Random Forests for Medical Applications
Olivier Pauly · 2012
Machine learning incarnates a key component for integrating the knowldege and experience of physicians into medical imaging applications such as computer aided diagnosis, detection and segmentation. In the last decade, random forests became a popular ensemble learning algorithm, as they achieve state-of-the-art performance in numerous computer vision tasks. Consisting in an ensemble of independent decision trees, random forests are very intuitive models, that offer a flexible probabilistic framework for solving different learning tasks. In this thesis, we formalize random forests models as ensemble partitioning approaches and propose novel related techniques for classification, regression and clustering. We introduce new task-specific forest models and demonstrate their great potential in different medical applications such as organ localization, segmentation, lesion detection and image categorization.