Detection of Parkinso’s disease using Convolutional Neural Networks and Data Augmentation with SPECT images

Parkinson’s Disease or PD, is syndrome related to humans’ brains which mostly has impact on the neurons producing dopamine inside the substantia nigra area. Despite the fact that this disease has been known for many years, accurate detection of PD in its initial stages is still a challenge for physicians and researchers. In this study, a deep neural network based on CNN is used to diagnose the disease, which is able to differentiate between patients with PD from healthy individuals based on specific type of images, namely SPECT images. The proposed method consists of these phases: preprocessing, training and testing/evaluation. 650 SPECT images were investigated in this study, taken from the PPMI database. Since the number of images in the dataset may not be sufficient for the training phase, a data augmentation phase was also added to the whole process. The architecture of the CNN used and the augmentation step on SPECT images are the novelties of this study. Simulation results compared with other classification methods, show an accuracy of 97.01%, recall of 96.61%, specificity of 96.61%, and an f1-score of 96.61%. Results of adding data augmentation also show an accuracy of 95.50%, recall of 98.88%, specificity of 97.82%, and an f1-score of 98.32%, which are promising compared to previous work.