pyREM: a crowd trained machine learning approach to automatic analysis of EEG data
pyREM: a crowd trained machine learning approach to automatic analysis of EEG data
Quentin Geissmann, and Giorgio F Gilestro
EEG data are at the basis of a plethora of neuroscientific questions: from sleep to consciousness and attention, many aspects of neuroscience heavily rely on electrophysiological correlates of brain activity. Yet, EEG analysis heavily relies on subjective scoring and interpretation to the point that many neuroscientists consider it an art, more than a systematic tool.
Can we teach this art to a computer?
Attempts at creating an objective way of scoring EEG data have been less than perfect so far, mainly because humans are reluctant about trusting the judgement of a machine, programmed according to hard-coded values and thresholds.
pyREM aims at solving this issue, using a machine learning approach to automatically analyse EEG data. pyREM learns how to classify EEG directly from humans, mimicking all the human’s principles and criteria without any apriori knowledge of what an EEG means. The overall goal of the project is to teach pyREM how 1, 10, 100 or 1000 laboratories score EEG so that the software will be able to automatically grasp and isolate the key fundamental criteria and become, in this way, the universal scorer.
If you are a laboratory interested in being part of this, please get in touch.
If you want to know more, you can
- read pyrem_poster
- read an early draft of the manuscript
- read the pyREM documentation
- get in touch!