DBIS

The cell is the basic structural and functional unit of organisms. With the development of microscopy imaging, studies based on cell morphology (cell shape, size, nucleus shape, size, etc.) are widely used in disease research, drug discovery, etc. Cell instance segmentation is the foundation task for biomedical research.

In the past few years, we have witnessed the great success of deep learning-based approaches in the field of biomedical research.
Convolutional neural networks (CNN) play an important role in computer vision tasks, which have been also widely used for cell instance segmentation.

However, segmentation of cell instances from micrographs has long presented various challenges. First, segmenting cellular pictures begins with the identification of several image objects. The shapes of the cells are often heterogeneous and prone to dynamic changes, making it practically impossible to build mathematical shape models. Second, cell compartmentalization and intra- and inter-cell heterogeneity lead to non-homogeneous marker distributions within and between cells, resulting in undesired picture characteristics such as intensity gradients. Moreover, expanding cell populations typically result in dense cell areas, making it difficult to identify picture attributes to the correct cell, particularly among groups of geographically adjacent cells.

In recent studies, Stringer et al. proposed a unified cell segmentation framework called Cellpose [1], which shows the powerful ability of deep learning in cell segmentation. The goal of this thesis is to develop methods based on Cellpose to improve the instance segmentation performance.

If you are interested in this thesis, a related topic, or have additional questions, please do not hesitate to send a message to mou@dbis.rwth-aachen.de

References
[1] Stringer C, Wang T, Michaelos M, Pachitariu M. Cellpose: a generalist algorithm for cellular segmentation. Nature methods. 2021 Jan;18(1):100-6.