Reconocimiento de patrones relacionados a la enfermedad de Parkinson usando sensores inerciales e Inteligencia Artificial

Abstract

Parkinson's disease (PD) is a currently incurable disease that affects 1.5% of people over the age of 65. Early detection of the disease would improve the quality of life of patients and their families, as it would allow for more timely preventive actions than detection at an advanced stage. One of the symptoms that usually appear in the early stages of PD are modified walking patterns, which alter the movement of arms and legs. Based on this, the analysis of these movements using Artificial Intelligence techniques is proposed to capture patterns to classify patients with PD from healthy subjects. For the development of this solution, data taken with IMU sensors during the walking of healthy subjects and patients with PD, located on the wrists and base of the spine of the subjects, from the ICESI University of Colombia, were used. Features, scalograms and temporal signals were extracted from this data, which were the inputs for ML Shallow and Deep Learning models, of which, the Red MLP model using the feature extraction method gave the best result, having 95% accuracy. This model was integrated into a Web application for medical use, where IMU signal files of patients can be uploaded and the probability of having PD can be predicted according to the model. Keywords: Parkinson's disease, Machine Learning, Deep Learning, IMU, Walking.