During his PhD, Logan developed we use a novel deep learning framework to generalise the Human Connectome Project’s multimodal cortical parcellation to UK Biobank data, which were acquired with a substantially different imaging protocol. These parcellations capture participant-specific variations in cortical organisation (Figure 2) that are stable and discriminable over a timescale of years. Accounting for this individual variability improved the performance and interpretability of multivariate brain-wide associations with cognitive scores, and supports new mechanistic insights into whole-body action planning by the recently somato-cognitive action network. Logan and the rest of the team anticipate this framework will generalise to other largescale neuroimaging datasets, enabling the scientific community to investigate cortical organisation with improved neuroanatomical precision. Most importantly, these individualised parcellations will be made available via the UK Biobank. 53 Logan Williams Computer Vision News Computer Vision News Figure 1. Multimodal fingerprint (top row) and gradient (bottom row) of an example cortical area (area 55b) in HCP. Gradient maxima across cortical features overlap and help define the border of all areas in the multimodal parcellation. Figure 2. Examples of atypical (‘split’) topography of area 55b in the UKB. (a) group average parcellation and multimodal features; (b) single participant with typical 55b topography; (c - e) 3 example participants with split 55b topography.
RkJQdWJsaXNoZXIy NTc3NzU=