Recognizing Cardiac Magnetic Resonance Acquisition Planes

In this paper we propose a method for automatic wrangling of missing or noisy acquisition plane information of cardiac magnetic resonance images in order to simplify case filtering and image lookup in large collections of cardiac data. To recognize standard cardiac planes we use features based on image miniatures combined with a decision forest classifier. We show that augmenting the dataset with a set of nondestructive transformations can improve classification accuracy. Our approach compares favorably to the state of the art while requiring fewer manual annotations.