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.
We generated two datasets, U2OS and NIH3T3, named after the cell type was visible.
| 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 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 in 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, 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