Lazy Resampling: Fast and information preserving preprocessing for deep learning

Benjamin Murray, Richard Brown, Pengcheng Ma, Eric Kerfoot, Daguang Xu, Andrew Feng, Jorge Cardoso, Sébastien Ourselin, Marc Modat

Abstract: Highlights•Lazy Resampling can improve the performance of deep neural networks.•Lazy Resampling mitigates the information loss typical in preprocessing pipelines.•Lazy Resampling’s non-destructive operations enable simpler pipelines.•Lazy Resampling can preserve label accuracy, particularly when mapping labels back to source data.•Lazy Resampling is designed to be simple to use, is integrated into MONAI, and is enabled with a single flag.