Adaptive optics optical coherence tomography (AO-OCT) has enabled in vivo cellular imaging of the human retina. This technology has excellent resolution but suffers from noise, making it difficult to see individual cells. However, because imaging noise (e.g., speckle) makes it difficult to see RPE cells from a single volume acquisition, many 3D volumes are typically averaged to improve contrast, significantly increasing acquisition time and lowering overall imaging throughput. To tackle this challenge, researchers developed a novel approach known as the parallel discriminator generative adversarial network (P-GAN). P-GAN is an AI-based approach that can effectively recover cellular features from a single AO-OCT image without requiring image averaging, which is time-consuming.
Researchers demonstrate that P-GAN increases RPE cell contrast by 3.5-fold and reduces the end-to-end time necessary to see RPE cells by 99-fold, allowing for large-scale imaging of cells in the real human eye. RPE cell spacing was assessed over a wide batch of AI-recovered pictures from three people and was within expected normative values.
The experimental results show that P-GAN can increase picture contrast by 3.5 times while reducing imaging time by a startling 99-fold. It paves the way for large-scale retinal imaging in clinical settings, which could revolutionize ophthalmic diagnosis. The findings show that AI-assisted imaging has the potential to overcome one of RPE imaging’s major limitations and make it more accessible in a clinical environment.
Related Content: Large Language Models In Ophthalmology