Ensemble of convolutional neural networks for Parkinson’s disease diagnosis from offline handwriting

Abstract

This paper proposes the ensemble of deep convolutional neural networks for diagnosing Parkinson’s disease from offline handwriting. The advantage of the offline approach lies in the fact that handwriting acquisition can be performed without any specialized equipment by using only a smartphone camera. The convolutional neural networks ensemble relies on pre-trained networks where the diversity is achieved through the multiple-fine-tuning of individual networks. The experimental results on two handwriting datasets showed that the proposed approach currently provides the highest classification accuracy compared to other strategies for diagnosing Parkinson’s disease based on offline handwriting.