AI for automatic tumour response measurement on CT scans
Plain Medical, an AI spin-off from Radboudumc and Fraunhofer MEVIS, has joined forces with Radboudumc in this newly established public-private partnership to bring 'Next Generation RECIST' technology to the clinic. The project develops artificial intelligence that automatically measures and tracks tumours over time on thorax-abdomen CT scans, enabling reliable automated response assessment for cancer patients.
Cancer is a leading cause of death and a growing burden on healthcare. In the Netherlands more than 2.5 million CT scans are performed each year, a five-fold increase since 2000, driven largely by oncology follow-up. At the same time up to 46% of radiologists report symptoms of burnout and 3-5% of radiology reports contain errors. Manually applying RECIST – the standard method to judge whether tumours shrink, remain stable or grow – is laborious, subjective and a recognised bottleneck in oncology care.
Building on Radboudumc's Universal Lesion Segmentation model and Plain Medical's auto-RECIST prototype, the partners will train the AI on a unique large-scale dataset of around 150,000 longitudinal CT studies from three Dutch hospitals. New modules for longitudinal lesion tracking, automated detection of new lesions and optimal selection of target lesions will be added and integrated into Plain's clinical reporting platform. Radboudumc contributes clinical and methodological expertise; Plain Medical leads AI development and product integration.
Deliverables include a curated longitudinal lesion dataset, an interactive segmentation model, an improved auto-RECIST algorithm with temporal consistency, dedicated AI modules for target-lesion selection and new-lesion detection, and validation reports from controlled usability studies. Together they bring the integrated system to TRL 6, ready for subsequent clinical validation and CE certification.
