Classification of chemically modified red blood cells in microflow using machine learning video analysis

R. K. Rajaram Baskaran, A. Link, B. Porr and T. Franke

Prerequisites

• Python 3.10
• Tensorflow & Keras 2.13.0
• OpenCV
• NumPy
• Matplotlib
• tqdm

Usage

1. run main.py to train, create, validate and test the model.
2. run plots.py to generate the plots as seen in the paper.

main.py <option_name>

Train, validate and test (native vs chem. mod.) RBCs.

Options:

• FA: Classification of native vs formaldehyde
• DA: Classification of native vs diamide
• GA: Classification of native vs glutaraldehyde
• MIX: Classification of native vs random mix of formaldehyde, diamide, glutaraldehyde

This generates all results in the directory results_<option>.

runall.sh

Runs all option: FA, DA, GA and MIX.

• Foreground: Shows the accuracy and loss.
• Background: nohup ./runall.sh &. You can log out and it will continue.

Modules

plots.py

Loads accuracy_and_loss_values.json and plots accuracy, loss and probability predictions.

video_processor.py

Labels the videos, subtracts the background, and returns them as NumPy arrays.

Tests

test_get_videos.py

Tests loading videos from the file directory.

test_bg_sub.py

Performs background subtraction, displays processed video.