3D AI-Powered Retina Imaging Platform Reveals Nerve Fiber Health
University of Pittsburgh researchers have developed a non-invasive imaging and analysis platform that uses visible-light optical coherence tomography (vis-OCT) combined with artificial intelligence to reconstruct and quantify the three-dimensional architecture of retinal ganglion cell (RGC) axon bundles in living eyes. It enables repeated, quantitative tracking of nerve fiber health over time — something existing imaging approaches cannot reliably achieve — opening the door to better monitoring of neurodegeneration, regeneration, and treatment response.
Description
The platform combines optical imaging, deep learning, and graph-based topology analysis into a single pipeline. Raw 3D vis-OCT volumes of the retina are first processed by a self-supervised despeckling module that improves image contrast without needing pre-existing "clean" reference scans. A deep-learning segmentation model then identifies the axon bundles and separates them from retinal blood vessels, improving the reliability of all downstream measurements. The segmented bundle structures are converted into a topological graph, where skeleton nodes and edges represent the bundle network, allowing the system to identify individual bundles, endpoints, branch points, trunks, and side branches. Because OCT-based segmentation is often fragmented, an optimization-based reconstruction step — using an integer optimization model with anatomical constraints and topology-based scoring — regroups disconnected pieces that belong to the same underlying bundle. This produces a more anatomically accurate reconstruction than segmentation alone, and enables calculation of quantitative 3D biomarkers describing bundle organization.Applications
- Preclinical drug and gene therapy evaluation in animal models of optic nerve injury or glaucoma- Longitudinal monitoring tools for ophthalmic research laboratories studying retinal neurodegeneration
- Analysis software for OCT device manufacturers
- Research platforms for evaluating axon regeneration or reconnection following retinal or optic nerve therapies
- Translational research tools supporting future clinical assessment of retinal neurodegenerative disease
Advantages
- Enables non-invasive, in vivo, 3D quantitative analysis of individual RGC axon bundles, rather than relying on ex vivo histology- Improves image contrast and downstream segmentation accuracy without requiring clean ground-truth training data
- Reconstructs fragmented bundle segments into anatomically accurate, bundle-wise structures, reducing false measurement inflation
- Generates a rich set of quantitative 3D biomarkers
- Supports repeated measurements in the same subject over time, enabling high-throughput, non-terminal longitudinal studies