A major aspect of research design that contributes to the efficacy of source localization is the number of EEG channels you are recording. I've always heard that 64 channels is the minimum necessary to get any reliable and valid source localization information from your EEG data, but more is desirable - the more channels you have, the better the localization algorithms work. You also want to keep in mind that EEG is recording cortical signals, so it is unwise to try to model any deep brain structure activity (e.g., basal ganglia, amygdala). Some people have done this, but they often have to show some very convincing corroborating evidence via fMRI. Do you have more information on what you are trying to do, what your experiment is targeting, etc.?

Source localization with EEG/ERP has bad spatial resolution. If you add more channels you can increase the quality of the localization but from 128 to 256 channels the gain is minimum. If you want to know where 'things happen' use fMRI, if you want to know when they happen use EEG.

Depends drastically on the source configuration. If there would be just a single local source ("current dipole"), its site could be found with just a few well-placed electrodes, with the spatial resolution depending on eg source depth and orientation, as well as on the SNR of the data. But in real case we never know in advance the number of sources... Thus I recommend you to read first some basics of EEG & source localization. Slightly biased, I would recommend our MEG–EEG Primer https://www.amazon.com/MEG-EEG-Primer-Riitta-Hari/dp/0190497777 ;-)

Good luck!

Good question. If you are exploring source localisation you could try different algorithms and different starting constraints to see how consistent the solutions are. Once you have got some experiment data and decided on an algorithm and constraints you could assess split-half reliability (e.g. split your electrode array by taking alternate electrodes, or randomly divide your participants into 2 groups and compare source solutions). Or ideally as João Miguel Castelhano implied you could use fMRI localisation to constrain your ERP sources.

I share my colleague Hari’s view regarding the Variability of the source licalization results. Some sources are complex such as the P300 where several variables in the Localization pricedure can keadnti different results. Others are relatucely simpler, such as the SI sources which can be reliably localized and with excellent singke trial s/n. Appropriate use of preprocessing algorithms can make a huge difference in the Objectivity, Reliability (Both within and across participant), and computational efficiency, and real time processing feasibility. You can find more at https://nfe.edu.hku.hk/publication/papers/ (scroll down to EEG source imaging). Th

Brittany K. Taylor

Dear Dr Brittany Taylor

Thanks for your answer. I used a 64-channel Active-Two acquisition system (BioSemi). At present, the EEG system at my school is an elasticised 128-channel HydroCel Geodesic Sensor Nets (Electrical Geodesics). There are not enough channels to get a reliable and valid source localisation. In addition, thanks for your reminder that an EEG records cortical signals. It is a valid point. I am planning to employ N300 and P300 to investigate the body with emotional valence. Moreover, I will examine the lateralisation of N300 and P300. The source localisation is just a sub-purpose of the research that I am considering.

Best regard

Chen

João Miguel Castelhano

Dear Dr João Miguel Castelhano

Thanks for your answer. So far, the maximum of EEG channels in my lab is 128 channels. Even if I can do source localisation analysis, the quality of the result will not be good. Additionally, I think your comment is like my supervisor: the quality of the EEG source localisation is not good enough to know where ‘things happen’; it can only demonstrate a lateralisation of the brain.

Best regards

Chen

Riitta Hari

Dear Professor Hari

Thanks for your answer. Depending on your comments, I suppose I could use source localisation to investigate my research, but I have to consider various issues of analysis. Additionally, I will borrow the book.

Best regards

Chen

Michael John Wright

Dear Professor Wright

Thanks for your answer. I did not consider the issue of algorithms and constraints. I am learning the method of source localisation. Thus, I have not yet understood ERP source localisation entirely. I realised that regardless of algorithms, the feature of EEG leads the low reliability of ERP source localisation. I will account for them when I do the research. In addition, it will be helpful to conduct the source localisation depending on your offering method. I will adopt the former methods to assess the reliability because I could not use fMRI in my school.

Best regards

Chen

Akaysha Tang

Dear Professor Tang

Thanks for your answer. I did not consider the diversity of analysing localisation among different ERPs. I will take it into account when I do source localisation. Additionally, only consideration of the reliability of ERP source localisation simplifies the problem. I will read the paper that you offered to figure out more about EEG source imaging.

Best regards

Chen

I recently used a combination of group-wise PCA and multiple sparse priors for source localisation of ERF in MEG. MSP was run on the spatial weights of first PC, but you could run separately on 1st, 2nd etc. I chose MSP because I thought there would probably be multiple source dipoles contributing to the ERF. This worked very well in Auditory cortex. See here (Price et al, 2017, Nature Comms; https://www.nature.com/articles/ncomms15671). The visual cortex looked perhaps too high in hindsight, but this is probably because of the strange shape of the visual cortex in the MNI brain (MSP uses the MNI space colin brain mesh for dipole reconstruction). I suspect things would have been better had I used individual surfaces etc. but that is not currently supported in SPM, and would have taken some considerable hacking.

I would also agree with Riitta Hari's reply. There are also a number of papers specifically addressing the question of source localization error (e.g. Michel et al., 2004 Clinical Neurophysiology; Andino et al. 2005 Experimental Brain Research, and many others) as well as issues of data preprocessing prior to source estimation (e.g. Murray et al., 2008 Brain Topography; Tivadar and Murray 2019 Organizational Research Methods and many others). This is a non-trivial issue indeed. However, I would not forcibly restrict source reconstructions to fMRI activations, as there are a number of strong assumptions behind such a practice (though there are also situations where such assumptions may be warranted). I would also not personally advise splitting your electrode montage in half as source estimations can be strongly influenced by the number and distribution of electrodes. Again, these and many other issues are nicely covered in Riitta Hari's and Aina Puce's primer as well as in many review articles too (e.g. Michel and Murray 2012 Neuroimage).

Happy to continue discussing these and related points. Good luck!

Darren Price

Dear Dr Price

Thanks for your reply; sorry I took so long to get back to you. I am glad you share your research with me, but I am unfamiliar with ERF. I am reading your journal, and your study is gripping. If I could approach MEG, I will go with your way to approach my research.

Regards

Chen

Micah M Murray

Dear Dr Murray

Thanks for your reply; sorry I took so long to get back to you. I will read these journals which you mentioned. Indeed, the issue of source localisation should not be simplified by reliability. Concerning source localisation of fMRI and EEG, I will read more journals to comprehend them. I am sure the book ‘MEG-EEG Primer’ is a good choice.

Regards

Chen