EARLY PARKINSON’S DISEASE DETECTION USING MULTI-MODAL FEATURES AND MACHINE LEARNING ALGORITHMS

Parkinson’s Disease (PD) is a chronic neurodegenerative disease, which affects millions of people worldwide. Accurate detection of early PD can assist clinicians implement medical and therapeutic interventions for PD patients in a timely manner, and effectively improve the quality of patients’ life. In this study, we utilize statistically significant features extracted from Movement Disorder Society-Unified Parkinson’s Disease Rating Scale (MDS_UPDRS) and DaTscan ([Formula: see text] I-Ioflupane) Single Photon Emission Computed Tomography (SPECT) imaging to develop the early PD detection models based on machine learning algorithms, such as Logistic Regression (LR), Random Forest (RF), and Support Vector Machine (SVM). The proposed early PD detection method comprises following steps: data collection, feature selection, classification modeling and early PD detection. The feature importance is also estimated using the RF model. Totally, 61 multi-modal features composed of 57 MDS_UPDRS items and 4 Striatal Binding Ratio (SBR) values are statistically significant in early PD and Healthy Control (HC) groups and considered as input to construct early PD classification models. The constructed models with stratified 10-fold cross-validation achieved promising classification performances, with mean accuracy, sensitivity, and Area Under Curve (AUC) all exceeding 98.30% and mean specificity over 96.40%. The proposed method in this study outperforms the previous state-of-the-art early PD detection methods and can be considered as an effective way to assist clinicians in the diagnosis of early PD from the clinical perspective.