Neuroparknet: A New Deep Neural Network Model for Classification of Parkinson's Disease

Abstract

In recent years, the volume and variety of biological data being acquired have increased significantly. Among these data types, the diagnosis of Parkinson's disease holds a critical place in medical research. For this study, speech signals were recorded from patients and healthy controls in a controlled environment at the Neurology Department of F & imath;rat University Hospital. 28 healthy controls, 22 Med Off patients and 30 Med On patients constituted our data set. Participants were asked to read a standardized text in a quiet room using a high-quality H1N Zoom microphone. 19 features were extracted from the obtained sounds. The dataset was categorized into three distinct classes: Healthy Control, Med Off (patients without medication), and Med On (patients medication). To evaluate classification performance, we used a three-layer deeep neural network (DNN) model as well as classical machine learning algorithms in MATLAB. Various classification scenarios have been considered, including many different combinations. For benchmarking, the DNN results were compared with those from commonly used algorithms in the literature: K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Decision Tree (DT), and Naive Bayes (NB). Furthermore, the DNN model's performance was assessed using the NeuroParkNet architecture. The comparative analysis revealed that the DNN model generally provided a more accurate and efficient classification process. However, in some specific cases, its performance was partially outperformed by traditional classification algorithms. These findings highlight the DNN's potential while also underscoring areas for optimization in Parkinson's disease classification systems. In addition, the effects of pharmacological treatments were also evaluated in this study.