Nuclear Segmentation

The goal of this project was to quantitatively evaluate a series of nuclear segmentation algorithms for use in our work. We hand-segmented several images from two datasets and used those as gold standards to evaluate methods.

Example image from the dataset

We generated two datasets, U2OS and NIH3T3, named after the cell type that was imaged.

  U2OS NIH3T3
Pixel size 1349 × 1030 1344 × 1024
Nr. Cells 1831 2178
Avg. Cover 23% 18%
Min Nr. Cells 24 29
Max Nr. Cells 63 70

A model based method developed by B. Roysam's group (Lin et al., 2003) was the best of the methods we evaluated at the time. In the meanwhile, however, other groups have reported better results on our dataset (list of citations at Google Scholar).

Code & Data

You can get the exact version used for the paper from the MurphyLab's reproducibility repository or, from my github page, a version that has been updated to work better in newer installations.

If you want to use the image above or any of the images in the dataset, feel free, as long as you cite our paper (see citation below).

Citation

Nuclear segmentation in microscope cell images: A hand-segmented dataset and comparison of algorithms by Luis Pedro Coelho, Aabid Shariff, and Robert F. Murphy

Digital Object Identifier: 10.1109/ISBI.2009.5193098

Open access PubMed Central version.

Full citation (use this if you use this code/dataset in a paper):

@inproceedings{Coelho2009,
    title = {Nuclear segmentation in microscope cell images: A hand-segmented dataset and comparison of algorithms},
    author = {Coelho, Luis Pedro and Shariff, Aabid and Murphy, Robert F.},
    booktitle = {2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro},
    doi = {10.1109/ISBI.2009.5193098},
    isbn = {978-1-4244-3931-7},
    keywords = {segmentation},
    pages = {518--521},
    year = {2009},
    publisher = {IEEE},
    url = {http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=5193098}
}

Article filed in categories: Work