Deep learning based optic nerve sheath diameter characterization and structure quantification on transorbital...

Area of Science:

• Medical imaging
• Neuro-ophthalmology
• Artificial intelligence in medicine

Background:

• Optic nerve quantification is vital for assessing intracranial pressure and neuro-ophthalmic conditions.
• Manual segmentation of optic nerve structures is time-consuming and resource-intensive.
• Automated segmentation accuracy relies heavily on ultrasound image quality.

Purpose of the Study:

• To develop a deep learning model for precise optic nerve segmentation, especially in sub-optimal ultrasound images.
• To improve optic nerve sheath diameter quantification using an uncertainty-aware deep neural network.
• To enhance the robustness of automated segmentation through shared and specific feature extraction.

Main Methods:

• Proposed a deep neural network with shared and specific feature extraction branches.
• Incorporated an uncertainty-aware loss function to promote robust object structure learning.
• Validated the model on a multi-center, publicly available dataset.

Main Results:

• Achieved a 73.3% Dice score for optic nerve segmentation.
• Obtained an 84.5% Area Under the Receiver Operating Characteristic curve (AUROC) for quantification.
• Demonstrated superior performance compared to state-of-the-art models.

Conclusions:

• The proposed deep learning model offers superior performance in optic nerve segmentation.
• The model shows strong potential for accurate optic nerve sheath diameter quantification.
• This approach is effective for neuro-ophthalmic condition assessment, particularly with challenging image quality.