Supervised learning and dimension reduction techniques for quantification of retinal fluid in optical coherence tomography images