A novel technique for classifying Parkinson's disease using structural MRI scans

Parkinson's is a movement-related progressive disorder that causes irreversible damage and heavily derails the patient's life. PD is generally detected via inspection of symptoms and family history by neurologists which may be inefficient, time-consuming and costly to the patient. Although the motor-related signs act as the major contributor to Parkinson's detection, these occur in the later stage of the disease when the symptoms have already intensified. Henceforth, there is a necessity for a Computer Aided Diagnosis (CAD) for an early yet accurate classification of Parkinson's. This work proposes a fusion of 3D- Discrete Wavelet Transform (DWT) and a variant of 3D Local Binary Pattern (LBP) on 3D T1-weighted structural Magnetic Resonance Imaging (sMRI) scans acquired from two publically available databases namely: PPMI and SWADESH for Parkinson's diagnosis. The International dataset has been retrieved from the PPMI database and the Indian dataset has been acquired from SWADESH database. Features have been extracted from the gray matter. Fisher Discriminant Ratio (FDR) has been utilized for selecting relevant features. Support Vector Machine, Logistic Regression, Random forest and K-NN have been used for performing classification. The proposed technique achieved a performance accuracy of 92.45% and 90.57% for data acquired from the PPMI database and SWADESH database respectively. It outperformed the state-of-art techniques with respect to accuracy, sensitivity and specificity. From the results, it has been concluded that the suggested technique has the potential for an early yet effective identification of Parkinson's and can be utilized for clinical diagnosis by neurologists.