Parkinson's disease (PD) is an overwhelming neurodegener-ative disorder caused by deterioration of a neurotransmitter, known as dopamine. Lack of this chemical messenger in the brain impairs several brain regions and yields to various movement and non-motor symptoms. The incidence of PD is considered to be doubled in the next two decades and this urges more researches on its early diagnosis and treatment. In this paper, we propose an approach to diagnose PD using magnetic resonance imaging (MRI) data. We first introduce a joint feature-sample selection method to select the optimal subset of samples and features for a reliable training process. This procedure selects the most discriminative features and discards poor sample (outliers). Then, a robust classification framework is proposed that can simultaneously de-noise the selected subset of features and samples, and learn a classification model. Our model can further de-noise the test samples based on the cleaned training data. Experimental results on both synthetic and a publicly available PD dataset show promising results.
展开▼
