Main

Astronauts face unique hazards from spaceflight, including ionizing radiation, altered gravitational fields, altered day–night cycles, confined isolation, hostile closed environments, distance-duration from Earth1, planetary dust-regolith2,3 and extreme temperatures and atmospheres4,5. As astronauts experience these hazards, the body responds by adapting and deconditioning, with the potential for synergistic effects as the exposures persist1,2,6. As humanity readies itself for exploration of deep space and planetary-class missions (Mars and cis-lunar), astronauts would benefit from optimally scoped biomedical systems.

From the mid-twentieth century through to the present day, humanity has conducted missions in low Earth orbit (LEO) with access to nearly real-time support from mission-control-centre (MCC) flight surgeons. However, upcoming deep-space crews will face (1) high-latency communications that prohibit real-time support7,8,9,10, (2) data bandwidth and power constraints7,11, (3) infrequent resupply7,8,12, (4) the carrying of only essential and effective medications, which may degrade over time13, (5) an inability to evacuate or be quickly rescued7,8,12 and (6) greater exposure to solar and galactic cosmic radiation11,14,15,16. To address these challenges, there is an opportunity for spaceflight biology, engineering, health and medicine to leverage the powerful emerging computer-science approaches of artificial intelligence (AI) and machine learning (ML). AI/ML uses sample data to create a representation of a system that can predict an outcome of interest on future, previously unseen data17.

The basis for this Review is content from a workshop held in June 2021 organized by NASA (National Aeronautics and Space Administration) entitled ‘Workshop on Artificial Intelligence and Modeling for Space Biology’18. Biologists, clinicians and AI/ML experts in the workshop articulated the requirements necessary to create an AI-driven, proactive ‘precision space health’ (PSH) system to ensure that the future of space health is predictive, preventative, participatory and personalized19,20. In this Review, we first introduce the basic requirements for the PSH system, focusing on the importance of multilayered monitoring in support of astronaut health and habitat. We then summarize the workshop discussions, focusing on two essential topics for a successful PSH implementation: (1) biomarkers and habitat metrics necessary to monitor health and prevent health crisis using AI/ML; (2) AI/ML methods and hardware that can be used or that need further development to integrate and enable this stream of heterogeneous data. A parallel Review from the workshop reviews AI/ML challenges and opportunities to enable fundamental space biological research21.

Overview of Precision Space Health

In 2014, the National Academy of Sciences recommended an ethical framework for long-duration and deep-space exploration with health standards continually evolving, improving and informed by data22. The workshop discussion concluded that this framework will be best enabled by principles of Precision Medicine (PM) and Precision Health (PH)23. PM refers to personalized medical treatment tailored to the individual characteristics of each patient24, including genetic predisposition, behavioural influences (exercise, nutrition, stress) and environmental influences (physical and social determinants). Leveraging PM principles, PH emphasizes disease prevention and early detection via individualized, longitudinal monitoring25,26,27.

The workshop recommended a PSH system maximize AI/ML usage to integrate longitudinal, clinical, biomarker, ’omics, behavioural and microbiome data about an individual in a healthy state so as to facilitate automated and early detection of pathogenic changes (Fig. 1)28,29,30,31,32,33,34,35. It was noted that NASA and space agencies/companies are well-suited to lead in this domain, which will advance technology transfer for healthcare implementation on Earth36,37,38. In 2015–2016, the ‘Twins Study’39 monitored two genetically identical individuals on Earth and in space. It was a successful proof of concept for the ability to collect the type of rich data necessary to enable a PSH system. The experiment used several bioassays to measure effects from environmental space stressors (for example, ionizing radiation and carbon dioxide levels). There is constant monitoring of these and other stressors in the cabin/habitat during different phases of the mission (for example, pre-launch, launch and in-habitat) and these data, used as cofactors in AI/ML algorithms, can predict the onset of a medical crisis versus the temporary healthy adaptation of an individual to changes resulting from environmental stressors. One important finding in the study39 was that longitudinal data from the astronaut twin were more informative about his own health than comparisons to his twin’s synchronized data on Earth. Such a finding clearly highlights the importance of establishing a personalized baseline for each astronaut.

Fig. 1: The PSH system.
figure 1

The AI/ML workshop generated this overall view of an ideal PSH system, with the requirement that it is a constantly learning system and has a feedback loop for improving the system (shown as bold arrows). Briefly, the research informs health and environmental monitoring, which informs appropriate countermeasures. The feedback loop shows how research and terrestrial support can be used to inform real-time health support and countermeasures, with outcome information flowing back into future monitoring strategies and biological research priorities.

