Automatic Heart Segmentation

From LRDE

Abstract

Atrial fibrillation is a common illness, which can be detected easily if you can localize the human atria in a MRI image. Manual segmentation is labor intensivetherefore here we adapt an automatic method developped for brain image segmentation. Using transfer learning, we retrain a convolutional neural network used for natural image categorization (VGG), and compare the advantages of using a pseudo-3D technique. We also explore ways to preprocess input data to improve final results.