2024: CO-VISION – testing of Co-registration of the scans lesion tracking and cancer dynamics assessment
The CO-VISION project addressed the challenge of comparing multiple CT scans taken from the same patient over time which is a time-consuming task for radiologists. The project aimed to validate the use of AI for automating image co-registration, enabling more accurate and efficient tracking of tumour progression. We implemented and tested three co-registration approaches: CorrField, a classical algorithm; a U-Net-based model trained on CorrField outputs; and a GAN-based model (Vox2Vox) that learns deformation fields directly. To benchmark performance, we adjusted a synthetic data generation pipeline that simulates longitudinal scan pairs across different organs. While initial testing was conducted on lung data, we extended the solution to kidney scans, demonstrating its scalability and cross-organ applicability. The project confirmed that AI-based co-registration is both technically feasible and clinically valuable, laying the groundwork for broader adoption in radiological workflows.