Novel imaging biomarkers for prognosis of developmental outcomes in babies born preterm using machine-learning
Lead Research Organisation:
King's College London
Department Name: Imaging & Biomedical Engineering
Abstract
Research Question: The main aim of this PhD project is to design a deep learning framework capable of uncovering features of abnormal brain development associated with neurocognitive disorders.
Methodology: For this, we propose:
1. To characterize the developmental trajectory of a healthy brain, while also taking into account the rapid maturation process that undergoes during the neonatal period and early life. These changes affect both the brain's shape and the MRI tissue contrast. This will allow comparison of an individual to the healthy gestational-age-adjusted pattern estimated from the population. Towards this aim, we will develop a multi-channel (structural and diffusion) registration framework that will combine intensity-based registration of structural data with metrics that align white matter tracts in diffusion data.
2. To predict neurocognitive outcomes and investigate effects of gestational age at birth, sickness after birth (lesions, ventilation) and genetic variants likely related to altered microstructure in the developing brain. Towards this aim, we will develop an interpretable machine learning method that connects the evolving macro- and micro-structural properties of the brain with clinical and genetic data.
Methodology: For this, we propose:
1. To characterize the developmental trajectory of a healthy brain, while also taking into account the rapid maturation process that undergoes during the neonatal period and early life. These changes affect both the brain's shape and the MRI tissue contrast. This will allow comparison of an individual to the healthy gestational-age-adjusted pattern estimated from the population. Towards this aim, we will develop a multi-channel (structural and diffusion) registration framework that will combine intensity-based registration of structural data with metrics that align white matter tracts in diffusion data.
2. To predict neurocognitive outcomes and investigate effects of gestational age at birth, sickness after birth (lesions, ventilation) and genetic variants likely related to altered microstructure in the developing brain. Towards this aim, we will develop an interpretable machine learning method that connects the evolving macro- and micro-structural properties of the brain with clinical and genetic data.
Organisations
People |
ORCID iD |
| Maria Deprez (Primary Supervisor) | |
| Irina Grigorescu (Student) |
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| EP/R513064/1 | 01/10/2018 | 30/09/2023 | |||
| 2125388 | Studentship | EP/R513064/1 | 01/10/2018 | 30/09/2022 | Irina Grigorescu |
| Description | As a result of our work so far we have found that: 1. Interpretable deep learning methods can be used to reveal anatomically plausible features in distinguishing preterm neonates from term ones. 2. Using structural and microstructural data in a deep learning registration network can achieve superior alignment in subcortical regions and an improved alignment of white matter tracts. |
| Exploitation Route | Newer interpretable machine learning techniques could be investigated to further the findings of term vs preterm differences, as well as different preprocessing techniques applied to the data (such as removing the tissue type which was initially found in our work). For the second finding as future work we aim to work out a solution regarding how to weigh the contribution of different modalities to the registration. |
| Sectors | Healthcare |