CNTSeg

CNTSeg segments cranial nerve (CN) tracts from diffusion magnetic resonance imaging (dMRI) and complementary modalities to enable quantitative analysis of CN morphology and trajectories.

Key Features:

• Automated CN tract segmentation: Performs voxel-wise segmentation of cranial nerve tracts without relying on reference streamlines.
• Multimodal inputs: Integrates T1-weighted images, fractional anisotropy (FA) images, and fiber orientation distribution function (fODF) peaks.
• Deep-learning multi-class network: Uses a multi-class deep neural network to discriminate and segment multiple CN classes.
• Interphase feature fusion: Fuses complementary information across modalities through interphase feature fusion to improve segmentation accuracy.
• Tractography-independent workflow: Circumvents tractography-based methods that depend on regions-of-interest (ROIs) placement or clustering.
• Empirical evaluation: Validated through comparisons and ablation experiments demonstrating performance in challenging segmentation scenarios.

Scientific Applications:

• Quantitative CN morphology and course analysis: Enables measurement and comparison of cranial nerve morphology and trajectories from imaging data.
• Segmentation of specific cranial nerves: Applied to segment five pairs of cranial nerves, including optic nerve (CN II), oculomotor nerve (CN III), trigeminal nerve (CN V), and facial-vestibulocochlear nerve (CN VII/VIII).
• dMRI-based CN tract analysis in complex anatomy: Suited for studying slender CN structures within complex anatomical environments where tractography may fail.

Methodology:

Multimodal inputs (T1-weighted, FA, fODF peaks) are fused via interphase feature fusion and processed by a deep-learning multi-class segmentation network, with performance assessed through comparisons and ablation experiments.