Active shape model registration of ocular structures in computed tomography images

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer-Aided Diagnosis

Background:

• Accurate segmentation of ocular structures in computed tomography (CT) scans is crucial for diagnosing globe injuries.
• Existing methods may lack the precision needed to differentiate subtle shape changes associated with specific injury types.

Purpose of the Study:

• To develop an active shape model (ASM) for segmenting five key ocular regions (cornea, sclera, vitreous cavity, anterior chamber, lens) from CT scans.
• To utilize the ASM to differentiate between normal globes and those affected by open globe injury (OGI).

Main Methods:

• Manual segmentation of 78 normal globes across five regions by two graders.
• Creation of a statistical shape model from segmented regions.
• Training and validation of an active shape model using leave-one-out cross-validation.
• Application of the trained ASM to CT scans of globes with OGI for shape analysis.

Main Results:

• The ASM demonstrated high segmentation accuracy, with Dice similarity coefficients ranging from 83.5% (anterior chamber) to 92.5% (sclera).
• Globes with OGI showed statistically significant shape differences in the sclera and vitreous cavity compared to normal globes.
• Zone 1 and Zone 2 OGI cases exhibited distinct shape variations in the cornea and anterior chamber, aligning with injury definitions.

Conclusions:

• The developed ASM provides reproducible and accurate segmentation of ocular structures, correlating well with manual segmentations.
• Quantitative shape metrics derived from ASM analysis of OGI cases are consistent with established medical knowledge of structural deformation.