Electroencephalography and science

Every experienced electroencephalographer has his or her personal approach to EEG interpretation. (...) there is an element of science and element of art in a good EEG interpretation; it is the latter that defies standardization

...writes prof. Ernst Niedermayer in the recent edition of a fundamental reference [10]. On the contrary, physical sciences are based upon strictly repeatable experiments and unquestionable mathematical procedures. In the biomedical sciences, the experimental part is far more difficult, due to inherent inter-subject variability and the inability to control all of the experiment's conditions, especially in the studies of complex systems like the human brain. Therefore, an objective and standardized methodology of analysis is of paramount importance for the evaluation of results. Any subjective factor in the procedure of data analysis may turn the unavoidable noise into methodology-induced bias, where variation of results may reflect differences in EEG interpretation rather than inter-subject variability. Lack of a standardized and widely accepted methodology of EEG parametrization may be one of the reasons¹ for the fact, which we would hereby like to highlight: judging from the perspective of time, we can hardly notice a steady and coherent progress in the field of automatic EEG analysis. This statement may be considered controversial--however, trying to enumerate the widely accepted and applied methods of EEG analysis, it's hard to name more than:

1. visual analysis of raw EEG traces,
2. Fourier estimation of spectral power in selected frequency bands,
3. description of evoked potentials, averaged in the time domain.

In the following we'll assume that the knowledge base of EEG analysis was formed via applications of these methods. To use this indispensable experience in the clinical practice and research, on a large scale and with repeatability required by the scientific methodology, we need a generally applicable algorithm for signal description (i.e. parametrization), which would encompass the above listed approaches. But how can we relate the visual analysis to mathematics? At least a part of it can be linked with a time-frequency description. Let's consider an example from a classical reference [13]:

The presence of sleep spindle should not be defined unless it is of at least 0.5 sec duration, i.e., one should be able to count 6 or 7 distinct waves within the half-second period. (...) The term should be used only to describe activity between 12 and 14 cps.

Mathematical implementation of such criteria allows for a scientific approach to the experience gained in decades of clinical applications, without the bias introduced by the inherent arbitrary factor, present in visual analysis. The following section presents an algorithm, which might serve as a basis for such attempt.

Footnotes

... reasons¹

At this point we should also indicate one more factor: non-reproducible research. In an attempt to reproduce some EEG research published in a journal article, the influence of inter-subject variability may be small compared to differences in details and parameters of the applied algorithms, which are usually not published, and hence impossible to reproduce [2].