CNTSeg: A multimodal deep-learning-based network for cranial nerves tract segmentation.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Diffusion magnetic resonance imaging (dMRI) is crucial for analyzing cranial nerve (CN) tracts.
• Current dMRI-based segmentation methods struggle with the slender structure and complex anatomy of CNs, leading to low accuracy.
• Tractography, ROI-based, and clustering approaches have limitations in individualized CN segmentation.

Purpose of the Study:

• To develop a novel multimodal deep-learning-based network for automated cranial nerve tract segmentation.
• To overcome the limitations of single-modality dMRI and existing segmentation algorithms.
• To achieve accurate and automated segmentation of multiple cranial nerve pairs.

Main Methods:

• Proposed CNTSeg, a multimodal deep-learning multi-class network for automated CN tract segmentation.
• Integrated T1w images, fractional anisotropy (FA) images, and fiber orientation distribution function (fODF) peaks.
• Designed a back-end fusion module for interphase feature fusion to enhance segmentation performance.

Main Results:

• Successfully segmented 5 pairs of cranial nerves: optic nerve (CN II), oculomotor nerve (CN III), trigeminal nerve (CN V), and facial-vestibulocochlear nerve (CN VII/VIII).
• Demonstrated promising and anatomically convincing results, even for challenging tracts.
• Achieved accurate segmentation without relying on tractography, ROI placement, or clustering.

Conclusions:

• CNTSeg offers an effective and automated solution for cranial nerve tract segmentation using multimodal MRI data.
• The multimodal approach and fusion module significantly improve segmentation accuracy and robustness.
• The developed method holds potential for quantitative analysis of CN morphology and course in clinical and research settings.