Before I build this: real-time sLORETA/eLORETA with 3D visualization, is this a real need?
Hi all,
I'm a software engineer beginning work on an open-source project and I'd like to pressure-test the idea with people who actually work in this space before I commit to building it.
The project: a standalone desktop application that performs real-time EEG source localization (sLORETA/eLORETA) using a template head model and renders estimated cortical source activity as a color-mapped overlay on an interactive 3D brain mesh. The intended tech stack is Rust, wgpu for GPU-accelerated signal processing and rendering, and Tauri for the application shell. Data acquisition via BrainFlow, with BIDS dataset support for offline replay and analysis. No MATLAB dependency, no cloud, runs locally on commodity hardware.
The gap I'm trying to fill: source localization algorithms are well-validated and the computational feasibility of running them in real time on a GPU has been demonstrated in published work. But as far as I can tell, no usable open-source standalone application exists that does this end-to-end — ingesting live EEG, solving the inverse problem, and rendering source estimates on a 3D cortical surface at interactive frame rates. The existing tools either do source localization offline (MNE-Python, Brainstorm), operate only in sensor space in real time (NeuroSkill, OpenBCI GUI), or require MATLAB.
My background is in systems programming, not neuroscience. I'm investing significant time in domain knowledge (working through Cohen's Analyzing Neural Time Series Data and the Nunez & Srinivasan text, and studying MNE-Python's inverse solution pipeline as a reference implementation). I plan to validate against the Localize-MI ground-truth dataset before making any claims about accuracy.
What I'd like from this community:
Does this project address a real need in your work, or is it solving a problem that doesn't meaningfully exist in practice?
For those who do source localization: is a template-based approach (ICBM152, no individual MRI) useful enough for your purposes, or is it too imprecise to be worth visualizing in real time?
What channel counts and devices would this need to support to be useful to you? Is there value in supporting consumer devices (Muse, OpenBCI Cyton) for source imaging, or is that misleading given their limited spatial sampling?
Are there existing tools or projects I've missed that already do what I'm describing?
What features would make you actually use this versus your current workflow?
I'm not trying to replace MNE-Python or Brainstorm for offline research analysis. The goal is specifically the real-time visualization layer that currently doesn't exist as a standalone application. If this turns out to be a solution in search of a problem, I'd rather hear that now than six months from now.
Appreciate any candid feedback — critiques included.
Comments
Hello Day,
I believe your proposed project would be a valuable open source contribution, so please continue. I suggest you turn your post above into a pdf and post it (along with some introductory comments) on the NeuroTechX Slack, where many more research professionals will see it and comment. As you might know Slack is not conducive to posting large texts, so that would better be presented as your pdf. Register using the tab at the top, on the page below: "Community/Slack" selection.
https://neurotechx.com/
As far as other live real-time source localization apps, I do know that the NeuroPype package supports this, but distribution is limited to university-only or high-priced paid. Not sure if they provide source access, once admited. It's not an open-source project.
https://www.neuropype.io/
I have done work in the field of neurofeedback and QEEG, and various clinician tools allow real-time 3D rotatable source localization maps. Such as Brainmaster Avatar. I believe there are others as well. But all of these clinical packages have astronomical pricing, so out of reach for many.
https://brainmaster.com/brainavatar-software/
In the field of neurofeedback, having ability to reward or inhibit activity at certain voxels / Brodmann areas is part of the advantage of sLoreta neurofeedback. Frequently these packages combine the real-time values, with lookup in a normative database to provide 'LZT' Live Z-Score Training.
As far as applicability to 'consumer' devices, you likely know that clinical QEEG / neurofeedback applications use 19 channel caps. You cannot go much below this without losing significant resolution. However 16 channels could probably work. I have seen very very fuzzy 3D renders with small numbers of channels, such as the 4 channel devices. It's possible, but applications probably limited.
I suggest you reach out to Dr. Cody Rall, a former US Navy psychiatrist who now works with a wide variety of EEG and neurofeedback consumer devices on his Youtube channel and website. I'm sure he would support your effort and possibly network you with related projects.
https://www.youtube.com/channel/UCMIs6RbQ1sAr_RmYyfc4iQg
https://www.youtube.com/@CodyRallMD/videos
https://www.sharpereveryday.com/
https://techforpsych.com/
Best regards, William
Oh, hey @u_Day, thanks for the mention, we've s(e)LORETA on our roadmap, but we are testing devices with more electrodes now, before we release it to public.
I added some logic for s(e)LORETA to the exg rust repo, feel free to check it out and share your feedback.