cell-segmentation

To reproduce the Deep Learning results

To run and reproduce the Deep Learning results, we have created a colab notebook CytoNet_DL_reproduce.ipynb. You need to login with the credentials for [email protected] to get access to data and models.

To visualize the results

To visualize the results from Image processing, machine learning and deep learning, we have created a colab notebook Visualization_notebook.ipynb. You need to login with the credentials for [email protected] to get access to data and models.

Image processing algorithms

Alive cells detection

The method for alive cells detection is in alive_cells/alive_cells_bboxes.py.

Dead cells detection

The method for dead cells detection is in dead_cells/dead_cells_bboxes.py.

Inhibited cells detection

The method for inhibited cells detection is in inhib_cells/inhib_cells_bboxes.py.

Running feature extraction pipeline

Cropping and glcm features

1. Make sure that the paths image_dir and bbox_dir in feature/extract_features.py are correct. Path bbox_dir should contain .txt files with bounding boxes for all cell types in the same dir.
2. Run python -m feature.extract_features in your terminal.
3. Cropped images will be saved to data/cropped and the glcm features will be saved to data/output.

Gabor features

1. Run python -m feature.gabor_filters in your terminal.
2. Index of top 1000 best features selected by AdaBoost will be saved to feature/output/gabor_index.csv.

Training the Machine Learning Cell State classifier

The training pipeline for cell state classification is in feature/training.py. The model will be saved to model_path specified in settings.py.

Inference Pipeline for Cell state classifier

To run the inference pipeline, add path to the chosen bounding boxes from Deep Learning in settings.py, and run python -m src.run_nms_before in your terminal. The inference pipeline is in src/run_nms_before.