Computer Vision News - November 2019

2 Summary Poster Presentation 0 Christian Baumgartner is a postdoctoral fellow at ETH Zürich. He speaks to us ahead of his poster session at MICCAI. This work is about learning a method that can segment anatomical structures from multiple experts. In medical image analysis, there is often the problem that expert annotations disagree with each other. This method manages to capture this variability of annotations, and after training, you can then sample plausible segmentations rather than just predicting a single one. Thi s inherent uncertainty needs to be ref lected because in downstream tasks such as radiotherapy planning i t i s very important that you know where annotat ions are certain and where they are uncertain. Us ing thi s method, you can analyse the var iabi l i ty of predicted segmentat ion maps, and say, for example, there i s a big di fference in that corner, but in that corner, al l experts agree. Christian tel ls us the idea for this work came about because he had been thinking about a closely related work cal led the Probabi l istic U-Net. He had some ideas for improving it, and in the end, fundamentally rethought theproblem and came up with a new solution. He is very pleased to add that this new solution outperforms the Probabilistic U-Net paper. In fact, the method outperforms state-of-the-art on multiple metrics. Not only are the numbers better, but also, it is statistically significant. A surprising side effect is that, even though they have the full probabilistic model, the segmentation accuracy does not suffer. In fact, the results are slightly better than for a standard U-Net. He explains the practical appl ications of the research: “Uncertainty is a very important PHiSeg: Capturing Uncertainty in Medical Image Segmentation "Thisinherentuncertainty needs to be reflected" Best of MICCAI