Retinal feature analysis from AI-driven OCT to OCTA translation

Our study investigates the potential of using generative machine learning to translate Optical Coherence Tomography (OCT) data into Optical Coherence Tomography Angiography (OCTA) images by retinal feature analysis using a public dataset, OCT500. OCT is widely used for imaging retinal structures but lacks the ability to capture blood flow information, unlike OCTA. The high cost and technical requirements of OCTA limit its accessibility in clinical settings, creating a need for alternative solutions. Here, we explore the use of a generative adversarial network (GAN) based framework that translates OCT into OCTA, providing detailed vascular information traditionally obtained from OCTA scans. This approach leverages quantitative vascular features like blood vessel density, vessel caliber, tortuosity, and perimeter index, using them as potential biomarkers for detecting retinal pathologies as well as quality metrics: SSIM (Structural Similarity Index Measure), FID (Fréchet Inception Distance) and PCQI (patch-based contrast quality index) scores to compare the generated OCTA quality. In this study, translated OCTA (TR OCTA) showed high image quality in both 3 and 6mm datasets (high-resolution and contrast quality, moderate structural similarity compared to ground truth OCTAs) in terms of SSIM, FID and PCQI scores and vascular features like tortuosity and vessel perimeter index exhibits more consistent trends compared to density features which are affected by local vascular distortions. By transforming readily available OCT data into high quality OCTA images, this method can potentially reduce dependence on costly OCTA technology, thereby broadening access to advanced retinal diagnostics.