Automated tremor amplitude and frequency determination from power spectra.

Analysis of tremor amplitude and frequency is often done in frequency domain from tremor power spectra. Several methods for tremor amplitude estimation from power spectra were described but have not been systematically compared and evaluated. In addition, current methods for tremor amplitude estimation from power spectra do not correct tremor amplitude when power spectra of adjacent peaks overlap and this could contribute to an overestimation of tremor amplitude. We developed a computer application for automated analysis of tremor power spectra that corrects tremor amplitude, when power spectra of adjacent peaks overlap, with a Gauss curve fitting method. The accuracy of previously described methods and new methods for tremor amplitude estimation from power spectra was evaluated by 78 simulated power spectra and 900 tremor power spectra records of 25 patients with Parkinson's disease. Thus the influence of peak amplitude, number of peaks and degree of peak overlapping on tremor amplitude estimation could be systematically evaluated and compared. We conclude that the Gauss curve fitting method (Gauss 1) enables a more accurate estimation of tremor amplitude than previously described methods.