Full size image

The transition from LEO to deep-space missions (and cis-lunar) will present novel operational health requirements. Table 1 provides an overview of five key operational elements discussed at the workshop: a PSH must assess and aggregate and analyse multilayered data, adapt to out-of-distribution data, and provide actionable feedback. Although there have been initial onboard International Space Station (ISS) demonstrations of autonomous medical activities40, the proposed PSH system is a tremendous change from the current LEO medical planning model, which focuses on estimating the likelihood of specific medical conditions41 and provides weekly in-flight assessments in real time with MCC through audio-video streaming42, with limited or no onboard data analytics for crew-centred decision-making.

Table 1 Five key operational requirements for a Precision Space Health system for deep space travel
Full size table

A multilayered, efficient and automated biomedical monitoring system, with predictive and clinical decision support, needs development and integration into an intelligent deep-space PSH system43,44. Redistribution of medical decisions, biological data and data management responsibilities must occur among the participants in this process (that is, the onboard Crew Health and Performance system, the crew themselves, the crew medical officer (CMO), the Environmental Control and Life Support System (ECLSS) and MCC45,46,47,48). The system must enable human crew members to assess themselves, and evaluate and act on the resulting data. It will need to identify, predict and provide health solutions to problems before they arise, as well as continuously manage and analyse the expanding accumulation of environmental, biological and health data. It should provide the CMO with explainable insights into the complexities of the biological environment, crew health and mission-specific requirements (for example, destination, duration, vehicle and habitat), as well as offer predictive outcomes depending on courses of treatment (or no treatment).

The workshop discussed repurposing and expanding terrestrial AI/ML approaches for PM/PH for spaceflight. Data collection, analysis and interpretation of health-related data and biomarkers are making substantial headway when coupled with AI/ML approaches37. In Table 2, we summarize the status of current technologies. Although in varying stages of maturity and requiring additional investment in development on Earth, many show relevance for detecting and mitigating spaceflight risks. These technologies need parallel development and integration toward a common goal so they can collectively support space exploration. We discuss these key technologies and science in the following sections.

Table 2 Pertinent terrestrial or space/analogue AI/ML biomedical projects and approaches with promise for adaptation and/or implementation in a PSH system
Full size table

Biomarkers and health-status assessment

A PSH system will require robust sets of biomarkers for early disease detection and prediction26,27,49,50,51. Of note, we need to identify markers of dysfunction and also of countermeasure success. However, clinically actionable biomarkers are still being discovered and implemented terrestrially52,53. Space agencies should invest in research to identify biomarkers to predict specific disorders. These biomarkers must also be safe, reliable and reproducible so as to assess biomedical function and assess environmental impacts on health54, and must be distinguishable between those that are causal versus indicative of a disease55. AI/ML can be leveraged in two ways. First, it can aid the optimization of prediction capabilities, facilitating the acceptance of monitoring thresholds on a personalized basis56. Second, AI/ML can learn from a constant stream of data, which opens the door to identifying new biomarkers with more predictive power than established ones. Another important aspect of biomarker discovery is the usage of model organism reference experiments and missions. Small animals provide access to many more data points, which is essential for AI/ML. Tissue-on-a-chip research provides an alternative to animal experiments, and remains simpler in its biological understanding compared to a full organism57,58,59. In the following we summarize the several data types the workshop identified as having the most potential for spaceflight health biomarkers.

Predictions using multi-omic, paired and phenotypic data

Considering the known altered immune function effects from spaceflight, the immune system is a key biological system to monitor and evaluate. Single-cell immune monitoring approaches such as single-cell RNA sequencing60,61 and mass cytometry62 are currently used to monitor populations in terrestrial clinical settings where the immune system is modulated63,64. These transcriptomics and proteomics data are challenging due to the high dimensionality (for example, there are over 20,000 genes measured in one transcriptomic sample) and low sample number. Dimensionality reduction will be key for combining and learning from these datasets. In contrast, physiological data such as levels of circulating cytokines have a much lower number of dimensions. The first step is to map a baseline immune configuration of individual astronauts. AI/ML models can then use departure-from-baseline to detect the most likely causal genes or proteins enabling physiological changes from spaceflight. Future approaches will most likely integrate comprehensive immune measurements65, combining omics, physiological, microbial66,67 and other relevant data such as microRNAs68, exosomes, cell-free DNA69, clonal haematopoiesis70 and DNA damage biomarkers71. Pairing of biomarkers based on behavioural phenotypes (speech patterns, semantic/sentiment breakdown, facial expressions) with multi-omics data can create useful omic-phenotypic connections72.

