study, validates historically described morphological characteristics of different conditions. Although not yet feasible as a real-time diagnostic tool, this work demonstrates the future potential of this methodology to aid ultrasound assessment and guide patient counselling.
Aims: To develop and test an automatic algorithm for segmentation of the fetal face from 3D ultrasound volumes, to build a statistical modelling algorithm to describe normal fetal face variability between 24-and 34-weeks’ gestation and characterise structural abnormalities and/or dysmorphic facial features.
Methods: 3D fetal face US volumes were prospectively acquired between 24- 34-weeks’ gestation. An atlas-based semi-automatic segmentation algorithm was developed, tested, and a shape model of the fetal face built. Size normalisation and rescaling was performed using a growth model giving the average size at every gestation. The model’s ability to objectively describe ultrasound normal fetal faces and characterise dysmorphic features was tested in cases of trisomy 21, trisomy 18, facial cleft and skeletal dysplasia and in individual examples of genetic conditions.
Results: 176 fetal face 3D ultrasound volumes (117 appropriately-grown, 18 growth-restricted, 41 abnormal) were included. The segmentation algorithm successfully segmented the fetal face, reducing the need for manual refinements. The shape model objectively described normal fetal face variability, and the use of shape scores introduced a methodology to extract more detailed information regarding facial morphology in known structural abnormalities and genetic syndromes. Global facial assessment only identified trisomy 18 as different to ultrasound normal cases. However, the use of regional scores evaluating specific facial zones, highlighted distinct abnormalities in different facial areas depending on the underlying diagnosis when compared to the ultrasound normal cohort.
Conclusion: Fetal facial morphology can be objectively characterised from 3D ultrasound analysis and in this study, validates historically described morphological characteristics of different conditions. Although not yet feasible as a real-time diagnostic tool, this work demonstrates the future potential of this methodology to aid ultrasound assessment and guide patient counselling.
Version
Open Access
Date Issued
2024-05-14
Date Awarded
2025-06-01
Copyright Statement
Attribution-NonCommercial 4.0 International Licence (CC BY-NC)
License URL
Advisor
Lees, Christoph
Publisher Department
Department of Metabolism, Digestion and Reproduction
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)