Can a lightweight dry-electrode EEG workstation reproduce a visual decoding experiment that normally depends on lab-grade EEG hardware?
That was the question behind this project.
EEG-based visual decoding is usually associated with well-funded neuroscience labs, wet electrodes, long setup time, and expensive amplifier systems. We wanted to test whether a more portable 7-channel dry-electrode setup could support a similar workflow: visual stimulation, EEG recording, signal preprocessing, feature extraction, and image-level decoding.
In this project, we used the NeuraDock EEG Workstation to replicate a NeurIPS 2024 visual decoding task with 7 dry EEG channels.
Visual decoding asks a difficult question: can we infer what a person is looking at from their EEG signals?
Recent research has shown impressive results using EEG embeddings, multimodal alignment, and image-generation pipelines. However, many of these experiments still rely on lab-grade wet-electrode systems, controlled environments, and hardware that is not easy for independent developers or smaller research teams to access.
NeuraDk was built to make EEG development more accessible. Instead of requiring a full lab setup, the system uses a wearable dry-electrode headset, a compact acquisition unit, and a software workflow designed for rapid BCI prototyping.
For this experiment, we wanted to evaluate whether this kind of portable EEG setup could support a serious visual decoding workflow.
System OverviewThe experiment follows a four-step pipeline:
Present visual stimuli to the user
- Present visual stimuli to the user
Record EEG responses through 7 dry electrodes
- Record EEG responses through 7 dry electrodes
Preprocess the EEG signal and extract features
- Preprocess the EEG signal and extract features
Use the decoding model to retrieve or reconstruct the image most aligned with the EEG response
- Use the decoding model to retrieve or reconstruct the image most aligned with the EEG response
The goal was not to claim that 7 dry electrodes can replace full lab-grade systems in every setting. Instead, we wanted to test how far a compact EEG setup could go when paired with careful preprocessing, timing calibration, and additional data collection.
Hardware SetupThe experiment used the NeuraDock EEG Workstation, a 7-channel dry-electrode EEG platform designed for BCI research and prototyping.
The headset uses dry comb-style electrodes, so no conductive gel is required. This reduces setup time and makes repeated data collection easier outside of a traditional lab environment
Compared with wet-electrode lab systems, the biggest advantage is accessibility: setup is faster, the hardware is easier to wear, and experiments can be run in ordinary environments. The tradeoff is lower signal-to-noise ratio and lower spatial resolution, which must be handled through engineering choices.
Experiment DesignWe used a visual decoding task inspired by recent EEG-based image decoding research. The participant viewed natural images while EEG signals were recorded. The model then attempted to identify which image best matched the recorded EEG response.
The main technical challenge was that dry electrodes produce noisier signals than wet electrodes. To compensate, we focused on three areas:
signal filtering
- signal filtering
timing calibration
- timing calibration
increasing the amount of training data
- increasing the amount of training data
Because the setup used software-based markers instead of a dedicated hardware trigger line, we also needed to estimate and compensate for timing delay between the visual stimulus and the recorded EEG response.
We tested multiple filtering ranges and delay offsets to find a stable processing configuration.
The best overall tradeoff came from preserving visual features while reducing power-line interference and high-frequency muscle artifacts. Timing calibration also proved important: the strongest performance clustered around a short post-stimulus correction window, which helped compensate for system latency.
This part of the work was critical. With a portable EEG device, the model performance does not come from hardware alone. It depends on the full pipeline: acquisition quality, preprocessing, calibration, data scale, and model design.
The 7-channel dry-electrode setup was able to reproduce the core visual decoding workflow and reach comparable performance on several retrieval tasks.
In our benchmark, NeuraDock achieved approximately 20.5% Top-1 accuracy in a 7-channel visual decoding setting. Top-5 accuracy reached around 45%. For 2-way and 4-way retrieval tasks, the performance was comparable to the reference benchmark.
These results suggest that dry-electrode EEG can be a viable signal source for visual decoding research when combined with sufficient data collection and careful preprocessing.
The system does not eliminate the gap between dry-electrode and wet-electrode EEG. Instead, it shows that part of the gap can be bridged through engineering.
What Worked WellThe most encouraging result was that a compact dry-electrode system could support an advanced EEG decoding workflow in a non-lab environment.
Three things mattered most:
A visual-area-focused 7-channel layout
- A visual-area-focused 7-channel layout
Careful timing and filtering calibration
- Careful timing and filtering calibration
Additional training data to compensate for lower SNR
- Additional training data to compensate for lower SNR
This suggests that portable BCI research does not always need to start with large, expensive hardware. For many early-stage experiments, a smaller system can be enough to validate workflows, test models, and build interactive prototypes.
LimitationsThere are still clear limitations.
Fine-grained image reconstruction remains difficult with only 7 dry EEG channels. Spatial resolution is limited compared with 64-channel systems, and high-N retrieval tasks become harder as the number of candidate images increases.
Reconstructed images can capture semantic structure and basic contours, but they are not yet comparable to higher-resolution neural decoding systems such as fMRI-based pipelines.
This project should be understood as an engineering validation, not a claim that dry EEG fully replaces lab-grade EEG. The value is that advanced BCI experimentation becomes more portable, repeatable, and accessible.
What This Means for DevelopersFor developers working on BCI, neuro-HCI, XR, AI interaction, or EEG-based cognitive systems, this experiment shows that meaningful visual decoding work can begin with a much smaller hardware setup.
A 7-channel dry-electrode EEG workstation can support:
EEG signal acquisition
- EEG signal acquisition
visual stimulation experiments
- visual stimulation experiments
preprocessing and feature extraction
- preprocessing and feature extraction
AI-assisted decoding workflows
- AI-assisted decoding workflows
reproducible BCI prototyping
- reproducible BCI prototyping
This lowers the barrier for teams who want to experiment with EEG-based AI interaction but do not have access to a full neuroscience lab.
NeuraDock on Crowd SupplyNeuraDock EEG Workstation is now in pre-launch on Crowd Supply.
We are building NeuraDock as an accessible EEG development platform for researchers, developers, and neurotechnology teams. The goal is to make BCI experimentation easier to start, easier to repeat, and easier to integrate into real-world workflows.
Follow the Crowd Supply campaign here:
https://www.crowdsupply.com/neuradock/neuradock-eeg-workstation
Official website:
GitHub:
Comments