Predictions using longitudinal, individualized and baseline data

For biomarker identification and real-time space health monitoring, longitudinal measurements will be essential to detect individualized health changes. For example, continuous monitoring of temperature with wearables has shown that fever thresholds change between individuals, or with age, gender or ethnicity73. Similarly, systems-level analyses have shown variations in immune setpoints63 and microbiome composition74. Previous efforts have successfully demonstrated individualized monitoring of changes to self-baselines using blood31, digital devices32 and non-invasive saliva sampling34, enabling personalized coaching of individuals75 and microbiomes33. Such approaches can first establish baselines of various biomarkers for each astronaut individually on Earth. AI/ML methods for time-series analysis can identify potentially adverse medical events by monitoring deviations from a healthy baseline using longitudinal data31,32,33,34,75. When digital twin technology is mature, having an AI-enabled digital twin of each astronaut may eventually be a better health predictor76 than what we can achieve with real biological twins. When and how to label a biological response as unhealthy is important. For example, when exposed to ionizing radiation, an expected healthy response is the induction of DNA damage. However, when DNA damage is persistent or there is poor induction of DNA damage, there is a long-term radiation toxicity associated71,77,78. Characterizing the genetics of each astronaut by whole-genome sequencing will be useful to better interpret longitudinal data.

Non-invasive health-status assessment with AI/ML

The workshop discussed ways in which health-status assessment systems can incorporate AI/ML in a non-invasive manner. For example, AI/ML voice analysis can monitor stress or fatigue, with privacy considerations mitigated by avoiding semantic analysis of dialogue79. AI/ML can analyse sleep and locomotion activity, and can assess how inferred health status is affected. To mitigate stress, AI-generated personalized and private therapy programmes can be included in crew health resources80,81. A further example is the pairing of pre-flight structural ‘biomarker’ (that is, anatomical) analysis for monitoring vascular and tissue ocular/vision changes82,83 with assessments of adverse headward fluid shifts occurring in microgravity. Such spaceflight-based results linked with other physiological monitoring of blood, nutritional, immunological and performance measures would require AI/ML analysis and equipment miniaturization37,84,85. Assessments can also motivate the importance of acquiring more space data, rather than relying on terrestrial analogues that may not replicate key features of human responses86.

Space missions should expect the emergence of novel phenotypic manifestations and situations in deep space1,6. Reinforcement learning and n-of-1 studies49,50,87,88 may help provide statistical power to derive multi-targeted treatment, behavioural interventions, or activity interventions (specific to individuals) during a mission in order to address novel phenotypes.

Monitoring the spectrum of layered and integrated flight data

The workshop recommended an essential component of the PSH system be a multilayered monitoring-collection system, including both spacecraft/habitat/ecological environments and individual astronauts (Fig. 2). Its recommendations span from both contact and non-contact devices to monitor the health of individual astronauts, to novel semi-automated assays to monitor the entire spacecraft and all habitable environments (plants, crops, animals, microbes, physicochemical environment and so on). When realized, this holistic multilayered PSH monitoring system would provide, for the first time, a continuous picture of the health of the entire spacecraft/habitat, and the entire living ecosystem inside.

Fig. 2: Layered and integrated data acquisition and monitoring for deep-space missions.
figure 2

The integrated biological and health monitoring system is shown as a pyramid, with layers of increasingly invasive and granular monitoring, where data flow from both experimental models as well as astronauts, and are put into the context of environmental monitoring via AI/ML algorithms.

Full size image

The initial layer of monitoring would be the continuous environmental sensing of physical (for example, vibrations, temperature, airflow, sounds and radiation), chemical (for example, carbon dioxide, dust particles, volatile organic and inorganic compounds) and biological components (for example, general microbiota and specific species with known risks). Many of these sensors are already standard instruments for LEO missions, but the data are not yet integrated into one database, nor analysable in situ. The NASA Open Science Data Repository (which includes both GeneLab and the Ames Life Sciences Data Archive) houses molecular ’omics and phenotypic data from spaceflight life sciences experiments, and has begun remediation of this issue89,90,91,92. Integrating sensor data will enable AI/ML models while keeping human in-the-loop inputs46. For example, radiation instruments currently in use on the ISS and planned for lunar missions gather real-time data on absorbed radiation dose and dose rates93,94,95. However, extrapolation from the absorbed dose to specific biological effects requires more detailed knowledge of the components of the space radiation field (for example, protons versus high-energy particles and neutrons). Also, of crucial importance for missions beyond Earth’s radioprotective Van Allen Belt, molecular and sophisticated dosimetry will be essential for the high-resolution detection of both DNA damage71 and gauging pharmaceutical stability96,97. Additional post-processing can achieve this on the ground with various types of active dosimeter93,98.

