README.md

Nucleus detection module

Usage

1. train.py – to train the detector.

2. threshold_opt.py – to generate predicted masks with one of the trained weights at a certain threshold.

   1. trained_model.hdf5 – weights of the model trained on LBC slides to detect all nuclei.
   2. trained_model_strictdata.hdf5 – weights of the model trained on LBC & Smear slides (with more annotated samples) to detect only free-lying nuclei.

3. particle_analysis.ijm – an ImageJ macro script to exact the centroids of blobs from generated binary masks and save to Results/*.csv.

4. test.py – to read Results/*.csv and evaluate performance at a certain threshold.

Other files

• draw_sample.py – to draw detection markers on a generated mask.
• PerformanceTest/ – the directory to evaluate performance as threshold changes (similar to the code above but more concise and with all needed data included).

Reference

• Microscopy cell counting and detection with fully convolutional regression networks (code: cell_counting_v2)