Morphological Neural Networks for Parkinson Detection through Speech Signals

This paper presents the implementation of morphological neural networks in the identification of subjects with Parkinson's disease. We use bio-markers from “Oxford Parkinson's Disease Screening”, which contains a total of 195 sustained voice donations with 32 patients male and female, of which 24 o them were diagnosed with Parkinson's disease and eight correspond to people healthy. Although different algorithms of machine learning have treated this problem, the use of dendrites morphological neural networks proved to have an excellent ability to identify subjects with Parkinson; the stochastic gradient descent learning algorithm obtained an accuracy of 94.74%, a precisión of 91.32%, a sensitivity of 86.98% and a specificity of 97.28%. These results are better than other sophisticated and proposed algorithms show in the results.