Identification and localisation of individual cells and nuclei (segmentation) in images of biological samples is crucial for quantitative biological research. In recent years, neural networks have become the state-of-the-art to perform this task. A few of the networks implemented in popular softwares such as Cellpose/Omnipose or Stardist are based on the prediction of a distance map. While these yields unprecedented accuracy, they rely on fully annotated datasets. This can be a serious limitation for generating training sets and performing transfer learning, especially in 3D.
In this talk, I will present the work of the team towards using partial annotations for neural network cell segmentation. In particular, I will focus on the idea to use neural networks to solve the eikonal equation to predict the distance map with sparse training data.
