Deep learning-based diffusion MRI tractography: Integrating spatial and anatomical information.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Diffusion MRI tractography visualizes brain white matter pathways, crucial for neuroscience and clinical studies.
• Current deep learning methods improve tractogram coverage but often generate false positives due to reliance on local information.
• Accurate reconstruction of white matter tracts remains a significant challenge in the field.

Purpose of the Study:

• To develop a novel deep learning framework to enhance the accuracy of diffusion MRI tractography.
• To improve the prediction of long-range streamline propagation by integrating spatial and anatomical information.
• To address fiber class imbalance during training using a weighted loss function.

Main Methods:

• A novel deep learning framework integrating image-domain spatial information (convolutional layers) and anatomical information (Transformer-decoder).
• Utilized a weighted loss function to mitigate fiber class imbalance during model training.
• Evaluated on the ISMRM 2015 Tractography Challenge and Tractoinferno datasets.

Main Results:

• Achieved 66.2% valid streamline rate and 63.8% white matter coverage on the ISMRM dataset, reconstructing 24/25 bundles.
• Demonstrated improved performance on the Tractoinferno dataset, with a 5.7% increase in white matter coverage and 4.1% decrease in overreach compared to RNN-based methods.
• The framework effectively handles diverse diffusion MRI acquisition schemes.

Conclusions:

• The proposed deep learning framework significantly improves the accuracy and coverage of diffusion MRI tractography.
• Integrating spatial and anatomical information enhances streamline prediction, overcoming limitations of local information reliance.
• This method offers a promising advancement for studying brain connectivity and neurological disorders non-invasively.