Neuroengineering of hallucinations through fMRI neurofeedback: Understanding and modulating neural dynamics for the control of hallucinations in health and disease

Recent years have seen considerable progress in understanding brain function, and in modulating it through real-time fMRI neurofeedback (NF). Inspired by these advances, paired with the unique possibility of inducing the clinically relevant Presence Hallucination (PH: strange sensation of having someone behind when no one is there) through MR-compatible robotics, I set out to: (1) understand the brain dynamics that lead to PH and have participants regulate them as to modulate PH; (2) study the neural underpinnings of this type of hallucinations in Parkinsonâ s disease (PD); and (3) advise on the optimal setup to induce PH.

In Part I of my thesis (Study 1), I investigated the neural correlates of PH-induction, hypothesizing that underlying hallucinations were sporadic dysfunctions in temporal processes of brain activity. Using Co-Activation Pattern (CAP) analysis I identified patterns of activity and studied their occurrence and transition probabilities, while participants experienced robot-induced PH (riPH). With this I showed that sensitivity to riPH depended on a temporary shift in transition probabilities that caused all CAPs to increase transitions to a specific brain pattern (PH-network)

With this knowledge, I then paired the MR-compatible PH-induction system with fMRI-NF (Part I â Study 2), to provide informative feedback on the PH-network activity as PH was induced, and allow participants to achieve volitional control over it. During three NF-training days, participants learned to up-regulate and down-regulate the PH-network, which lead to an increase in sensitivity to riPH post-training, as compared to pre-training. Moreover, for participants that were successful during NF and became sensitive to riPH, we noted lasting changes in brain activity marked by an increased occurrence of the PH-network during induction

In Part II (Study 3), I investigated the neural correlates of PH in PD, as it is a common hallucination in this condition that might predict cognitive decline and persistent psychosis. I investigated riPH, cognitive impairment and fMRI neural correlates, in a cohort of patients stratified based on the severity of hallucinations: no hallucinations, minor hallucinations (subgroup including PH), and structured hallucinations (mostly visual). I showed increased sensitivity to riPH across the axes of hallucination severity and cognitive impairment. Studying multivariate patterns of brain activity and behavior, I then identified that antagonistic activations and deactivations between large brain networks important for cognition, self-related processing, and vision, underpinned preserved cognitive capabilities, decreased sensitivity to riPH, and no hallucinations

Finally in Part III (Study 4), motivated by the translation of riPH from healthy to clinical populations, I analyzed all experiments that used riPH, and quantified the effect of various experimental parameters with a Bayesian meta-analysis. With this I was able to propose