centroid for each slice along one axis in a .nrrd file

Hi All

I have some medical images, saved as .nrrd files. (segmented)

I would like to calculate the centroid for each slice along a single axis.

Then create a line of best fit for this.

Any suggestions for this please,(I am very much a beginner in matlab and not of engineering background.)

Thanks heaps

Answer 1

[Ergin Sezgin]

Hey,

You can isolate the layers, binarize them with imbinarize function, manipulate the regions by morphology if necessary, and then get the region properties by regionprops function. This will yield the centroid data for each region. Then you can extract centroids into an array and fit a line in 3D space using these values.

Good luck.

[Mel A]

Thanks Ergin,

It seems that imbinarize works with .pngs?

I have segmentations , saved as .nrrd.

Its a 3D structure,

I am not able to convert them to 2D slices of pngs

I can use nrrdread to read the nrrd file, which generates a 3D volume

Then it think it gives me a single value for the enitre bounding box I think.

What I am looking for is slice by slice centroids for the partcular volume. Any thoughts pls

Thanks

[Ergin Sezgin]

Hello again,

imbinarize works with 2D and 3D grayscale images.

After importing .nrrd data into your workspace, I presume you will see a 3 dimensional single class array. It represents consecutive 2D slices of volume data. You can access each slice and save them into different variables by indexing such as slice_1 = nrrdData(:,:,1), slice_2 = nrrdData(:,:,2). A loop would be a better way to extract and process these slices. After you get slices, you can perform binarization and morphological processes which will enable you to have centroids data.

Is it possible for you to save the data you get after importing by nrrdread function, into a .mat file and share here?

Regards

[Mel A]

Thanks I think , slice function extarct the entire bounding box,

Not sure what I am doing sorry,

Shared the mat file

[Ergin Sezgin]

Hello,

You can use the following code to access and examine each layer after your import the .nrrd data. centroidsArray should yield the output you want. Try it yourself and please let me know the outcome.

centroidsArray = [];

for i=1:1:size(V,3)

currentLayer = V(:,:,i);

currentLayer = logical(currentLayer);

regionProperties = regionprops("table", currentLayer, "Centroid");

if ~isempty(regionProperties)

if height(regionProperties) > 1

for j=1:height(regionProperties)

rowData = [regionProperties.Centroid(j,:), i];

centroidsArray = [centroidsArray; rowData];

end

else

rowData = [regionProperties.Centroid(1,:), i];

end

centroidsArray = [centroidsArray; rowData];

end

end

figure

scatter3(centroidsArray(:,1),...

centroidsArray(:,2),...

centroidsArray(:,3), '.')

axis equal

grid on

You can then fit a line using centroids.

Good luck

[Mel A]

Perfect @Ergin Sezgin

Thanks so much,

This worked,

Now my problem is converting this to mm/ world coordinates rather than pixels.

I get the line I want but I cant align it on my STL model created with the same exact .nrrd file. I think the problem is that the numbers in the centroid array are in pixels/voxels.

Please let me know if you a way for this

I am currently reading and trying, will update if I can work this out.

[Ergin Sezgin]

Hey Mel,

I'm glad that it worked.

Normally, I use calibration plates and tools such as a checkerboard plates to compute intrinsic parameters which enable me to get distortion free images. Knowing the real size of checkerboard squares and getting an image of the board, it is possible to get an analytical relation between real world dimensions and pixels. This is useful for 2D data. For the third dimension along Z axis, you need some kind of encoder. You can assume every layer has an equal distance between each other or you can get real, precise distances between layers from the hardware you use to get .nrrd data.

I hope this helps

Good luck