O uso de modelos auto regressivos para a classificação de indivíduos acometidos pela doença de Parkinson

Abstract

The Parkinson's disease is the second most common neurodegenerative disease affecting up to six million people in the world. For several years the diagnosis of individuals with this disease has been obtained through scales and questionnaires. These methods are characterized as subjective measurements because the results depend on the experience of health professionals which increases the probability of errors. The correct evaluation of the disease is substantial to obtain information that contributes to the decision of the appropriate treatment for each subject. In this study, objective analysis methods were used to visualize and differentiate the characteristics of the movement between groups of individuals with Parkinson's disease submitted to drug treatment and neurologically healthy subjects. Signals collected from 26 subjects by an accelerometer were obtained from a study database (MACHADO, 2016) [4] and used as measures for the development of this work. Among the total of individuals, 10 of them are healthy and 16 are carriers of Parkinson's disease being treated with Levedopa. From this study (MACHADO, 2016) [4], we used only the signals resulting from the static task, in which the subject remained with the elbow in extension without voluntary movements and from the accelerometer belonging to the device located in the hand. From these data, the present research created a system to discriminate individuals affected by Parkinson's disease from healthy individuals using the parameters of an Auto Regressive model. Mathematical equations were modeled from 2 to 10 parameters in order to find which one best classifies the subjects in the corresponding groups. To obtain this result we used a pattern recognition method with performance indexes. The method proposed in this study was validated, therefore it is able to discriminate the two groups of individuals most effectively with a certain amount of parameters.