May 2023 Cycle Report

Published 30 May 2023

Support for DICOM-WSI images

DICOM-WSI images are large, high-resolution digital images of pathology slides that are used for diagnostic and research purposes. It is quickly becoming the standard for pathology images. Until recently, this format was not supported by Grand Challenge. This was a significant limitation for researchers and medical professionals who wanted to use the images in reader studies and algorithms. To address this limitation, we have added support for DICOM-WSI images. Users can now upload DICOM-WSI images to the platform, and the images will be automatically converted to TIFF format for internal use.

Support for TIFF file segmentations

Previously, TIFF files could not be created for interfaces of type Segmentation. The interface defines labels for all unique voxel values, and the image is validated to ensure it only contains voxel values defined in the interface. For TIFF files this is an operation that uses a lot of compute power, therefore it was not supported. Now we implemented support for these files by extracting the minimum and maximum voxel values from the TIFF tags (MinSampleValue, MaxSampleValue, SMinSampleValue, SMaxSampleValue). The values are used to validate that the segmentation contains a definition for all values ranging from the minimum to the maximum voxel value. So in order to use this, the TIFF file should be a grayscale image with integer datatype and contain at least one of each of the minimum and maximum voxel value tags.

3D mask editing

We are pleased to highlight the recent improvements to the mask-creation tool. In this cycle, we've enabled users to create masks using 3D sphere and cube shapes as a brush. Alongside this, we have improved the preview of where a mouse stroke will paint the image canvas: the preview will now snap to the voxels. The updated visual indicator simplifies creating masks and ensures more accurate results with a single click.

Annotation statistics plugin works for big pathology images

The annotation statistics can now reliably be used for computing statistics of overlays on pathology scans. The plugin is faster, more memory efficient, and allows annotating while it is computing new results. This allows iterating, refining, and evaluating regions of interest more quickly.

Algorithm creation limits for challenge participants

Up until now, type 2 challenge participants were able to create an infinite amount of algorithms for a challenge. They are often not aware that they can simply upload a new container image or reactivate an older container image for an existing algorithm instead. This has resulted in a lot of one-time-use-only algorithms and inefficient use of our resources.

To remedy this issue, we introduced a limit to the number of algorithms a user can create per challenge phase. The limit is set to 3 algorithms per phase. Once a user has reached that limit they will not be able to create new algorithms, and will instead be instructed to use one of their existing algorithms and upload a new container image there. We clearly state this new limitation on the algorithm creation page for challenges and warn users about this limit when creating their first algorithm. Nonetheless, we do encourage challenge admins to explain this to their participants as well.

Cover photo by Zetong Li on Unsplash