For the extensive background literature, please refer to the menu item “Further information”. A good overview is provided by the article in Wikipedia, see https://en.wikipedia.org/wiki/Electroencephalography.
The signal analysis of the EEG is a quantitative analysis of the group of “brain waves” created by the respective derivative assembly. In principle, both brain hemispheres should be recorded in the same suitable way.
The signal analysis can take place in the time domain, frequency domain or phase space. The high time resolution (in ms) allows de facto analysis in “real time”. A distinction is made between scalp electrodes (on the surface of the head), local surface electrodes on the neocortex of the brain and individual electrodes for recordings in deeper regions of the brain.
For the assembly and division of the electrodes in the internationally standardized 10/20 system, see more details under the above-mentioned Wikipedia hyperlink.
In the IASON project, we are principally only dealing with scalp electrodes, but their spatial resolution is limited. The measured electrical activity results from the synchronisation of the electrical oscillations of individual local neuron groups, which may well number in the millions. Only if this synchronized neuron activity, the firing (= discharging) of the neurons in a temporally coordinated system and sequence, occurs for large neuron groups, is it large enough that it can be measured on the head surface at this point as a temporally varying voltage. This is a very simplified representation. The book “Klinische Enzephalographie” (2012) by Zschocke et al. (see menu item “Further information” under “Books & Papers”) provides information about the details in an outstanding way in chapter 1.2 "The EEG as the sum of cortical field potentials".
From what has been said so far, it already follows that the EEG signal is not a stationary signal, i.e. one whose average value remains constant over a defined period of time. This is why one likes to speak of “quasi-stationary”, which, however, only describes the fact that the methods that can be used for stationary signals (time series), such as the Fast Fourier Transformation (FFT), see https://en.wikipedia.org/wiki/Fast_Fourier_transform, also want to be used (and indeed do) for the EEG in order to be able to make statements about the superimposed frequency components in the signal by means of spectral analysis. This is of course possible, but requires that the subject is in the so-called “resting state”, in which the brain acts as a kind of “autopilot”. All other cases of EEG analysis, with the exception of the sleep EEG, are expressions of conscious or pre-conscious processes, i.e. “events”, whose processing is reflected in the EEG in so-called ERPs, i.e. "event related potentials" (https://en.wikipedia.org/wiki/Event-related_potential). Therefore, when considering the EEG as an electrical reflection of mental states to which a meaning can be attributed as an event, caution is required when looking for EEG sections that are not “distorted” by any artifacts in order to be able to perform a quantitative analysis of the EEG signals.
An example by Prof. N. Birbaumer (University of Tübingen) from his video lecture "Medical Psychology and Medical Sociology I, 3rd hour in the winter semester 2001, at minute -32.31" (see https://timms.uni-tuebingen.de/tp/UT_20011112_001_medpsych_0001) should clarify the problem. There, the EEG signal of a flutist performing a piece by M. Ravel is shown, at the same time the time representation of the audio signal of the Ravel piece is shown. The exact timing of the flute solo is decisive for the performance of all other musicians, and the success of the whole performance depends on it.
One can now see in the video at the quoted position that, 2.5 s before the flautist begins his virtuoso solo, his EEG signal first drops off briefly, then rises sharply, then drops off again very sharply, and finally runs roughly as “normal” as before this excitation in the EEG signal. However, the flutist then plays an extremely virtuoso flute solo, the performance of which apparently does not result in any visible precipitation in the EEG signal, which looks just as “calm” as before the excitation phase. Actually, the event “flute solo” in the EEG should be marked as a composition of excitation phase and execution phase, if one wants to attach any significance to the event.
However, it can also be seen very clearly that it is not sufficient to perform a quantitative analysis with quiet EEG segments alone, without considering their background of meaning, because then quiet segments would actually be mixed with those that are also “quiet” (i.e. their amplitude does not exceed a certain threshold), but which actually belong to events with meaning, such as the above-mentioned flute solo.
The problem of artifacts (blinking eyes, muscle twitching etc.) is often solved automatically by subjecting all EEG channels to an "independent component analysis (ICA, see http://arnauddelorme.com/ica_for_dummies/), identifying the artifact by determining a component, subtracting this component and then, after recalculation to the modified “original signal”, eliminating the influence of the artifact. The application of this method requires preconditions which are certainly not always met by the EEG signal.
It is already assumed that the EEG signal adjusted by ICA has all the characteristics of the original signal. But it is not that simple, see the example of the flutist above. With ICA the influence of the disturbing artifact is determined and removed. This results in a de facto “new” signal which, according to the assumption, still fulfills all the requirements to contain “hidden information”, e.g. about the presence of Alzheimer's disease. However, the question must be asked here whether this procedure does not also obstruct ways to obtain the desired information.
Nevertheless, the “event problem” remains. This is also the point of view of G. Ulrich, who in his work The spontaneous resting EEG” (see menu item “Further reading” under “Books & Papers”) takes a comparable point of view when he claims that the waking EEG should be regarded as a reflection of different events. For the time being, we do not modify the original signal with ICA, but we look for “quiet” segments in the original EEG signal, whose identification is done according to certain heuristic criteria, see results of work package 4.1.
In the quiet areas, should be enough information available to determine AD specifics, especially when using complex quantitative methods to analyze the EEG signal. In the long run, however, we need a “meaningful (semantic) EEG”, subdivided into segments of identified “events” and “resting state” epochs whose existence is best guaranteed in the “default mode” of the brain, see https://en.wikipedia.org/wiki/Default_mode_network.
Even short-term EEG segments should provide meaningful information about whether the patient suffers from AD and in which stage he or she is, or whether signs of AD can already be seen in the EEG in the preclinical stage of AD, which these are and to what extent they allow a valid prediction of the further development of the disease. Thus, there is a great need for procedures that go beyond the methods usually used in EEG analyses, which are mostly illustrative and based on individual experience.
In order to go beyond spectral analysis in data analysis, quantitative data analysis is required, which is currently very poorly represented in clinical practice. The representations of the qEEG (quantitative EEG) on the internet are mostly incomplete, too specific (e.g. only related to neurofeedback) or generally unsatisfactory, see e.g. https://de.wikipedia.org/wiki/Quantitatives_EEG.
A very good summary of the qEEG is the dissertation by W. Tirsch: "Biomedical Relevance of Quantitative EEG Analysis", see "Further information" under menu item "Links".