Publications highlights

• A predictive role of the old cortex in general intelligence (npg Artificial Intelligence, under review)

  This Perspective paper reconciles contemporary concepts in neuroscience with AI, focusing on learning and memory. We argue that the hippocampus helps compute and maintain predictions over multiple time spans thereby increasing the number of comparisons between predicted and real latent sensory experiences. This may help explain why animals with a hippocampus are able to learn so efficiently compared to popular, data-hungry AI architectures.

• HippoMaps: Multiscale cartography of human hippocampal organization (Nature Methods, accepted)

  Here we present a toolbox and data repository of hippocampal maps spanning structural, functional, and diffusion MRI, histology, intracrantial EEG at multiple spatial scales. These maps can be aligned and directly compared by spatial correlation with robust permutation correction. Such comparisons can give context to maps, informing us whether they are organized into subfields, anterior-posterior difference, and which other feature maps they resemble. This cancontextualize, for example, fMRI task maps or temporal lobe epilepsy abnormality maps by linking them to microcircuit or electrophysiological features.

• Hippocampal connectivity patterns echo macroscale cortical evolution in the primate brain (Nature Communications)

  The hippocampus is structurally conserved across primates, but its functional organization has evolved—likely due to integration with higher-order brain networks.

• Evaluation of surface-based hippocampal registration using ground-truth subfield definitions (eLife, 2023)

  Here we examine a set of seven 3D histology samples at microscopic resolution under our previously established hippocampal surface-based coordinate system. We generate a new ground-truth maxprob subfields atlas and introduce intersubject registration to witin 0.5mm into the HippUnfold pipeline.

• Automated hippocampal unfolding for morphometry and subfield segmentation with HippUnfold (eLife, 2022)

  Here, we present software to model the hippocampus as a 2D folded sheet instead of as a subcortical nucleus or volume as in other research software. This topology provides constraints on image interpretation, and can help with technical issues like subfield parcellation, inter-subject alignment, and making inferences about incomplete or low-resolution data. This is made fully automatic using detailed deep-learning based image segmentation.

• Surface-based hippocampal subfield segmentation (TICs, 2021)

  The hippocampus, like the neocortex, is composed of a thin, folded tissue. This is often hard to appreciate in MRI with low resolution, or in histology where only one plane of view is available. However, we argue that optimal intersubject registration of hippocmapal tissue should follow this topology. More broadly, hippocampal imaging and histology can be better interpreted with this prior knowledge.

• Hippocampal subfields revealed through unfolding and unsupervised clustering of laminar and morphological features in 3D BigBrain (NeuroImage, 2020)

  Here, we manually delineated hippocampal folding structure and subfields in the fully 3D ground truth BigBrain histology dataset. We then computationally extracted morphological and laminar features within this folding configuration based on popular neocortical analysis methods. Unsupervised clustering of these hippocampal features revealed subdivisions which closely resembled ground truth subfield definitions. This highlights both the validity of subfield definitions, which remain contested both in MRI and in histology, as well as the sensitivity of the methods.

Media

• Recent 45min talk as the inaugural speaker for The Neuro's postdoc and graduate seminar series. Find it here
• Featured scholar in the BigBrainProject for in-depth and data-driven analyses of the fully 3D, 40 micron resolution reconstruction of the human hippocampus. See story here
• CTV News London interview alongside PhD supervisor Dr Ali Khan for our Trends in Neuroscience paper advocating the 2D folded geometry of the hippocampus as its primary organizing structure and the ideal basis for subfield parcellation. See the story here.

Areas of collaborations

• Simulation and neural mass models of hippocmapal activity
• Mapping of hippocampal abnormalities in epilepsy
• Resting-state and task-based mapping of hippocampal fMRI data
• Extension of hippocampal analysis methods to other species, including mouse, macaque, and marmoset
• Microstructural circuitry mapping using DWI
• Research software development, including MicaPipe, BigBrain Warp, Z-Brains, and other large pipeline and data management projects