Multi-atlas learner fusion: An efficient segmentation approach for large-scale data.

摘要: We propose multi-atlas learner fusion (MLF), a framework for rapidly and accurately replicating the highly accurate, yet computationally expensive, segmentation based on fusing local learners. In largest whole-brain study reported, segmentations are estimated training set of 3464 MR brain images. Using these estimates we (1) estimate low-dimensional representation selecting locally appropriate example images, (2) build AdaBoost learners that map weak initial to result. Thus, segment new target image project into space, construct segmentation, fuse trained, selected, The MLF cuts runtime modern computer from 36 h down 3-8 min - 270× speedup by completely bypassing need deformable atlas-target registrations. Additionally, describe technique optimizing learning parameters, quantify ability replicate result with mean accuracies approaching intra-subject reproducibility testing 380 (3) demonstrate significant increases in when compared state-of-the-art separate dataset, (4) show under large-scale data model significantly improve over small-scale framework, (5) indicate has comparable performance as algorithms without using non-local information.