search

AICAN-Research / FAST-Pathology

⚡ Open-source software for deep learning-based digital pathology
BSD 2-Clause "Simplified" License
122 stars 26 forks source link

Not able to upload .tif files #106

Open Bsingstad opened 5 days ago

Bsingstad commented 5 days ago

Describe the bug I have previously uploaded .svs files to FastPathology with success, but when uploading .tif files the system fails and gives me this error: FAST exception caught in Qt event handler OpenSlide was unable to open the file: C:/Users/Bjorn/Downloads/img.tif

To Reproduce

andreped commented 5 days ago

Hello, @Bsingstad :]

FastPathology only supports pyramidal images. So if you try to import a nonpyramidal TIFF, that will not work.

Are you trying to import an image patch from a WSI? Is there a reason why you are not using the full original WSI? Or which type of image is this TIFF that fails to import?

Bsingstad commented 4 days ago

Hi @andreped , thanks for your fast response. This was what we were able to extract from our PACs system at the hospital this time, but then I understand that we have to find a way to extract .svs or other pyramidal images.

Additionally, I wonder what kind of software you use to annotate the images (as far as I understand, FastPathology cant be used for annotation)?

andreped commented 4 days ago

This was what we were able to extract from our PACs system at the hospital this time, but then I understand that we have to find a way to extract .svs or other pyramidal images.

@sahpet likely could best assist you on how to get appropriate images exported from whichever solution you are trying to export from. But the images you got exported, are they just snapshots from the WSI, or are they the actual full resolution WSI in TIFF format? Which options do you have for export? it is possible that there is a format that is the appropriate one (for instance an option to enable BigTIFF which may produce the right pyramidal format - not sure).

Additionally, I wonder what kind of software you use to annotate the images (as far as I understand, FastPathology cant be used for annotation)?

We use QuPath for annotating our WSIs (see here) and so we have for several years. I assume you want to train a segmentation model, or maybe a patch-wise classifier? To train both, perhaps it is of interest to you to check our NoCodeSeg repository, which was described in this paper.

We have a video here demonstrating how you can annotate your data using QuPath, export these to MIB for training, then exporting the trained model and using it in FastPathology for deployment (inference on new WSIs): https://www.youtube.com/watch?v=9dTfUwnL6zY

We have had several research teams using this pipeline with great success. Both for training segmentation models as well as patch-wise classification models :]