Using deep learning for a diffusion-based segmentation of the dentate nucleus and its benefits over atlas-based...

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Radiology

Background:

• The dentate nucleus (DN) is crucial for motor control and cognitive functions.
• Accurate DN segmentation is vital for its use as a neuroimaging biomarker.
• Current manual segmentation is time-consuming, and atlas-based methods lack precision.

Purpose of the Study:

• To develop and validate a deep learning (DL) algorithm for automated dentate nucleus segmentation.
• To assess the performance of DL-based segmentation against manual tracing and atlas-based methods.
• To enable scalable and accurate DN volume estimation from common MRI sequences.

Main Methods:

• A deep learning algorithm was trained to automatically segment the dentate nucleus.
• The algorithm utilized T1-weighted, T2-weighted, and diffusion MRI sequences.
• Segmentation accuracy was evaluated using the Dice Similarity Coefficient (DSC) compared to manual labels.

Main Results:

• The DL approach achieved high agreement with manual DN segmentation.
• Fractional Anisotropy (FA) maps from diffusion MRI yielded the highest DSC (0.83).
• The DL method significantly outperformed single-atlas (DSC=0.23) and multi-atlas (DSC=0.33) segmentation.

Conclusions:

• Automated DL segmentation provides accurate and reproducible dentate nucleus delineation on clinical MRI.
• This method offers a scalable alternative to manual tracing and improves upon existing atlas-based techniques.
• The developed technique enhances the utility of the dentate nucleus as a biomarker in neurological research.