MICCAI 2022 Daily – Tuesday

data for algorithmic training, while their data are never shared and instead we bring the algorithms to the data. The focus of the FeTS challenge revolves around the benchmarking of methods for federated learning (FL) and is specifically a two-fold focus across its two main tasks, i.e., training and evaluation. Task 1 (“Federated Training”) seeks the engagement of our scientific community in the development of effective FL weight aggregation methods for the creation of a consensus model, given a pre-defined segmentation algorithm for training, where we already see very interesting approaches being developed. Task 2 (“Federated Evaluation, in-the- wild”) seeks algorithmic generalizability on out-of-sample data based on federated evaluation of segmentation algorithms. In this 2 nd task, algorithms are evaluated during the testing phase on >10,400 scans by being securely distributed across 32 independent institutions from the collaborative network of the FeTS initiative (https://www.fets.ai/, Fig.2), using the MedPerf platform, released by the ML Commons Medical working group. MedPerf is an open-source framework for benchmarking machine learning models to deliver clinical efficacy, while prioritizing patient privacy and mitigating legal and regulatory risks. The FeTS challenge builds on the success of our International Brain Tumor Segmentation (BraTS) challenge, and particularly its 2021 instance, where we brought together the Radiological Society of North America (RSNA) and the American Society of Neuro-Radiology (ASNR), to collaborate and contribute in the RSNA-ASNR-MICCAI BraTS 2021 challenge. This collaborative effort led to i) the creation of the BraTS 2021 centralized dataset of >8,000 clinically-acquired multi-institutional MRI scans provided for the training, validation, and testing of algorithms, and ii) the volunteering of more than 60 clinical expert neuroradiologists to create the ground truth labels in the largest publicly-available curated dataset for such a multi-label multi- parametric problem (Fig.1). In addition to the imaging data from BraTS 2021, the FeTS challenge provided the real-world data partitioning according to the originating institution. Being the first of its kind, the success of the FeTS challenge is a product of the tremendously collaborative effort from all the members of its organizing committee, who contributed to the challenge design, and software development, as well as monetary awards, and included members of the University of Pennsylvania, the German Cancer Research 25 DAILY MICCAI Tuesday Federated Tumor Segmentation