Researchers at the National Institutes of Health have significantly enhanced a technique for generating high-resolution images of cells in the eye using artificial intelligence (AI). They report that AI accelerates the imaging process by 100 times and increases image contrast by 3.5 times. This advancement, they believe, will equip researchers with a more effective tool for assessing age-related macular degeneration (AMD) and other retinal diseases.

Johnny Tam, PhD, who heads the Clinical and Translational Imaging Section at NIH’s National Eye Institute, is at the forefront of advancing a technology known as adaptive optics (AO) to refine imaging devices that use optical coherence tomography (OCT). OCT, akin to ultrasound, is a noninvasive, quick, painless method that is routinely used in most eye clinics.

However, imaging RPE cells with AO-OCT presents new challenges, notably a phenomenon known as speckle, which disrupts image clarity in a manner comparable to how clouds can obscure aerial photographs. To manage this, researchers traditionally had to repeatedly image cells over extended periods. As the speckle pattern shifts over time, different aspects of the cells become visible. Researchers would then undertake the painstaking task of compiling these multiple images to create a clear, speckle-free image of the RPE cells.

To address this, Tam and his team introduced a groundbreaking AI-based approach known as the parallel discriminator generative adversarial network (P-GAN)—a sophisticated deep learning algorithm. By training the P-GAN network with nearly 6,000 manually analysed AO-OCT images of human RPE cells, each paired with its speckled original, the team taught the network to identify and restore features obscured by speckle.

When applied to new images, the P-GAN successfully clarified RPE images by recovering intricate cellular details. Remarkably, this system achieved results with a single image capture that traditionally required the acquisition and analysis of 120 images. P-GAN surpassed other AI techniques in various objective performance metrics, assessing aspects such as cell shape and structure.

Vineeta Das, PhD, a postdoctoral fellow in the same section at NEI, noted that P-GAN cut down the time for image acquisition and processing by approximately 100 times. It also enhanced the image contrast significantly—by about 3.5 times.

“Artificial intelligence is pivotal in surpassing a major limitation of retinal cell imaging, which is the time consumption,” said Tam. “Adaptive optics elevate OCT-based imaging, akin to moving from a balcony seat to the front row for retinal imaging. With AO, we can achieve 3D views of retinal structures at a cellular level, allowing us to detect very early signs of disease.”

While AO enhances the clarity of OCT images, processing these images post-capture has traditionally been time-consuming. Tam’s recent innovations target the retinal pigment epithelium (RPE), a layer of tissue vital for supporting the retina’s metabolically active neurons, including photoreceptors. The retina, which lines the back of the eye, plays a crucial role in converting light into neural signals transmitted to the brain. The RPE is key in studying retinal diseases, which often begin when the RPE deteriorates.

By integrating AI with AO-OCT, Tam believes they have removed a significant barrier to routine clinical imaging for conditions affecting the RPE, which has traditionally been challenging to image.

“Our findings indicate that AI can revolutionize the way we capture images,” Tam added. “Our P-GAN AI technology will make AO imaging more accessible for routine clinical use and research focused on understanding the structure, function, and diseases of the retina. Viewing AI as an integral part of the imaging system, rather than just a post-processing tool, represents a paradigm shift in the field.”

Stephanie Price. HT World. (2024). AI-based adaptive optics takes retinal imaging to the next level, finds study. Healthcare Technology World. https://www.htworld.co.uk/news/ai/ai-based-adaptive-optics-takes-retinal-imaging-to-next-level-finds-study/