Automated thresholding algorithms outperform manual thresholding in macular optical coherence tomography angiography...

Area of Science:

• Ophthalmology
• Medical Imaging Analysis
• Retinal Perfusion Quantification

Background:

• Optical Coherence Tomography Angiography (OCTA) is vital for assessing retinal vasculature.
• Vessel Density (VD) and Vessel Skeleton Density (VSD) are key OCTA parameters.
• Current algorithms for VD/VSD calculation lack standardized comparability and reliability assessment.

Purpose of the Study:

• To evaluate the comparability and test-retest reliability of common OCTA analysis algorithms.
• To assess the diagnostic performance of different algorithms in distinguishing healthy versus pathologic macular perfusion.

Main Methods:

• Acquisition of 3x3mm OCTA en face images from superficial and deep retinal layers using swept-source OCTA.
• Calculation of VD and VSD using manual thresholding and six automated algorithms (Huang, Li, Otsu, Moments, Mean, Percentile) via ImageJ.
• Comparison of algorithms using intra-class correlation coefficients, measurement differences, repeatability coefficients, and Receiver Operating Characteristic (ROC) analyses (Area Under the Curve - AUC).

Main Results:

• Significant differences in binarization thresholds, VD, and VSD were observed across algorithms and manual thresholding (p < 0.0001).
• Automated algorithms demonstrated higher reproducibility compared to manual thresholding.
• The Mean algorithm showed a significantly higher AUC than manual thresholding for the superficial retinal layer.

Conclusions:

• Automated thresholding algorithms enhance OCTA parameter reproducibility and improve macular pathology diagnosis sensitivity.
• Current algorithms are not interchangeable, necessitating further standardization for clinical application.
• The Mean algorithm warrants further investigation; automated methods are preferred but require standardization.