The UAMS Department of Biomedical Informatics (DBMI) is the only U.S. institution participating in a $10 million European grant project that aims to harness the power of artificial intelligence (AI) algorithms for cancer care.
There is a reason for that.
“UAMS is widely recognized as the very best in the field of cancer imaging data management,” said Dr. Karim Lekadir, project coordinator of EuCanImage project. “The UAMS team, led by Fred Prior, built and currently manages the world-renowned Cancer Imaging Archive (TCIA) and hence, they bring a unique experience and a lot of expertise to the project.”
The EuCanImage project was recently featured in a promotional video celebrating its progress and success .
The EuCanImage Project
EuCanImage aims to create a database of medical images from Europe used in cancer diagnosis and treatment. The goal is to create a highly secure, federated and large-scale database, open to researchers who can advance the role of AI in cancer research and treatment.
Specifically, they plan to:
- Collect a database of more than 25,000 imaging studies, along with the specific clinical pathological data associated with each image
- Develop eight use cases in breast, liver and colorectal cancer.
- Provide tools to enable for future AI research in cancer imaging.
DBMI Professor and Chair Fred Prior, Ph.D., received a $894,074 four-year grant in 2020 for UAMS’ part of the project. Funding for the $10 million project comes from the European Commission.
UAMS is one of 20 partners from 11 countries working on EuCanImage. The coalition includes experts that are at the top of their fields worldwide, including bioinformaticians, data scientists, medical doctors, radiologists, surgeons, software developers and AI innovators from the private sector.
The other partners represent diverse expertise from across Europe: from academia and hospitals to industry and geographically from Spain to Lithuania and from Sweden to Italy. The project’s coordination home is at the University of Barcelona in Spain, where Dr. Lekadir is Director of the Artificial Intelligence in Medicine Lab. Six hospital sites in Europe are contributing the data, along with TCIA.
“Concretely, UAMS DBMI are helping us create a high-quality cancer imaging repository in Europe, by contributing to a number of new tools, adapted for European needs and data regulations, to facilitate the workflow of cancer image de-identification, curation, annotation and storage,” Dr. Lekadir said. “They are also contributing to the interoperability of the annotated cancer imaging data, to ensure new and existing data from around the world can be pooled together to enhance the quality of future research and to increase international cooperation in the field.”
UAMS DBMI has the experience to provide such unique expertise because it is home to The Cancer Imaging Archive (TCIA). Funded by the National Cancer Institute’s (NCI) Cancer Imaging Program (CIP), the archive launched in 2011 and is the largest publicly available archive of medical images of cancer in the U.S. and quite possibly the largest in the world. Such large datasets are necessary for research, specifically for scientists who are attempting to “train” computer algorithms to diagnose and treat cancer using AI and machine learning (ML) techniques.
Many believe this is the future of medicine.
In the meantime, the challenging task of creating and maintaining such massive databases of medical images is top of mind for researchers at DBMI. In the last 10-plus years, the TCIA project has given the DBMI team valuable insights into how to make the big ideas of tomorrow into reality today.
DBMI’s specific role with EuCanImage is to provide across-the-board guidance and advice based on its accumulation of experience from working on TCIA. DBMI is also on a working group that is focused on how to prepare the data, ensuring the images are properly curated by being anonymous, enhanced and annotated in a way that is useful for researchers. DBMI is also contributing to a working group focused on interoperability between TCIA and EuCanImage, allowing data from both archives to be used in the same research.
In-Person Data-Annotation Workshop
The EuCanImage project team members met in person for the first time in July 2022 at the University of Barcelona in Spain, where the promotional video was filmed. DBMI Assistant Professor Jonathan Bona, Ph.D., represented the UAMS team at the three-day Annotation Workshop.
Dr. Prior explains that the data preparation, known as curation, is truly the key element that turns a database of images into something useful for training algorithms.
“There’s a huge amount of work to create annotated data, which you need for training and test sets for machine learning,” Prior said. “So that’s the reason for the large group – we’re collecting data from all over Europe and doing a lot of work to make that data consistent and available for machine learning, as well as supporting big teams that are working on machine learning algorithms themselves.”
Dr. Bona was pleased with the productivity of the in-person meetings. The pandemic has prevented the groups from meeting sooner.
“These were groups that we have been meeting with virtually every week, and it was incredible to finally meet in person and have the kinds of conversations needed to push the project forward and come up with some new directions to pursue,” Bona said. “This meeting specifically was focused on data annotation, so that meant there were live, hands-on sessions where everyone in the room got to try out new data annotation software that radiologists on the project will be using to annotate cancer images. That allowed us to test out workflows in real time, which was quite valuable.”
Dr. Bona said the meeting also gave everyone a chance to talk through some of the “big picture” issues – such as how to bridge language gaps in the electronic health record data and other challenges related to working with so many different data sources.
Drs. Prior and Bona, along with programmer lead and DBMI Ph.D. candidate Michael Rutherford, will return in October to visit clinical sites in Spain and Poland and attend the first, in-person, whole team project meeting.
“The idea is that we’ll design tools that are based on our expertise but built specifically for them to work with the systems they already have at the clinical sites,” Dr. Bona said. “This will help transform that data to the common data model that is needed to bring it all together into a central repository.”
Dr. Prior said that with projects like these, each challenge represents an opportunity to learn something that could be applicable to other similar projects in Arkansas and across the world.
“These tools, although they’re being developed in this context, are highly applicable to other projects and will be used by use in other contexts quite frequently,” Dr. Prior said.
Other countries and groups are catching on.
“That’s why I recently spent a week in India, and why another team member went to China,” Dr. Prior said. “We’ve been invited all over the world to help people try to figure out how to do this.”
Other members of the UAMS EuCanImage team besides, Prior, Bona and Rutherford include Lawrence Tarbox, Ph.D.; Kirk Smith, William Bennett, and Tracy Nolan.