Home

Welcome to MeshFL!

Technical Overview

1. MeshNet Model

• A specialized deep learning architecture for volumetric brain MRI segmentation.

• Flexible configuration for input channels, output classes, and training hyperparameters.

2. Federated Learning

• NVFlare ensures privacy-preserving training across decentralized data sites.

• Custom gradient aggregation methods optimize distributed model updates.

3. Data Pipeline

• MRI data is stored in an SQLite database (mindboggle.db) with zlib compression.

• Automatic dataset partitioning into training, validation, and testing subsets using splits.json.