Identification of Early-stage Parkinson's Disease Utilizing Graph Theory and Machine Learning

A new attempt for combining machine learning and node parameters based on graph theory is described in this paper in order to find effective features in the early stage of Parkinson's disease (PD). A classifier based on graph measures provides informative predictions. A classification model using support vector machines (SVM) algorithm was constructed to differentiate PD from healthy controls (HC). Six node parameters all achieved a high classification power with accuracy of 96.97% for betweenness centrality, 99.00% for degree centrality, 97.92% for cluster coefficients, 94.06% for local efficiency and 95.84% for shortest path lengths and 96.74% for nodal efficiency. Left median cingulate and paracingulate gyri and right posterior cingulate gyrus were related to non-motor aspects of experiences of daily living. The left supplementary motor area, right posterior cingulate gyrus and right superior occipital gyrus involved with motor experiences of daily living. Left hippocampus and left insula were associated with motor symptoms. Our results suggest that node parameters based on graph theory have high diagnostic efficiency in distinguish early-stage PD from HC, and machine learning methods may provide a new perspective in PD network patterns.