Fractal analysis of the EEG and clinical applications

Most of the knowledge about physiological systems has been learned using linear system theory. The randomness of many biomedical signals has been traditionally ascribed to a noise-like behavior. An alternative explanation for the irregular behavior observed in systems which do not seem to be inherently stochastic is provided by one of the most striking mathematical developments of the past few decades, i.e., chaos theory. Chaos theory suggests that random-like behavior can arise in some deterministic nonlinear systems with just a few degrees of freedom. One of the most evocative aspects of deterministic chaos is the concept of fractal geometry. Fractal structure, characterized by self-similarity and noninteger dimension, is displayed in chaotic systems by a subset of the phase space known as strange attractor. However, fractal properties are observed also in the unpredictable time evolution and in the 1/f^? power-law of many biomedical signals. The research activities carried out by the Author during the PhD program are concerned with the analysis of the fractal-like behavior of the EEG. The focus was set on those methods which evaluate the fractal geometry of the EEG in the time domain, in the hope of providing physicians and researchers with new valuable tools of low computational cost for the EEG analysis. The performances of three widely used techniques for the direct estimation of the fractal dimension of the EEG were compared and the accuracy of the fBm scaling relationship, often used to obtain indirect estimates from the slope of the spectral density, was assessed. Direct estimation with Higuchi's algorithm turned out to be the most suitable methodology, producing correct estimates of the fractal dimension of the electroencephalogram also on short traces, provided that minimum sampling rate required to avoid aliasing is used. Based on this result, Higuchi's fractal dimension was used to address three clinical issues which could involve abnormal complexity of neuronal brain activity: 1) the monitoring of carotid endarterectomy for the prevention of intraoperative stroke, 2) the assessment of the depth of anesthesia to monitor unconsciousness during surgery and 3) the analysis of the macro-structural organization of the EEG in autism with respect to mental retardation. The results of the clinical studies suggest that, although linear spectral analysis still represents a valuable tool for the investigation of the EEG, time domain fractal analysis provides additional information on brain functioning which traditional analysis cannot achieve, making use of techniques of low computational cost.