Decoding the Self: Single-Trial Prediction of Self-Boundary Meditation States From Magnetoencephalography Recordings.

Abstract

The sense of self is a multidimensional feature of human experience. Different dimensions of self-experience can change drastically during altered states of consciousness induced through meditation or psychedelic drugs, as well as in a variety of mental disorders. Some experienced meditation practitioners are able to modulate their sense of self deliberately, which allows for a direct comparison between an active and suspended sense of self. Meditation therefore has the potential to serve as a model-system for alterations in the sense of self. The current study aims to identify a neural marker of such meditation-induced alterations in the sense of self based on magnetoencephalography (MEG) recordings of meditation practitioners (N = 41). Participants alternated between a state of reduced sense of self, termed self-boundary dissolution, a resting state and a control meditation state of maintaining their sense of self. Machine learning methods were used to find multivariate patterns of brain activity which distinguish these states on a single-trial basis. Source band power and Lempel-Ziv complexity features allowed to predict the mental state from MEG recordings with significantly above-chance accuracy (> 0.5). The highest performance was obtained for the self-boundary dissolution versus rest classification based on Lempel-Ziv complexity, which showed an average accuracy of ~0.64 when training and testing were performed on data from the same individual (within-participant prediction) and ~0.57 when models trained on one group of individuals were tested on different participants (across-participant prediction). Potential applications include decoded neurofeedback, for example, for clinical treatments of disorders of the sense of self, or for assistance in meditation training.