Do computers know EEG reading?

This is how neurologists check frequencies and amplitudes of EEG signals in the old days - I found this card from the back cover of a Japanese book (2006) about EEG reading.

The method is matching boxes, e.g., 1-30, to EEG waves. If the number of peaks matches box n, then the frequency of the wave is n Hz. Of course, a wave can have many frequency components and thus matches several boxes on different amplitudes, like lower frequencies modulated onto higher frequencies - just imaging how sin(x)*sin(4x) look like in your high school math class.

Now, with computerized EEG recordings, neurologists still are using this way to diagnose. The only difference is that the boxes on the card are replaced by different background rulers on computer screens. So are amplitude rulers.

The question is, whether representing the knowledge and methods from physicians is the only way we can teach computers to read EEG? If this is the only possible track, then how we can teach computers to identify different wave patterns, in time domain?  If the answer is yes, then a lotta work done in these years about spike detection is a killer approach.

But, "do submarines swim?" "They just don't swim in the way you swim." Dr. Dijkstra, who served as Computer Science Department chair at UT Austin between 1984- 2000.

So, another approach is to teach computers a whole new way to mathematically analyze EEG waveforms, not following the way that physicians use which identifies particular EEG patterns, like spike-slow waves. For example, we may teach computers that the skews in this EEG is different from skews in another EEG.

No comments: