Professional Summary
I am a postdoc specializing in how patterns of brain activity and connectivity change over time. I develop mathematical and computational tools for analysing functional neuroimaging data, with the broader aim of helping to understand complex brain dynamics. My research combines unsupervised machine learning, directional statistics, signal processing, and neuroscience. I work with fMRI, EEG/MEG, PET, SPECT, and ASL data from healthy participants, consciousness-altering experiments, and patient groups. My current application areas include psychedelic states, sleep and glymphatic physiology, coma and other disorders of consciousness, and methodological development on healthy reference cohorts such as the human connectome project.
During my PhD, I developed methods for studying phase coherence and multimodal brain states in fMRI, EEG, and MEG data. Much of this work involved directional statistics. At the Neurobiology Research Unit, I currently analyse human neuroimaging data from psilocybin, LSD, and sleep experiments in healthy controls and comatose patients, as well as EEG, fMRI, and SPECT data from porcine models of the glymphatic system and human patients with disorders of the glymphatic system.
I maintain a close collaboration with the lab of Prof. Morten Mørup, who supervised my BSc, MSc, and PhD, and with the lab of Prof. Dimitri Van De Ville in Geneva, where I have worked on graph signal processing and brain structure–function relationships. My ongoing projects include collaborations with Profs Kanti V. Mardia and Daniel Kondziella.
I have first-authored in PNAS, Nature Communications, NeuroImage, and Frontiers in Neuroscience, as well as several IEEE conference contributions and co-author contributions to various neuroscience journals.
My methodological experience includes:
- Unsupervised and probabilistic modeling: K-means, mixture models, Hidden Markov models, and stochastic block models for univariate, multivariate, and matrix-variate data.
- Matrix, tensor, and low-rank modeling: PCA and sparse PCA, ICA, NMF, sparse coding, archetypal analysis, coupled generator decompositions, Tucker and CP decompositions, generalized eigendecompositions, low-rank reparameterization.
- Directional and manifold-valued statistics on spaces including the real and complex hyperspheres, Grassmann manifold, torus, SPD/SPSD manifolds.
- Time-series and time–frequency analysis: Phase-coherence analysis, multitaper spectral and cross-spectral estimation, empirical and variational mode decomposition, time-shift- and time-stretch-invariant modeling.
- Graph and spatial analysis: Graph analysis, graph signal processing, structural-connectome and cortical-geometric eigenmodes, Procrustes alignment.
- Multisubject and multimodal data fusion: Joint modeling across subjects and functional neuroimaging modalities, particularly fMRI and combined EEG/MEG data.
- Neuroimaging analysis software: fMRIPrep, ASLPrep, MRIQC, TEDANA, SPM, FSL, MNE-Python, EEGLAB, and Connectome Workbench.
- Proramming in Python (primary language since 2021), Matlab (since 2015), R (only minor experience)
- Pulse detection in fast fMRI during sleep data using, e.g., ICA or sparse coding with a potential temporal shift invariance. Significance: To show the existence of norepinephrine spikes in the sleeping brain, signifying a potential important working mechanism of the glymphatic system, which is thought to cleanse the brain during sleep by propagating cerebrospinal fluid through the brain parenchyma. Suggested prerequisites: Coding experience in python/pytorch, interest in unsupervised machine learning and the brain.
- Interregional brain phase coherence alterations as a function of psychedelic drugs. Two datasets: One on the acute effects of psilocybin (n=28) and one on the long-term effects of psilocybin (n~80 ish) (possible to include even more datasets). We have previously shown that a certain network of frontoparietal and default mode regions appear to be reduced in activity acutely following psilocybin. We now have improved methods and access to further datasets, allowing us to conduct a confirmatory study on the previously shown effects. Significance: The subjective effect of psychedelics are clear but previous analyses on fMRI data have shown differing results between analysis method and research groups. With the phase coherence analysis, we believe to be able to show consistent effects across datasets. Suggested prerequisites: Interest in psychedelic drugs and the brain, some coding experience (less method development, more analysis and data handling).
- Requirements for the Hilbert transform in fMRI data. The Hilbert transform is used to assess phase coherence between regional brain signals, but generally requires data to be “monocomponent”, i.e., contains only one “type” of oscillation. Real data, both M/EEG and fMRI do not generally uphold this but nevertheless the Hilbert transform is still used to assess phase coherence. In this project we want to figure out the implications of this, i.e., 1. how much signal do we capture with existing methods and 2. what can we do to capture more (artefact removal, filtering etc). Significance: This study will enable a more correct characterization of brain networks by reducing the heuristics needed to model such. Suggested prerequisites: An interest in math and/or signal processing and coding in matlab or python. The project will mainly be theoretical / use synthetic data but application to brain networks is a possibility.