GAIT SPECTRAL ANALYSIS: AN EASY FAST QUANTITATIVE METHOD FOR DIAGNOSING PARKINSON'S DISEASE

Abstract

At present, there is no quantitative test to definitely diagnose Parkinson's disease (PD). For this purpose, we computed the power spectra of stride and swing signals of normal persons and patients. The evaluation of power spectra in stride on normal group shows that the main peak of the frequency range is in the range of 0.018 to 0.02 Hz. In contrast, the main peak frequency is different in different PD patients. Our studies on swing signal and its power spectra show that there is a significant difference between the amplitude of frequency components between normal and PD groups. Patients show power spectra amplitude even more than 10 times that of normal patients. The clinical data were obtained from http://www.physionet.org. For measuring time intervals, force sensors were used in the plantar portion of the foot. Power spectra of left stride, right stride, and left swing were computed. Frequency domain of power spectra was divided into 10 parts and then the surface area under each part was calculated. We used artificial neural network for classification of these groups. The clinical data was divided into two parts, training and test sets. An accuracy of 93.75% was obtained during training. The test data was used for validation of the classifier and an accuracy of 92.86% was obtained. The proposed classifier may be used as a tool for helping the clinicians to diagnose PD. Surely the final diagnosis should be obtained by an expert neurologist.