Read in part of large histological (ndpi) file

I want to read in a very large histological image file (file size: 2.05 GB, *.ndpi file, pixel size: 57856 x 1.111e+05) and don't have enough RAM.

I tried blockedImage and thought it would fit great but still, I am unable to get my data in highest resolution.

I tried:

% p,f are path and filename for the 2.05 GB *.ndpi file

A = blockedImage(strcat(p,f));
 
levelSizeInfo = table((1:length(A.Size))', ...
    A.Size(:,1), ...
    A.Size(:,2), ...
    A.Size(:,1)./A.Size(:,2), ...
    'VariableNames',["Resolution Level" "Image Width" "Image Height" "Aspect Ratio"])

and obtained

levelSizeInfo =
9×4 table

Resolution Level    Image Width    Image Height    Aspect Ratio
________________    ___________    ____________    ____________

1               57856        1.111e+05        0.52074   
2               28928            55552        0.52074   
3               14464            27776        0.52074   
4                7232            13888        0.52074   
5                3616             6944        0.52074   
6                1808             3472        0.52074   
7                 904             1736        0.52074   
8                 407             1188        0.34259   
9                 203              594        0.34175   

So I want to assess the resolution level 1 information and therefore tried to split it up in smaller patches:

bls = selectBlockLocations(A,"Levels",1,"BlockSize",[4096, 4096]);
bimds = blockedImageDatastore(A, "BlockLocationSet", bls);

However, I cannot assess these 4096 x 4096 blocks.

I tried to work with:

>> while hasdata(bimds)
    [data, info] = read(bimds);
    disp(info);
end

Error using images.blocked.TIFF/getIOBlock (line 139)
Out of memory.

But get the above error message. Is there any possibility that I read in my large image and split it up until my memory is sufficient and then assess the data? I don't want to use lower resolution, since I am really interested in some specific histological information only reliable visible as highest resolution.

I would be also happy, when I would provide voxel coordinate start and end position and split it up manually such that it fits my RAM. I can do it in python but I would like to work with this dataset in Matlab as well.

Similar to this (which was not working either):

>> img = bigimage(strcat(p,f));
%Starting coordinates
coordStart = [250 250];
%Ending Coordinates
coordEnd = [750,750];
%Extract the region between the aforementioned coordinates
blk1 = getRegion(img,1,coordStart, coordEnd);
%Display the image
bigimageshow(bigimage(blk1))

Error using imread (line 440)
Unable to read TIFF file "SSES-1_MAX.ndpi". File is corrupt or image does not contain any readable strips.

Thanks in advance!

Answer 1

[Ashish Uthama]

NDPI look like TIFF file, but unfortunately, they dont stick to the spec. It internally stores the full res image likely compressed that make it hard for TIFF tools to extract just a part of it.

Do you have any tools to convert it to a TIFF file?

e.g you could try: https://pperso.ijclab.in2p3.fr/page_perso/Deroulers/software/ndpitools/

Another option is to try this custom adapter which uses OpenSlide to interepret this non-standard TIFF file:

https://www.mathworks.com/help/images/read-whole-slide-images-with-custom-blocked-image-adapter.html

Then blockedImage ought to work as expected.