Optimizing real-time fMRI neurofeedback design through simulations: how learning mechanisms and feedback timing can affect results in a paradigm designed to alleviate depression

Abstract

This study presents a novel computational model of a real-time fMRI neurofeedback paradigm that is currently being tested in a controlled experiment as a potential therapy for alleviating symptoms of depression in adults. A simulation-based approach was used to simulate neurofeedback sessions using this model. Two variations of the model were constructed. One variation of the model contains a learning mechanism by which this neurofeedback paradigm could alleviate the biased attentional engagement towards negative-valence stimuli in depressed simulated subjects. The other variation of the model uses a learning mechanism by which neurofeedback could improve the ability of depressed simulated subjects to disengage from negative-valence stimuli. These two variations of the model were tested in simulated neurofeedback sessions using four different schedules of neurofeedback delivery: continuous, smoothed continuous, intermittent and interblock. The results of these simulated neurofeedback sessions showed that smoothed continuous feedback was the most beneficial feedback schedule when using the former variation of the model, and intermittent feedback was the most beneficial when using the latter variation. Simulations were also run to model the case of more severely-depressed subjects. For this severely-depressed case, better outcomes were observed using the second variation of the model, regardless of which feedback schedule was used. The implications of these results for the future optimization of the experimental real-time fMRI neurofeedback paradigm are discussed, as well as implications for the field of clinical real-time fMRI neurofeedback more generally. Future directions for building on the modeling work presented in this study are also discussed.