Spike detection in EEG through random line segments
Erik Bølviken, Geir Olve Storvik, Gutorm Høgåsen, Pål G. Larsson · NORA - Norwegian Open Research Archives · 1996
A major difficulty in automatic spike detection in EEG is variability in the patterns of interest and in the frequencies of the background signal. The paper presents a methodology which attempts to deal with this through detailed stochastic modelling. Spikes are defined mathematically by superimposing sequences of random line segments on the running background process. The slope and duration of these line segments are allowed to vary randomly around average values. At each point in time the likelihood of such a pattern being present is evaluated recursively. Real time processing is possible. The approach enables us to feed information about randomness in shape, and to estimate tuning parameters empirically. Preliminary test results are given.