Skip to content

example cell tracking

Jump to bottom
Jian Tay edited this page Jun 15, 2021 · 3 revisions

Example of tracking

This document will walk you through an example of tracking bacteria growing in a microscope time-lapse image.

[[TOC]]

Prerequisites

  • MATLAB R2016a or later with the Image Processing Toolbox installed
  • The Linear Assignment toolbox is installed
  • Download the example image file celldemo.tif

celldemo.tif is a multi-page TIFF file containing time-lapse images of a growing cyanobacteria cell. We thank the Cameron Lab for supplying us with the cells to make this image.

Overview

After reviewing this document, you will be able to:

  1. Write code to segment cells using intensity threshold and watershed algorithms
  2. Change LAPLinker properties to set up tracking
  3. Generate an output image to validate the tracking
  4. Save the tracked data

Glossary

  • Segmentation - Identification of cells in the image
  • Tracking - Identifying and grouping data from the same object between frames

The following are common definitions used to describe object-oriented code:

  • Class, e.g. LAPLinker class - The type of object.

  • Class file - The m-file that contains the code describing the object, e.g. LAPLinker.m

  • Object, e.g. LAPLinker object - An "instance" or concrete realization of a class. You create an object by assigning a variable name to the class.

    Example: obj = LAPLinker results in obj which will be a LAPLinker object.

  • Object properties, e.g. LAPLinker properties - Properties contain data for the object. For example, resulting time series data is stored in the property tracks of a LAPLinker object. You access properties of an object using dot notation: obj.tracks.

  • Method - A function defined within the class file that (typically) acts on an object of that class. Methods are called using the object as the first input argument: treeplot(obj).

Step-by-step

Set up

This section is optional, but recommended if you are new to MATLAB

  1. Create a new folder in your MATLAB directory and change your working directory to that folder 
    mkdir exampleLAPLinker
cd exampleLAPLinker
  2. Download and copy the demo image file into this directory
  3. Start a new script 
    edit exampleCells.m
  4. Copy or type the commands in the following sections into the script

Segmenting the cells

Tip: To get help on a particular function, use help function_name. For example, help imread.

To identify the cells in this image, we will use a simple intensity thresholding algorithm. The basic principle is to identify a threshold intensity that identifies cells in the image (e.g. pixels above or below the threshold are cells and the rest are background).

  1. Read and display the first frame of the image using imwrite:

    I = imread('celldemo.tif');
imshow(I, [])

  2. Use the datatips tool to determine a suitable threshold.

    Datatip demo

    For this example, since the cells have a dark border, it is easier to set a maximum threshold value. We will choose 1.65e4 for this example.

  3. Generate a mask (a binary image where true/white pixels indicate a cell and false/black pixels indicate the background)

    mask = I < 1.65e4;

  4. Display the mask to see the initial results (feel free to run this function after each of the following steps)

    imshow(mask, [])

  5. Fill in the interior of the cell using imfill

    mask = imfill(mask, 'holes');

  6. Use morphological opening to smooth the edges of the cell

    mask = imopen(mask, strel('disk', 3));

  7. Thicken the cell masks to get a better fit

    mask = bwmorph(mask, 'thicken', 3);

  8. Remove any objects less than 300 pixels

    mask = bwareaopen(mask, 300);

  9. Display the final mask to check

    imshow(mask)

Measuring cell properties

Once we have a mask, we can use MATLAB's regionprops function to measure data about the cell. For this example, we will measure the center coordinate (centroid) and the length of the cell (MajorAxisLength).

For tracking, since these cells are relatively stationary within the image, we can use the pxintersect metric to compute the ratio of overlapping pixels to identify the same object in different frames. To do so, we also need to get a list of pixels in the cell (PixelIdxList).

Putting it all together, we get:

data = regionprops(mask, 'Centroid', 'PixelIdxList', 'MajorAxisLength');

Separating clusters of cells

As the cells grow and divide, it will be necessary to separate the different cells. For example, try reading in and segmenting frame 10.

I = imread('celldemo.tif', 10);

mask = I < 1.65e4;
mask = imfill(mask, 'holes');
mask = imopen(mask, strel('disk', 3));
imshow(mask)

MATLAB will treat connected regions in the mask as a single object, so we have to split the cells apart. To do so, we will use an algorithm known as the watershed transform. MATLAB has a great explanation of the algorithm here.

We can follow the steps outlined in the documentation for the watershed function:

%Compute the distance transform
dd = -bwdist(~mask);
dd(~mask) = -Inf;

%Perform an h-minima transform to suppress minima with a depth
%less than 1. This will prevent oversegmentation (cell divided
%into too many separate objects)
dd = imhmin(dd, 1);
LL = watershed(dd);

%Update the mask
mask(LL == 0) = 0;

Checking the final mask should show that there are now two objects clearly separated by a line.

imshow(mask)

Set up tracking using a LAPLinker object

  1. Create a new LAPLinker object

    %Create a LAPLinker object
L = LAPLinker;

  2. Set up tracking parameters

    %Set up the tracking parameters to use overlapping pixels
L.LinkedBy = 'PixelIdxList';
L.LinkCostMetric = 'pxintersect';
L.LinkScoreRange = [1 12];
L.TrackDivision = true;
L.DivisionParameter = 'PixelIdxList';
L.DivisionScoreMetric = 'pxintersect';
L.DivisionScoreRange = [1 12];
L.MinFramesBetweenDiv = 5;

    Here we are setting up the object to link objects by the number of overlapping pixels in each frame, as well as tracking division using the same metric. After division, each cell must grow for a minimum of 5 frames before dividing again.

