Computer Vision News - October 2024

Computer Vision News Computer Vision News 52 Congrats, Doctor Logan! For over a century, it has been recognised that parcellating the human cerebral cortex is a prerequisite for understanding its function. Typically, modern parcellations define a set of cortical regions from images that have been brought into a common reference space using non-linear volumebased registration, or surface-based alignment of cortical shapes. Although accessible and straightforward to implement, cortical shape does not fully reflect the underlying cytoarchitectural and functional organisation of the cerebral cortex, producing misalignments that blur fine-scale aspects of cortical organisation during template creation. To address these issues, the Human Connectome Project (HCP) developed and extensively validated Multimodal Surface Matching (MSM), a flexible surface registration tool that allowed registration to be driven by multiple, complementary features that better correspond with cortical areas (Figure 1). This registration configuration, known as `MSMAll', improved inter-participant alignment and enabled the HCP to generate version 1.0 of their group average multimodal cortical parcellation. To account for variations in cortical organisation that could not be captured by MSMAll, the HCP generated participant-specific parcellations using a set of fullyconnected neural networks that assigned labels to each point on the cortical surface based on their multimodal fingerprint. To date, individual HCP multimodal parcellations have been generated using high resolution structural images and 1 hour of resting-state functional imaging. However, many studies are unable to acquire such large amounts of highquality data per participant, potentially limiting the generalisability of this approach to other important datasets with more standard neuroimaging protocols. A prime example is UK Biobank, which has acquired lower resolution multimodal brain imaging data46 from >60,000 individuals. Logan Williams defended his PhD a few months ago. He works as part of the MeTrICS lab at King’s College London, who develop methods for processing and analyzing the cerebral cortex, the outermost layer of the brain. Under the supervision of Emma Robinson and David Edwards, Logan developed and validated a surface CNN that accurately generated individualized maps of the cerebral cortex in the UK Biobank. Since completing his PhD, Logan has returned to New Zealand to complete his medical training in radiology.

RkJQdWJsaXNoZXIy NTc3NzU=