search

NKI-AI / dlup

Dlup are the Deep Learning Utilities for Pathology developed at the Netherlands Cancer Institute
Apache License 2.0
25 stars 7 forks source link

Anisotropic mpps #17

Closed lromor closed 2 years ago

lromor commented 3 years ago

Is your feature request related to a problem? Please describe.

Support anisotropic mpps

Describe the solution you'd like read_region can be adapted to rescale the sampled region in the highest resolution direction.

Describe alternatives you've considered N/A

Additional context N/A

jonasteuwen commented 3 years ago

I think the simplest fix is to allow for a bit more slack in the non-uniformity, and raise a warning. We can always look into it if it is really off?

DennisHaijma commented 3 years ago

For CAMELYON16 (recorded at (0.243x0.243 μm and 0.226x0.226 μm) I mitigated this by bringing in some slack, setting the NumPy tolerance to 1e-3 at line: https://github.com/NKI-AI/dlup/blob/0dfa2e4bb6ec40b9c6fb1fb3849786015b7a4fb2/dlup/_image.py#L100 So still needs attention I think.

jonasteuwen commented 3 years ago

I am not sure this is the best approach perhaps we should just forward this in the whole pipeline. I cannot imagine that there are non-square pixels though. Perhaps we can allow a 0.1% deviation. Would that work?

On 7 Sep 2021, at 14:23, DennisHaijma @.***> wrote:

 For CAMELYON16 (recorded at (0.243x0.243 μm and 0.226x0.226 μm) I mitigated this by bringing in some slack, setting the NumPy tolerance to 1e-3 at line: https://github.com/NKI-AI/dlup/blob/0dfa2e4bb6ec40b9c6fb1fb3849786015b7a4fb2/dlup/_image.py#L100.

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub, or unsubscribe.

DennisHaijma commented 3 years ago

Yeah the MPP's are slightly different after 3 decimals for the WSI's in CAMELYON16. Not sure how the mpp_x and mpp_y is generated by the digital slide scanner.

Excerpt from paper about imaging details:

The whole-slide images were acquired at 2 different centers using 2 different scanners. RUMC images were produced with a digital slide scanner (Pannoramic 250 Flash II; 3DHISTECH) with a 20x objective lens (specimen-level pixel size, 0.243 μm × 0.243 μm). UMCU images were produced using a digital slide scanner (NanoZoomer-XR Digital slide scanner C12000-01; Hamamatsu Photonics) with a 40x objective lens (specimen-level pixel size, 0.226 μm × 0.226 μm).

github-actions[bot] commented 2 years ago

This issue is stale because it has been open 60 days with no activity. Remove stale label or comment or this will be closed in 15 days