Skip to main content
Go to the home page of the European Commission (opens in new window)
English en
CORDIS - EU research results
CORDIS

A Foundation Model for Next-Generation Generalizable AI in Neuro-Radiology

Project description

Smarter AI for brain imaging

Radiology stands to benefit from AI, but progress has been slowed by the need for vast labelled datasets. Additionally, the adaptability of conventional AI tools is limited. Most models struggle to generalise across tasks or patient populations, holding back clinical adoption. With this in mind, the ERC-funded project will develop a visual foundation model trained on over 200 000 brain scans using self-supervised learning. Focused on neuro-radiology, this model will support a wide range of applications, from emergency triage to dementia risk prediction, using minimal labelled data. With open-source tools and large-scale training, the project is paving the way for more flexible, accurate, and accessible AI in medical imaging.

Objective

Medical artificial intelligence (AI) holds immense promise for transforming radiology by introducing advanced diagnostic capabilities. Yet, traditional AI models face challenges like extensive data annotation needs and are task-specific with limited generalizability to different scenarios. This problem of robust and label-efficient generalization continues to be a key translational challenge for medical AI models and has prevented their broad uptake in real world healthcare settings. AI-Next will build on the recent advent of foundation models, providing a unique opportunity to rethink the development of medical AI and overcome the challenges of traditional AI models. At the core of AI-Next stands the development of a state-of-the-art visual foundation model (VFM) that learns generalizable representations from unlabelled radiology scans and provides a basis for label-efficient model adaptation in several applications. The VFM will focus on neuro-radiology and be trained with data at unprecedented scale including >200,000 brain computed tomography and magnetic resonance imaging scans from population-based and disease specific cohorts, leveraging self-supervised learning. The VFM will be adapted with a limited set of explicit labels to a range of tasks with clinical significance. This includes the screening and triage of scans for emergency findings, longitudinal disease activity assessment, risk prediction of dementia, forecasting of disease evolution and finally automated radiology reporting. Moreover, the VFM will be employed for image processing tasks, including generating super-resolution images to enhance diagnostic capabilities. Overall, AI-Next will represent a crucial step towards more generalisable, accurate and label-efficient AI in neuro-radiology, offering significant potential for improving diagnostics, clinical decision-making as well as patient outcomes, and its open-source innovations will serve as a blueprint for the broader field of radiology.

Fields of science (EuroSciVoc)

Keywords

Programme(s)

Topic(s)

Funding Scheme

HORIZON-ERC - HORIZON ERC Grants

See all projects funded under this funding scheme

Call for proposal

(opens in new window) ERC-2024-COG

See all projects funded under this call

Host institution

UNIVERSITATSKLINIKUM BONN
Net EU contribution
€ 2 268 799,59
Address
VENUSBERG-CAMPUS 1
53127 Bonn
Germany
Region
Nordrhein-Westfalen Köln Bonn, Kreisfreie Stadt
Activity type
Higher or Secondary Education Establishments
Links
Total cost
€ 2 268 799,59

Beneficiaries (2)