EEG connectivity features associated with fibromyalgia revealed by machine learning.

Introduction: We present connectivity-based features associated with fibromyalgia, derived from raw EEG data at the sensor level. Methods: These connectivity features were identified through a data-driven method, employing machine learning. We carried out some automatic, moderate pre-processing and extracted spectral connectivity features. Machine learning experiments then followed, employing feature importance analyses and feature selection techniques for building high-performing classification models; finally, based on robust cross-validation and test evaluation, we obtained the features associated with fibromyalgia. The raw EEG signals from 463 participants are used in the primary analysis. An external dataset that consists of 48 participants is used to validate the identified connectivity features. Results: Five features in the gamma band (Fz-Cz, Pz-P4, Fz-C3, Cz-P4, and Cz-Pz) were able to objectively detect the presence or absence of fibromyalgia with an accuracy of 99.57%. The identified connectivity features associated with fibromyalgia also show promising results on EEGs that are collected using a different type of device. Discussion: EEG-based functional connectivity features associated with fibromyalgia, identified using machine learning in the gamma band at the sensor level, can distinguish between fibromyalgia participants and healthy controls with 99.57% accuracy. These findings advance our understanding of the brain-based mechanisms of fibromyalgia and provide novel targets for future non-invasive neuromodulation and neurofeedback trials. However, future studies need to replicate these findings in independent EEG datasets in people with fibromyalgia as well as compare with other clinical populations.