Mask the Unknown: Assessing Different Strategies to Handle Weak Annotations in the MICCAI2023 Mediastinal Lymph Node Quantification Challenge

Stefan M. Fischer1,2,3,4Orcid, Johannes Kiechle1,2,3,4Orcid, Daniel M. Lang1,3Orcid, Jan C. Peeken2Orcid, Julia A. Schnabel1,3,4,5Orcid
1: School of Computation, Information and Technology, Technical University Munich, Germany, 2: Department of RadioOncology, Klinikum rechts der Isar, Technical University Munich, Germany, 3: Institute of Machine Learning in Biomedical Imaging, Helmholtz Munich, Germany, 4: Munich Center of Machine Learning (MCML), Germany, 5: School of Biomedical Engineering and Imaging Sciences, King’s College London, UK
Publication date: 2024/06/14
https://doi.org/10.59275/j.melba.2024-8g8b
PDF · Code · arXiv

Abstract

Pathological lymph node delineation is crucial in cancer diagnosis, progression assessment, and treatment planning. The MICCAI 2023 Lymph Node Quantification Challenge published the first public dataset for pathological lymph node segmentation in the mediastinum. As lymph node annotations are expensive, the challenge was formed as a weakly supervised learning task, where only a subset of all lymph nodes in the training set have been annotated. For the challenge submission, multiple methods for training on these weakly supervised data were explored, including noisy label training, loss masking of unlabeled data, and an approach that integrated the TotalSegmentator toolbox as a form of pseudo labeling in order to reduce the number of unknown voxels. Furthermore, multiple public TCIA datasets were incorporated into the training to improve the performance of the deep learning model. Our submitted model achieved a Dice score of 0.628 and an average symmetric surface distance of 5.8~mm on the challenge test set. With our submitted model, we accomplished the third rank in the MICCAI2023 LNQ challenge. A finding of our analysis was that the integration of all visible, including non-pathological lymph nodes improved the overall segmentation performance on pathological lymph nodes of the test set. Furthermore, segmentation models trained only on clinically enlarged lymph nodes, as given in the challenge scenario, could not generalize to smaller pathological lymph nodes. The code and model for the challenge submission are available at https://gitlab.lrz.de/compai/MediastinalLymphNodeSegmentation

Keywords

deep learning · lymph node quantification · weakly supervised learning · image segmentation

Bibtex @article{melba:2024:008:fischer, title = "Mask the Unknown: Assessing Different Strategies to Handle Weak Annotations in the MICCAI2023 Mediastinal Lymph Node Quantification Challenge ", author = "Fischer, Stefan M. and Kiechle, Johannes and Lang, Daniel M. and Peeken, Jan C. and Schnabel, Julia A.", journal = "Machine Learning for Biomedical Imaging", volume = "2", issue = "MICCAI 2023 LNQ challenge special issue", year = "2024", pages = "798--816", issn = "2766-905X", doi = "https://doi.org/10.59275/j.melba.2024-8g8b", url = "https://melba-journal.org/2024:008" }
RISTY - JOUR AU - Fischer, Stefan M. AU - Kiechle, Johannes AU - Lang, Daniel M. AU - Peeken, Jan C. AU - Schnabel, Julia A. PY - 2024 TI - Mask the Unknown: Assessing Different Strategies to Handle Weak Annotations in the MICCAI2023 Mediastinal Lymph Node Quantification Challenge T2 - Machine Learning for Biomedical Imaging VL - 2 IS - MICCAI 2023 LNQ challenge special issue SP - 798 EP - 816 SN - 2766-905X DO - https://doi.org/10.59275/j.melba.2024-8g8b UR - https://melba-journal.org/2024:008 ER -

2024:008 cover