SANS CNN system Kamran, S.A., Hossain, K.F., Ong, J. et al. SANS-CNN: An automated machine learning technique for spaceflight associated neuro-ocular syndrome with astronaut imaging data. npj Microgravity 10, 40 (2024). SANS-CNN: An automated machine learning technique for spaceflight associated neuro-ocular syndrome with astronaut imaging data | npj Microgravity
4 Likes
Thanks @Wilester2025 for sharing those papers ,
Great resources and we will certainly approach their model and data in detail. 
Hi everyone,
Here is the CDSS subgroup summary from 5th May 2026.
Thanks to everyone who contributed and made this discussion and brainstorming possible.
There will be a forum where an engineer/developer will be at the center, and other scientists will help them build the CDSS-specific layer.
Here is the Docs link to the framework: CDSS Project Idea Sheet - Google Docs Thanks @robertsb21
Also, the diagram was generated from the framework by ChatGPT Image 2.
@Ritika_Saha will put the Excel sheet here.
Let’s build the first SANS-CDSS! @AIMLawg
Summary :
Meeting Summary: CDSS Sub-Group Meeting – 5th June 2026
Key Discussion Points:
-
Psychological support through plants – Jian and Nic explored how growing plants (not for food) could provide psychological benefits for astronauts on long missions. Nic suggested using Likert-scale surveys for measurement; Wayne added biometric data (heart rate, blood pressure) as potential metrics. Elizabeth mentioned Antarctic expeditions as analog environments.
-
Main framework for CDSS – Alireza presented a comprehensive architecture for a Clinical Decision Support System focused on SANS (Space-Associated Neuro-ocular Syndrome), the leading eye problem for Moon/Mars missions. The framework includes:
- Input data: OCT, fundus images, visual acuity, physiology (heart rate, etc.), mission data (CO₂ levels, mission day), astronaut medical history.
- Dashboard: Space-friendly UI (Ritika’s focus).
- Intelligence layer: Agents with Bayesian networks and RAG (retrieval-augmented generation) for safe, rational reasoning.
- Memory/knowledge: Knowledge graphs linking medical literature and astronaut records.
- Governance & safety: Federated learning, anonymization, cybersecurity.
- Infrastructure: Offline/edge computing for space conditions.
- Validation: Benchmarks, human expert scores, calibration methods (e.g., AUC).
-
Project scope & skills – Alireza encouraged members to align their expertise with framework components (e.g., his own work on foundation models for eye images). Jian suggested adding a preventive medicine loop where the AI recommends actions or additional data collection.
-
Feasibility – Despite complexity, Alireza believes AI coding agents can accelerate development significantly (e.g., six‑month model now possible in two days).
Next Steps: Members to identify which layer(s) of the framework they can contribute to, with continued integration into the digital twin project (coordinated by Jian).
2 Likes
Hi Alireza -
I am thinking to volunteer working on a “synthetic data generator or simulator”, to model the data input layer that feeds into everything else. Without enough real data yet, we can focus on this synthetic data to build the rest of the framework. The framework, the UI and the rest of the knowledge graph will be the true innovation but the data input can be the initial foundation that we can build. I can pull in from our DT group’s focus — generative AI frameworks to start exploring. As you said during the meeting, validation and verification will be challenging. I will start looking into foundation models, as well.
The plan is to start from “small real data, large synthetic data” first, but gradually transition to “large real data, small synthetic data” later to bridge the gap, to prepare the framework to become mission ready.
This is just an idea I had after reading your post and thinking about it over the weekend :). I hope to show some prototypes next meetings and will work with Ritika on the UI elements. Certainly welcome more new & fresh ideas!
3 Likes
@AliReza-H Sorry for having been unable to attend the meeting due to a schedule conflict.
