davco6 / segnu

Automatic cell nuclei segmentation assisted by CNN

SegNu

SegNu is an open-software that automatically detects cell nuclei assisted by CNN.

Briefly, SegNu classify each detected object in three categories: single cell nucleus, non-cell nucleus or cell nuclear aggregate. Afterwards, each cell nuclear aggregate is segmented recursively by watershed segmentation. In order to prevent oversegmentation (during the watershed segmentation), it is performed a graph analysis and later optimization. The first step, graph analysis, is done merging near segmented regions and then calculated the softmax function. After that it is selected the optimized graphs and generated a graph cluster containing optimally a cell nuclei.

Further information about SegNu can be found in the Master Thesis (wrote in spanish):

http://openaccess.uoc.edu/webapps/o2/handle/10609/82134

Requirements:

• Inside the weights folder you must include the weights file, name as "model_weights.h5", generated in the training of the CNN with TensorFlow. This file could be download in: https://drive.google.com/open?id=1Giing5bmtw81OXW2yEwxhBs2MC8PIcO-

• SegNu was written in Python 3.0, and use the following external modules:

  • numpy >=1.13.3
  • scikit-learn >= 0.13.0
  • scikit-image >= 0.19.1
  • keras >= 2.1.5
  • tensorflow >= 1.6

Email contact:

[email protected]

Important note:

This software is distributed in the hope that it will be useful for research use only, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.