Researchers from the National Institutes of Health have developed an AI tool that enhances optical coherence tomography (OCT), a common retinal imaging technique, by improving image resolution and reducing the time required for imaging. OCT uses a beam of light to generate detailed images of internal tissues, and when combined with adaptive optics (AO), it produces 3D retinal structures at cellular-scale resolution.

AO-OCT can visualize retinal structures, including the retinal pigment epithelium (RPE), but it faces challenges such as speckle, or noise, inherent to the imaging process. Traditionally, researchers would image cells repeatedly over time to create a speckle-free image, which is time-consuming.

To address this, the team developed an AI tool called parallel discriminator generative adversarial network (P-GAN), which can recover speckle-obscured cellular features from a single AO-OCT volume. By training P-GAN on nearly 6000 analyzed AO-OCT images of human RPE paired with their speckled originals, the tool successfully removed speckle and uncovered cellular details.

P-GAN produced results comparable to the manual method that required acquiring and averaging 120 images, reducing imaging acquisition and processing time 100-fold and improving image contrast 3.5-fold. The development of P-GAN represents a significant advancement in AI’s role in imaging and its potential to make AO imaging more accessible for clinical applications and research on blinding retinal diseases.