Deep CNN Based Classification of the Archimedes Spiral Drawing Tests to Support Diagnostics of the Parkinson’s Disease
Sven Nõmm, Sergei Zarembo, Kadri Medijainen, Pille Taba, Aaro Toomela
Abstract
Application of the deep convolution neural networks to distinguish between the Archimedes spiral drawing testes produced by the Parkinson’s disease patients and healthy control subjects is discussed in the present paper. Majority of the existing results for the spiral test analysis are based on the kinematic and geometric features, whereas the application of deep learning techniques did not get much attention. Such approach excludes the shape of the drawn curve as the entity to be analysed. The approach proposed in this paper combines both kinematic and pressure features on the one side and shape of the drawn line on the other side. The present research is spanned around the following novel points. Test drawing is enhanced to incorporate kinematic and pressure parameters of the drawing. Data augmentation procedure is then applied to provide a sufficiently large dataset to train a convolution neural network. The goodness of the trained model exceeds those of shallow classifiers.