Prenatal face and brain 3D ultrasound
Research type
Research Study
Full title
Prenatal facial and brain analysis from 3D ultrasound
IRAS ID
244945
Contact name
Christoph Lees
Contact email
Sponsor organisation
Imperial College London
Duration of Study in the UK
2 years, 0 months, 31 days
Research summary
Research Summary
The diagnosis of fetal facial abnormalities is challenging even for experienced investigators. Despite advances in prenatal tests, many genetic syndromes or chromosomal abnormalities are still diagnosed only postnatally, when the diagnosis is more straightforward. Recent reports have suggested that antenatal three-dimensional ultrasound (US) can contribute in the identification of abnormal features of the fetal face and lead to in utero diagnosis of abnormalities by targeting genetic tests.
Structural abnormalities of the fetal brain may also carry an adverse prognosis as they can be associated with abnormalities of the brain development or genetic syndromes and chromosomal abnormalities which may impact on the cognitive and motoric functions regardless of the underlying cause. Automated detection of disease and classification of normal development could potentially assist Fetal Medicine Specialists in more accurate diagnosis of challenging conditions by highlighting abnormal image features.
Our team has recently described the feasibility of the analysis of the facial shape using prenatal ultrasound-acquired 3D volumes and an automated algorithm called "statistical shape modelling". Similarly, algorithms have evolved to enable fully automatic classification of the brain with the development of the so-called "deep learning" method, which has been successfully applied to medical image analysis.
In this project we aim to: (1) improve the statistical modelling technique for the antenatal analysis of 3D ultrasound volumes obtained between 24 and 34 weeks, (2) prospectively evaluate 3D prenatal ultrasound of normal cases and those with facial dysmorphism to establish whether it is possible to prenatally diagnose genetic syndromes through computed analysis of fetal facial features and (3) test the performance of the "deep learning" method for the distinction between normal and abnormal brain development.Summary of Results
In the UK, 2–3% of babies are born with abnormalities, many of which are linked to underlying genetic conditions. Identifying the genetic cause during pregnancy can provide crucial information to guide decisions about pregnancy, delivery, and postnatal care planning, as well as offer long-term predictive information for parents. However, making a definitive genetic diagnosis before birth is challenging—currently, only around one-third of fetuses with an underlying genetic condition receive a diagnosis in utero.
Many genetic syndromes are associated with distinctive facial features, which can serve as important clues for diagnosis. However, these features are often subtle and difficult to detect using routine 2D ultrasound imaging during pregnancy. Although 3D ultrasound provides more detailed images of the baby’s face, interpreting these scans relies heavily on the subjective judgment of clinicians, making it less reliable for consistent, definitive diagnosis.
This study, led by Professor Christoph Lees and managed by Dr Anna Clark at Imperial College London in collaboration with Great Ormond Street Hospital, aimed to develop an automatic method for analysing 3D ultrasound scans to objectively describe fetal facial features. The work was supported by the National Institute for Health Research (NIHR) Biomedical Research Centre based at Imperial College Healthcare NHS Trust and Imperial College London (grant number P74419).
Researchers analysed 176 fetal faces between 24- and 34- weeks’ gestation. Of these, 117 were appropriately grown and considered normal, 18 had fetal growth restriction, and 41 had dysmorphic facial features identified on ultrasound. The dysmorphic features included those due to a structural abnormalities like cleft lip and also those due to an underlying genetic or chromosomal condition such as Trisomy 18, Trisomy 21, skeletal dysplasia, and Cornelia de Lange syndrome.
For the first time, researchers successfully implemented a semi-automatic segmentation algorithm to extract the fetal face from 3D ultrasound images—a critical step that converts the 3D ultrasound volumes into a format suitable for computerised analysis. Previously, this process would have taken approximately 12 hours per scan if done manually; the method described in this work reduces this to around 40 minutes of computational time plus 10 minutes of manual landmarking. Although further reductions in processing time are needed for routine clinical use, this represents a significant breakthrough in fetal facial analysis.
To assess facial abnormalities, it was first necessary to establish a reference for normal fetal facial shape in the third trimester. By estimating a growth model for fetuses between 24- and 34-weeks’ gestation, the researchers accounted for age-related size differences, allowing meaningful comparisons between fetuses of different gestational ages. Using a statistical shape model, they were able to objectively describe normal morphological facial features and track how these features change with gestational age—an important foundation for future work in assessing suspected differences.
The researchers then analysed the 41 fetal faces identified as dysmorphic. Using shape analysis at 412 control points across the face, they compared individual cases and groups with known genetic conditions against the normal reference cohort. This analysis not only identified areas where facial morphology differed but also objectively characterised how those features deviated from expected facial development.
The study demonstrates that differences in fetal facial morphology can be identified, described, and characterised using automated methods. While real-time analysis with current ultrasound equipment is not yet feasible, this approach holds promise as an offline diagnostic tool, particularly in cases where facial abnormalities might prompt further testing or invasive procedures.
The technique was not applicable to analysis of the fetal brain.
Although the methodology is not yet ready for clinical application, this proof-of-principle study allowed researchers to understand the differences in facial morphology with different genetic conditions and gestational age changes. With larger studies and advances in processing technology, automated facial analysis using the methodology described in this work could one day alert clinicians and parents to early signs of genetic conditions, guiding counselling, diagnostic testing, and care planning during pregnancy.REC name
West Midlands - Edgbaston Research Ethics Committee
REC reference
18/WM/0370
Date of REC Opinion
2 Jan 2019
REC opinion
Further Information Favourable Opinion