Adaptive frequency decomposition of EEG with subsequent expert system analysis

We present a hybrid system for automatic analysis of clinical routine EEG, comprising a spectral analysis and an expert system. EEG raw data are trans formed into the time-frequency domain by the so-called adaptive frequency d ecomposition. The resulting frequency components are converted into pseudo- linguistic facts via fuzzification. Finally, an expert system applies symbo lic rules formulated by the neurologist to evaluate the extracted EEG featu res. The system detects artefacts, describes alpha rhythm by frequency, amp litude, and stability and after artefact rejection detects pathologic slow activity. All results are displayed as linguistic terms, numerical values a nd maps of temporal extent, giving an overview about the clinical routine E EG. (C) 2001 Elsevier Science Ltd. All rights reserved.