Regensburg 2000 – scientific programme
Parts | Days | Selection | Search | Downloads | Help
DY: Dynamik und Statistische Physik
DY 46: POSTER II
DY 46.62: Poster
Thursday, March 30, 2000, 15:00–18:00, D
Analyzing brain tumor related EEG signals using a generalized overcomplete ICA algorithm — •M. Habl1, Ch. Bauer1, Ch. Ziegaus1, F. Schulmeyer2, and E. W. Lang1 — 1Universität Regensburg, Institut für Biophysik und physikalische Biochemie, 93040 Regensburg — 2Klinikum der Universität Regensburg, Klinik für Neurochirurgie, 93042 Regensburg
Automatized, non-invasive and low-cost methods for detecting and characterizing brain tumors would be highly desirable. Scalp EEG has been used as a clinical tool for the diagnosis and treatment of brain deseases. Its use has been severely hampered by the dominance of artifactual signal components. Recently a statistical method, called ICA, which can extract statistically independent features from biomedical signal patterns has been shown to be easily implementable on artifical neural networks. ICA generally relies on two main assumptions: a) that source signals are linearly mixed to the observed senosr signals and b) that there are no more sources than sensors. The latter assumption has been relaxed by overcomplete ICA which seems to provide a method for dealing with more sources than sensors. In this study an overcomplete ICA algorithm modified by a kernel-based density estimation procedure is investigated to separate EEG signals from tumor patients into spatially independent source signals. The algorithm allows artifactual signals to be removed from the EEG and isolates brain related signals into single ICA components. Their backprojection onto the scalp sensors provides topographic relations useful for a meaningful interpretation by the experienced physician.