A new AI model called OCTCube-M could help improve patient selection, predict disease progression and streamline clinical trials by extracting insights from 3D optical coherence tomography (OCT) eye scans. Published in Nature Biomedical Engineering, the model outperformed older 2D approaches in detecting age-related macular degeneration and diabetic retinopathy.
Seeing the full picture, not just slices
Developed using over 1.62M retinal images from 26,000 OCT scans, OCTCube-M is a multi-modal foundation model that reads the full 3D structure of the eye rather than relying on individual image slices. Researchers at Washington University and Genentech trained it to extract richer signals from volumetric data that 2D models miss.
Smaller trials, faster answers
In collaboration with Genentech, the team tested OCTCube-M against randomized controlled trial data and found it could significantly reduce the number of patients needed while achieving equivalent statistical power. Researchers are also exploring whether AI models like this could enable digital twins — virtual patient models that predict disease progression without an experimental therapy — an approach the FDA has shown increasing interest in for phase 2 studies.
Beyond the eye
The technology supports oculomics, an emerging field that uses eye imaging to detect systemic disease. Because retinal blood vessels mirror changes throughout the body, AI analysis of OCT scans could eventually flag early signs of cardiovascular disease, diabetes and kidney dysfunction.
