A novel GA-ELM approach for Parkinson's disease detection using brain structural T1-weighted MRI data

Parkinson's disease is the second most common neurodegenerative disorder caused by progressive loss of dopamine in substantia nigra. Various techniques like Magnetic Resonance Imaging (MRI), functional MRI (fMRI), and Positron emission tomography (PET) could be used to enumerate the loss of neurons in different parts of brain. In this paper we present a novel approach for detecting PD using brain MRI scans. Because of non-invasiveness and high resolution property, MRI is preferred over other techniques. For this study, the MRI images (healthy/PD patients) have been collected from Parkinson's Progression Markers Initiative (PPMI) organization. Research efforts have stated that Extreme Learning Machine (ELM) has better and accurate diagnosis ability. In this paper, PD diagnosis based on ELM-based method along with Genetic Algorithm feature subset selection has been proposed. The classifier uses voxel based morphometric features extracted from MRI. Since, the feature extracted are large in number, a feature subset selection technique using Genetic Algorithm is implemented. The performance of GA-ELM method is evaluated using classification accuracy, sensitivity, specificity. The results show that the classification accuracy obtained for ELM model is higher than the one obtained using SVM approach. Also GA-ELM classifier model produces a better generalization performance with high sensitivity and low misclassification rate.