I would like to know if there is any tool or code suitable with Brainstorm that extracts ERPs from one trial?
I am studying epoch around an erroneous response from a memory task. Many subjects have only one or two errors in a category (aware or unaware).
Do you have any suggestion on how to deal with it since I can not average? I found methods in papers with features extraction and classification, some using ICA, some with LDA, deep learning, machine learning, SVM. I don't know if there is any integrated tool for this issue in Brainstorm.
I found topic about sLoreta but how can it help? Do I need to associate Brainstorm with any other software like EEGlab?
The EEG signal was recorded on 256 channels, rereferenced offline to LM/RM (Mastoid) and the frequency sampling is 250Hz. Pre-processing includes band-pass filter [0.1-40Hz], one epochs time window is [-200,600ms]
An event-related potential (ERP) is the electrophysiological response to a stimulus. It is present in all your trials, just buried in noise. The goal of averaging multiple trials is to remove all the "noise", ie. the components of the signals that are not time-locked to the stimulus.
EEG is a technique that is most of the time too noisy to allow single-trial analysis. Except maybe with newborns, who still have a very conductive skull.
The amplitude of your brain signal of interest is most likely much lower than all the other physiological and instrumental noise recorded at the same time. With advanced signal processing techniques (ICA, spatial filtering...) you could maybe get the impression you extract one component of the signals that is representing your brain activity of interest, in order to measure the amplitude of one well-known ERP component. You can probably find articles justifying that single-trial EEG for cognitive studies is a valid approach, and that you can do great statistics with only N=4 subjects.
But in my opinion, you should not rely on a single EEG trial to assess anything. I would not trust any of these extreme denoising techniques, you have no ground truth to decide whether what you select makes sense or not.
There are many publications about the minimum number of repetitions necessary to get a decent SNR in your ERPs. Start with the review of this literature, and you might realize that there is a methodological issue in the design of your experiment.
Thank you for your answer.
In our case, we have 28 subjects in a group so we can do grand average from all subjects but many of the subjects have 1 or 2 epochs on the studied response category.
So if applying a minimum number of epochs from one subject to be included, there would be only 4 subjects (having >=8 epochs).
We didn't design the experiment, we work on data from it.