A novel magnetic resonance technique to characterise grey matter microstructure in the human brain

Accurate characterisation of the microstructure of the human cerebral cortex is important fora number of applications. It guides anatomical parcellation, the functional correlates of whichinform neurosurgical decision making. It also enables detection of subtle abnormalities suchas focal cortical dysplasia (FCD), an important cause of epilepsy. Many studies have beenconducted in this area over the past century using different techniques ranging fromqualitative histological to quantitative in vivo Magnetic Resonance (MR) studies. However,achievement of precise whole-cortex microanatomical mapping has been hindered by thelack of a comprehensive mapping method that takes into account a wide range ofmicrostructural (cytoarchitectonic and myeloarchitectonic) tissue properties.To bridge this gap, I conducted my research with two main aims. First, I investigated thepresence of information in the Magnetic Resonance Imaging (MRI) signal about tissuemicrostructure over and above MR relaxometry-based tissue properties (e.g. longitudinalrelaxation time T1, effective transverse relaxation time T2*), on which MR-basedmicrostructural mapping methods are based. I proposed a novel quantitative framework thatemploys Magnetic Resonance Fingerprinting (MRF) and statistically characterises the MRFresiduals, after accounting for the relaxometry-based tissue properties. I showed thepresence of area-specific characteristics in the MRF residual signals from three corticalareas of individuals, suggesting that the framework could reveal more information about themicrostructural variations between cortical areas. This method could especially be helpfulwhere the information derived from the MR-relaxometry tissue properties is not sufficient fordelineating the distinction between two cortical areas.The second aim of my research was to propose an automated microanatomical mappingmethod for parcellating the cortex at the voxel level. I used the MRF residual analysisframework to characterise voxels from seven cortical areas. I then developed a feature-based supervised machine learning classification model that takes the statisticalcharacterisation of each voxel MRF residual as its input feature vector. At the level of theindividual subject, the average parcellation accuracy of the model was >80% for sevencortical areas. The proposed quantitative in vivo voxel-wise cortical parcellation methodcould be further expanded to cover the whole brain. The framework might also be used forpurposes other than cortical parcellation, including accurate lesion detection and delineationin neurosurgery and staging of neurological diseases.