As noted on the sheet, I added several additional architectural layers and capability areas to the document from an AI governance, knowledge management, and human-centered systems perspective. These additions include:
• Knowledge & Memory Layer (RAG, longitudinal astronaut memory, knowledge graphs)
• Governance & Safety Layer (auditability, uncertainty quantification, validation, safety policies)
• Human Factors Layer (explainability, cognitive-load-aware interfaces, multimodal interaction)
• Infrastructure Layer (privacy-preserving learning, cybersecurity, interoperability)
• Additional inputs related to evaluation, mission readiness, human factors, validation datasets, and risk taxonomy
• Additional skills covering AI systems, governance, mission assurance, knowledge architecture, and cross-disciplinary integration
My intention was to help ensure that the CDSS is approached not only as a clinical decision-support tool, but as a complete socio-technical system that can operate safely, transparently, and effectively in long-duration spaceflight environments.
I would be very happy to help lead or co-lead the architecture, AI governance, knowledge systems, and human-AI integration aspects of the effort, and to support the engineer/developer team in translating these concepts into an operational framework.
Looking forward to collaborating with everyone as the project moves forward.
2 Likes
Hi All,
My apologies, I couldn’t make the last meeting. I reviewed the CDSS document and have requested edit access. The architecture plan is good, but its quite a large scope (not sure how many members are actively involved in this and what timeline we are working with). Is SANS the current focus?
My expertise is in the cardiovascular and critical care realm and am building cardiovascular digital twins through my startup. But SANS is a real problem and happy to help contribute as possible.
2 Likes
Dear @jgong
Wonderful !
This is a foundational part of the project, I think.
Please create the repository in your GitHub and share it here. You will be at the center of this layer—2–3 others can approach you as a team. We’ll build a minimal viable product (MVP) and test it against current benchmarks from the literature. In 70–90 days (about 5–6 CDSS meetings) then we’ll submit a preprint to bioRxiv.
P.S. I’ll set up my own GitHub for the Eye Foundation model layer and will be the developer there. Let’s build more cool things
1 Like
Hi everyone,
Those who are interested in contributing to a particular role are requested to add their names to the Excel sheet and specify the role(s) they would like to contribute to.
Thank you.
2 Likes
Dear @jgong
I would be happy to collaborate on the UI and visualization components as the prototypes evolve. It will also be interesting to explore how the synthetic data layer can interact with the knowledge graph and adaptive decision-support modules to create a more integrated system.
Looking forward to the upcoming meetings!
1 Like
Hi all, before I forget just dropping some fundamental publications - if haven’t already(? can;t remember)
Validation of the NASA_Integrated_Medical_Model.pdf (634.5 KB)
4 Likes
not sure, if the git repository has been created yet? @jgong yet, here is a sample GitHub repository with an architecture-first structure, hopefully it would be helpful:
space-cdss/
│
├── docs/
│ ├── architecture
│ ├── requirements
│ ├── risk-analysis
│ ├── evaluation-framework
│ └── use-cases
│
├── knowledge-layer/
│ ├── rag
│ ├── knowledge-graph
│ └── memory
│
├── reasoning-layer/
│ ├── bayesian-network
│ ├── inference
│ └── agent-orchestration
│
├── safety-governance/
│ ├── audit
│ ├── policy-engine
│ └── validation
│
├── data-connectors/
│ ├── biosensors
│ ├── vehicle-systems
│ └── simulators
│
└── ui/
1 Like
Hello everyone,
Following our recent discussion, I’ve set up an initial GitHub repository to start structuring the CDSS effort for long-duration spaceflight.
The aim is to create a shared working space where we can gradually shape the architecture, governance layer, reasoning/RAG components, and broader system design in a transparent and collaborative way.
Here is the repository: GitHub - aka79/NASA-CDSS-for-Long-Duration-Spaceflight: Clinical Decision Support System (CDSS) for Long-Duration Spaceflight · GitHub
The current structure is intentionally lightweight, designed to support early-stage contributions across:
-
System architecture
-
AI governance frameworks
-
Reasoning and RAG layers
-
Security and federated learning considerations
-
Documentation and research notes
This is still in its formative stage, so contributions, critiques, and ideas are all very welcome as we refine the direction together.
Looking forward to building this out collectively.
