MRIgRT real-time target tracking: TrackRAD2025 challenge report

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Artificial Intelligence in Medicine

Background:

• Magnetic resonance imaging (MRI)-guided radiotherapy (MRIgRT) enhances real-time motion management during treatment using cine-MRI.
• Current target localization methods in MRIgRT, such as template matching and deformable image registration, have limitations.
• Advanced localization algorithms are crucial for improving the precision and efficiency of radiotherapy delivery.

Purpose of the Study:

• To establish a benchmark dataset and challenge (TrackRAD2025) for evaluating algorithms in MRIgRT target tracking using 2D+t cine-MRI.
• To compare the performance of various algorithms in propagating target segmentation masks across dynamic MRI sequences.
• To assess the impact of different factors, like field strength and anatomical site, on tracking accuracy and speed.

Main Methods:

• The TrackRAD2025 challenge provided a multi-institutional dataset of cine-MRI scans from 585 cancer patients.
• Participants developed algorithms to track target segmentation masks across frames in sagittal cine-MRI data.
• Submitted algorithms were evaluated on the challenge platform using geometric accuracy, surrogate dose accuracy, and execution speed metrics.

Main Results:

• The top five algorithms achieved high performance, with mean Dice similarity coefficients >0.87 and Euclidean center distances <2.1 mm, comparable to interobserver variability.
• Foundation models, with or without finetuning, were prominent among the top-performing solutions.
• MRI field strength had minimal impact; tracking performance was better in the pelvis (reduced motion) than in the thorax and abdomen.

Conclusions:

• TrackRAD2025 successfully benchmarked MRIgRT target tracking algorithms on diverse, multi-institutional cine-MRI data.
• Foundation models show significant promise for clinical translation in MRIgRT target tracking.
• The results provide a foundation for developing more precise and efficient radiotherapy delivery systems.