Publication

Abstract : Machine learning models are being increasingly proposed for the automated classification of Parkinson’s disease from brain imaging data such as magnetic resonance imaging (MRI). However, the problem of class imbalance is a major setback in deriving the maximum benefit from using these techniques. We propose in overcoming the class imbalance between Parkinson’s disease (PD) and normal cohorts (NC) subjects by using variational autoencoders (VAEs) as the generative model that follows a probabilistic regeneration and aim to experiment over the effect of latent variables in generating new MR images of the subjects to improve the detection of PD. The efficiency of the proposed method with and without data augmentation is compared and evaluated using a deep learning classifier model at the subject level. The results obtained using a model trained with data augmentation show a significant increase of 6% in the performance of the classification model.