Four Radboud AI Innovation vouchers granted

Date of news: 18 May 2021

In May 2021, four project teams originating from different faculties have received a voucher of a maximum of €20.000 for collaborative projects that contribute to the mission of Radboud AI: strengthening industrial collaboration and impact in the field of AI. The projects have been selected by a committee consisting of Radboud University and Radboudumc experts. In view of the high quality and quantity of the proposals, the Radboud AI Management Board has approved four proposals, instead of the planned three.

The following projects have been granted:

How do people feel about AI-driven decision-making

Applicants: Frederik Zuiderveen-Borgesius, Tibor Bosse, Gabi Schaap, Marvin van Bekkum, Iris van Ooijen, Iris van Rooij

External parties: Maaike Harbers (Rotterdam University of Applied Sciences, UAS) and Tjerk Timan (TNO).

Our society can benefit tremendously from artificial intelligence (AI). AI advances important goals, such as efficiency and economic growth. AI can help to predict where car accidents will occur, who will commit fraud, who can repay a mortgage, etc. But AI-driven decisions about people can also have unfair or discriminatory effects. The interdisciplinary team (computer science, AI, law, communication science, psychology, and science and technology studies) will in this project assess what the Dutch public thinks about AI-driven decision-making. This will be done via 1) discussion sessions of focus groups, to provide a deeper understanding about people’s knowledge and reasoning behind opinions about AI-driven decision-making than is possible with only quantitative methods. 2) a large-scale representative survey among a sample of the Dutch public. The applicants will assess the knowledge of and opinions about AI, among various social segments in the Netherlands. They also examine how specific aspects of AI-driven decision-making affect people’s opinions.

AI into practice: prostate biopsy diagnostics

Applicants: Geert Litjens, Jeroen van der Laak, Katrien Grünberg

External parties: Patrick de Boer, David Tellez, Wouter Bulten (Patholyt), Utrecht Medical Center, Linköping University Hospital, Sectra Medical, Qserve Group.

The field of pathology is undergoing a revolution: from analog microscopes to digital imaging. This transition requires heavy upfront investments and is partly driven by the promise of AI to improve the workflow through an increase in quality, consistency and efficiency, and a decrease in costs. The growing worldwide shortage of pathologists and a focus on reducing healthcare costs raises concerns about the accessibility of high-quality pathology diagnostics. Radboudumc and Patholyt aim to prospectively validate the algorithm for AI-assisted prostate cancer grading within clinical diagnostics in multiple centers in Europe. To build toward this goal, this AI voucher project will consist of three phases: 1. The AI solution for prostate cancer will be integrated into the Sectra PACS clinical diagnostics system installed in Radboudumc. 2. Radboudumc will conduct a pilot reader study to test the previous integration and provide feedback from technicians and pathologists. 3. An international public-private consortium will be created to design a study protocol to investigate the impact of AI-powered solutions for prostate cancer grading integrated into the clinical workflow.

The right dose at the right time: AI for safe and effective paediatric drug use

Applicants: Jolien J.M. Freriksen, Saskia N. de Wildt

Collaborators: Martha Larson, Arjen de Vries

External parties: Rens van de Schoot (Utrecht University), Tjitske van der Zanden (Erasmus MC),

Inge Holsappel (Royal Dutch Pharmacy Association and Dutch Pediatric Formulary).

Many medicines are not registered for use in children and therefore the majority of the drug labels contain no information on paediatric dosing. Prescribing drugs to children ‘off-label’ is common practice. It is essential that clinicians can consult reliable and up-to-date information to make well-informed decisions with respect to paediatric dosing. This project assesses the feasibility of using AI-software for searching, screening and selecting relevant scientific literature at the Dutch Paediatric Formulary. First, currently available benefit-risk analysis documents from the formulary will be converted into AI-software compatible datasets. Second, simulation studies with different feature extraction techniques will be performed to assess the performance of AI and to find out which model can best be used for future unlabeled data. Finally, a paper will be prepared not only showing results of the current project, but also describing possible directions for further research to get AI-software implemented in the current workflow of the Dutch Paediatric Formulary. The results of the project will be widely disseminated through the international network of the project participants as the Formulary is used across Europe with wider plans for dissemination in collaboration with the Bill and Melinda Gates Foundation.

Detecting manmade structures from space

Applicants: Aafke Schipper, Konrad Mielke, Mark Huijbregts, Tom Claassen

External parties: Gertjan Geerling (Deltares), Johan Meijer (PBL Netherlands Environmental Assessment Agency)

Manmade structures like roads, railways, mines, dams and (aquaculture) farms provide a multitude of essential societal benefits, including accessibility, transport, resources, energy generation, water security and food. However, they also have negative consequences for the environment. Applied environmental research institutes like Deltares and PBL Netherlands Environmental Assessment Agency need up-to-date georeferenced information on these artificial structures in order to evaluate the trade-offs among their benefits and impacts. Currently, however, this information is not comprehensively available across the globe. The ongoing developments in machine learning and remote sensing open up promising opportunities for alternative approaches to manual geo-referencing for mapping manmade structures. This project will prepare a grant proposal that focuses on developing and applying novel machine-learning approaches for the global mapping of manmade structures based on remote sensing imagery. In parallel, the applicants will provide a proof of concept by building and evaluating a prototype of a convolutional neural network (CNN) trained to detect dams from satellite imagery.