Parkinson Disease Identification using Residual Networks and Optimum-Path Forest

• Autor(a) principal: Passos, Leandro A.
• Data de Publicação: 2018
• Outros Autores: Pereira, Clayton R. [UNESP], Rezende, Edmar R. S., Carvalho, Tiago J., Weber, Silke A. T. [UNESP], Hook, Christian, Papa, Joao P. [UNESP], IEEE
• Tipo de documento: Artigo de conferência
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://hdl.handle.net/11449/186245
• topic: Parkinson's Disease; Residual Networks; Machine Learning

Known as one of the most significant neurodegenerative diseases of the central nervous system, Parkinson's disease (PD) has a combination of several symptoms, such as tremor, postural instability, loss of movements, depression, anxiety, and dementia, among others. For the medicine, to point an exam that can diagnose a patient with such illness is challenging due to the symptoms that are easily related to other diseases. Therefore, developing computational methods capable of identifying PD in its early stages has been of paramount importance in the scientific community. Thence, this paper proposes to use a deep neural network called ResNet-50 to learn the patterns and extract features from images draw by patients. Afterwards, the Optimum-Path Forest (OPF) classifier is employed to identify Parkinson's disease automatically, being the results compared against two well-known classifiers, i.e., Support Vector Machines and the Bayes, as well as the ones provided by ResNet-50 itself. The experiments showed promising results concerning OPF, reaching over 96% of identification rate.