Rede Neural Multicamadas para Classificação de Doenças Neurodegenerativas a partir de Sinais de Marcha

Abstract

Neurodegenerative disease is a general term to describe various diseases that affect the neurons, causing motor and physiological symptoms, such as involuntary movements, weakness, and gait problems. Neurodegenerative diseases do not have cure, and diagnosing them, especially in early cases, is a hard task, due to the lack of definitive tests for many of them. In this paper, we investigate the problem of identifying subjects with Parkinson’s Disease (PD), Huntington’s Disease (HD), Amyotrophic Lateral Sclerosis (ALS), and healthy control subjects using gait dynamics. This multiclassification problem is tackled with a multilayer feedforward neural network fed with statistical and signal features extracted from time series obtained from the left and right foot using force-sensitive resistors. In contrast to other approaches, we have used only one minute of gait signal, focusing on minimizing the patients’ effort, who may already have difficulties walking continuously and without aid, even for short periods of time, including those in the initial stages of the neural disease. Our results show that 96.88% of accuracy may be achieved for this multiclassification problem, surpassing other results in the literature, and pointing to a step close to aid the diagnosis of neurodegenerative diseases.