Stroke detection based on the scaling properties of human EEG

We propose a new method of detecting stroke by use of electroencephalogram (EEG) time series. When detrended fluctuation analysis is applied to the data, it is found that there exist two scaling regions for every channel. Thus with the geodesic sensor nets used there are as many as 128 paris of scaling exponents for each subject. We then determine a stroke index S that is based on the normalized variances of those scaling exponents. It is shown that S=1.3 distinctly separates the 28 normal and stroke subjects we have studied. We also show that the effect of stroke on EEG signals is global, in contrast to the local effect revealed by radiological studies such as MRI.