The Federated Tumor Segmentation (FeTS) Challenge 2024

International challenges have become the standard for validation of biomedical image analysis methods. We argue, though, that the actual performance even of the winning algorithms on ``real-world`` clinical data often remains unclear, as the data included in these challenges are usually acquired in very controlled settings at few
 institutions. The seemingly obvious solution of just collecting increasingly more data from more institutions in such challenges does not scale well due to privacy and ownership hurdles.
 
We build upon the first-ever proposed federated learning challenge, Federated Tumor Segmentation (FeTS) 2021 and its follow-up in 2022, to introduce FeTS 2024, intending to address these hurdles, for the creation of tumor segmentation models. Specifically, the FeTS 2024 challenge will use clinically acquired, multi-institutional
multi-parametric magnetic resonance imaging (mpMRI) scans from the RSNA-ASNR-MICCAI BraTS 2021 challenge, and BraTS 2023 Glioma challenge, as well as from various remote independent institutions included in the collaborative network of a real-world federation (www.fets.ai). The FeTS 2024 challenge focuses on innovating at
the level of federated aggregation where locally trained models combine to form the consensus model for the segmentation of intrinsically heterogeneous (in appearance, shape, and histology) brain tumors, namely gliomas (and particularly the radiographically appearing glioblastomas). Compared to the BraTS challenge [1-4], the ultimate goal of FeTS is the creation of a consensus segmentation model that has gained knowledge from data of multiple institutions without pooling their data together (i.e., by retaining the data within each institution).
 
Since the conception of FeTS 2021 and FeTS 2022, Federated Learning has matured to a more active research field in biomedical AI. What separates FeTS 2024 challenge from those of previous years is that while FeTS 2021 and FeTS 2022 focused on semantic segmentation, now focuses on instance segmentation which in line with the
findings of [11-12] stands superior due to the added value of evaluating multiple individual tumors per patient.
 
This year we plan to further broaden the final evaluation of the aggregation methods, by testing their generalizability beyond segmentation tasks to a hidden (to the participants) task. This added evaluation will be a significant part of the final challenge paper which will provide detailed meta-analysis and inform us further insights about the developed aggregation methods. It will also inform further expansion of the tasks planned for next year's challenge proposal. For fairness, since the participants will only have access to the segmentation data, this added evaluation on a new task will neither be considered for the monetary awards, nor for the ranking. We will however announce performance on this hidden task during the challenge results presentation at MICCAI.