A second layer is traditional non-invasive physiological metrics, collected by ‘wearables’, point-of-care devices (for example, ultrasound, blood pressure, breath-analysis, ocular/visual and respiration), videos showing behavioural health, and self-administered tests, such as cognitive tests, exercise routines and sleep data54. Platforms should be minimally intrusive, and data should be easily decipherable by the crew and CMO. An example of non-intrusive monitoring shown recently used active sonar (speaker-microphone) to monitor heart rate and heart rhythms remotely99.

A third layer would be based on molecular physiological biomarkers and/or truly ‘invasive’ measures obtained from various swabs, blood draws, saliva sampling and other molecular assays. A ‘smart toilet’ (as well as a smart shower booth and a smart mirror) could preserve and prepare waste specimens for biochemistry assays and microbiome profiling25,100,101. Such platforms hold promise in expanding to include in situ and real-time analytical capabilities. Similarly, non-invasive high-frequency monitoring of molecular components from saliva over time can also provide immune signatures that may monitor deviations from a healthy immune baseline, using anomaly detection algorithms to assess change points as potential adverse medical events34.

Adaptations in computing, model-training and data communications in deep space

To enable real-time monitoring, in situ operations will play an increasingly important role in enabling immediate responses (Fig. 3). In addition, as the volume of health-related data grows until it is no longer viable to downlink datasets, AI/ML applications will offer real-time health capabilities, requiring full in situ computation for maximal autonomy. In this section we will dive deeper into the various aspects of this paradigm shift.

Fig. 3: Space biology and PSH system AI/ML life cycle.
figure 3

The life cycle of AI/ML in space biology research and spaceflight health enables full utilization of Open and FAIR space biological datasets through a data management environment. AI/ML methods leveraging neuromorphic processors and edge computing facilitate a layered and integrated spectrum of in-flight data acquisition to power a PSH system for deep-space missions.

Full size image

Translational science and the AI/ML biomedical life cycle

Translational knowledge and data from both fundamental and applied space biomedical research are part of a crucial translational pipeline to inform a PSH system. Such research builds a wealth of evidence and statistical power upon which AI/ML biomedical predictions rely. The life cycle of AI/ML and the cross-cutting relationship between space biological research and a PSH system are shown in Fig. 3.

Current limitations for data communications and computing

There are several limitations to deploying computer, data-storage and communication systems on board any spacecraft102. These factors include volume, weight and power constraints, resilience to launch vibrations, chemically reactive planetary dusts and ionizing radiation103, and reliability in autonomous maintenance and repair. The pioneering development of the Spaceborne Computer-1 and -2104 aboard the ISS has made inroads towards addressing these challenges. Specialized hardware could reduce the footprint of computer needs while delivering very high AI/ML performance. For example, onboard graphics processing units would accelerate AI/ML capabilities for DNA sequencing requiring real-time base calling for detecting modified nucleic acids105. The workshop recommended more investment in the development of hardware enabling AI/ML. The following are strategies the workshop identified to circumvent the unique constraints of spaceflight: (1) developing software that relies less on accessing Earth databases, (2) using ML-based data compression106 and (3) deploying active learning systems based on pre-trained models on large Earth databases but constantly learning from the growing stream of data collected on board the spacecraft/habitat107. As missions move further from Earth, access to computing power will need to transition from terrestrial-based to spacecraft and habitat-based. Many challenges of building space-ready AI/ML are difficult to foresee, and there will be more challenges for Mars missions108.

