Aprendizado de máquina aplicado ao auxílio do diagnóstico da doença de Parkinson

Abstract

Currently, it is not a trivial task to point out a test that can diagnose accurately enough a patient with Parkinson’s Disease, as well as it is quit difficult to assess the level of the disease. Experts recommend the application of different types of tests, many of them based on signs and biomedical imaging, such as electroencephalogram, computed tomography and magnetic resonance to aid the detection of the disease process, since as the age ranges, symptoms such as fatigue and weakness can hide diagnosis. In order to provide a more effective clinical information to doctors aiming at diagnosis with greater confidence, methodologies to perform the fusion of different imaging modalities have become increasingly popular and promising. Recently, the use of forms containing some activities using a biometric pen with multi-sensors have been applied for the detection of Parkinson’s Disease by means of handwriting analysis. However, information derived from the scanned image of the form itself, and the one obtained by same pen have not been used together for this purpose. Thus, this proposal aims using pattern recognition techniques and image processing aimed at using the information from the form together with data from the pen. We believe a possible improvement in the medical diagnosis of Parkinson’s Disease can be archived. Another contribution of this proposal, is the design of a multimodal database to aid in the diagnosis of Parkinson’s Disease.