Abstract
Medical image registration is a fundamental task in medical image analysis, with applications ranging from image-guided surgery to longitudinal studies. However, the field lacks standardized benchmarks and evaluation frameworks, making it difficult to compare different approaches. In this work, we present Learn2Reg, a comprehensive multi-task medical image registration challenge that provides standardized datasets and evaluation metrics.
Our challenge encompasses multiple registration tasks including inter-patient, intra-patient, and multi-modal registration scenarios. The framework consists of three main components: (1) a diverse dataset collection covering various anatomical regions and imaging modalities, (2) standardized evaluation metrics that assess both accuracy and robustness, and (3) a comprehensive leaderboard system that enables fair comparison of different approaches.
We evaluate the challenge on a large-scale dataset involving multiple institutions and imaging protocols. Results demonstrate the effectiveness of deep learning approaches in medical image registration while highlighting the importance of standardized evaluation frameworks. The challenge provides valuable insights into the current state of the field and identifies areas for future improvement.
Learn2Reg represents a significant contribution to the medical image registration community, providing a standardized platform for method comparison and advancement of the field.
BibTeX
@article{hering2022learn2reg,
title={Learn2Reg: comprehensive multi-task medical image registration challenge, dataset and evaluation in the era of deep learning},
author={Hering, Alessa and Hansen, Lasse and Mok, Tony CW and Chung, Albert CS and Siebert, Hanna and H{\"a}ger, Stephanie and Lange, Annkristin and Kuckertz, Sven and Heldmann, Stefan and Shao, Wei and others},
journal={IEEE Transactions on Medical Imaging},
volume={42},
number={3},
pages={697--712},
year={2022},
publisher={IEEE},
doi={10.1109/TMI.2022.3213983}
}