How Many EEG Channels Should I Use In My Research? [Less Than You Think!]

Wondering how many EEG channels you should use in your experiment? For a long time, researchers considered 32 EEG channels the “gold standard”. However, times are changing. Read on to find out more.

Why are 32 EEG channels the gold standard?

The reason so many researchers use 32 EEG channels is partly historical.

Scientists believed that, if applied correctly (using the 10-20 or 10-10 positioning system), the minimum number of EEG channels needed to reliably localize brain activity during Evoked Response Potentials (ERPs) was 32.

Even today, although there are many applications that require fewer channels (e.g., resting state potentials), and other applications that require more (e.g., spatiotemporal resolution in visual processing), most clinicians and researchers use 32 electrode set-ups.

While there is nothing technically wrong with this decision, research suggests that substantially fewer channels can produce similar results. So why would you use more?

Imagine hanging a picture on your wall using four nails on each corner. Perhaps you do this out of habit. If someone showed you how to hang the same picture on the same wall with only a single nail, that would be the preferred method.

It is the same in EEG research.

Why fewer EEG channels are better

Each additional EEG channel comes with an associated cost. Namely, it increases set-up time, material costs and cleaning time between sessions.

Long story short: using as few EEG electrodes as possible will save you time and money, and reduce participant discomfort. So, how can you reduce the number of EEG channels you use?

Soler et al. (2020) decided to find out. The researchers conducted a source localization study using an 8 and a 32 EEG channel set-up. The results were surprising.

Using advanced mathematical pre-processing techniques (see our article on EEG signal noise reduction for more), they found that they were able to achieve comparable results from each set-up.

How can you achieve the same data quality with fewer EEG channels?

Conventionally, there are several ways to achieve ERP source localization, all of which solve the EEG inverse problem. Common approaches include Minimum Norm Estimates (MNEs), low-resolution tomography (LORETA), and beamforming.

However, multiple sparse priors (MSP), which utilizes Bayesian statistics, is particularly effective here. For more information on how to use MSP for the EEG inverse problem, see Friston et al. (2008). 

Novelly, Soler et al. (2020) opted for Multivariate Empirical Mode Decomposition (MEMD) before applying their MSP analysis.

MEMD identifies independent modes in raw data. That is, it identifies independent natural frequencies in the EEG signals.

By using this MEMD technique, the raw EEG data is reconstructed by combining a few so-called Intrinsic Mode Functions. This eliminates a lot of noise from the recording, but retains enough information for the MSP-technique to function.

Testing the validity of pre-processing EEG channels

To test the validity of their approach, Soler et al. (2020) considered 18 EEG signal configurations. They compared the results of raw-MSP with MEMD-MSP in each configuration. The configurations were generated by considering:

• One, three and five active sources,
• Each of which generated a synthetic EEG signal using 8, 16, or 32 electrodes,
• Before two noise levels were added to each synthetic signal: 10 and -5 dB.

As well as simulated ERPs, Soler et al. (2020) used real-world data. The researchers used a multi-subject, multi-modal human neuroimaging dataset to evaluate the MEMD method.

The researchers used the Wasserstein metric to analyze the accuracy of the raw-MSP and MEMD-MSP approaches for all scenarios.

Importantly, the MEMD-MSP method permitted half the number of channels without substantial loss of accuracy. In fact, in several scenarios, the researchers used 8 channels instead of 32 and had sufficient accuracy to reconstruct the source.

Additionally, MEMD pre-processing significantly reduced the number of so-called ghost sources.

Figure 1 demonstrates how MEMD processing allowed the researchers to reduce the number of EEG channels they needed to use.

Conclusion

It is important to use as few EEG channels as possible when conducting research. The fewer the number of channels the more time and cost savings.

Soler et al. (2020) showed that, with pre-processing, it is possible to use far fewer EEG channels than previously thought.