Texture-based Classification for the Automatic Rating of the Perivascular Spaces in Brain MRI

Perivascular spaces (PVS) relate to poor cognition, depression in older age, Parkinson's disease, inflammation, hypertension and cerebral small vessel disease when they are enlarged and visible in magnetic resonance imaging (MRI). In this paper we explore how to classify the density of the enlarged PVS in the basal ganglia (BG) using texture description of structural brain MRI. The texture of the BG region is described by means of first order statistics and features derived from the co-occurrence matrix, both computed from the original image and the coefficients yielded by the discrete wavelet transform (WSF and WCF, respectively), and local binary patterns (LBP). Experimental results with a Support Vector Machine (SVM) classifier show that WCF achieves an accuracy of 80.03%.