Pool-Based Active Learning with Proper Topological Regions