University Hospital Tübingen, University of Tübingen, German Center for Mental Health (DZPG)

Exploring underlying mechanisms of real-time single region neurofeedback, functional connectivity neurofeedback and support vector machine neurofeedback

Beatrix Barth (1,2,3), Thomas Dresler (1,2,3), Mohit Rana (2), Betti Schopp (1), Sandra Ladegast (1), Andreas J. Fallgatter (1,3), Ann-Christine Ehlis (1,2,3)

(1) University Hospital Tübingen, (2) University of Tübingen, (3) German Center for Mental Health (DZPG)

In neurofeedback (NF), participants can learn to modulate their own brain activity (e.g., electrical brain activity, hemodynamics). To date, the majority of NF studies have focused on regional brain activation, but there are emerging approaches that target brain network properties (i.e., functional connectivity (FC) estimates) or brain activation patterns (using machine learning algorithms, e.g., support vector machine (SVM)). We compared three NF approaches for the motor cortex (MC) with functional near-infrared spectroscopy in 57 healthy participants regarding the underlying processes. In all groups, participants were able to regulate their MC activation, but without learning progress across sessions. FC-NF and single-region neurofeedback (SR-NF) resulted in a higher proportion of learning than SVM-NF. Regarding the underlying processes, no specific effects for the respective training parameters emerged from the different protocols (e.g., FC-NF members did not modulate connectivity but MC amplitude, similar to SR-NF members). In a pre-post motor task, reaction times decreased more in SR-NF and learners compared to non-learners. Accuracy increased after NF and was higher in NF learners. MC activation increased over time for learners and decreased for non-learners. Task load decreased over the sessions, but was higher in SVM-NF than in FC-NF. Motivation was higher among non-learners than among learners. In summary, SR-NF and FC-NF led to comparable results that were superior to the SVM-NF implemented here. Unexpectedly, the training parameters were not specifically modulated depending on the training protocol. However, NF learners differed from non-learners with respect to several outcome measures.