tracking_model: Instant registration and tracking of segmented microscopy images

Installation

Install on your own machine

You can run the following commands to install the tool in your own conda environment.

Windows Install

1. Download and install Python 3.9 version of Miniconda for Windows: https://docs.conda.io/en/latest/miniconda.html#windows-installers

2. Login, download and install Visual Studio 2022 Professional to build pyklb: https://visualstudio.microsoft.com/vs

3. Open "Command Prompt" and create a conda environment and activate it:

conda create -n trackcells python=3.9
conda activate trackcells

4. Install the tracking_model:

pip install git+https://github.com/eleniadam/tracking_model.git
track_cell --help

Example

1. Crop the 3D images using roi_convertor

2. Relabel the timestamp i+1 3D image (according to timestamp i 3D image):

Commandline Options

track_cell generate-relabelledimage --help

Example Command

track_cell generate-relabelledimage 
--tree_file edges_1_100_st6.csv 
--image_i klbOut_Cam_Long_00084.lux_SegmentationCorrected.klb 
--image_ii klbOut_Cam_Long_00085.lux_SegmentationCorrected.klb 
--output_dir out_data

The CSV file must depict each tree edge in the form: timestampN_labelA,timestampM_labelB where timestamp{N,M} and label{A,B} are three digit numbers.

image_i, image_ii can be in klb/h5/tif/npy formats with these extensions and the filename must contain the three digit number of the timestamp.

Output image will be in tif format and saved as relabel_timestampii.tif

3. Superimpose two 3D images into one 4D image:

Commandline Options

track_cell generate-superimposedimage --help

Example Command

track_cell generate-superimposedimage 
--image_i klbOut_Cam_Long_00084.lux_SegmentationCorrected.klb 
--image_ii klbOut_Cam_Long_00085.lux_SegmentationCorrected.klb 
--output_dir out_data

image_i, image_ii can be in klb/h5/tif/npy formats with these extensions.

Output 4D image will be in tif format and saved as imagepair_timestampi_timestampii.tif

4. Train the model:

Commandline Options

track_cell train-model --help

Example Command

track_cell train-model 
--original_dir original 
--groundtruth_dir groundtruth 
--output_dir out_data

Input 4D images must be in tif format. The output folder contains the model and optimal weights information saved as trained_model.h5 and weights_file.h5

5. Predict the image:

Commandline Options

track_cell predict-image --help

Example Command

track_cell predict-image 
--image imagepair_076_077.tif 
--model_file trained_model.h5 
--weights_file weights_file.h5 
--output_dir out_data

Input 4D image must be in tif format. The model and weights files are the output files of the train-model command.