Simulated Neurophysiological Data

GIF references: Screenshots by H Muzart. Datacode scripts modifications by H Muzart, from original files by SPM DEM developers and MathWorks third-party developers. Showing curves drawn in Matlab GUI.

Simulating EEG/fMRI signals

Generating neural data without scanning/recording an actual real brain can be done with various software tools. There is the Matlab-based (FIL's) SPM software's standard functions (e.g. hrf, etc) and supplementary toolbox DEM which incorporates mathematical models from the Friston et al [ref pubs]. These are demos with open-source code and adjusted parameters. Essentially, one may simulate very simplified LFPs (local field potentials constituting micro and macro electrophysiology - in terms of ERPs and oscillations) and BOLD (blood oxygen level dependent - hemodynamic response) signals. These don't have to be just measured from humans/animals, but can be quasi-reconstructed via reverse engineering in simple terms, using simple functions. This can mapped onto psychological stimuli human data, but also human-like AI agents (e.g. Cullen et al supplem.)

For example, here my endeavour is to use many of these tools to simulate what kind of response I may get with all the other cognitive models I developed as part of CognTech, and fit my data/models to those simulated responses.

Put very simply:

Sensory stimulus X --> Neurophysiological activity Y --> LFPs (EEG-basis) --> BOLD (fMRI) signal

Pattern of external cognitive stimuli, X{n} <--> Pattern of internal neural activity, Y{n}

Since we know that experimentally, it can be mathematically modelled too.

One may also use data from the literature for functional validation and confirming anatomical localisation.

Thus, novel neural data can be generated without any recourse to real participants. This for now only remains for illustrative purposes. Moreover, in the future, it is something I am considering delving deeper into.

Another application, but in reverse, would be generating psychophysical stimuli from neural (eg. EEG/fMRI) signals (e.g. visual cortex activity to reproduce visual stimuli (UCB-lab), etc).

-----------

These tools may ultimately provide powerful enough predictions so that we may use those for all sorts of applications (clinical or otherwise), and could reduce the need for actual animal/human experimentation, as neuro-cognitive physiological processes equivalents of engineered synthetic samples.

Below are just other demos of the other sub-programs...