Modelling with deep-space biomedical data

Most current terrestrial AI/ML techniques use large numbers of observations (and usually a large feature set) to train a model. The application of these methods to human health in spaceflight is challenging for several reasons. First, biomedical data collected and reusable from astronauts in-flight are historically and currently limited. Over 600 people worldwide have gone to space, and only 24 beyond LEO. Missions have had an average duration of fewer than 30 days, making it almost impossible to train sophisticated AI/ML models using only spaceflight human data. Second, the data collected during spaceflight have been narrow and inconsistent compared to what is typically available in clinical or research settings on Earth. Until recently, there was no standardized set of biomedical measures taken during NASA human spaceflight explicitly for research purposes. Finally, NASA’s mission of exploration means that the models needed for a PSH system must be extrapolated beyond the context in which we have spaceflight experience. Even though the laws of physics do not change over the course of a deep-space mission, the human body does, and potentially in nonlinear ways, thus lowering the accuracy of the ‘approximation’ training data. A key consideration for AI/ML medical applications is whether the system needs to be trained in situ using locally collected data, or whether the models can be trained using ground data prior to the mission, during the mission with an uplink of the updated model, or gradually developed into active learning systems107 to add autonomy degrees. These distinctions are important, because training a model is typically computationally intensive and requires large amounts of data, whereas performing inference with a trained model is far less demanding.

To address these considerations and limitations, Table 3 summarizes the workshop recommendations for methods to address the key requirements of an AI-enabled PSH system, and gives examples of applications. Neuromorphic processors are one exciting emerging technology to provide in situ computing that is well suited for performing AI/ML tasks in a spaceflight context, specifically, because of their radiation resilience, low power requirements and high efficacy for deep learning tasks. Indeed, NASA has recognized the potential of neuromorphic computing for these reasons109. Edge computing is a resilient approach that creates a redundant network to distribute computing needs based on sophisticated job management that prioritizes onboard activities. Finally, computing hardware, model software, and data management and communication strategies for PSH need to be maximally adaptive to accommodate constant reassessment of newly acquired data. Active learning algorithms expedite this approach, where humans can help AI/ML algorithms learn faster by injecting new data types and new labels into the system.

Table 3 Computing, hardware and model development requirements and recommendations for AI modelling for health in deep space
Full size table

Federated learning was developed to deal with limited abilities for data transfer, where multiple model components are trained at the location of the data, and then merged into a single federated model. This would deal with large datasets generated in space, or NASA astronaut health data containing considerable privacy constraints. Regardless of how small and efficient computing resources and algorithms become, AI/ML in space works under limited conditions compared to that on Earth. Transfer learning is also a highly effective way to reduce spaceborne computing and data demands. Using this technique, AI/ML models are first trained on Earth using analogue datasets (for example, animal data or synthetic data). Subsequently, in space, the final computation to ‘transfer’ the model to the actual context of interest is performed, such as a classification model using astronaut health data. This reduces the computing and data requirements needed to achieve a model with acceptable efficacy. Dimensionality reduction is a central way to reduce the amount of data to consider in the prediction model, but there is more work to be done at the research level to be smarter about what features drive a response and thus only consider these factors.

Finally, probably the biggest challenge in a PSH system is the fact that such a system will operate using out-of-distribution data. Any training done with Earth analogues will only partially resemble what is truly happening in space. Methods that are more resilient to operating under different assumptions will be essential. The workshop listed several such methods, summarized into three categories: translation from animals to humans110,111; generalization across domains using approaches such as risk extrapolation or domain invariant methods55; and adapting models trained on Earth health data to perform well on space health data where the baseline normal is persistently different.

Further challenges for deep-space health

Integrated biomedical flight data acquisition, AI/ML modelling tools and techniques, as well as a PSH system will be crucial pillars in bridging the gap between the current LEO operational paradigm and that which is needed for successful cis-lunar and planetary-class missions. Discussed below are several other space biomedical challenges involving AI/ML covered by the workshop.

Crew confidence and fidelity of AI/ML

Through the lens of clinical significance, data are used to inform clinical decisions about preventive and acute care. It is important that we shift the role of diagnostic AI/ML from simply predicting labels, to interpreting context and providing iterative cues that guide diagnosticians112. The crew, flight surgeons and all related staff must wholeheartedly establish training and build confidence in an AI-based PSH system. Tools and techniques for AI/ML and modelling (or any type of data system) additionally ought to include comprehensive and potentially continuous assessment of its credibility, ethics and trustworthiness. This includes methods that address reference data or model prediction benchmarks. Fidelity and ethics assessments of AI modelling extend past specific technique validation. They encompass broader aspects of credibility, including a full provenance of life-cycle development and evaluation of the systemic assumptions, biases and deployment limitations (that is, toward predictions or knowledge-gained)113,114.

