Photo Credit: iStock.com/jes2ufoto

Deep learning models using OCT imaging successfully predicted SANS with moderate-to-high accuracy.   

The study was published in the June 2025 issue of American Journal of Ophthalmology; researchers conducted a retrospective study to develop deep learning artificial intelligence (AI) models for predicting Spaceflight Associated Neuro-ocular Syndrome (SANS) using optical coherence tomography (OCT) images of the optic nerve head (ONH).  

They trained AI deep learning models to predict SANS onset using 2 OCT datasets: astronaut pre- and inflight images (flight data) and pre- and in-bedrest images from participants undergoing head-down tilt bedrest (HDTBR) as a terrestrial model (ground data). The datasets were split by participants for training and testing. ResNet50-based models were trained separately on flight data, ground data, and a combined dataset. The model performance was evaluated using only preflight or pre-bedrest OCT images. The receiver operating characteristics (ROC) area under the curve (AUC) were used to measure prediction accuracy. Class activation maps (CAMs) were generated to highlight key image regions contributing to SANS prediction.  

The results showed that the model trained on flight OCT data achieved a ROC AUC of 0.82 (95% CI: 0.54 – 1.0) on flight data and 0.67 (95% CI: 0.51 – 0.83) on ground-based HDTBR data. The model trained on HDTBR data reached an AUC of 0.71 (95% CI: 0.50 – 0.91) on ground data and 0.76 (95% CI: 0.51 – 0.91) on flight data. The combined model produced an AUC of 0.81 (95% CI: 0.53 – 0.95) on flight data and 0.72 (95% CI: 0.52 – 0.92) on ground data. The CAMs highlighted the peripapillary nerve fiber layer, retinal pigment epithelium, and anterior lamina surface as key predictors.  

Investigators concluded that AI models demonstrated moderate-to-high accuracy in predicting SANS from preflight OCT imaging, and the consistent performance across datasets supported HDTBR as a valid Earth-based model for SANS. 

Source: ajo.com/article/S0002-9394(25)00298-3/abstract