TractoInferno - A large-scale, open-source, multi-site database for machine learning dMRI tractography.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Tractography is crucial for mapping white matter pathways in the brain.
• Developing robust machine learning (ML) algorithms for tractography requires large, standardized datasets.
• Existing datasets often lack diversity in acquisition sites and processing protocols.

Purpose of the Study:

• To introduce TractoInferno, the world's largest open-source multi-site tractography database.
• To provide a standardized resource for training and evaluating ML tractography algorithms.
• To facilitate advancements in diffusion MRI analysis.

Main Methods:

• Compiled 284 human diffusion MRI (dMRI) datasets from 6 different 3T scanner sites.
• Included T1-weighted images, dMRI data, spherical harmonics, fiber ODFs, and reference streamlines for 30 bundles.
• Generated masks for tractography algorithms and performed manual quality control throughout the pipeline.

Main Results:

• The TractoInferno database comprises 350 GB of data.
• It includes reference streamlines from 4 tractography algorithms and benchmarked ML models.
• Database creation involved significant computational resources (20,000 CPU-hours, 3,000 GPU-hours) and manual effort (200 man-hours).

Conclusions:

• TractoInferno provides a valuable, standardized resource for the ML tractography community.
• It addresses common challenges in ML tractography by offering diverse data and a consistent evaluation protocol.
• This database will accelerate the development and validation of novel tractography algorithms.