PRIVASA research group from Health Technology lab at Turku University of Applied Sciences (TUAS), Finland has won 2nd position in the International Federated Tumor Segmentation challenge. FeTS2021 is the first international competition on Federated Learning across domains. The challenge focused on application of a machine learning technique called Federated Learning on medical imaging.
In the healthcare domain, privacy of patient data is critically important making it challenging to train machine learning models on data from multiple geographically distinct institutions. Federated learning offers a privacy preserving solution to train a global model across multiple institutions while keeping all of the sensitive data on the respective local sites.
The task of FeTS2021 challenge was to improve federated learning by identifying an efficient way to aggregate brain segmentation information coming from individual health organizations. The HT-TUAS team developed two novel and robust methods for adaptive weight aggregation, says Irfan Khan, a Research Engineer in PRIVASA.
The algorithm was trained for the objective of segmenting glioblastoma brain tumors in MRI scans. Glioblastoma is the most aggressive brain tumor with a life expectancy of ~14 months post treatment. Computer aided segmentation of the tumor can assist the Radiologist in improved prognosis. The novel method can be used for generalizable machine learning in real-world clinical practice and production environments in geographically distinct institutions, says Dr. Suleiman Khan, PRIVASA collaborator.
The HT-TUAS team was the only team to win a top position in both the challenge leaderboards of the global machine learning competition with 125 globally registered participants. In the competition sponsored by Intel Corporation, more than 55 worldwide institutions attempted to tackle the brain tumor segmentation challenge including University of Pennsylvania (USA), MD Anderson Cancer Center (USA), German Cancer Research Center (DKFZ), and National Institutes of Health (NIH, USA).
The challenge was hosted in a virtual BrainLes Workshop held in Strasbourg, France, in the 24th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI). MICCAI is one of the leading conferences in medical imaging attracting world leading biomedical scientists, engineers, and clinicians from a wide range of disciplines associated with medical imaging and computer assisted intervention.
The aggregation techniques developed in PRIVASA bring agility and flexibility to accelerate the product development of AI enterprises operating in Finland and international healthcare businesses and markets. PRIVASA creates a privacy-protecting ecosystem that allows enterprises to develop their software products on sensitive data without sharing with the third party, says Dr. Mojtaba Jafaritadi, Principal Lecturer in TUAS and Principal Investigator in PRIVASA.
PRIVASA – Privacy Preserving AI for Synthetic and Anonymous Health Data is a three-years research project funded by Business Finland and coordinated by the University of Turku. It aims at developing secure data analysis frameworks that empower multiple institutions, hospitals, and clinics to harness remote, private and encrypted patient data without sharing the patient data.
International challenges have become a standard for validation of biomedical image analysis methods. As a top performing team in the FeTS2021 challenge, TUAS will join hands with the global leaders to contribute in the scientific consortium studies for collaborative learning.
HT-TUAS team in FeTS2021 Challenge:
Muhammad Irfan Khan, TUAS, Finland
Suleiman Khan, PRIVASA collaborator, University of Helsinki, Finland
Esa Alhoniemi, TUAS, Finland
Elina Kontio, TUAS, Finland
Mojtaba Jafaritadi, TUAS, Finland
Interested in learning more?
Watch the Research Engineer Irfan Khan’s presentation at MICCAI Conference.