Multi-scale deep networks and regression forests for direct bi-ventricular volume estimation

摘要: Direct estimation of cardiac ventricular volumes has become increasingly popular and important in function analysis due to its effectiveness efficiency by avoiding an intermediate segmentation step. However, existing methods rely on either intensive user inputs or problematic assumptions. To realize the full capacities direct estimation, this paper presents a general, fully learning-based framework for bi-ventricular volume which removes unreliable We formulate as general regression consists two main learning stages: unsupervised image representation multi-scale deep networks random forests. By leveraging strengths generative discriminant learning, proposed method produces high correlations around 0.92 with ground truth human experts both left right ventricles using leave-one-subject-out cross validation, largely outperforms larger dataset 100 subjects including healthy diseased cases twice number used previous methods. More importantly, can not only be practically clinical but also easily extended other organ tasks.