The reliability and accuracy of artificial intelligence (AI) algorithms can degrade when such algorithms are applied to new, varied datasets — a common challenge known as “distribution shift”. For example, an AI model trained to detect tumors in MRI scans from a single hospital may not perform well on scans from another hospital. To address this issue, we aim to adapt the internal representations of “foundation models”, which are pre-trained on diverse data without specific labels. These adapted representations may allow us to partially mitigate distribution shift and boost the model’s performance in a broader environment.

Dutch cancer care struggles with escalating patient numbers, costs, and staff shortages. IMAGINE tackles this by advancing Imaging and Image-Guided Interventions (IGI). Like GPS for navigation, IGI incorporates medical imaging into cancer care, ensuring precision and efficiency. Minimally invasive, IGI minimizes side effects, hastens recovery, and reduces hospital stays, alleviating healthcare strain. To unlock IGIs potential, IMAGINE will develop new treatment concepts by integrating imaging with scientific insights from key technologies from the fields of medical physics & artificial intelligence. Demand-driven, IMAGINE accelerates IGI development and implementation through an open innovation lab, fostering collaboration among academics, clinicians, and industry.