Performance evaluation of automated segmentation software on optical coherence tomography volume data.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Numerous Optical Coherence Tomography (OCT) segmentation methods exist, but often use limited datasets.
• Existing validation lacks a common ground truth, hindering unbiased algorithm comparison.
• Clinical OCT datasets with comprehensive ground truth are scarce.

Purpose of the Study:

• To review research-oriented automated retinal segmentation tools for OCT images.
• To evaluate and compare the performance of these tools.
• To establish a common ground truth for objective performance assessment.

Main Methods:

• Literature review of automated OCT retinal segmentation software.
• Development and application of a common ground truth dataset.
• Performance evaluation of selected segmentation tools against the common ground truth.

Main Results:

• Identified limitations in existing OCT segmentation datasets and validation protocols.
• Demonstrated the feasibility of comparing algorithms using a unified ground truth.
• Provided a comparative analysis of different automated segmentation tool performances.

Conclusions:

• A standardized ground truth is crucial for unbiased evaluation of OCT segmentation algorithms.
• This work facilitates more reliable performance comparisons in retinal image analysis.
• Highlights the need for robust datasets reflecting clinical realities in OCT research.