Is Biofeedback Training of Ownership Perceptions Possible? EEG Classification of Volitional Hand-Ownership using Common Spatial Patterns

Abstract

‘Ownership’ is the natural feeling that our body parts belong to us when we attend to them. And yet, this innate sensation of ownership is susceptible to disorder which has led to interventions involving ownership of mirrored and imagined limbs. In contrast, states of ‘detachment’ from limbs are also desirable in cases where amputees seek to rid themselves of an illusory, although painful presence of a ‘phantom’ limb. In this thesis, because ownership is influenced by one’s control of internal attention, the possibility of optimising current ownership interventions through biofeedback is proposed and assessed through an ‘ownership–detachment’ calibration paradigm and classification analysis of the electroencephalography (EEG).

Specifically, 36 participants underwent a task involving internal control of ownership and detachment perceptions of virtual hands. Trials began with an ‘internal guidance’ phase with the same virtual-hand stimuli across the ownership–detachment conditions and differed only by pre-trial instruction. Trials then transitioned towards an ‘external guidance’ phase where visuals were specific to ownership or detachment condition and served as a reinforcing materialisation of the pre-trial instructions. Analyses began with the creation of predictive algorithmic ‘classifiers’ using Common Spatial Patterns (CSP) and Linear Discriminant Analysis to categorise unlabelled EEG into its ‘ownership’ or ‘detachment’ label. Classifiers were created for electrode clusters of frontal (representing internally-controlled neural processes), centro-parietal (representing hand sensorimotor and proprioception neural processes) and occipital (as a control) regions. To estimate online classification performance, cross-validation analysis of the accuracy at which classifiers could distinguish between ownership–detachment condition produced the definitive metric of the paradigm.

The results show strong support for the hypothesis that ownership intentions are discriminable in the EEG with the highest internal-guidance mean accuracy of 63.90%, range 32.35% – 96.15%; significantly higher than expected chance classification (p < .00001). As a corollary, the hypothesis that the intention to alter perceived ownership produces distinct EEG signals is supported. The frontal cluster produced the highest accuracy and reliability with the CSP scalp plots characteristic of horizontal saccadic patterning even under internal guidance where visuals were neutral between ownership–detachment conditions. A classification regime using electrodes most susceptible to saccadic activity further improved performance. Subsequent analysis through Independent Component Analysis showed consistent differences in horizontal saccadic activity across the top-two performing participants – the intent to own the virtual hand increased saccadic activity when compared with the intent to detach.

It is suggested the increase in saccadic activity stems from a volitional increase in visual attention due to the ownership intention having an explicit relationship to the visual stimulus of a virtual hand. This implies saccadic activity is possibly the most robust physiological metric of intention of ownership of body parts and tools. This implication taken in conjunction with the perceptual-scales findings of a correlated increase in ownership perception with internal engagement suggest saccadic training may have a role in enhancing perceived ownership of visual objects such as virtual hands.