Turku UAS team receives third position in International Federated Tumor Segmentation challenge
The PRIVASA research group from the Health Technology lab at Turku University of Applied Sciences (HT-TUAS), Finland has received an honorary award at 3rd position in the International Federated Tumor Segmentation challenge (FeTS2022).
The aim of task 1 in the FeTS2022 challenge was to identify tumor regions in the brain and efficiently train a machine learning model in distributed settings on >8,000 clinically acquired multi-institutional MRI images and design an algorithm that performs well on an unseen test dataset.
– After successful contribution of the HT-TUAS team in FeTS2021, which resulted in the 2nd position in two leaderboards, we developed yet another novel method for a regularized weight aggregation, says Irfan Khan, Research Engineer at Turku UAS.
The team proposed a mathematical solution to regularize model weight aggregation in brain lesion segmentation. Their principled approach is scalable, cost-efficient, and adaptable to heterogeneous data.
The research presented at the FeTS2022 is related to the PRIVASA project funded by Business Finland .
– Our goal is to design and develop a distributed data processing framework that allows enterprises to develop their software products on encrypted and decentralized data from multiple institutions, hospitals, and clinics without sharing the patient data. The consistent performances of our federated learning methods in FeTS2021 and FeTS2022 clearly indicate that the developed models are robust and generalizable; this is a big achievement, says PRIVASA project leader and Principal Lecturer Mojtaba Jafaritadi from Turku UAS.
HT-TUAS team is very satisfied with the results. The Health Technology research group leader Elina Kontio says that health tech researchers aspire to work in clinical informatics academic projects with direct potential implications in translational medicine.
– PRIVASA is on course to become a center of excellence in federated machine learning research, says Esa Alhoniemi, AI Specialist at Turku UAS.
The organizers acknowledged that considering the competition as a pure computational challenge and using FeTS2021 evaluation criterion, team HT-TUAS achieved third position in the final rankings of FeTS2022. The team was invited to present the approach developed in the challenge in the 8th BrainLes workshop in 25th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) in Singapore in September.
– The fact that other competing algorithms took inspiration from our proposed algorithm from last year. This testimony advocates for being a “seminal work” in the field, says Turku UAS Fellow Suleiman A. Khan.
The Federated Tumor Segmentation (FeTS) is the largest “real-world” federated machine learning initiative with 71 healthcare organizations, institutions and research centers collaborating across six continents. A total of 84 teams registered globally in the FeTS2022 challenge.
– Distributed, collaborative and Federated Learning is crucial for health-care research in the future, says Research Engineer Irfan Khan.