This document summarizes an automated method for detecting microaneurysms, hard exudates, and cotton wool spots in retinal fundus images. 130 retinal images were used from a public database to train and evaluate classifiers. Features were extracted from pre-processed images and candidate lesions were classified using support vector machines and k-nearest neighbors algorithms. The proposed method achieved 95.6% sensitivity, 94.87% specificity, and 95.38% accuracy in detecting lesions, outperforming random sampling. Active learning was shown to select more informative training samples than random sampling, improving classifier performance.