TrackRAD2025: Real-time tumor tracking for MRI-guided radiotherapy

The use of magnetic resonance imaging (MRI) to visualize and characterize motion is becoming increasingly important in treating cancer patients, especially with radiotherapy. In particular, motion management is crucial for tumors affected by respiratory motion such as liver, pancreatic or thoracic tumors, to ensure a high radiation
dose to the tumor and the sparing of neighboring organs. The recent development of MRI-guided radiotherapy, based on hybrid MRI-linear accelerator (linac) systems (1), called MRI-linacs, offers the possibility to adapt to changes in tumor position during treatment. 2D cine-MRI allows real-time tumor motion visualization and allows
closely following the tumor with the radiation beam, but requires tumor segmentation on all time-resolved frames. This needs to be done in real-time, with frame rates of up to 8 Hz or more, with high accuracy and robustness to ensure the sparing of critical organs. Currently, clinically available solutions rely on conventional deformable image registration (DIR) or template matching to propagate contours from a labeled frame and struggle with large non-rigid motion. This limits treatment to beam gating, where the beam is turned off for large motion. The fast inference of artificial intelligence (AI) methods, obtained by shifting computation time to the training phase, is promising for this task (2). TrackRAD2025 will impact the field of MRI-guided radiotherapy by providing cine-MRI data from multiple MRI-linac institutions to test competitive real-time tumor tracking methods based on a unified platform for comparison.

The objective of TrackRAD2025 will be real-time tumor tracking on time-resolved sagittal 2D cine-MRI sequences. Algorithms will be provided with a template tumor segmentation on the first frame, and the remaining 2D cine-MRI sequence requiring real-time segmentation. The submitted algorithms should produce a tumor segmentation mask on each frame.

TrackRAD2025 will provide the first public multi-institutional dataset and evaluation platform to compare the latest developments in cine-MRI-based tumor tracking methods competitively. Both unlabeled (477 patient cases) and labeled datasets (108 patient cases with 2D cine-MRI frame sequences where the tumor is manually
segmented) will be provided for model development and testing. Six international centers will provide data (3 Dutch, 1 German, 1 Australian, and 1 Chinese). The data from 0.35 T and 1.5 T MRI-linacs will be divided into a public training set and a private test set to calculate evaluation metrics. The challenge will feature a preliminary
testing (validation) phase with 8 cases and a final testing phase with 50 cases. Submitted algorithms will be rated for their ability to reproduce ground truth segmentation labels on the test set using the Dice similarity coefficient, Hausdorff and average surface distance, error of the tumor center of mass, estimated radiation dose delivery
accuracy, and the inference speed.

TrackRAD2025 will allow determining the most promising methods to improve clinical tumor tracking on cine-MRI at MRI-linacs, which will benefit patients suffering from various motion-affected tumor entities with more accurate dose delivery. This will also lead the way to multi-leaf collimator tracking at MRI-linacs instead of gating, where the radiation beam continuously follows the movement of the tumor to deliver radiation more efficiently and shorten treatment times for an increased number of treatments per day.

1. Keall PJ, Brighi C, Glide-Hurst C, Liney G, Liu PZY, Lydiard S, et al. Integrated MRI-guided radiotherapy - opportunities and challenges. Nat Rev Clin Oncol. 2022;19(7):458-70
2. Lombardo E, Dhont J, Page D, Garibaldi C, Kunzel LA, Hurkmans C, et al. Real-time motion management in MRI-guided radiotherapy: Current status and AI-enabled prospects. RadiotherOncol. 2024;190:109970.