Segmentation of single-cells in microscopy images is one of the major challenges in computational biology. It is the first step of most bioimage analysis tasks, from drug discovery to tissue classification, and most importantly for several advanced machine and deep learning based phenotyping techniques.
The most challenging part in single cell segmentation is the extraction of individual cells when several cells share common boundaries. In the classical approach, a post processing algorithm is applied to separate single cells in the clusters. Recently, however a great variety of 2D deep learning systems were developed reaching the segmentation precision of the human. However, further development is needed to analyze 3D biological structures (e.g. spheroids, organoids, embryos) at single cell level. Since the deep learning methods usually require huge training datasets to perform well on unseen images, the ground truth data set generation is highly needed with the possible most accurate way and with as less human interaction as possible.
3D-Cell-Annotator solves this task using active contours with shape descriptors as prior information for true single cell annotation in a semi automatic fashion. The software uses the convenient 3D interface of the Medical Interaction Toolkit (MITK) a popular medical image analysis software and it requires a recent version of the NVidia drivers and a CUDA enabled GPU.
The results show that the 3D-Cell-Annotator precision reaches the level of an expert human annotator on images acquired with confocal, multiphoton and light-sheet microscopes.
HOW TO CITE US:
Ervin A. Tasnádi, Tímea Tóth, Maria Kovacs, Akos Diosdi, Francesco Pampaloni, Jozsef Molnar, Filippo Piccinini*, Peter Horvath*, (2019), 3D-Cell-Annotator: an open-source active surface tool for single cells segmentation/annotation in microscopy 3D images. Under revision.
Graphical User Interface (GUI) of 3D-Cell-Annotator
Mouse embryo single cell with an interesting phenotype