Choice of the Conditions of Interest

As shown in Fig. 2, the space medical conditions in consideration have been categorized based on their likelihood of occurrence and consequences on crew health and mission objectives. “Red” risks are given the highest research priority due to their significant impact on crew health and performance. “Yellow” risk can be those: i) accepted due to a very low probability of occurrence, ii) that require in-mission monitoring to be accepted, or iii) that require refinement of standards or mitigation strategies to be accepted. “Green” risks are considered sufficiently controlled either due to low likelihood and consequence or because sufficient mitigation strategies are available to manage the risk to an acceptable level.

In this article, given their need for further research, we have only included “red” and “yellow” risks and the final list of conditions is shown here:

• Psychological and Mental Health includes HRR risks:

  • Adverse Cognitive or Behavioral Conditions and Psychiatric Disorders

  • Performance Decrements and Adverse Health Outcomes Resulting from Sleep Loss, Circadian Desynchronization, and Work Overload

• Acute Radiation Sickness

• Surgical Emergencies and Anesthesia includes HRR risks:

  • Renal Stone Formation

  • Bone Fracture due to Spaceflight-induced Changes to Bone

  • Injury and Compromised Performance Due to Extra-Vehicular Activities

  • Injury from Dynamic Loads

• Cardiovascular Deconditioning including Post-Flight Orthostatic Intolerance

• Altered Sensorimotor/Vestibular Function Impacting Critical Mission Tasks

• Infections and Sepsis includes HRR risks:

  • Adverse Health Effects Due to Host-Microorganism Interactions

  • Adverse Health Event Due to Altered Immune Response

• Spaceflight Associated Neuro-ocular Syndrome Performance Decrement

• Crew Illness Due to Inadequate Food and Nutrition

• Reduced Physical Performance Capabilities Due to Reduced Aerobic Capacity

• Impaired Performance Due to Reduced Muscle Size, Strength and Endurance

• Adverse Health and Performance Effects of Celestial Dust Exposure

• Ineffective or Toxic Medications During Long-Duration Exploration Spaceflight

Psychological and Mental Health

A journey into deep space places exceptional challenges on the psychological health of the crew. A 2001 National Academy of Sciences commissioned report named psychological health among the most significant risks for crewmembers during a Mars mission. ⁹ During such missions, crewmembers will be exposed to many stressors, including extended periods of social isolation and close-quarter confinement, chronic stress from carrying out high-caliber work, solving unexpected emergencies with little help from the ground, ⁵⁸ and structural brain changes due to microgravity and ionizing radiation. ⁵⁴ As such, conditions such as depression and anxiety disorders are likely to arise, as shown from previous long-duration missions on Mir and similar ground analog studies. ⁷⁷ Fortunately, more acute conditions such as delirium have only been described on rare occasions. ⁷⁷

Without appropriate and effective mitigation, such conditions could derail a mission and even threaten crewmembers’ lives. Most countermeasures developed for short-duration and International Space Station (ISS) missions involve keeping crews connected to home, such as video conferences with family members, mission control and healthcare professionals, and resupplying cargos of letters, food, and surprises. ⁷⁷ Unfortunately, these will be restricted, or indeed wholly unavailable, as crews journey toward Mars and communication latency increases up to 22 min. ⁵⁷

Various AI systems are being developed to combat this unusual form of “homesickness”; for instance, closed loop communication systems that integrate behavioral monitoring and real-time feedback. The Crew Interactive Mobile Companion system has been deployed on the ISS since 2018 to provide artificial companionship and work assistance. ²¹ It is equipped with natural language processing and computer vision capabilities to interpret speech and facial expressions. Therefore, it can carry out simple mission tasks and converse empathetically based on the derived emotional state of astronauts. ¹⁰⁶ NASA and Ejenta have also collaborated recently to develop a conversational AI that provides emotional support for mission crews. ¹⁰⁶

