FeTS

FeTS enables federated training of deep learning models for brain tumor sub-compartment segmentation from magnetic resonance imaging (MRI), harmonizing preprocessing and generating reference labels across institutions while keeping primary patient data local.

Key Features:

• Decentralized Data Analysis: Implements federated learning to train tumor sub-compartment delineation models across multiple sites without centralizing primary patient data.
• Harmonized Processing: Provides consistent curation and preprocessing protocols for MRI data to ensure uniform identification of regions of interest.
• Gold Standard Reference Labels: Supports generation of reference labels for tumor sub-compartments used to train and validate deep learning segmentation models.
• Software Architecture and Functionality: Built on open-source frameworks including the Insight Toolkit and Qt and supports both centralized and federated model training environments.

Scientific Applications:

• Neuro-oncology segmentation: Segmentation of brain tumors into sub-compartments to support diagnosis, treatment planning, and quantitative imaging analyses.
• Distributed model development: Enables multi-institutional federated model development using distributed MRI datasets to improve robustness and generalizability of AI models.
• Reference label generation: Facilitates creation of gold-standard labels for training and validating deep learning segmentation models.
• Personalized medicine research: Supports development of models that can contribute to individualized diagnostic and therapeutic decision-making by leveraging heterogeneous institutional data.

Methodology:

Implements federated learning to train deep learning segmentation models on distributed MRI datasets while maintaining data locality and applying harmonized curation and preprocessing protocols.