January 2024 Cycle Report

Published 29 Jan. 2024

Grand Challenge features¶

New Runtime Environment for Algorithms¶

This cycle we have switched the execution backend for algorithm jobs and challenge evaluations to Amazon SageMaker Training. This has several benefits for you:

The maximum runtime for algorithm jobs or challenge evaluations can now be set up to 28 days
The size of /dev/shm is now 50% of the system memory
K80, V100, A10G and A100 GPUs are now available alongside the existing T4

Please see the updated algorithm documentation for more details.

⚠️An important change is that the /tmp directory will be empty at runtime. This means that you may need to update your container images as anything that you placed in /tmp in your Dockerfile at build time is no longer available at runtime. It is best to place those files in a subdirectory of /opt.

Editing archive items and display sets¶

This cycle we enhanced the user experience around creating and updating archive items and display sets:

You no longer need to be familiar with our API to create complex archive items (and display sets). You can now create them through the UI.
You can now update multiple values of an archive item or display set in one go.
You can now delete archive items yourself, no need to contact support any longer. For display sets this was already possible.
Validation and upload errors are reported more clearly now, either directly on the form or (for files and images) as a notification.

We hope these changes make managing your cases easier.

Cirrus features¶

Display text, video content, and static images in Cirrus¶

It is now possible to put the grand-challenge component interface values of types:

MP4 file
Thumbnail jpg
String

into the View content section of a reader study, archive or algorithm configuration to display them inside view items alongside other image data.

Strings will be displayed as scrollable text, video data using a video player, and thumbnail images as static images that cannot be interacted with.

Streamlined educational reader studies¶

Running an educational reader study can be streamlined by enabling the Instant verification option in the reader study settings. This will allow a user to go through the reader study more efficiently. The Save and continue button will be replaced by a Verify and continue button which will show the answer verification pop-up and allow the user to save and go to the next case upon dismissal.

Maintain selected location and zoom when changing orientation¶

When changing orientation (e.g. axial to sagittal) the images used to be re-fitted and centered to the view area. Now, the viewed image location and zoom are maintained and the orientation change will be akin to a rotation around the center point. This ensures, for instance, that when viewing a nodule in 3D it will be kept at the centre of the viewing area.

Limit view area to image volume¶

When viewing a 3D image, panning to the edges of the image, or changing orientation, can cause the image to partly fall outside of the viewing area. This results in a non-optimal amount of voxels being displayed:

A new optional behaviour limits the view area to image volume (when appropriate) and will 'shift' the view area to view a larger part of the image:

This is a behaviour common in the PACS that many radiologists use. This new behaviour is an opt-in and can be enabled by checking the Default limiting the view-area to an image volume flag in the viewer configurations.

Additional configuration options for annotations¶

When viewing images in CIRRUS on very high resolution monitors the annotations can sometimes be hard to distinguish. Users can now configure both default annotation colour and line width (in pixels) via the viewer configurations.

In a Reader Study, the default annotation colour (from either the viewer configuration or Cirrus) may lead to poor visibility when contrasted against the underlying image. We've now implemented the option to assign a colour per question. This enhancement facilitates the adoption of a consistent colour scheme across segmentations or annotations.