On Earth, there is unprecedented attention in developing AI chatbots for mental health conditions. In their systematic review, Abd-Alrazap and colleagues identified 41 unique mental health chatbots and 14 randomized controlled trials that assessed the effectiveness of chatbots in mental health. ¹ A number of chatbots relevant to spaceflight-induced psychological stressors have also been validated in clinical trials. They include Woebot, which delivers personalized psychotherapy for depression and anxiety, ³⁸ Tess, which teaches coping skills and provides support for social isolation, ³⁰ and an interview-style conversational system developed by Welch et al. to cope with the psychological and health stress due to the COVID-19 pandemic. ¹²⁵ To better emulate the patient-clinician experience, these trials drew similar conclusions on the importance of appropriately expressed empathy; assigned human traits (e.g., being helpful, caring, open to listening, and nonjudgmental); minimization of speech interpretation errors; closer simulation of natural conversation; and involvement of trained and experienced clinicians in the design process. ^{30 , 38 , 125} To adapt these systems for aerospace research, tests designed specifically for the high-performing astronaut population against long-duration mission stressors should be adopted, such as the Cognition Test Battery that has been tested in NASA’s ground-based Human Exploration Analog project. ⁸⁵

Acute Radiation Sickness

As Mars crews leave Earth’s magnetosphere and venture into deep space, they will be exposed to space radiation originating from background galactic cosmic radiation and highly energetic solar particle events. ⁹¹ Both induce high levels of oxidative stress on the human body. ⁹¹ Among the associated health effects, the most serious and unpredictable is acute radiation sickness related to solar particle events. ³⁴ Chronic exposure to space radiation also exposes the crew to long-term risks, including cardiovascular diseases, ¹⁶ cataracts, ¹³ cognitive dysfunction, ⁸⁸ and carcinogenesis; ¹²⁴ however, these are beyond the scope of this review article. The combined risk of radiation-induced mortality for a Mars mission was estimated to be around 5%, with an upper 95% confidence interval near 10%. ²⁷ Morbidity is dependent on the level of radiation exposure, and early mortality is often due to gastrointestinal mucosa destruction leading to fluid loss via extensive vomiting and diarrhea, and bone marrow depletion leading to significant anemia, thrombocytopenia, and immunosuppression. ⁷⁴ Acute physiological support entails aggressive control of vomiting, fluid and electrolyte replacement, antibiotics, and transfusion of blood products. ⁸⁷ Adequate treatment can only be achieved through accurate quantification of exposure levels and clinical markers, such as the rate of decline of absolute lymphocyte count and the time to onset of emesis and chromosomal changes, to discern disease severity and anticipate future complications. ⁸⁷

Although in-flight radiation dosimeters and radiation sensors are commonplace on the ISS, quantifying changes in chromosomal structures and molecular signatures offers a more accurate estimation of the radiation dose absorbed by the body. However, these techniques are often laborious and require dedicated expertise. Considerable success has been achieved in automating and streamlining the dicentric assay, the current gold standard of radiation dosimetry. ^{55 , 70 , 104} For example, Jang et al. ⁵⁵ proposed a new deep learning system, and doses ranging from 1–4 Gray agreed well within a 99% confidence interval of gold-standard measurement methods. In their recent review, Ainsbury et al. ² have also highlighted the potential for AI to predict personalized risks and responses toward radiation exposure and the significance of translating such technology to aerospace research. A note of caution for translation toward space application is that the dose–effect correlation is different between terrestrial ionizing radiation and space radiation. ⁶⁹

Point-of-care automated quantification of full blood count is limited on both the ISS ⁵⁶ and Earth, ⁷ likely due to poor extrapolation from small volume samples. To overcome this, new systems incorporating AI have been developed and validated. For example, the Sight OLO system, a cubic-foot-sized device developed by Bachar et al., was shown to achieve accuracy comparable to clinical hematology analysers ⁷ ; the Hilab system developed by Gasparin et al. also provides low-cost and accurate analysis of full blood count parameters. ⁴¹

Despite all preventative and protection measures, crews may still be affected by acute radiation sickness. In this situation, it may be necessary to provide the affected individuals with supportive treatment, including hemodynamic resuscitation, intravenous (IV) fluid restoration, and bone marrow transplant. ³⁴ AI tools could assist in this task, with systems dedicated to hemodynamic optimization, for example, in the perioperative context or in sepsis (see following sections).