    Please see configuring LAPLinker for more details on these parameters.

  3. Insert the code into a for loop.

    Now that we have the foundation of the code, we can create a for loop to loop over each frame of the image. Note that the example image has 17 frames.

    The full code is posted below. Here, we show a minimal version with comments showing where the different parts of the code would go. However, note the position where assignToTrack is called.

    %Create a LAPLinker object
L = LAPLinker;

%Set up the tracking parameters
L.LinkedBy = 'PixelIdxList';
%...etc.

%Process each frame
for iT = 1:17
 
   %Read in image
   I = imread('demo_cell.tif', iT);
 
   %Segment the cells
   mask = mask = I < 1.65e4;
   %...etc.
 
   %Measure data
   data = regionprops(mask, 'Centroid', 'PixelIdxList', 'MajorAxisLength');
 
   %Assign the data to tracks
   L = assignToTrack(L, iT, data);

end

Visualizing the tracking results

Visualizing tracking results is one way to validate that the code is operating as expected. The following describes how to generate an output showing both the cell, an outline of the mask, and the object IDs.

  1. Compute the perimeter of the mask

    maskPerimeter = bwperim(mask);

  2. Make the line thicker by dilating the perimeter mask

    maskPerimeter = imdilate(maskPerimeter, ones(2));

  3. We want to display the outlines in green, so generate an RGB matrix

    maskDisplay = cat(3, zeros(size(maskDisplay)), maskDisplay, zeros(size(maskDisplay)));

  4. Generate a composite image of both image and mask

    Iout = double(imfuse(imadjust(I), maskDisplay, 'blend'));
Iout = uint8(Iout ./ max(Iout(:)) * 255);

  5. Label the tracks. Since we only want to look at tracks which are in the current frame, we can use the activeTrackIDs property.

     %Insert the IDs of active tracks
 for iActive = L.activeTrackIDs
     trackData = getTrack(L, iActive);
     Iout = insertText(Iout, trackData.Centroid(end, :), trackData.ID, ...
         'BoxOpacity', 0, 'TextColor', 'white', 'AnchorPoint', 'Center');
 end

  6. Show the image

    imshow(Iout)

Putting it all together

We have now covered all the basic concepts. The following is the full code listing of the example. You can copy and paste the code as a new script.

%Create a LAPLinker object
L = LAPLinker;

%Set up the tracking parameters to use overlapping pixels
L.LinkedBy = 'PixelIdxList';
L.LinkCostMetric = 'pxintersect';
L.LinkScoreRange = [1 12];
L.TrackDivision = true;
L.DivisionParameter = 'PixelIdxList';
L.DivisionScoreMetric = 'pxintersect';
L.DivisionScoreRange = [1 12];
L.MinFramesBetweenDiv = 5;

for iT = 1:17
    
    %Read in image
    I = imread('demo_cell.tif', iT);
    
    %Segment the cells
    thLvl  = 1.65e4;
    
    %Compute mask and fill holes
    mask = I < thLvl;
    mask = imopen(mask, strel('disk', 3));
    mask = imfill(mask, 'holes');
    
    %Separate the cell clumps using watershedding
    dd = -bwdist(~mask);
    dd(~mask) = -Inf;
    dd = imhmin(dd, 1);
    LL = watershed(dd);
    mask(LL == 0) = 0;
    
    mask = bwmorph(mask, 'thicken', 3);
    mask = bwareaopen(mask, 300);
        
    %Measure data
    data = regionprops(mask, 'Centroid', 'PixelIdxList', 'MajorAxisLength');
    
    %Assign the data to tracks
    L = assignToTrack(L, iT, data);
    
    %Generate an output image to validate
    maskDisplay = imdilate(bwperim(mask), ones(2));  %Get outline of the cells
    maskDisplay = cat(3, zeros(size(maskDisplay)), maskDisplay, zeros(size(maskDisplay)));
    
    Iout = double(imfuse(imadjust(I), maskDisplay, 'blend'));
    Iout = uint8(Iout ./ max(Iout(:)) * 255);
    
    %Insert the IDs of active tracks
    for iActive = L.activeTrackIDs
        trackData = getTrack(L, iActive);
        Iout = insertText(Iout, trackData.Centroid(end, :), trackData.ID, ...
            'BoxOpacity', 0, 'TextColor', 'white', 'AnchorPoint', 'Center');
    end
    
    %Write the output as a GIF
    [imind, cm] = rgb2ind(Iout,256); 
    if iT == 1
        imwrite(imind, cm, 'output.gif', 'Loopcount',inf, 'DelayTime', 0.2);
    else
        imwrite(imind, cm, 'output.gif', 'writeMode', 'append', 'DelayTime', 0.2);
    end
    
end

This code should produce an output GIF that looks like:

Tracked cells

Saving the tracked data

Now that tracking is complete, it is a good idea to save the data.

trackdata = L.tracks;
save('demodata.mat', 'trackdata')

Continue to data analysis

< Back to User Guide

Clone this wiki locally