1st Supervisor: Robert Leech, King’s College London
2nd Supervisor: Tomoki Arichi, King’s College London
Aim of the PhD Project:
- To conduct the first ever studies with newborn infants and a new fMRI method: Neuroadaptive Bayesian Optimisation
- This will combine real-time fMRI analysis, machine learning, and MR-compatible robotics.
- These methods will be used to characterise the development of activity within the developing sensori-motor system of preterm and term newborn infants.
Project Description / Background:
During early life, human motor behaviour rapidly evolves to enable goal-directed behaviour and purposeful environmental interaction. Across this crucial period, the human brain is undergoing a dramatic but highly programmed sequence of maturation as the cortex forms and adjacent white matter fibres branch and refine to establish a dense framework of connectivity that subserves brain function throughout the lifespan. In recent years, the widespread use of non-invasive neuroimaging methods such as MRI have enabled us to make enormous strides in our understanding of these critical processes. In this context, functional MRI (fMRI) represents a powerful tool as it can map systems wide patterns of spontaneous or induced neural activity with excellent whole brain spatial specificity.
Using a combination of state-of-the-art fMRI methodsand custom-built MR-compatible robotics (for review see Allievi et al. 2014, Front. Neurol.), we have been able to characterise how patterns of brain activity and functional networks related to the sensori-motor system evolve in the newborn period. These studies confirm that sensori-motor activity in the newborn brain is much more organised and complex than traditionally envisaged and confirm that across this juncture there are rapid maturational changes as connections and region-specific functional roles are established. Understanding this process has important clinical implications, as despite great advances in neonatal medicine, cerebral palsy (the life-long posture and movement difficulties resulting from an early brain injury) remains common and still cannot be effectively treated.
Although extremely informative, previous fMRI studies have used a traditional “task-based” approach, in which patterns of activity induced by pre-specified and highly structured stimuli are identified using methods like statistical parametric mapping. However, a major limitation to this approach is that it can only consist of a single or a few stimuli, and constrains study populations to only those who have met (or can meet) specific conditions (such as cooperation or a particular behavioural state like sleep). This then greatly limits the generalisability and reproducibility of any findings. In this project, we will therefore plan to develop and use a new methodology: Neuroadaptive Bayesian optimisation (Lorenz et al. 2017, Trends in Cognitive Sciences) for the first time in the challenging newborn population. This approach combines real-time fMRI with machine learning to efficiently and automatically search through experimental conditions to identify those that are optimal for an individual infant and therefore is particularly suitable for this inherently uncooperative subject group. This will allow us to adapt the pattern of stimulation to individual infants and efficiently map brain responses across a large space of stimuli, allowing for greater understanding and generalisation across different infant populations and ages.This project would be suitable for a student with a background in bioengineering and/or imaging sciences with a particular interest in their application to neuroscience. The student would work closely with the supervisory team and a multi-disciplinary team of clinicians and imaging scientists based at the KCL Centre for the Developing Brain with access to a dedicated 3 Tesla MRI scanner located on the neonatal intensive care unit at St Thomas’ Hospital.