1st Supervisor: Dr Maria Deprez, King’s College London
2nd Supervisor: Dr Jonathan O’Muircheartaigh, King’s College London
Clinical Supervisor: Prof Grainne McAlonan, King’s College London
Aims of the Project:
- This project will investigate infant brain development using deep-learning image analysis tools applied to multimodal MRI.
- Using conditional deep generative models and new deep-learning registration techniques, we will de-couple brain growth (volume) from tissue maturation (image intensity).
- We will investigate how infant brain development is impacted by factors which raise the likelihood of neurodevelopmental difficulties.
Lay Summary:
Brain development during the first year of life is characterised by rapid and regionally asynchronous maturation of white matter. The result is a flip in T1 & T2 image contrast, most visible in the first 12 months, reflecting the progressive myelination of brain white matter against the steady values seen in grey matter [1]. Measurements of brain growth and maturation are important for characterising differences in developmental trajectories in various conditions that have a higher likelihood of having neurodevelopmental difficulties later in life, such as infants with a family history of autism [2] or early onset serious organ diseases. Most brain imaging analysis tools focus on the childhood or adult brain. In these periods, brain maturation is slow and individual brains have very similar image characteristics. In infants however, two MRI scans separated by as little as a month can be dramatically different (see Figure 1). Existing image analysis tools simply cannot cope with the age-dependent local tissue intensity changes resulting from this rapid maturation.
In this project we propose to build deep learning tools for accurate measurements of brain growth and maturation during the first year of life. We will build on Brain Imaging in Babies project that offers a database of high quality multi-modal MRI in infants with a family history of neurodevelopmental conditions such as autism or attention deficit hyperactivity disorder. In addition, we will investigate a unique cohort of infants born with serious organ disease. In both cases, these infants have a higher likelihood of having neurodevelopmental difficulties. In these datasets, we will use artificial intelligence to
- Design a deep conditional generative model [3-5] to predict multi-modal MRI contrasts at different time-points for each individual baby
- Combine the deep conditional generative model with deep learning image registration [6] for accurate estimation of local brain growth and tissue maturation
- Develop spatio-temporal brain growth and maturation models [7,8] and/or interpretable deep learning [9] for detailed assessment of brain development in ASD and liver disease during the first year of life.
The project would suit a candidate from a quantitative discipline, interested in image analysis, machine learning, deep learning as well as early brain development.
[1] Dubois, J.: The early development of brain white matter: review of imaging studies in fetuses, newborns and infants. Neuroscience 276 (2014) 48–71
[2] Hazlett, H.C.: Early brain development in infants at high risk for autism spectrum disorder. Nature 542(7641), 2017.
[3] Krebs, J.: Learning a Probabilistic Model for Diffeomorphic Registration. IEEE Transactions on Medical Imaging 38(9), 2019.
[4] Park, T.: Semantic Image Synthesis with Spatially-Adaptive Normalization. IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2019.
[5] Lu, Y.: Attribute-Guided Face Generation Using Conditional CycleGAN. ECCV 2018.
[6] Grigorescu, I.: Attention-Driven Multi-channel Deformable Registration of Structural and Microstructural Neonatal Data. Perinatal, Preterm and Paediatric Image Analysis. PIPPI 2022.
[7] Uus A.:Multi-Channel 4D Parametrized Atlas of Macro- and Microstructural Neonatal Brain Development. Frontiers in Neuroscience 15, 2021.
[8] O’Muircheartaigh, J: Modelling brain development to detect white matter injury in term and preterm born neonates. Brain : a journal of neurology 143(2), 2020.
[9] Smilkov D.: SmoothGrad: removing noise by adding noise. arXiv:1706.03825, 2017.V