Modelling individual variations in brain structure and function using multimodal MRI

Thesis

Every brain is different. Understanding this variability is crucial for investigating the neural substrate underlying individuals’ unique behaviour and developing personalised diagnosis and treatments. This thesis presents novel computational approaches to study individual variability in brain structure and function using magnetic resonance imaging (MRI) data. It comprises three main chapters, each addressing a specific challenge in the field.

In Chapter 3, the thesis proposes a novel Image Quality Transfer (IQT) technique, HQ-augmentation, to accurately localise a Deep Brain Stimulation (DBS) target in low-quality clinical-like data. Leveraging high-quality diffusion MRI datasets from the Human Connectome Project (HCP), the HQ-augmentation approach is robust to corruptions in data quality while preserving the individual anatomical variability of the DBS target. It outperforms existing alternatives and generalises to unseen low-quality diffusion MRI datasets with different acquisition protocols, such as the UK Biobank (UKB) dataset.

In Chapter 4, the thesis presents a framework for enhancing prediction accuracy of individual task-fMRI activation profiles using the variability of resting-state fMRI. Assuming resting-state functional modes underlie task-evoked activity, this chapter demonstrates that shape and intensity of individualised task activations can be separately modelled. This chapter introduced the concept of "residualisation" and showed that training on residuals leads to better individualised predictions. The framework’s prediction accuracy, validated on HCP and UKB data, is on par with task-fMRI test-retest reliability, suggesting potential for supplementing traditional task localisers.

In Chapter 5, the thesis presents a novel framework for individualised retinotopic mapping using resting-state fMRI, from the primary visual cortex to visual cortex area 4. The proposed approach reproduces task-elicited retinotopy and captures individual differences in retinotopic organisation. The proposed framework delineates borders of early visual areas more accurately than group-average parcellation and is effective with both high-field 7T and more common 3T resting-state fMRI data, providing a valuable alternative to resource-intensive retinotopy task-fMRI experiments.

Overall, this thesis demonstrates the potential of advanced MRI analysis techniques to study individual variability in brain structure and function, paving the way for improved clinical applications tailored to individual patients and a better understanding of neural mechanisms underlying unique human behaviour.

Authors: Zheng, YQ

• DOI: 10.5287/ora-qqxekgkba
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English