Winter newsletter 2023
Published 20 Feb. 2023
🛠️ New features of Grand Challenge
Greetings, on behalf of the Grand Challenge team, we would like to extend our warmest wishes for the New Year. In 2022, we focused on making the platform more approachable, hands-on, and applicable. We accomplished this by updating our documentation on how to participate in and host challenges and reader studies, submit and use algorithms, and use our archives. Additionally, our software engineers worked diligently to improve medical imaging viewer options, personalized challenges to meet the requests of challenge hosts and ensured that Grand Challenge was available 24/7 for all users. As we move into 2023, we focus on ensuring that the outcomes of challenges are more likely to succeed in clinical practice. To achieve this, we emphasise the objectiveness and generalizability of algorithms submitted to challenges. By default, challenges now require using a private test set to evaluate and compare the performance of submitted algorithms in the most objective way, to prevent overfitting and move towards generalizability. If your challenge design does not fit our provided format, please do not hesitate to contact us to discuss the possibilities.
"CoNIC Challenge: Pushing state-of-the-art for automatic nuclear recognition"
The Colon Nuclei Identification and Counting CoNIC Challenge was organized to help drive forward research and innovation for automatic nuclei recognition in computational pathology. Sponsored by AWS, TIGER was organized in collaboration with an international network of clinical and academic partners. The challenge encouraged researchers to develop algorithms for segmentation, classification and counting nuclei within the current largest publicly available nuclei-level dataset in CPath, containing around half a million labelled nuclei. CoNIC ran until March 2022. This blog post summarizes the challenge and presents some of the main findings. Additionally, a summary of the Grand Challenge submission format is provided, which required participants to submit algorithms - keeping the test set completely unseen.
“Grand Challenge at RSNA 2022"
Radboudumc aims to bridge the gap between clinic and research departments around the world. At RSNA, we have showcased Grand Challenge Connect (GCC) on Amazon Web Services (AWS), which builds on Amazon HealthLake Imaging to deliver Medical AI capabilities and enables Researchers and Clinicians to collaborate on the development of machine learning solutions closely. Learn more about the launch of GCC on AWS at RSNA 2022 in this blog post!
🔦 Highlighted algorithms
Calcium scoring in non-contrast CT showing the heartThis algorithm performs calcium scoring in non-contrast-enhanced computed tomography (CT) scans. Additional to a total score for coronary calcium, it provides scores per anatomical location, i.e. left anterior descending, left circumflex, right coronary artery and thoracic aorta. Try out this algorithm with your own data!
JawFracNetJawFracNet processes 3D volume patches sampled from the input CBCT scan to predict a mandible segmentation and mandibular fracture segmentations. The input is a head CBCT scan, and the outputs are the final mandible and fracture segmentations. Try out this algorithm with your data!
🚀 Upcoming challenges and running challenges
Grand challenge currently offers two types of submissions: prediction submission and algorithm container submission. The algorithm container submission type has the advantage of producing reproducible algorithms that remain accessible to the research community long after the challenge has ended. This allows for continued use and exploration of the algorithms by the community. Therefore, it should be noted that we are phasing out the prediction submission procedure in favour of the algorithm submission procedure to ensure that challenges always produce reproducible algorithms. We will also require an organizational base cost for hosting a challenge. To learn more about hosting a challenge on our platform, go to our documentation!
🥅 Goal: An AI challenge to compute tumour cellularity in pancreatic cancer and colon cancer with tumour cell nucleus segmentation with supervised- and transfer learning, respectively
✍️ Register: Accepting submissions from December 19th, 2022
⏰ Deadline: The development phase ends on February 8th
🏆 Leaderboard of finished challenges
The goal was to automatically segment brain tumour and resection cavities in intraoperative ultrasound (iUS) in 23 patients with pre-operative 3T MRI images, iUS images and segmentations of brain tumours in pre-operative iUS and resection cavities in during- and post-resection iUS available for training. The overall winner is Moona Mazher. Congratulations!
The goal was to localize triplet action by including bounding box localization of the regions of action triplets. The performance is evaluated by the average precision (AP) for recognition and localization of surgical action triplets. The winners are Wintegral, SDS-HD, and CITI. Congratulations!