Parkinson's disease classification by residual network type 1-d CNN using vocal datasets

Abstract

For the diagnosis and early detection of Parkinson's disease, a noninvasive method based on observed abnormal motor signs is desired. Recently, Gundu'z successfully reported a novel Parkinson's disease classification using vocal signal datasets with deep learning techniques. We have tried to improve classification result of Gunduz's one (previous paper). In this paper, a novel-residual-network-type 1-d CNN were introduced for Parkinson's disease classification using vocal feature datasets that was also used by the previous paper. Whereas the previous paper provided classification result with an accuracy of 0.857, F-measure of 0.910, and MCC of 0.594, our resulting residual-type network gives better results of accuracy of 0.888, F-measure of 0.928, and MCC of 0.692.Reference[1] Hakan Gunduz, "Deep learning based Parkinson's disease classification using vocal feature sets," IEEE Access, vol. 7, pp. 115540-115551, August 2019.This work was partly supported by JSPS KAKENHI Grant Number JP18K19844 and JP18H03514.