dylanmuir / readimagejroi Goto Github PK

There are always more pixels included in PixelIdxList than the area of ROI measured by ImageJ.

Sample.zip

Hi Dylan
Thank you for your useful tools.

It seems ROIs2Regions.m have bugs around handling x-y coordinates.

I tried to visualize my ('freehand') ROI data with the code below.
But it shows "fragmented" ROIs.

cvsROI = ReadImageJROI('RoiSet.zip');
sRegions = ROIs2Regions(cvsROI , vnImageSize);
mnLabelMatrix = labelmatrix(sRegions);
imagesc(label2rgb(mnLabelMatrix));

I checked ROIs2Regions.m and found 2 points.

[1] x-y was wrongly swapped.

ROIs2Regions apply transpose (') to the mbThisMask.
That will cause swapping of x and y dimension.

   sRegions.PixelIdxList{nROIIndex} = find(mbThisMask');

So I think it should be removed.

   sRegions.PixelIdxList{nROIIndex} = find(mbThisMask);

With this change, the ROI shape in mnLabelMatrix looks correct.

[2] Unnecessary +1

Additional validation, I compared the result with the image manually output from ImageJ.
I applied "Fill" to background and ROI with two different colors.
For background: Edit > Fill
For ROI: ROI Manager > More > Fill

mnImageJBinary= imread('Sample_ImageJ_Binary.tif');
imshowpair(mnImageJBinary>0, mnLabelMatrix>0);

And found that there is a shift between 2 images.

I guess that the addition of "1" to mnCoordinates is not necessary.

mbThisMask = poly2mask(sThisROI.mnCoordinates(:, 1)+1, sThisROI.mnCoordinates(:, 2)+1, vnImageSize(1), vnImageSize(2));

So it should be like below.

mbThisMask = poly2mask(sThisROI.mnCoordinates(:, 1), sThisROI.mnCoordinates(:, 2), vnImageSize(1), vnImageSize(2));

With editing these 2 points, ROIs2Regions.m works well for "Freehand" ROI in my environment.
I did not check for other cases "rectangle", "oval", "polygon".

First of all, I want to know this is generally happening or caused by my environment or somehow my misunderstanding.

My environment for this issue is below.
Matlab: 2019a
ImageJ: 1.52v on Fiji

Best,
Gaku