Best regards,
Ayse
2 Likes
Wonderful start @ayse
GitHub is a great place to start building these AI based projects,
But i highly suggest limit scope your project , For example on Governance layer , so other whom are interested in collaboration with you can reach out.
Each project,in each layer will have a Github repo from it’s own developer .
Keep Building
1 Like
Hi ,
It’s wonderful that you have interest inthis project and i am very happy about
If you have engineering skills , feel free to define a project related to those layers of CDSS system,
If you have knowledge , share the idea and ask for engeeniers here to help you building it !
Please reach out if anything we can provide to make you build more and faster
sure, thanks for your suggestion…this is the coordination / governance umbrella repo for early-stage design, I’ll update it accordingly.
1 Like
I highly recommend everyone read those three papers from Google , although they are very straight forward about findings , i made a NotebookLM slides so i can understand the context better , which i share it here :
“A multi-agent system for automating scientific discovery”
“An AI system to help scientists write expert-level empirical software”
“Accelerating scientific discovery with Co-Scientist”
Link :
4 Likes
Hi all,
This is a good use case and novel method for the input layer of CDSS for SANS.
I think there is a great opportunity to start a project if anyone is interested.
1 Like
I wanted to share a very recent Google Health/DeepMind paper, Towards Conversational AI for Disease Management, alongside the related GitHub repository
I think this paper could be useful for our subgroup, not because we would directly use AMIE or RxQA as-is, but because it provides a strong methodological blueprint for how a clinical decision support system could be evaluated and strengthened.
The paper moves beyond simple diagnostic AI and focuses on management reasoning: how an AI system follows a patient over multiple visits, updates its recommendations as symptoms, test results and treatment responses change, and grounds its outputs in clinical guidelines and drug formularies. This is highly relevant to lunar and deep-space exploration, where a CDSS would need to support not just “what might this be?”, but “what should the crew do next, given limited resources, delayed communication, mission constraints and evolving clinical risk?”
Several elements could be useful for enhancing our CDSS model:
- A management-reasoning layer
AMIE separates conversational interaction from deeper management planning. A similar structure could help our CDSS distinguish between crew-facing guidance and more detailed reasoning over evidence, protocols, risks and mission constraints. - Guideline- and formulary-grounded recommendations
The system retrieves and reasons over authoritative medical guidance and drug information. For our project, this could translate into grounding recommendations in NASA medical protocols, exploration medical kit contents, medication constraints, and operational procedures. - Structured, citation-backed outputs
AMIE generates structured management plans with references to source documents. This could be valuable for our CDSS by making recommendations more transparent, auditable and safer for review by crew medical officers or ground support. - A spaceflight-specific medication benchmark
RxQA is a benchmark for medication reasoning. We could adapt this idea into a “SpaceRxQA” benchmark focused on medications, countermeasures and constraints relevant to long-duration missions, including limited formularies, contraindications, monitoring limitations, drug interactions and escalation thresholds. - Multi-visit astronaut case simulations
The paper evaluates AI using multi-visit clinical scenarios. We could develop similar spaceflight-specific cases, such as infection risk, SANS, renal stone risk, radiation-event follow-up, musculoskeletal injury, behavioural health, sleep disruption or dental emergencies. This would allow us to test whether the CDSS can reason longitudinally rather than only respond to one-off inputs. - Safety and guardrail testing
The approach could help us formalise “never events” for the CDSS, such as recommending unavailable resources, failing to escalate red flags, giving overconfident advice, hallucinating medications, or ignoring communication delay and onboard capability.
Overall, I think the main opportunity is to use this paper to strengthen our evaluation and validation framework. A possible next step would be to create an evaluation/ section in the subgroup GitHub repo (I’m happy to branch and add this) containing:
- a SpaceRxQA concept;
- multi-visit astronaut case templates;
- safety/never-event test cases;
- a management-reasoning rubric;
- human factors/usability criteria for crew-facing recommendations.
Sorry for the long text - we can look into it further at the meeting this afternoon.
1 Like