Partnership for Online Personalized AI-driven Adaptive Radiation Therapy (POP-AART) is a public-private collaboration between The Netherlands Cancer Institute, the University of Amsterdam and Elekta. The lab focuses on the use of artificial intelligence for precision radiotherapy.
It is a major challenge to give patients the right dose of radiation, at the right spot with least damage to healthy tissue, and while the patient and the tumor move and change shape during radiation and over time. And this should happen at each and every treatment session (which varies from 3 to 35). Within the POP-AART lab six PhD researchers develop novel AI strategies for improving the images on which the radiation treatment is based, predicting changes over time of the tumor and incorporating them in automatic treatment planning and adaptation.
All 6 of the work packages involve deep learning and cover fundamental research topics. These range from improving CT images obtained just before radiation to the level of diagnostic quality CT images, predicting deformations and segmentations of the tumor and organs at risk and incorporate these data in online and automated treatment plan optimization for each patient individually at each radiation session.
Deep Generative Learning for Cone Beam Computed Tomography
Goal: to design deep generative models that improve CBCT image quality while enforcing geometric and pathological integrity
Learning Inverse Models for Cone Beam Computed Tomography Reconstruction
Goal: to learn deep inverse models to infer high quality 3D and 4D CBCT from measured projections that optimally exploit the existing knowledge from the physical acquisition processes
Deep Learning Geometry for 3D Medical Image Registration
Goal: to learn models that optimally register the varying geometries in pairs or series of images of deformable anatomy
Interactive Deep Learning for Medical Segmentation
Goal: to replace static segmentation models with interactive model based approaches that can, therefore, integrate the feedback provided by the experts interactively and can minimize or eliminate even violations of the constrains are necessary
Learning to Forecast for Adaptive Radiation Therapy
Goal: to learn forecasting models that predict dose distributions for a given anatomy and series of future anatomies and associated dose distributions in adaptive radiation treatment
Reinforcement Learning for Radiation Treatment Plan optimization
Goal: to improve and accelerate treatment plan optimization using novel reinforcement learning algorithms for guiding radiation in adaptive radiotherapy treatment
Efstratios Gavves is assistant professor of Computer Vision and Deep Learning at the Informatics Institute at UvA.
Jan-Jakob Sonke is is theme leader image guided therapy at the Netherlands cancer institute and professor by special appointment of Adaptive Radiotherapy at the Faculty of Medicine at the UvA.
Elekta, headquartered in Stockholm, Sweden, is a leader in precision radiation medicine with more than 4,000 employees worldwide.
Netherlands Cancer Institute is among the top 10 comprehensive cancer centers, combining world-class fundamental, translational, and clinical research with dedicated patient care.
University of Amsterdam is the Netherlands’ largest university, offering the widest range of academic programmes.