21.07.2023

New publication

New publication on sex-related patterns in the electroencephalogram and their relevance in machine learning classifiers.

Our algorithm for detecting the sex from EEG curves highlights the relevant areas. In unfiltered EEG data, there is an increased relevance of the time periods around the heartbeat. This part of the sex differences in the EEG signal probably does not come from the brain at all, but from signal artifacts of electrical heart activity.

Deep learning for EEG analysis is intriguing as it can utilize patterns that us humans are not even aware of. It can uncover new features and evaluate much more complex patterns in these multichannel zig-zag curves. However, deep learning is shameless in exploiting illegitimate features. When we replicated sex detection from EEG, our CNN first focused on sex differences in heart artifacts, rather than actual brain activity.
Implications: neural networks for detecting pathology from raw EEG are easily sex-biased. Recommendations: thoroughly remove artifacts, balance your dataset, and test for biases when using deep learning for EEG.

Jochmann, T., Seibel, M. S., Jochmann, E., Khan, S., Hämäläinen, M. S., & Haueisen, J.:

Sex-related patterns in the electroencephalogram and their relevance in machine learning classifiers.
(2023) Human Brain Mapping, 1– 11.

https://doi.org/10.1002/hbm.26417

Contact:    Dipl.-Phys. Thomas Jochmann