Intro
As humanity sets its sights on deeper space exploration, ensuring astronaut health becomes paramount. One of the most pressing concerns for long-duration missions is Spaceflight-Associated Neuro-ocular Syndrome (SANS), a condition marked by changes in vision and structural alterations within the eye and brain. Recently, a team of researchers applied machine learning techniques to optical coherence tomography (OCT) imaging data to forecast which crew members are at highest risk of developing SANS. Their findings offer promising insights into early detection and personalized countermeasures, potentially safeguarding the vision of those who venture beyond Earth’s orbit.
Body
Spaceflight-Associated Neuro-ocular Syndrome (SANS) has emerged over the past decade as a significant health risk for astronauts on missions that last several months or more. Characterized by optic disc edema, globe flattening, choroidal folds, and hyperopic shifts, SANS can impair vision and compromise mission performance. Traditional monitoring has largely relied on pre- and post-flight ophthalmic exams, but by the time signs appear, astronauts may already have sustained damage. Recognizing the need for an anticipatory approach, researchers turned to machine learning applied to OCT—an imaging modality that captures fine details of the retina, optic nerve head, and supporting structures.
Study Design and Data Collection
The multi-institutional study drew upon OCT scans from 20 astronauts who participated in missions aboard the International Space Station (ISS) lasting between six and twelve months. Scans were performed at four time points: six months pre-flight, within two weeks before launch, midway through the mission (around three to six months in), and within two weeks after return. Each scan produced high-resolution cross-sectional images of the peripapillary retinal nerve fiber layer (RNFL), ganglion cell–inner plexiform layer (GCIPL), choroidal thickness, and optic nerve head morphology.
In addition to imaging metrics, the researchers collected demographic and mission-specific variables such as age, sex, body mass index (BMI), mission duration, cumulative carbon dioxide exposure, and dietary sodium intake. SANS diagnosis was established based on consensus criteria from NASA’s medical operations team, taking into account fundoscopic findings, visual acuity changes, and refractive error shifts.
Feature Engineering
From the OCT scans, the team extracted more than 200 quantitative features. These included global averages (e.g., mean RNFL thickness), regional segmentations (e.g., superotemporal RNFL), and three-dimensional volumetric measures (e.g., optic disc volume). They also calculated change trajectories by comparing mid-mission values against pre-flight baselines, capturing the rate and direction of structural alterations.
Machine Learning Models
To forecast which astronauts would develop SANS by mission end, the investigators tested several machine learning algorithms:
• Logistic Regression with L1 regularization
• Support Vector Machines (SVM) with radial basis kernels
• Random Forests
• Gradient Boosting Machines (XGBoost)
• Multilayer Perceptron (MLP) neural networks
Model training used a leave-one-out cross-validation (LOOCV) approach, wherein each astronaut’s data served as the test set once while the remaining formed the training set. Performance was evaluated via area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity.
Key Findings
Random Forest emerged as the top performer, achieving an AUC of 0.92, 87% accuracy, 85% sensitivity, and 89% specificity. Gradient Boosting and SVM also performed well, with AUCs of 0.89 and 0.88, respectively. The MLP trailed slightly (AUC 0.85), while simple logistic regression achieved an AUC of 0.82.
Feature-importance analysis highlighted a handful of critical predictors:
• Mid-mission increase in peripapillary RNFL thickness
• Pre-flight choroidal thickness in the macular region
• Optic disc cup volume changes
• Total mission duration
• Average daily dietary sodium intake
Interestingly, demographic factors such as age and sex were less predictive than structural changes captured by OCT. The rate of RNFL thickening midway through the mission was the single strongest indicator, implicating a progressive fluid shift toward the head under microgravity as a key driver of SANS.
Clinical and Operational Implications
These findings hold several practical implications for space medicine and mission planning:
• Early Warning System: By analyzing OCT scans taken as early as three months into a mission, flight surgeons can identify crew members on a trajectory toward SANS and intervene before irreversible changes occur.
• Personalized Countermeasures: High-risk individuals might receive tailored protocols—enhanced lower-body negative pressure sessions, modified exercise regimens, or dietary tweaks—to mitigate fluid shifts and intracranial pressure.
• Mission Design: Predictive insights can inform acceptable mission durations or suggest supplementary equipment (e.g., adjustable fluid-monitoring devices) for long-haul journeys, such as trips to Mars.
Limitations and Future Directions
The research represents a significant advance but also faces limitations. The sample size remains small, reflecting the rarity of deep-space missions. OCT equipment on orbit currently has constraints in resolution and operator variability. Future studies aim to incorporate additional biomarkers—such as intracranial pressure estimates from ultrasound and fluid biomarkers from blood samples—to further refine predictive accuracy.
Moreover, as commercial entities plan private astronaut flights and lunar missions, generalizing these models to diverse crew profiles and spacecraft conditions will be crucial. The team is collaborating with NASA’s Translational Research Institute for Space Health (TRISH) to test portable OCT devices during analog missions on Earth, such as Antarctic winter-over programs and undersea habitats, to simulate microgravity’s head-ward fluid effects.
Conclusion
By harnessing machine learning and high-resolution OCT data, researchers have crafted a predictive tool that can forecast SANS with remarkable accuracy. As space agencies gear up for Artemis missions to the Moon and crewed expeditions to Mars, such proactive health monitoring will be indispensable. Early detection and intervention not only protect vision but also preserve mission performance and crew well-being on humanity’s next giant leaps.
3 Takeaways
1. Machine learning applied to OCT imaging can predict SANS risk mid-mission with up to 92% accuracy, offering an early warning system for astronaut eye health.
2. Key predictive features include rates of RNFL thickening, pre-flight choroidal thickness, and mission duration, overshadowing traditional demographic risk factors.
3. Implementing these predictive models enables personalized countermeasures—such as targeted exercise protocols and dietary adjustments—to mitigate neuro-ocular risks during long-duration spaceflight.
3-Question FAQ
Q1: How soon into a mission can SANS be predicted?
A1: Models identify risk as early as three to six months into flight, allowing for interventions before significant vision impairment occurs.
Q2: What countermeasures can reduce SANS risk once high-risk individuals are identified?
A2: Personalized interventions include lower-body negative pressure sessions, modified exercise routines, and dietary sodium adjustments to control intracranial fluid shifts.
Q3: Can these machine learning models be adapted for Earth-based patients?
A3: While designed for microgravity environments, similar approaches could assist with conditions involving elevated intracranial pressure or fluid imbalances, such as idiopathic intracranial hypertension on Earth.
Call to Action
Curious about the future of space medicine and how AI is safeguarding astronaut health? Subscribe to our newsletter for the latest updates on aerospace research, machine learning breakthroughs, and innovations that push the boundaries of human exploration. Stay informed, stay ahead—because when we protect our explorers, we secure our place among the stars.
