Cardiac MRI segmentation with sparse annotations: Ensembling deep learning uncertainty and shape priors

Fumin Guo, Matthew Ng, Grey Kuling, Graham A. Wright

Abstract: Highlights•Globally optimal label fusion (GOLF) algorithm for combining segmentation predictions generated by deep learning ensembles.•Deep learning uncertainty-guided coupled continuous kernel cut (ugCCKC) that integrates GOLF, pixelwise segmentation uncertainty, shape priors.•UNet++ deep ensemble combined with GOLF and ugCCKC for short-axis cardiac cine MRI segmentation.•Algorithm training using small datasets with sparse annotations and adaptation of a pretrained model to previously unseen datasets.•Strong correlations and agreement between algorithm and manual analyses of LV functional indices for a number of diverse cardiac MRI datasets.

External IDs:dblp:journals/mia/GuoNKW22