Comparing Image Segmentation Techniques for Determining 3D Orbital Cavernous Hemangioma Size on MRI

Area of Science:

• Ophthalmology
• Medical Imaging
• Computational Biology

Background:

• Orbital cavernous hemangiomas are benign vascular tumors.
• Accurate size measurement is crucial for diagnosis and treatment planning.
• Current segmentation methods vary in subjectivity and agreement.

Purpose of the Study:

• To compare the interobserver agreement of three segmentation methods for measuring orbital cavernous hemangioma size.
• To evaluate user-dependent versus parameter-dependent segmentation techniques.

Main Methods:

• Retrospective analysis of T2-weighted MRI from 14 patients with orbital cavernous hemangiomas.
• Two observers used ellipsoid (user-dependent), GrowCut, and k-means clustering (parameter-dependent) segmentation methods.
• Interobserver agreement assessed using Lin's concordance correlation coefficients.

Main Results:

• K-means clustering showed the highest interobserver agreement (0.95), followed by ellipsoid (0.92).
• GrowCut exhibited poor agreement (0.12).
• Tumor size measurements varied significantly between methods and observers.

Conclusions:

• K-means clustering provides reliable and reproducible size measurements for orbital cavernous hemangiomas.
• Parameter-dependent methods with low subjectivity, like k-means clustering, are preferable for consistent results.
• K-means clustering effectively delineates complex tumor structures.