PREPROCESSING OF CAMELYON16 DATASET USING CLAM

[routsourav1729]

Hi, i was trying to reproduce the results u have shown with CAMELYON 16 dataset. i have used the configuration u have provided in 0-extract_patches to run with create_pathes_fp.py

seg_level,sthresh,mthresh,close,use_otsu,a_t,a_h,max_n_holes,vis_level,line_thickness,white_thresh,black_thresh,use_padding,contour_fn,keep_ids,exclude_ids
-1,8,7,4,TRUE,25,4,8,-1,100,5,50,TRUE,four_pt,none,none

but i am getting around 1200 patches at level 1 compared to 5771 reported in the paper. and similarly for level 2 i got 400 images compared to 1528 as mentioned in paper. am i doing something wrong? do i need to change he config file?

also i think the CLAM repo is modified so i am getting some error while using convert_h5_to_jpg.py with CLAM. can u look into that and help me?

[routsourav1729]

is this the correct command to get patch level 1(20x) .h5 coordinates?:
python create_patches_fp.py --source /home/thomas/mil4wsi/dataset/CAMELYON16/images --save_dir /home/thomas/mil4wsi/dataset/CAM16_PREPROCESS --patch_size 256 --patch_level 1 --preset camelyon.csv --seg --patch --stitch

similarly i am just using patch level 2 and 3. and at patch level 3 each iamge has just 113 patches

[Bontempogianpaolo1]

I see... I might have used the combination 0,1,2 instead of 1,2,3. To resolve this doubt, we can visually compare the patches I have with those you have. The cell nucleus (the black dots so to speak) are typically all the same size (30-50 nm). So just compare whether the space they occupy is similar or not.

Slide 001 ( not tumor)
x5

x10

x20

[routsourav1729]

yes the patches seems to be of similar scale but my initial patch are different from these. there might be some other issue. One more thing to note that i think CLAM repo is updated with a slightly different dataset_h5.py code. so i have used the new code to extract h5 cordinates but an older repo which matches your code for convert_h5_to_jpg.py ro convert .h5 coordinates into jpg. they have changed the code for Whole_Slide_Bag_FP class in dataset_h5.py. can u look into it.

[Bontempogianpaolo1]

Hin @routsourav1729,
I see. If your resolution is identical to mine with fewer patches, the signal could be more evident, and you could also obtain better results. So I wouldn't worry about that; check CLAM-generated segmentation masks to preprocess WSIs efficiently.

I have checked the current implementation of CLAM, but nothing unusual is happening. Could you tell me the commit of CLAM where it worked? In that case, I would refer to it in the readme