Neonatal brain MRI

Around 5% of all children are born prematurely and 40% of them will suffer neurological, behavioural or learning difficulties (such as hyperactivity) later in life. High-resolution magnetic resonance (MR) imaging enables us to detect the changes in brain development resulting from premature birth. The following MR images clearly show the difference in brain shape and volume of a full-term baby at birth and the baby born at 25 weeks imaged at term-equivalent age.

As a postoctoral research assistant at Imperial College London I worked on development of segmentation tools for neonatal brain MRI. The babies in this study were born prematurely and scanned at different ages equivalent to the last trimester of pregnancy. This task is extremely challenging due to rapid changes in brain shape and appearance of tissues on MRI. The result was development of dynamic 4D atlas of developing brain, which captures changes of brain anatomy in time as well as methodology for segmentation of brain structures during neonatal period which utilizes this atlas. The atlas is available for download at http://www.brain-development.org/. The construction of the atlas and the segmentation methodology that utilizes it is described in this NeuroImage paper.

I built a similar 4D model from structural MRI and fMRI data. This model has been used to visualise development of default mode network, which is responsible for brain activity when we are resting, such as processing information or day-dreaming. This work was done as a part of an fMRI study which detected this network for the first time in babies in age equivalent to third trimester of pregnancy. The 4D model beautifully visualized brain activity at early time-points (around 30 weeks of gestation) when the connection does not exist, middle time-points when the connection develops as neurons migrate to their final destinations and finally localizations of centres of the brain which communicate when this network is active. The discovery was described in this press release ans published in PNAS. Creation of the 4D model was described in this paper. Click on the image below to view the movie.

We also recently built and atlas of progressing myelination in prematurely born babies and showed that we can predict age at scan with accuracy of around 1 week. In future, we might be able to assess developmental delay with similar prediction tools.

We are now developing interpretable deep learning methods methods to find markers of preterm birth. We have been able to create maps of locations in neonatal brain which are different due to the preterm birth.

Relevant publications:

Quantitative assessment of myelination patterns in preterm neonates using T2-weighted MRI. Siying Wang, Christian Ledig, Joseph V. Hajnal, Serena J. Counsell, Julia A. Schnabel & Maria Deprez. Scientific Reports volume 9, Article number: 12938 (2019). open access

A dynamic 4D probabilistic atlas of the developing brain. M Kuklisova-Murgasova, P Aljabar, L Srinivasan, SJ Counsell, V Doria, A Serag, IS Gousias, JP Boardman, MA Rutherford, AD Edwards, JV Hajnal, D Rueckert. NeuroImage 54(4), 2011.

Emergence of resting state networks in the preterm human brain. V Doria, CF Beckmann, T Arichi, N Merchant, M Groppo, FE Turkheimer, SJ Counsell, M Murgasova, P Aljabar, RG Nunes, DJ Larkman, G Rees AD Edwards. Proceedings of National Academy of Sciences, 107(46), 2010.

Investigating Image Registration Impact on Preterm Birth Classification: An Interpretable Deep Learning Approach. Grigorescu I., Cordero-Grande L., David Edwards A., Hajnal J.V., Modat M., Deprez M. (2019) In: Wang Q. et al. (eds) Smart Ultrasound Imaging and Perinatal, Preterm and Paediatric Image Analysis. PIPPI 2019, SUSI 2019. Lecture Notes in Computer Science, vol 11798. Springer, Cham