Related Experiment Video
Updated: Jun 14, 2025
09:11
Assessment of Global Ocular Structure Following Spaceflight Using a Micro-Computed Tomography Micro-CT Imaging Method
Published on: October 27, 2020
5.7K
Artificial Intelligence Deep Learning Models to Predict Spaceflight Associated Neuro-Ocular Syndrome
Alex S Huang1, Jalil Jalili1, Evan Walker1
1From the Hamilton Glaucoma Center (A.S.H., J.J., E.W., R.N.W., and M.C.), The Viterbi Family Department of Ophthalmology, Shiley Eye Institute, University of California, San Diego, California, USA.
American Journal of Ophthalmology
|June 12, 2025
Summary
Artificial intelligence models can predict Spaceflight Associated Neuro-ocular Syndrome (SANS) using OCT imaging. This research validates head-down tilt bedrest as a suitable model for SANS development, aiding future spaceflight health research.
Area of Science:
- Ophthalmology
- Aerospace Medicine
- Artificial Intelligence
Background:
- Spaceflight Associated Neuro-ocular Syndrome (SANS) is a significant concern for astronauts.
- Optical Coherence Tomography (OCT) is a key imaging modality for ocular health assessment.
Purpose of the Study:
- To develop deep learning AI models for predicting SANS development.
- To utilize OCT imaging of the optic nerve head (ONH) for SANS prediction.
Main Methods:
- Retrospective analysis of OCT datasets from astronauts and head-down tilt bedrest (HDTBR) participants.
- Training Resnet50-based deep learning models on flight and/or ground data.
- Evaluating model performance using receiver operating characteristic (ROC) area under the curve (AUC) and class activation maps (CAMs).
Main Results:
- AI models demonstrated moderate-to-high performance in predicting SANS from preflight/pre-bedrest OCT images.
- Cross-trained models (flight and ground data) showed comparable performance, suggesting similarities between SANS in space and HDTBR.
- CAMs highlighted critical ONH regions, including the peripapillary nerve fiber layer and retinal pigmented epithelium.
Conclusions:
- Deep learning AI models can effectively predict SANS using preflight OCT imaging.
- The study supports HDTBR as a valid Earth-bound model for SANS research.
- Findings contribute to understanding and mitigating SANS risks in spaceflight.
