Improved method for optical coherence tomography angiography: from reconstruction to clinical indicator...

Area of Science:

• Ophthalmic imaging and diagnostics
• Medical image processing and analysis
• Biomedical engineering

Background:

• Optical coherence tomography angiography (OCTA) is vital for noninvasive ophthalmic vascular imaging.
• Existing OCTA reconstruction methods suffer from spectral leakage and motion artifacts, impacting image quality and clinical quantification accuracy.
• Enhancing OCTA reconstruction and automating clinical indicator extraction are critical for reliable diagnosis and disease monitoring.

Purpose of the Study:

• To propose and validate an improved OCTA reconstruction method using a smoothed Walsh window function to reduce spectral leakage while maintaining axial resolution.
• To enhance blood flow B-scan signals through retinal layer segmentation.
• To develop and evaluate a fully automated pipeline for calculating key clinical indicators, including foveal avascular zone (FAZ) parameters and vessel density, using local fractal dimension analysis.

Main Methods:

• Employed a spectral-domain OCT system for volumetric retinal data acquisition from healthy volunteers.
• Utilized a smoothed Walsh window for full-spectrum splitting to mitigate spectral leakage during OCTA reconstruction.
• Integrated deep learning-based retinal layer segmentation and Otsu's thresholding to improve blood flow B-scan signals.
• Applied local fractal dimension analysis for automated segmentation of vascular networks and FAZ regions for quantitative analysis.

Main Results:

• The proposed reconstruction method significantly improved vessel connectivity, contrast (up to ~0.74), and signal-to-noise ratio (SNR up to ~1.09 dB) compared to traditional algorithms.
• FAZ segmentation based on local fractal dimension achieved high similarity (0.9735 ± 0.0066) with manual ground truth, exhibiting low false positive (1.76% ± 0.75%) and false negative (3.48% ± 0.78%) rates.
• Intraclass correlation coefficients for FAZ area, perimeter, and circularity index exceeded 0.90, indicating high reproducibility.
• Sectoral vessel density measurements aligned with established normative data.

Conclusions:

• The combination of a smoothed Walsh window function and retinal layer segmentation substantially improves OCTA image quality and blood flow signal clarity.
• The automated local fractal dimension-based analysis pipeline accurately and reproducibly quantifies FAZ morphology and vessel density, showing strong agreement with manual annotations.
• This integrated approach provides a robust framework for advanced OCTA reconstruction and automated clinical indicator derivation, supporting enhanced ophthalmic diagnosis and longitudinal disease assessment.