Surgical Emergencies and Anesthesia

Conditions requiring surgical intervention are expected to have one of the highest impacts on a potential Mars mission. Within the NASA HRR, three standalone “red” risks have been identified for surgery-related conditions (renal stones, bone fractures, and injuries during extravehicular activities). ⁸⁶ Additionally, data from analog populations and previous spaceflight estimated that the risk of conditions potentially requiring general anesthesia to be around 2.56% on a 950-d mission to Mars with six crewmembers. ²² Some of the risks are directly increased by exposure to the space environment. For instance, significant bone mass loss in microgravity could increase the risk of fractures during extravehicular activities. ⁴⁸ Managing surgical complications will pose major challenges due to equipment and manpower constraints, such as limited perioperative imaging capabilities and lack of both surgical equipment and on-board surgical expertise. ⁸

Realistically, nonoperative options should be attempted first, with surgery reserved to conditions that threaten the loss of limb or life. ⁶⁶ Open surgery in weightlessness may be restricted as body fluids become hard to control in microgavity. ⁴⁸ For example, in the case of a femur fracture, temporary stabilization such as long leg plastering or external fixation may be considered. Regional anesthesia should also take preference over general anesthesia in the spaceflight setting, because of their lower risks and requirement for preoperative, intraoperative, and postoperative resources. ⁶⁶ If general anesthesia were necessary, adapted approaches such as the use of IV ketamine, which is commonly used in austere settings, and the use of video laryngoscopes for airway management could increase safety and likelihood of success. ^{66 , 108} These techniques could be carried out by the crew under the guidance of validated instructional videos, as real-time telecommunication and remote control of equipment will be precluded by transmission delays. ⁵⁷

Despite the enormous challenges involved, we anticipate that several AI applications could assist remote crewmembers in caring for a surgical patient. Preoperative imaging is a crucial step in surgical planning. Among the different imaging modalities, ultrasonography is currently and will likely remain the leading imaging modality for human spaceflight. ^{61 , 97} Among the high risk surgical conditions listed in the NASA HRR, ultrasonography-based AI tools have proven useful in the detection of renal stones ^{82 , 90} and soft-tissue injuries. ^{28 , 103} For bone fractures, AI has only been developed for the pediatric population, ¹³¹ in whom radiation exposure from plain radiographs or computed tomography is a significant concern. ¹¹⁹ However, given the increasing interest in implementing ultrasonography for point-of-care diagnosis of bone fractures, ^{18 , 89} it is reasonable to expect future AI research to extend its application to adults.

The correct identification of anatomical landmarks is also crucial to ensure safety and success during surgery, especially if emergency procedures were to be carried out by nonexpert crewmembers. Semantic segmentation, i.e., partitioning an image to locate underlying structures and boundaries, ¹¹⁴ has been applied to medical imaging to provide intraoperative guidance, such as avoiding biliary tract injury during laparoscopic cholecystectomy ⁷⁶ and ureteric injury during rectal cancer resection. ⁶² Although autonomous surgical robots are a long way away, some exciting progress has been made. For example, the Smart Tissue Autonomous Robot developed by Johns Hopkins University matched or outperformed human surgeons in autonomous bowel anastomosis in animal models. ¹⁰² However, researchers must develop ways to minimize the mass and volume of these advanced hardware until they can be carried on board.

The use of AI in anesthesia has been reviewed in several articles. ^{49 , 51 , 105} One of the most useful areas for translation into spaceflight is closed-loop systems designed for automated anesthesia delivery. ^{71 , 83} They are usually based on real-time measurement of electroencephalogram data and vital signs, which are surrogate markers for the depth of sedation. Perioperative hemodynamic optimization has also been developed to decrease postoperative morbidity and reduce both hospital length of stay and hospital costs in studies involving major surgery. ^{64 , 92 , 111} Most of these systems work by administering IV fluids and vasopressors at a rate guided by an AI controller, informed by physiological parameters which can be invasive or noninvasive, and include blood pressure, cardiac output, and/or estimators of fluid responsiveness. ⁶⁴

Finally, AI has seen its application for postoperative care to predict complications such as surgical site infections, ⁴⁴ bleeding, ¹⁹ intensive care unit admission, and mortality. ²⁰ This might have important implications for accurate planning and allocation of the limited resources, as well as optimization of the recovery process for crewmembers who require surgery in future spaceflight.

Spaceflight-Associated Neuro-ocular Syndrome

Spaceflight-associated neuro-ocular syndrome (SANS) is a constellation of structural and functional changes to astronauts’ vision after prolonged exposure to the space environment. ³ This syndrome was observed in around 23% of short-duration Shuttle crewmembers and 48% of long-duration ISS astronauts. ¹³⁰ Affected individuals presented with one or more of the following clinical signs: refractive error, optic disc edema, choroidal folds, cotton wool spots, and optic nerve thickening. ⁶⁷ The main contributor of SANS is thought to be microgravity-induced cephalad shift and the associated increase in intracranial pressure, venous congestion, and changes in intraocular pressure; however, its exact etiology remains unknown. ⁶⁷ This has prevented the development of effective countermeasures other than symptomatic relief, such as prescription glasses to adapt to refractive changes. ³ A few risk factors for SANS have been identified, including duration of spaceflight, high salt diets, intensive resistive exercise, increased ambient carbon dioxide concentrations, and nutritional deficits. ³

