Multi-fraction image reconstruction for radiotherapy on an MR-Linac

Motivation: Patients undergoing online-adaptive radiotherapy on an MR-Linac are imaged throughout treatment, but insufficient image quality can lead to long contouring times for pancreatic tumours. Goal(s): We aim to improve image quality by exploiting the accumulated MRI raw data across treatment fractions, to decrease contouring times and provide high-quality reference images for real-time algorithms. Approach: Deformation vector fields estimated from non-rigid image registration of daily images were used to reconstruct a midposition reference image combining the data across multiple treatment fractions. Results: Reference images presented fewer streaking artefacts, but new artefacts appeared in areas with large daily changes (stomach, bowels). Impact: Image quality on an MR-Linac can lead to slow contouring of abdominal tumours, limiting the potential of adaptive MR-guided radiotherapy. Combining raw data across fractions can improve image quality, potentially reducing clinical workloads and enabling novel real-time adaptive treatment workflows.