A Transfer-Learning Approach to Image Segmentation Across Scanners by Maximizing Distribution Similarity

摘要: Many successful methods for biomedical image segmentation are based on supervised learning, where a algorithm is trained manually labeled training data. For supervised-learning algorithms to perform well, this data has be representative the target In practice however, due differences between scanners such often not available. We therefore present in which does necessarily need data, allows use of from different studies than The assigns an importance weight all images, way that Kullback-Leibler divergence resulting distribution and minimized. In set experiments MRI brain-tissue with four substantially our method improved mean classification errors up 25% compared common approaches.