Discriminação do tremor em repouso do punho entre indivíduos com e sem a doença de parkinson por meio de sensores inerciais e classificadores

Abstract

Parkinson's disease (PD) is a neurological disease that affects the motor system. Associated motor symptoms are muscle stiffness, bradykinesia, tremors, and gait disorders. The correct diagnosis, especially in the early stages, is fundamental for the quality of life of the individual with PD. However, the methods used for the diagnosis of PD are still based on subjective criteria with the use of scales such as the Unified Parkinson Disease Rating Scale (UPDRS). Therefore, the aim of this study is to verify whether a combination of characteristics extracted from signals from inercial sensors, through tremor at rest of the wrist, can discriminate data from individuals with PD from data of individuals without PD (in this study considered healthy). This study has the participation of 27 individuals, 15 with PD in the early stages (Hoehn and Yahr score 1 and 2) and 12 individuals without PD. Two units of inercial measurement (IMU) were positioned in the most tremor-affected limb in the population with PD and in the dominant limb of individuals without PD. One IMU was positioned on the back of the hand and the other on the posterior region of the forearm. The IMUs used in this study are equipped with three sensors, each, an accelerometer, a gyroscope and a magnetometer, so the complete system consists of 6 sensors. All 3 sensors are capable of detecting movement along the 3 axes. The characteristics extracted from the data are related to signal amplitude, frequency, entropy, variability and distribution form. In total, 108 characteristics were extracted (18 for each sensor). To assist in the classification process, tests were performed with different percentages of the extracted characteristics, according to the importance of these characteristics for the classification of individuals. The importance of the characteristics was determined by the Relief function of the software R. The analysis started with 10% of the most important characteristics, with an increase of 10 in 10% until reaching 100%. During the increment process, the sensitivity, specificity, accuracy and accuracy of the classifiers used were calculated. The classifiers used were: Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Naive Bayes (NB). The highest percentages of sensitivity and accuracy (86.6% and 84.4% respectively) were obtained with a combination of 10% of the characteristics by the KNN classifier. For precision and specificity (95.3% and 92.8%, respectively) the best results were obtained with NB using 100% of the characteristics. In view of the results obtained, it was possible to conclude that patients in early stages of PD can be successfully diagnosed using inercial sensors and classifiers for tremor at rest of the wrist.