Automated Segmentation of Trigeminal Nerve and Cerebrovasculature in MR-Angiography Images by Deep Learning

Jinghui Lin, Lei Mou, Qifeng Yan, Shaodong Ma, Xingyu Yue, Shengjun Zhou, Zhiqing Lin, Jiong Zhang, Jiang Liu, Yitian ZhaoFrontiers in Neuroscience 2021 · DOI

Abstract

Trigeminal neuralgia caused by paroxysmal and severe pain in the distribution of the trigeminal nerve is a rare chronic pain disorder. It is generally accepted that compression of the trigeminal root entry zone by vascular structures is the major cause of primary trigeminal neuralgia, and vascular decompression is the prior choice in neurosurgical treatment. Therefore, accurate preoperative modeling/segmentation/visualization of trigeminal nerve and its surrounding cerebrovascular is important to surgical planning. In this paper, we propose an automated method to segment trigeminal nerve and its surrounding cerebrovascular in the root entry zone, and to further reconstruct and visual these anatomical structures in three-dimensional (3D) Magnetic Resonance Angiography (MRA). The proposed method contains a two-stage neural network. Firstly, a preliminary confidence map of different anatomical structures is produced by a coarse segmentation stage. Secondly, a refinement segmentation stage is proposed to refine and optimize the coarse segmentation map. To model the spatial and morphological relationship between trigeminal nerve and cerebrovascular structures, the proposed network detects the trigeminal nerve, cerebrovasculature, and brainstem simultaneously. The method has been evaluated on a dataset including 50 MRA volumes, and the experimental results show the state-of-the-art performance of the proposed method with an average Dice similarity coefficient, Hausdorff distance, and average surface distance error of 0.8645, 0.2414, and 0.4296 on multi-tissue segmentation, respectively.

Introduction

Trigeminal Neuralgia (TN) is a debilitating condition characterized by sudden, excruciating facial pain. The primary cause is typically neurovascular compression, where a nearby blood vessel presses against the trigeminal nerve at its root entry zone (REZ) near the brainstem. Microvascular Decompression (MVD) is the preferred surgical treatment, requiring precise preoperative visualization of the nerve and surrounding vasculature. However, the trigeminal nerve is minute and located in a complex anatomical region, making manual segmentation in 3D MR-Angiography (MRA) time-consuming and prone to variability. This study presents an automated deep learning framework to provide high-fidelity 3D reconstructions for surgical planning.

Methodology

The proposed framework utilizes a coarse-to-fine 3D segmentation strategy to handle the complex morphology and small scale of the target structures.

Two-Stage Architecture

  1. Coarse Stage: This stage employs a modified $CS^2Net$ backbone. Standard residual blocks are replaced with Res2Blocks, which utilize multi-scale feature extraction within a single layer. This allows the network to capture both large-scale vascular structures and the fine details of the trigeminal nerve fibers. The output is a preliminary confidence map.
  2. Refine Stage: A 3D U-Net processes the original MRA data concatenated with the confidence map from the coarse stage. This stage acts as a refinement mechanism, focusing on boundary accuracy and anatomical continuity.
The proposed two-stage coarse-to-fine segmentation framework, featuring Res2Blocks and a refinement stage.

Loss Function and Optimization

To address the extreme class imbalance—where the trigeminal nerve occupies a negligible fraction of the total volume—a hybrid loss function was implemented. This combined Weighted Cross-Entropy (assigning the nerve a weight 300 times higher than the background) with Dice Loss to optimize overlap accuracy.

Data and Results

The model was trained and validated on the TRINE (TRIgeminal NErve) dataset, consisting of 50 MRA volumes. Ground truth annotations were provided by a senior neurosurgeon. The proposed method achieved state-of-the-art performance, outperforming established models like 3D U-Net and AnatomyNet.

  • Trigeminal Nerve (TN): Dice Similarity Coefficient (DSC) of 0.8024.
  • Brainstem (BS): DSC of 0.9344.
  • Cerebrovasculature (CV): DSC of 0.8568.
Visualization of MRA data and the resulting 3D segmentation of the trigeminal nerve (red), vasculature (cyan), and brainstem (white).

The inclusion of the brainstem as a spatial "anchor" proved vital for locating the nerve. Furthermore, the multi-scale capabilities of the Res2Blocks prevented the small nerve structures from being lost during feature extraction. Qualitatively, the 3D renderings demonstrated superior continuity and surface smoothness compared to baseline methods, providing a clearer "roadmap" for neurosurgeons.

Conclusion

This automated framework provides a robust solution for the 3D mapping of the trigeminal nerve and surrounding vasculature. By leveraging a coarse-to-fine approach and multi-scale feature extraction, the model achieves high accuracy in a challenging anatomical domain. Future work will focus on improving generalization across different MRI scanner strengths and quantifying the degree of neurovascular compression to further assist clinical decision-making.

Summary updated

Abstract

Trigeminal neuralgia caused by paroxysmal and severe pain in the distribution of the trigeminal nerve is a rare chronic pain disorder. It is generally accepted that compression of the trigeminal root entry zone by vascular structures is the major cause of primary trigeminal neuralgia, and vascular decompression is the prior choice in neurosurgical treatment. Therefore, accurate preoperative modeling/segmentation/visualization of trigeminal nerve and its surrounding cerebrovascular is important to surgical planning. In this paper, we propose an automated method to segment trigeminal nerve and its surrounding cerebrovascular in the root entry zone, and to further reconstruct and visual these anatomical structures in three-dimensional (3D) Magnetic Resonance Angiography (MRA). The proposed method contains a two-stage neural network. Firstly, a preliminary confidence map of different anatomical structures is produced by a coarse segmentation stage. Secondly, a refinement segmentation stage is proposed to refine and optimize the coarse segmentation map. To model the spatial and morphological relationship between trigeminal nerve and cerebrovascular structures, the proposed network detects the trigeminal nerve, cerebrovasculature, and brainstem simultaneously. The method has been evaluated on a dataset including 50 MRA volumes, and the experimental results show the state-of-the-art performance of the proposed method with an average Dice similarity coefficient, Hausdorff distance, and average surface distance error of 0.8645, 0.2414, and 0.4296 on multi-tissue segmentation, respectively.
BibTeX
@inproceedings{lin2021automated,
  author = {Jinghui Lin and Lei Mou and Qifeng Yan and Shaodong Ma and Xingyu Yue and Shengjun Zhou and Zhiqing Lin and Jiong Zhang and Jiang Liu and Yitian Zhao},
  title = {{Automated Segmentation of Trigeminal Nerve and Cerebrovasculature in MR-Angiography Images by Deep Learning}},
  booktitle = {Frontiers in Neuroscience},
  year = {2021},
  doi = {10.3389/fnins.2021.744967},
}
Figure