This repository contains a Jupyter-lab Notebook showing image segmentation of HeLa cells' nucleus using TensorFlow and Keras. The notebook for image segmentation uses a U-net network. The default example uses the "mobilenet" backbone and "imagenet" weights.
See the project https://github.com/qubvel/segmentation_models for other backbones and weights.
All images in this project are DIC images of HeLa cells under normal conditions.
Training masks are generated from the DIC images and edited with ImageJ/Fiji. See the directory tree as an example and follow the following steps:
-
Open the image.
-
Select the contour of the body with the "Polygon Selection" button.
-
Create a mask as "Edit->Selection->Create mask".
-
Select the mask using the "Wand(tracing) tool".
-
Set a number for the body of the mask as: "Process->Math->Set", and select a number different than 0 (background). For more than two objects (background and nucleus) change "Labels" in the Notebook.
-
Save the mask using the same name of the input image, but inside the masks folder (see directory tree).