A Segmentation Method of Foramen Ovale Based on Multiatlas

Jiashi Zhao, Huatao Ge, Wei He, Yanfang Li, Weili Shi, Zhengang Jiang, Yonghui Li, Xingzhi LiComputational 2021 · DOI

TL;DR

This paper introduces a novel multi-atlas based segmentation method for the foramen ovale, a critical anatomical landmark for treating trigeminal neuralgia. By leveraging a curated dataset and a robust registration-fusion pipeline, the method achieves accurate segmentation, paving the way for improved computer-assisted surgical guidance and reduced procedural risks.

What Problem Does This Paper Address?

Trigeminal neuralgia, a debilitating neurological condition, is often treated with radiofrequency ablation. This procedure involves puncturing the trigeminal nerve through the skull's foramen ovale. However, the foramen ovale's variable morphology and complex surrounding anatomy make accurate targeting challenging for clinicians. Manual segmentation of the foramen ovale from medical images is time-consuming and prone to subjective errors. This paper aims to address this by developing an automated, accurate, and efficient method for segmenting the foramen ovale, thereby enhancing computer-assisted surgical planning and execution.

Methods

The core of this work is a multi-atlas segmentation approach. This method leverages prior anatomical knowledge embedded in multiple "atlas" images to segment a target image. The process involves three main steps:

  1. Atlas Selection: The method identifies the 10 atlas images from a larger set that are most similar to the target image (the CT scan to be segmented). Similarity is measured using normalized cross-correlation.
  2. Multi-Atlas Registration: The selected atlas images are then registered to the target image. This registration is performed in two stages:
  3. Coarse Registration: A multiresolution affine transformation is used to align the overall structures.
  4. Fine Registration: A multiresolution B-spline transformation is employed for non-rigid, local deformation to achieve precise alignment. This hierarchical approach improves accuracy and reduces computation time.
  5. Label Fusion: After registration, the segmented regions (labels) from the transformed atlases are combined to produce a final, consensus segmentation. The paper primarily uses the STAPLE (Simultaneous Truth and Performance Level Estimation) algorithm for this fusion, comparing its performance against majority voting (MV) and a selective iterative method (SIMPLE).

Datasets

The study utilized a dataset of 30 CT scans of the human skull.

  • 20 scans were used to create the foramen ovale atlas dataset.
  • 10 scans were reserved for testing and evaluating the segmentation method. The paper highlights the creation of a novel skull base foramen ovale atlas dataset, which is intended to support future research in this area.

Key Insights and Ideas

  • Multi-atlas superiority for limited data: The paper demonstrates that a multi-atlas approach can achieve high segmentation accuracy even with a relatively small dataset (10 test scans), which is a significant advantage in medical imaging where large annotated datasets are often scarce.
  • Hierarchical registration is key: The combination of multiresolution affine and B-spline transformations is crucial for accurately aligning complex anatomical structures like the foramen ovale, balancing global alignment with local deformation.
  • STAPLE algorithm's effectiveness: The STAPLE algorithm proves to be a robust choice for fusing segmented labels from multiple atlases, outperforming simpler methods like majority voting.
  • Clinical relevance of accurate segmentation: The accurate segmentation of the foramen ovale is directly linked to improving the success rate and safety of trigeminal neuralgia treatment, reducing patient pain and complications.

Key Results

The study evaluated the segmentation performance using Dice coefficient, 95% Hausdorff distance, and Average Surface Distance (ASD).

  • Dice Coefficient: STAPLE and SIMPLE algorithms achieved significantly higher average Dice scores (around 0.858-0.871) compared to MV (around 0.790-0.803), indicating better overlap with manual segmentations.
  • 95% Hausdorff Distance: STAPLE and SIMPLE showed lower average Hausdorff distances (around 3.227-4.274) than MV (around 3.639-5.054), suggesting better boundary agreement.
  • Average Surface Distance (ASD): Similar to Hausdorff distance, STAPLE and SIMPLE achieved lower ASD values (around 0.728-0.998) compared to MV (around 0.933-1.258).

Qualitatively, the MV algorithm produced fragmented and incomplete segmentations, while STAPLE and SIMPLE provided more coherent and accurate results, closely matching manual segmentations. The paper also noted that segmentation of the right foramen ovale generally performed slightly better than the left, though statistical significance was not found in existing literature to explain this difference.

Limitations and Future Work

  • Segmentation Time: While the method is accurate, the authors acknowledge that the preoperative segmentation process, particularly registration and fusion, can still be time-consuming (around 30 minutes). Future work will focus on optimizing the algorithm to reduce this time.
  • Segmentation Accuracy: Further improvements in segmentation accuracy are also a goal for future research.
  • Right vs. Left Foramen Ovale Performance: The observed slight difference in performance between the left and right foramen ovale warrants further investigation, as existing literature does not provide a clear anatomical explanation.

Summary updated

Abstract

Trigeminal neuralgia is a neurological disease. It is often treated by puncturing the trigeminal nerve through the skin and the oval foramen of the skull to selectively destroy the pain nerve. The process of puncture operation is difficult because the morphology of the foramen ovale in the skull base is varied and the surrounding anatomical structure is complex. Computer-aided puncture guidance technology is extremely valuable for the treatment of trigeminal neuralgia. Computer-aided guidance can help doctors determine the puncture target by accurately locating the foramen ovale in the skull base. Foramen ovale segmentation is a prerequisite for locating but is a tedious and error-prone task if done manually. In this paper, we present an image segmentation solution based on the multiatlas method that automatically segments the foramen ovale. We developed a data set of 30 CT scans containing 20 foramen ovale atlas and 10 CT scans for testing. Our approach can perform foramen ovale segmentation in puncture operation scenarios based solely on limited data. We propose to utilize this method as an enabler in clinical work.
BibTeX
@inproceedings{zhao2021segmentation,
  author = {Jiashi Zhao and Huatao Ge and Wei He and Yanfang Li and Weili Shi and Zhengang Jiang and Yonghui Li and Xingzhi Li},
  title = {{A Segmentation Method of Foramen Ovale Based on Multiatlas}},
  booktitle = {Computational},
  year = {2021},
  doi = {10.1155/2021/5221111},
}
Figure