shapeanalysis log

Shape analysis log

This document describes the script script_shapeanalysis.m, which focuses in the analysis of the shape of the independently detected cells into the disambiguation of the clumps detected in different frames.

The first step is the call to the script initscript.m which does a number of things:

• Loads the folder name structure from the /MACROPHAGES folder, whether it is from the first (/MACROPHAGES/MACROS1), second or third dataset.
• If no processing has been done on it, then the right functions are called and the _mat_ folders are created and filled.
• If all the _mat_ folders are found, then a quick check is made to the folders' paths to see if they are consistent with the actual names. This is helpful when working between OS.
• Load all relevant variables:
  • From the handles structure, the clump-development is built (or loaded). The following variables are created:
    • clumptracktable
    • clumpidcodes
    • timedfinalnetwork
    • tablenet

Select a clump by its unique label.

The next step involves choosing the uniquely identifiable clump from the table clumpidcodes, called wuc=clumpincodes(ix), where wuc refers to Which Unique Clump. Then the next code is run:

ix=10;
wuc = clumpidcodes(ix);
labelsinclump = getlabelsfromcode(wuc);

Then we get the clump life cycle (timeofclump) and the extension of one frame before and one frame after, to have the frames where the clump was not yet formed, or has ceased to exist:

timeofclump = unique(tablenet(clumptracktable.clumpcode==wuc,:).timeframe);
clumpnucleirange = (timeofclump(1)-1):(timeofclump(end)+1);
tabletracks = tablenet(ismember(tablenet.timeframe, clumpnucleirange),:);
tableclump = clumptracktable(ismember(tablenet.timeframe, clumpnucleirange),:);

Analyse the frames before and after the clump is formed

Select the frames that will have the analysis.

First of all, select the names of the files involved:

t = 1;
framet = clumpnucleirange(t);
frametplusT = clumpnucleirange(t+1);

The information is then accessed via the filename stored in cell variable filenames{framet}.

Then, load the data

Use getdatafromhandles, and to deal with less variables in the code, package them into structures. We do the single frame, sf, structure first:

[dataL, dataGL, clumphandles, dataR, dataGR] = ...
            getdatafromhandles(handles, singleframe);
% sf := single frame
sf.dataL = dataL;
sf.dataGL = dataGL;
sf.clumphandles = clumphandles;
sf.dataR = dataR;
sf.dataGR = dataGR;

Next, do the frame with a clump, cf, structure:

[dataL, dataGL, clumphandles, dataR, dataGR] = ...
            getdatafromhandles(handles, clumpframe);
% cf := clump frame
sf.dataL = dataL;
sf.dataGL = dataGL;
sf.clumphandles = clumphandles;
sf.dataR = dataR;
sf.dataGR = dataGR;

Notice that the generation of sf, cf only differ in the name of the structure variables. Also, do not forget to clear the helper variables used with the getdatafromhandles.m function:

clear dataL dataGL clumphandles dataR dataGR

A future version of this code should incorporate all this processing into function getdatafromhandles.m, to make scripts cleaner.

Populate sf and cf with important information

Information that will be included:

• Original image X

sf.X = cat(3, sf.dataR, sf.dataGR, zeros(size(sf.dataR)));
cf.X = cat(3, cf.dataR, cf.dataGR, zeros(size(cf.dataR)));

• Binary object properties from regionprops, and boundaries for single cells detected on framet.

cf.regs = regionprops(cf.clumphandles.overlappingClumps==10);
for jx=1:length(clumplab)
    auxbinmat = sf.clumphandles.nonOverlappingClumps==clumplab(1);
    sf.regs(jx) = regionprops(auxbinmat);
    sf.boundy{jx} = bwboundaries(auxbinmat);
end

• On cf, include the clump that is currently being worked on:

cf.thisclump = cf.clumphandles.overlappingClumps==10;

Follow independent cells with a cross correlation

In order to keep the size of the images and make localisation easier, the function imfilter will be used instead of xcorr2.

• Get initial position

for kx=1:length(clumplab)
    testImage = imfilter(sf.dataGR, imcrop(sf.dataGR, sf.regs(kx).BoundingBox));
    [maxval, mxidx] = max(testImage(:));
    [yinit, xinit] = ind2sub(size(sf.dataGR), mxidx);
    sf.xy(kx,:) = [yinit xinit];
end
clear testImage maxval mxidx yinit xinit;

Check the initial position of the points using function

plotBoundariesAndPoints(sf.X, [],sf.xy)

• Get the maximum position on frametplusT with the information of the independent cell in framet:

for kx=1:length(clumplab)
    testImage = imfilter(cf.dataGR.*cf.thisclump, imcrop(sf.dataGR, sf.regs(kx).BoundingBox));
    [maxval, mxidx] = max(testImage(:));
    [yinit, xinit] = ind2sub(size(sf.dataGR), mxidx);
    cf.xy(kx,:) = [yinit xinit];
end
clear testImage maxval mxidx yinit xinit;

• Move the boundaries from the independent cell in sf.boundy using the movement from the cross correlation:

for kx=1:length(clumplab)
    cf.movedboundy{kx} = sf.boundy{kx} + ...
        repmat(cf.xy(kx,:)-sf.xy(kx,:), size(sf.boundy{kx},1),1);
end

• The moved boundaries in cf.movedboundy will be used as initialised points for the SOM or ASM. This can be accessed here.

Some plots

figure(1)
clf
plotTracks(handles,2,clumplab);
title(sprintf('Nuclei that belong to clump %d, moving through time.', wuc));

figure(2)
plotBoundariesAndPoints(cf.dataGL, sf.boundy, cf.movedboundy, 'm-')
hold on
rectangle('Position', minibb);
title('Moved boundary');

Ideas on the move...

Anything I havetried but not finished...

• Analysis of the moved boundaries with active contours. I changed the parameters of n=25 and the pair name value of 'SmoothFactor' to 1.5

smoothf = 1.5; % 3 or 1.5 or 1...
BW1 = activecontour(cf.dataGR, cf.movedmask{1}, 25, 'Chan-Vese', 'SmoothFactor', smoothf);
BW2 = activecontour(cf.dataGR, cf.movedmask{2}, 25, 'Chan-Vese', 'SmoothFactor', smoothf);

This approach seems to have potential with the moving objects and the shape evolution. The SmoothFactor helps with the boundaries not deforming so much because of the intensity differences of the edges.