Advanced Fetal Brain Imaging using MRI-informed Ultrasound
Implementing effective evidence-based maternity care relies upon accurate estimation of gestational age (GA). Ultrasound (US) provides reliable estimates of GA if performed early in pregnancy. However, in low-income settings, the lack of appropriately trained sonologists and the tendency for women to present late in pregnancy are barriers to the use of US for dating purposes.
This project proposes to exploit dynamic structural features within US volumes of the fetal brain taken by non-specialist healthcare workers to estimate GA automatically at a single visit, thereby enabling clinically useful estimates of GA to be made even in the third trimester of pregnancy. This will be achieved with a bespoke tool, Autodate, which will be simple to use, will require minimal user training, and will be designed for use in hard-to-reach clinical settings across the world. Autodate has been developed based upon the project team’s first-hand experience of providing maternity care in rural Kenya, where problems due to a lack of reliable GA data are common.
Although portable US machines are increasingly sold in low-income settings, the lack of suitably trained sonologists, and the tendency for women to present late in pregnancy for care (i.e. after 24 weeks’ gestation) when US estimates of GA are less accurate, are significant barriers to delivering effective, evidence-based care. This solution addresses these two barriers.
Autodate was conceived on the basis that fetal brain development follows a precise spatial and temporal pattern, with folds emerging and disappearing on the surface of the brain (cerebral cortex) at fixed time points during pregnancy. This timing is so precise that post-mortem neuroanatomical and MRI evidence suggest that the ‘developmental maturation’ of the fetal brain is more closely related to GA than traditional estimates of brain size. Therefore, the US appearance of maturational markers in the brain offers the potential for estimating GA. Whilst it is possible in specialist centres to quantify fetal brain development subjectively, doing so on a large scale in routine clinical practice is impractical, especially in low-income settings. An obvious solution is to automate the process of quantifying fetal brain development using images already captured during routine US examinations.
Specifically, the project hypothesises that the precise spatiotemporal sequence of development of the cerebral cortex can be captured using machine-learning methods. By substituting size-based approaches with a range of structural features that are automatically identified and assembled by Autodate, the project aims to generate clinically valuable estimates of GA at 34 weeks’ gestation and beyond. Since the basis of the model depends on features and not size, it is anticipated that the estimates of GA late in pregnancy will be much more accurate than estimates based on other measurements. Results from a proof-of-concept study examining the feasibility of Autodate (n = 158 fetuses dated <14 weeks with US and rescanned at 34 weeks) provided estimates of GA at this late gestation with an accuracy of ±11 days.
In Phase I, Autodate will be developed using an existing dataset of US volumes from 4,321 ‘healthy’ fetuses scanned every 5±1 weeks between 14 and 41 weeks’ gestation to ‘learn’ which regions of the brain discriminate GA most effectively. For validation purposes, a model of ‘normal fetal brain development’ will be generated and applied as a reference for estimating GA in newly acquired brain images. In Phase II, the project aims to assess the clinical performance of Autodate in a cluster randomised controlled trial conducted with existing collaborators in India, Kenya, and South Africa.
This model of ‘normal fetal brain development’ would also enable the identification of novel biomarkers of neurodevelopment as a basis for defining and testing hypotheses that will provide a better understanding of the links between brain structure and function.