The goal is to better understand the properties of brain signals using various signal-processing techniques for decoding and brain-machine interfacing applications.
Decoding of the visual world from brain signals
We studied how much information is contained in different types of brain signals by showing natural images to monkeys and then using information-theoretic and machine-learning techniques. We found that surprisingly, ECoG is more informative than LFP or spikes, and of all frequencies, the gamma band is most informative (Kanth and Ray, 2020, Journal of Neuroscience). We are currently building models to predict the gamma response produced by natural images by studying various features of these images (Kanth and Ray, 2023, bioRxiv).
Detailed analysis of properties of various brain signals
Our setup allows a thorough investigation of the properties of macro-signals such as LFP, ECoG and EEG since we have simultaneous access to spikes from ~100 microelectrodes as well. For example, we used drifting visual stimuli to study whether LFPs represent inputs coming in a particular visual area or the outputs (Salelkar, Somasekhar and Ray, 2018, Journal of Neurophysiology). We also investigated the spatial spread of different brain signals and found that the spatial spread of LFP is band-pass (Dubey and Ray, 2016, Journal of Neurophysiology), and the spatial spread of ECoG is remarkably local (Dubey and Ray, 2019, Journal of Neuroscience). Tuning of gamma oscillations to various stimulus parameters was also remarkably consistent between LFP and ECoG (Dubey and Ray, 2020, Scientific Reports).
Signal Processing Techniques
Bridging brain signals across scales requires mathematical and signal processing tools (such as power spectral density and coherence), which are often studied only at one level, and their limitations are not well understood at other levels. To build links across scales, we need to thoroughly investigate the properties of such tools. For example, we have shown that a popular metric of phase consistency called coherence is also modulated by amplitude correlations between the signals (Srinath and Ray, 2014, Journal of Neurophysiology), and using reference schemes such as average referencing can change the phase differences between two signals by 90 degrees (Shirhatti, Borthakur and Ray, 2016, Neural Computation). We have developed a publically available toolbox to study brain signals using a technique called Matching Pursuit (Subhash Chandran, Mishra, Shirhatti and Ray, 2016, Journal of Neuroscience), which can better resolve transients as well as rhythms present in a signal, and used Matching Pursuit to determine the duration of gamma rhythm accurately (Subhash Chandran, Seelamantula and Ray, 2017, Journal of Neurophysiology). We have also studied how spike-gamma phase relationship varies with contrast and attention (Das and Ray, 2018, Frontiers in Computational Neuroscience), and have identified several issues with conventional spike-LFP phase analysis techniques (Ray, 2015, Current Opinion in Neurobiology; Ray, 2022, Annual Review of Vision Sciences).
Neurofeedback
Our brain generates several rhythms that wax and wane depending on the behavioral state. For example, a wave called “alpha” (~10 cycles per second) is generated when our eyes are closed and our mind is calm, but it wanes off with stress. Since the early 1970s, scientists have studied whether we can control our own alpha power if provided feedback. In such experiments, typically a tone is generated whose pitch varies with the ongoing alpha power. In these studies, people indeed learned to control their alpha and this “neurofeedback training” even reduced their anxiety levels. However, these results were subsequently challenged, since the reduction in anxiety could be due to familiarity with the task/environment, irrespective of feedback. In our study, we addressed this issue by providing “invalid” feedback (tone pitch independent of alpha) on a small fraction of trials without telling the subjects about it, which served as a “control” condition. While previous studies have used such “controls” on a different group of subjects, ours is the first in which a subject is his/her own control. We found that valid feedback indeed sustained alpha power more than invalid, especially for people who could not otherwise sustain their alpha activity. Future studies will test whether neurofeedback can be used in the treatment of anxiety and depression. For more details, see Biswas and Ray, 2019, eNeuro and Biswas and Ray, 2017, Journal of IISc. Source code for running the entire neurofeedback paradigm using an inexpensive openBCI amplifier is available here: https://github.com/supratimray/AlphaFeedbackProject. We have found that this inexpensive setup is also able to record gamma oscillations reasonably accurately (Pattisapu and Ray, 2023, PLoS One).
