Tractography and machine learning: Current state and open challenges.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Traditional tractography methods have limitations in reconstructing white matter pathways.
• Supervised machine learning (ML) algorithms offer potential solutions by incorporating anatomical priors and making non-local decisions.

Purpose of the Study:

• To address the lack of standardized frameworks for training, evaluating, and comparing ML-based tractography methods.
• To present datasets and evaluation tools beneficial for ML algorithms in neuroimaging.

Main Methods:

• Review of existing ML-based tractography techniques.
• Description of relevant datasets and evaluation tools for ML algorithms.
• Discussion of current evaluation strategies and their limitations.

Main Results:

• ML methods show promise in reconstructing larger white matter bundles with fewer false positives and improved robustness.
• No single ML method has yet achieved conclusive performance or widespread adoption.

Conclusions:

• Standardized frameworks are crucial for the advancement and adoption of ML in tractography.
• Future work should focus on addressing the specific needs of ML tractography methods with tangible solutions.