Efficient orbital structures segmentation with prior anatomical knowledge.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Anatomical Segmentation

Background:

• Accurate segmentation of orbital structures is crucial for diagnosing various ophthalmic and neurological conditions.
• Manual segmentation is time-consuming and prone to inter-observer variability.
• Automated methods often require extensive training data and may lack generalizability.

Purpose of the Study:

• To develop a fully automatic method for segmenting orbital structures in CT images.
• To leverage prior anatomical knowledge for efficient and accurate segmentation.
• To provide a user-friendly pipeline for orbital structure segmentation.

Main Methods:

• Utilized prior anatomical knowledge (shape, intensity, spatial relationships) to define a volume of interest (VOI).
• Employed predefined rules within the VOI for fast localization and segmentation of eye globes, optic nerves, and extraocular muscles.
• Developed an intuitive pipeline that does not require training data.

Main Results:

• Achieved high average Dice similarity coefficients: 0.81 for right eye globes, 0.79 for left eye globes, 0.72 for right optic nerves, 0.79 for left optic nerves, 0.73 for right extraocular muscles, and 0.76 for left extraocular muscles.
• Demonstrated accuracy and efficiency in segmenting multiple orbital structures.
• Validated the method on 30 publicly available CT datasets.

Conclusions:

• The proposed method offers an accurate and efficient solution for automatic orbital structure segmentation in CT images.
• The approach eliminates the need for training data, making it broadly applicable.
• The intuitive pipeline allows for user modification and extension to segment additional structures.