Parkinsonism is the second most common neurodegenerative disease in the world. Its diagnosis usually relies on visual analysis of Emission Computed Tomography (SPECT) images acquired using 123I − io f lupane radiotracer. This aims to detect a deficit of dopamine transporters at the striatum. The use of Computer Aided tools for diagnosis based on statistical data processing and machine learning methods have significantly improved the diagnosis accuracy. In this paper we propose a classification method based on Deep Ranking which learns an embedding function that projects the source images into a new space in which samples belonging to the same class are closer to each other, while samples from different classes are moved apart. Moreover, the proposed approach introduces a new cost-sensitive loss function to avoid overfitting due to class imbalance (an usual issue in practical biomedical applications), along with label information to produce sparser embedding spaces. The experiments carried out in this work demonstrate the superiority of the proposed method, improving the diagnosis accuracy achieved by previous methodologies and validate our approach as an efficient way to construct linear classifiers.