Although intracranial pressure changes underlie much of the pathology of SANS, it has never been measured during spaceflight. This is because the process is invasive, requiring insertion of a needle through the vertebral discs into the cerebrospinal space, and risks bleeding, infection, and injury to local structures such as the spinal cord. Novel AI tools or models that rely on surrogate markers for intracranial pressure have emerged in recent years, with the potential of accurate noninvasive estimation. ^{35 , 37 , 60}

Currently, SANS diagnosis relies on manual interpretation of in-flight ocular ultrasound, fundoscopy, and optical coherence tomography scans, which astronauts must collect with real-time remote guidance by ground experts. ³ However, this privilege will soon be lost with deep space missions. Proof of concept AI applications to automatically interpret these imaging modalities have been proposed. ¹¹⁶ The European Space Agency recently carried out the “Retinal Diagnostics” research project on the ISS, which made use of a portable retinal imaging device linked to an iPad, where images taken were fed into an AI model for rapid identification and monitoring of ocular abnormalities. ¹¹⁰ NASA has also announced plans to develop a multimodal visual assessment system, creating a “pooled” diagnosis from multiple diagnostic tests as they have different sensitivities and specificity at picking up the ocular signs associated with SANS. ⁷²

AI solutions could address the different shortcomings of SANS research and incorporate greater automation in the prediction, early diagnosis, and monitoring of SANS. However, no potential AI application dedicated to SANS treatment could be identified.

Infections and Sepsis

On Earth, sepsis, which refers to severe infection with organ failure, is a primary cause of mortality and the most expensive condition treated in hospitals. ^{39 , 42 , 118} Among the medical conditions that threaten astronauts, sepsis and infections are estimated to have the highest negative impact on mission success (ranked 1^st out of 30 conditions). ⁹⁶ The estimated incidence of infection during a 950-d Mars mission with 6 crewmembers (with 4 on the Mars surface) is a staggering 90.49 events. ²² The most likely sources of infections are acute respiratory infection and skin and subcutaneous tissue infection. ⁹

Evidence from spaceflight shows that the risk and severity of infections, both newly acquired and reactivated, are increased due to physiological changes in space. First, the immune system is downregulated by microgravity, radiation, and chronic stress. ¹¹⁵ This is due to significant changes such as lymphoid hypoplasia, ¹⁰⁷ decreased phagocytic activity, ⁹⁴ and decreased T cell activity. ⁷⁵ Second, the virulence of microbial pathogens, which is the ability to manifest and cause severe disease, is increased in microgravity. Microgravity allows aerosolized pathogen-containing particles to persist, increasing their likelihood of inoculating the human body. ⁸⁰ Once inside the host, microgravity-induced phenotypic changes, such as more extensive biofilm structures and enhanced antibiotic resistance, complicates effective pathogen clearance. ^{45 , 80}

On Earth, sepsis has received considerable attention in AI research. Various systems have been developed and tested in almost every aspect of sepsis management, including early diagnosis, phenotyping, and sepsis treatment. InSight, a prediction tool built upon six vital signs, used data from a multicenter cohort of more than 684,000 patients to predict which patients would develop sepsis and whether they would proceed into severe sepsis or septic shock. ⁷³ It achieved an area under the curve of 0.85–0.96 and outperformed traditional sepsis scoring systems. ⁷³ To guide fluid and vasopressor therapy, the cornerstone of sepsis treatment, ⁶⁸ AI algorithms have been developed to provide treatment guidance and retrospective data show that treatment regimens close to the AI recommendations resulted in the lowest patient mortality. ⁶³

A final issue with treating infections, as well as other conditions, is that drugs degrade faster in space. ³² A possible solution would be to generate medicines and antibiotics in situ at a Mars/Moon outpost, a concept that has been proven feasible in principle. ³³

