Using ScoreTiler with two different masks produces identical results.

[PinkEyeSurvivor]

Describe the bug

When using two different extraction masks (over two very different WSI regions) the output csv from ScoreTiler and the tiles extracted are the same when clearly the slide coordinates at a minimum should be different (and in this case the Nuclear score is anticipated to be different. Interestingly, the locate tiles function shows these tiles to be located in the correct positions and in the right numbers. Is there a file that is being leaked between class instances?

Here are my examples:

scored_tiles_extractor2 = ScoreTiler(
    scorer = NucleiScorer(),
    tile_size=(512, 512),
    n_tiles=20,
    level=0,
    check_tissue=True,
    tissue_percent=80.0,
    pixel_overlap=0, # default
    prefix="CA2scored/", # save tiles in the "scored" subdirectory of slide's processed_path
    suffix=".png" # default
) 

scored_tiles_extractor2.locate_tiles(
    slide=test_slide,
    extraction_mask=ca_2_poly_custom_mask,
) 

scored_tiles_extractor3 = ScoreTiler(
    scorer = NucleiScorer(),
    tile_size=(512, 512),
    n_tiles=20,
    level=0,
    check_tissue=True,
    tissue_percent=80.0,
    pixel_overlap=0, # default
    prefix="CA1scored/", # save tiles in the "scored" subdirectory of slide's processed_path
    suffix=".png" # default
)

scored_tiles_extractor3.locate_tiles(
    slide=test_slide,
    extraction_mask=poly_custom_mask,
)

Expected behavior
I would expect that given two different extraction masks I should at least pull tiles with different coordinates between masks and in cases where appropriate produce differing nuclear scores.

Some copy/paste from the .csv:

[from CA1]
CA1scored/tile_0_level0_13501-26400-14013-26912.png,0.7599788363195672,1.0
CA1scored/tile_1_level0_14013-25888-14525-26400.png,0.7598940221993982,0.999888185646619
CA1scored/tile_2_level0_12989-16672-13501-17184.png,0.7506675064022765,0.987724445743891
CA1scored/tile_3_level0_11965-16160-12477-16672.png,0.7480076281028041,0.9842178059115734
CA1scored/tile_4_level0_12477-16160-12989-16672.png,0.7333986936009114,0.9649581760096282
CA1scored/tile_5_level0_11965-15648-12477-16160.png,0.7279953669776805,0.9578347221404039
[truncated but 20 results reported]

[from CA2]
CA2scored/tile_0_level0_13501-26400-14013-26912.png,0.7599788363195672,1.0
CA2scored/tile_1_level0_14013-25888-14525-26400.png,0.7598940221993982,0.999888185646619
CA2scored/tile_2_level0_12989-16672-13501-17184.png,0.7506675064022765,0.987724445743891
CA2scored/tile_3_level0_11965-16160-12477-16672.png,0.7480076281028041,0.9842178059115734
CA2scored/tile_4_level0_12477-16160-12989-16672.png,0.7333986936009114,0.9649581760096282
CA2scored/tile_5_level0_11965-15648-12477-16160.png,0.7279953669776805,0.9578347221404039
[truncated but 20 results reported]

Screenshots showing correct tile positions:

[ernestoarbitrio]

Thanks for reporting it and sorry for the late reply. We're on it :D we will keep you posted

[alessiamarcolini]

Dear @PinkEyeSurvivor, would it be possible for you to share the WSI used for the example?

[alessiamarcolini]

@PinkEyeSurvivor Also, I think the important missing information is: how do you call the extract method? Are you correctly passing the extraction_mask?
If left to default, the tiler will use the BiggestTissueBoxMask, therefore explaining why you get the same tiles :)