Robust optic nerve segmentation on clinically acquired computed tomography.

Area of Science:

• Medical imaging
• Neuroscience
• Ophthalmology

Background:

• The optic nerve (ON) is crucial in various pathological conditions.
• Accurate segmentation of the ON aids in understanding its development and disease progression.
• Current automated segmentation methods for the ON have limitations.

Purpose of the Study:

• To optimize registration and fusion methods for a novel multi-atlas framework.
• To achieve automated segmentation of the optic nerve, eye globes, and muscles from CT data.
• To evaluate the robustness and accuracy of the proposed framework.

Main Methods:

• A multi-atlas approach involving affine and nonrigid registration.
• Statistical fusion of segmented results from multiple atlases.
• Utilized ANTS Symmetric Normalization registration and NMI-STAPLE fusion.
• Evaluated on 501 CT scan volumes from thyroid eye disease patients.

Main Results:

• The optimized multi-atlas framework achieved a median Dice similarity coefficient of 0.77 for ON segmentation.
• Performance is comparable to inter-rater human reproducibility (0.73).
• Identified optimal registration and fusion techniques among 18 compared methods.

Conclusions:

• The developed multi-atlas framework enables robust and accurate automated segmentation of the optic nerve and related structures on CT scans.
• This automated approach has significant potential for clinical applications in diagnosing and monitoring ON pathologies.
• The findings advance the field of medical image analysis for neurological and ophthalmological conditions.