× close
Credit: Unsplash/CC0 Public Domain
Researchers at the National Institutes of Health have applied artificial intelligence (AI) to a technology that produces high-resolution images of eye cells. They report that using AI, imaging is 100 times faster and image contrast is 3.5 times better. They say this advance will provide researchers with better tools to assess age-related macular degeneration (AMD) and other retinal diseases.
The work appears in communication medicine.
“Artificial intelligence can help overcome a key limitation of retinal imaging cells: time,” said Dr. Johnny Tam, director of clinical and translational imaging at the NIH's National Eye Institute.
Tam is developing a technology called adaptive optics (AO) to improve imaging devices based on optical coherence tomography (OCT). Like ultrasound, OCT is non-invasive, quick, and painless, and is standard equipment in most eye clinics.
Imaging RPE cells with AO-OCT presents new challenges, including a phenomenon called speckle. Speckles interfere with his AO-OCT, just as clouds interfere with aerial photography. At any time, parts of the image may become blurred. Managing speckle is similar to managing cloud cover.
Researchers repeatedly image cells over long periods of time. Over time, the speckles change and different parts of the cell become visible. Scientists then take on the painstaking and time-consuming task of stitching together many images to create a speck-free image of RPE cells.
Tam and his team developed a new AI-based method called Parallel Discriminative Generative Adverb Network (P-GAN), a deep learning algorithm. By feeding approximately 6,000 manually analyzed images of human RPE obtained with AO-OCT to his P-GAN network in combination with original images with corresponding spots, the team was able to uncover cellular features hidden by the spots. We trained a network to identify and recover.
When tested on new images, P-GAN successfully removed speckles in RPE images and restored cellular details. A single image capture yielded comparable results to manual methods that required acquiring and averaging 120 images. P-GAN outperformed other AI technologies on a variety of objective performance metrics that evaluate things like cell shape and structure. Dr. Vineeta Das, a postdoctoral fellow in NEI's Division of Clinical and Translational Imaging, estimates that P-GAN reduces imaging acquisition and processing time by about 100 times. P-GAN also has improved contrast, which is about 3.5 times larger than before.
“Adaptive optics takes OCT-based imaging to the next level,” Tam says. “It's like going from a balcony seat to a front-row seat to image the retina. With AO, he can reveal 3D retinal structure with cellular-scale resolution, making it very easy to see the disease. We can zoom in on early signs.”
Adding AO to OCT greatly improves cell visibility, but processing AO-OCT images after acquisition takes much more time than OCT without AO.
Tam's latest research targets the retinal pigment epithelium (RPE), the light-sensing tissue layer behind the retina that supports metabolically active retinal neurons, including photoreceptors. The retina runs along the back of the eye and captures light coming in from the front of the eye, processes it, converts it into signals, and sends them to the brain via the optic nerve. Scientists are interested in his RPE because many diseases of the retina occur when it is damaged.
Tam believes that the integration of AI and AO-OCT overcomes major hurdles in routine clinical imaging using AO-OCT, especially for diseases affecting the RPE, which have traditionally been difficult to image. thinking about.
“Our results suggest that AI has the potential to fundamentally change how images are captured,” Tam said. “Our P-GAN artificial intelligence will make AO imaging more accessible in routine clinical applications and in research aimed at understanding the structure, function, and pathophysiology of blinding retinal diseases. Instead of thinking of it as part of the whole imaging system, we need to think of it as part of the whole imaging system.'' Tools that are applied only after an image is captured are a paradigm shift in the AI field. ”
For more information:
Vinita Das, Hulu Zhang, Andrew Bower and others use artificial intelligence-assisted adaptive optical coherence tomography to reveal live human retinal cells hidden by speckle. communication medicine (2024). DOI: 10.1038/s43856-024-00483-1
Magazine information:
communication medicine