Classification of Patients with Dementia and Parkinson’s with Support Vector Machine

Clinical findings and magnetic resonance images are used to diagnose dementia and Parkinson’s, which are neurodegenerative diseases. Studies are carried out for prediction of neurodegenerative diseases and/or for the efficient treatment process. Although there are studies that address and classify neurodegenerative diseases such as dementia and Parkinson’s individually with healthy control groups to accelerate the diagnosis stage, there are no studies that include two or more diseases at the same time. We aimed to make an original contribution to the literature by considering dementia and Parkinson’s diseases together. In our study, where we classify dementia and Parkinson’s on the differences in brain volume and brain thickness, we used the voxel-based morphometry technique for the volumetric measurement of shrinkage and the support vector machine (SVM) algorithm for the estimation of the diseased/healthy state. In our study, we used data from 3082 participants, 1141 with Parkinson’s disease, 1145 with dementia, and 796 with healthy controls. Datasets were obtained from NFID and PPMI. Atlases in SPM12 and CAT12 libraries in the MATLAB program were used for brain-volume measurements, cerebral cortex thickness, and area calculations. Obtained volume measurements were classified by the support vector machine algorithm and 10-fold cross-validation method was used for validation. As a result of the study, the mean accuracy rate of 93.51% and the mean AUC-ROC value of 97.59% were obtained. These findings show that although there are two diseases and a healthy group, it gives better results than some studies in the literature with dementia and control group or Parkinson’s and control group. As a result, the SVM algorithm is in parallel with the studies conducted with dementia and control group or Parkinson’s and control group, and it has been proven that it can be an effective method in a database containing both disease and control data.