Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
Resting-state fMRI has proven to entail subject-specific signatures that can serve as a fingerprint to identify individuals. Conventional methods are based on building a connectivity matrix based on correlation between the average time course of pairs of brain regions. This approach, first, disregards the exquisite spatial detail manifested by fMRI due to working on average regional activities, second, cannot disentangle correlations associated to cognitive activity and underlying noise, and third, does not account for cortical morphology that spatially constraints function. Here we propose a method to address these shortcomings via leveraging principles from graph signal processing. We build high spatial resolution cortical graphs that encode each individual's cortical morphology and treat region-specific, whole-hemisphere fMRI maps as signals that reside on the graphs. fMRI graph signals are then decomposed using systems of graph spectral kernels to extract structure-informed functional signatures, which are in turn used for fingerprinting. Results on 100 subjects showed the overall superior subject differentiation power of the proposed signatures over the conventional method. Moreover, placement of the signatures within canonical functional brain networks revealed the greater contribution of high-level cognitive networks in subject identification.
Jean-Philippe Thiran, Gabriel Girard, Elda Fischi Gomez, Philipp Johannes Koch, Liana Okudzhava
Dimitri Nestor Alice Van De Ville, Friedhelm Christoph Hummel, Gabriel Girard, Takuya Morishita, Elena Beanato, Lisa Aïcha Mireille Julie Fleury, Maximilian Jonas Wessel, Philipp Johannes Koch, Philip Egger, Andéol Geoffroy Cadic-Melchior