Ethics, genomics and de-identification

The spaceflight community has a responsibility to meet ethical obligations involved in a PSH system and related data privacy28,115. Whole-genome sequencing will probably be a pre-flight component required to enable an effective AI-driven PSH system. Handling privacy and the wishes of all spacefarers (NASA astronauts, but also international and private-commercial space travellers) requires sincere care, clear governance, and policy. De-identified data systems with decoupled federated learning systemic firewalls are one approach to ensure data are explicitly not traceable116. Broad data analysis must occur to support deep-space health, with completely untraceable data to protect any impacts to multi-generational offspring regarding their privacy and quality of life. The development of modified Genetic Information Nondiscrimination Act (GINA) guidelines28,117 and waivers for utilization of spaceflight genomics data should protect astronauts and their relatives.

Make data AI/ML-ready

Metadata and data curation-processing standards for PSH and the field of space biology need to be determined, ensuring data are ‘AI-ready’ and modelling accessible. A space ‘data readiness level’ metric can be implemented as a tool to encourage reliable data quality118. Basic synthetic datasets and model libraries for space health and biology also need development. Adaptations and innovations in statistics, algorithms, data and medical informatics are almost certainly going to be required and modified for spaceflight health and biology21,119.

Interdisciplinary teams

Deep-space missions will have one-of-a-kind space biology and health requirements. The interdisciplinary breadth of the teams required to develop, collaborate, refine and implement these systems is novel. Communities of programmers, biologists, data architects, human factors, operational clinicians and hardware engineers need to be interwoven (non-siloed), so as to implement a PSH and biomonitoring/research system120. Research is likely to flow more freely between researchers and clinical operations during deep-space missions, which could enable novel discovery outcomes. For example, a recent research team utilizing an ultrasound device was interested in blood-flow behaviour in the internal jugular vein in long-term ISS mission crew. The team noticed anomalies in venous blood flow, which led to a clinical team handover, who then characterized (and treated for) flow stasis and venous thrombosis121,122.

Recommendations and conclusion

The workshop brainstormed a decadal vision of what should be developed to leverage AI/ML for both spaceflight health support and space biological research. Various AI/ML technological needs were identified, and the workshop brought together (for the first time) an array of interdisciplinary AI/ML experts with space-oriented biologists and clinicians. Workshop participants agreed that it is difficult to predict the needed space capabilities as there is no platform today to serve as a developmental testbed for verification or validation of real-time, actionable, AI/ML model testing. During the workshop, time was spent understanding NASA’s data systems and technologies as they stand today, and the fundamental data management developments and designs needed to begin implementing AI/ML. Although no detailed roadmap, nor strategic framework, was produced, the general workshop goal of producing a decadal vision was achieved, through the identification of several broad focus areas over the next decade. Based on current capabilities, and with applications of AI/ML still in their infancy terrestrially in relation to clinical health, workshop participants arrived at a number of reasonable avenues to begin enabling this field:

  • Develop hardware and data systems to enable efficient data collection for in situ analytics, with an emphasis on baseline and longitudinal data

  • Develop stronger data management, harmonization, AI-readiness, and a data-readiness-level metric to allow in-mission and terrestrial-analogue research

  • Invest in an AI/ML/modelling ‘Zoo’, repurpose current AI/ML approaches for space-specific needs and research into novel algorithms, software, hardware and models

  • Transition crew health from treatment to prevention, which is essential for deep space, and requires a holistic, multilayered, biomonitoring approach of the entire spacecraft/habitat ecosystem, including multi-modal, multi-hierarchical data and research into targeted biomarkers

  • Integrate hardware, data systems, software/AI/ML and biomonitoring to enable the development of a PSH system for real-time, adaptable, maximally autonomous decision-making and analysis, in conjunction with the CMO and crew participation.

The workshop generated several far-reaching questions that cannot yet be fully answered and require focus in the next decade of research and development. How much degradation of crew health and capability should we expect? How much decision support and data analytics capability should be built into a mission to maximize the chances of success? How much long-term health risk will astronauts face post-career? These questions can only be answered by scientific research, and this is reliant on strong data systems in future missions and experiments.

The development of the AI/ML modelling systems discussed in this Review will be multi-year, interdisciplinary and involve far-reaching collaborations. As humanity explores beyond LEO, it will leave the confines of an immediately accessible, large and continuously supportive cohort of mission-control health and science staff, with systems that have been developed over the past ~60 years. Today’s exciting new spacecraft, with expansive payload abilities, greatly enable the field of space health and biology to develop and deploy AI/ML/edge technologies, toward in situ analytical capabilities. The relatively recent emergence of terrestrial AI, ML and modelling brings a key contribution towards making humanity multi-planetary through spacecraft and habitat biomedical science support and a PSH system.