Talk by Dr. Dr. Jürgen Dammers (CSN Virtual Seminar)

We hereby announce the next talk in the 'CSN Virtual Seminar':

Neural Fingerprinting: Unveiling Unique Brain Signatures through Temporal Dynamics

by Dr. Jürgen Dammers, Medical Imaging Physics (INM-4), Jülich Research Centre

Abstract

Neural fingerprinting is an emerging technique that enables the identification of individuals based on unique patterns of brain activity derived from neuroimaging data, such as magnetoencephalography (MEG) and functional magnetic resonance imaging (fMRI). Traditional approaches have primarily concentrated on second-order statistical measures, such as connectivity matrices, to extract these individual signatures. However, recent advancements in machine learning have shifted the focus toward the direct analysis of the temporal dynamics of neural time series. In this presentation, we will explore how deep learning techniques can be applied to efficiently classify short segments of neuroimaging data, thereby revealing distinct neural fingerprints with high accuracy.

We propose a novel method that utilizes multivariate time series classification models, such as Rocket, to MEG recordings, achieving an impressive classification accuracy of over 99% in assigning 1-second time segments to individual subjects. Furthermore, we introduce contrastive learning techniques that facilitate robust fingerprinting, even with limited data, thereby addressing common challenges such as model retraining and data scarcity.

Our results demonstrate that temporal dynamics offer a valuable source of information for neural fingerprinting, exceeding traditional methods in both speed and accuracy. Additionally, we also highlight future challenges, such as scaling these methods for larger datasets and exploring their application in clinical settings for personalized diagnosis and treatment.