Cardiovascular Deconditioning and Postflight Orthostatic Intolerance

Exposure to weightlessness alters blood volume, blood flow, and pressure distribution, which manifests in structural and functional changes in the cardiovascular system, ⁶ contributing to orthostatic intolerance and decreased exercise capacity. ^{10 , 24 , 50} Postflight orthostatic intolerance (PFOI), the inability to maintain blood pressure while in an upright position after re-exposure to gravity, ¹⁰⁹ is one of the most well-documented symptoms in astronauts ¹² and was described as “the most significant operational risk associated with the cardiovascular system of astronauts.” ²⁴ The etiology of PFOI is complex and includes a hypoadrenergic response to the orthostatic position, alongside plasma volume losses and fluid shifts. ¹⁰⁹ These lead to an excessive fall in cardiac output and/or inadequate vasoconstriction to maintain cerebral blood supply, ¹²³ leading to symptoms of light-headedness, dizziness, presyncope, and syncope. ²⁴ This problem affects about 20–30% of crewmembers who fly short-duration missions and 83% of astronauts following long-duration missions. ¹²⁸ Symptom manifestation could prohibit an astronaut from standing and performing emergency egress from a spacecraft after landing. ²⁴ Identified risk factors for PFOI include: 1) being female; 2) lower norepinephrine response upon standing and subsequently insufficient increase in peripheral vascular resistance; 3) lower diastolic and supine systolic blood pressure; 4) duration of the orthostatic and heat stress; 5) length of the mission; and 6) participation in in-flight exercise countermeasures. ¹⁰⁹

Conventionally, during the prelanding phase, prevention of PFOI is done through oral fluid-loading, oral midodrine (an alpha-adrenergic agonist), and/or subcutaneous octreotide (a somatostatin analog which is currently the preferred option). ¹⁰⁹ The treatment regime is based on individual risk factors and determined by flight surgeons on the ground. To provide objective and more quantifiable guidance, an AI-based risk stratification score could prove useful. This could consider firstly, the vast number of risk factors identified in previous research, and secondly, the success of similar scores developed for related Earth-based diseases. ^{26 , 59 , 121} For instance, Costantino et al. ²⁶ tested artificial neural networks in the risk stratification of patients evaluated in the emergency department for syncope. It was based on ten variables, including sex, age, syncope during exertion, trauma following syncope, and presence of abnormal electrocardiogram, and was effective in predicting the short-term risk of patients with syncope.

After landing, a combination of IV fluids and vasopressors are administered by flight surgeons to treat hypotension, and rehabilitation is carried out in comprehensive medical facilities on Earth. ¹⁰ To guide hemodynamic optimization of crewmembers during re-entry and in the postlanding phase, inspiration can be taken from closed-loop systems developed for the hemodynamic management of surgical patients. ^{92 , 111} These systems require a controller that administers the drugs to maintain hemodynamic parameters (blood pressure, cardiac output) within a target range. Ideally, these closed-loop systems require continuous blood pressure monitoring, which is generally invasive. However, potential noninvasive or minimally invasive solutions exist, e.g., the Finapres system. ³⁶ Convertino ²⁵ also described a wearable finger photo-plethysmographic device with integrated AI, which could collect and perform real-time feature extraction of arterial waveforms for the early detection of central hypovolemia and circulatory collapse. In particular, the author described its potential use in PFOI for guiding the administration of IV fluids before the onset of a syncopal episode during re-entry and landing. These technologies still require future validation in operational missions and partial gravity. Finally, to guide postlanding hemodynamic optimization, the crew could also rely on ultrasound-based echocardiographic assessment, whose interpretation could be automated by an AI. ⁵

Medical Conditions of Interest with No Identified AI Applications

There are many medical conditions for which we could not identify relevant existing or potential AI applications, or that AI would not serve as the best mitigation technique. These include the issues of muscle and bone loss, nutritional aspects of spaceflights, exposure to toxic environments (including fire and pollution of the cabin atmosphere by lunar or Mars dust, irritants, or chemicals), space motion sickness, and decompression sickness during extravehicular activities.

A potential explanation is that conditions such as muscle and bone loss and the nutritional aspects of spaceflight do not necessitate immediate, critical decisions by a medical expert, unlike shock or induction of general anesthesia. ⁷⁷ Mitigation strategies could be planned well before commencing the mission. Prevention, diagnosis, and mitigation for other conditions, such as space motion sickness and decompression sickness during extravehicular activities, are relatively well understood and controlled, ^{14 , 23} even in the absence of dedicated medical expertise. In terms of exposure to toxic environments (including fire and pollution of the cabin atmosphere by lunar or Mars dust, irritants, or chemicals), the main focus of mitigation should be centered on preventative hardware and crew training. ⁷⁷ Taken together, AI technology would most likely not be very beneficial in the conditions listed above and will not be discussed in this paper.