An ensemble learning method for segmentation fusion

Abstract

The segmentation of cells present in microscope images is an essential step to automate many tasks, including cell counting, analysis of the cell-division cycle, determining protein concentration, and analysis of gene expression per cell. In single-cell genomics studies, cell segmentations are vital to assess the genetic makeup of individual cells and their relative spatial location. Deep learning models are currently the most promising approaches among the various techniques and tools that have been developed to provide robust segmentation. We propose a learning ensemble strategy that aggregates many independent candidate segmentations of the same image to produce a single consensus segmentation as an alternative to developing another cell segmentation targeted model. We are particularly interested in learning how to ensemble crowdsourced image segmentations created by experts and non-experts in laboratories and data houses. Hence, these image segmentations are subject to high potential annotation errors created on purpose or by chance. We compare our trained ensemble model with other fusion methods adopted by the biomedical community, such as SIMPLE and STAPLE, and assess the robustness of the results on three aspects: fusion with outliers, missing data, and synthetic deformations. Our approach outperforms these methods in efficiency and quality, especially when there is a high disagreement among candidate segmentations of the same image.