Multi-class & binary classification of Parkinson’s disease and SWEDD variants using SBR features derived from SPECT imaging

Abstract

Parkinson’s disease (PD) becomes the second most common disease and is caused by the loss of dopamine neurons. It is very challenging to detect the disease at an early stage and the chances of misdiagnosis are high due to the similarity of symptoms to other disorders. Therefore, this paper proposed a deep learning-based automatic deep neural network (DNN) model for diagnosing disease as early as possible. The binary and multiclass classifications have been done among three classes (PD, SWEDD, and Healthy people) by using SBR features of SPECT modality. Even also compared the results of the proposed model of both classification categories with other highly used machine learning algorithms (Support vector machines, Naive Bayes, k-nearest neighbors, and decision trees) in literature. From the outcomes, it has been observed that the proposed DNN model provided the highest accuracy in comparison with other classifiers i.e. 97.67%, 83.43%, 94.46%, & 83.18% for PD vs Healthy, SWEDD vs Healthy, PD vs SWEDD, and PD vs SWEDD vs healthy classification probabilities respectively. Hence, this automatic DNN classifier may aid professionals to diagnose the disease in an early stage.