NASA & Google build AI medical assistant for Mars-bound astronauts
As humanity ventures deeper into space, with aspirations of missions to the Moon and Mars, the challenge of maintaining astronaut health becomes increasingly complex. Unlike the International Space Station, where crews benefit from real-time communication with Earth, regular resupply missions for medicines, and the option of a relatively swift return, future long-duration voyages will demand an unprecedented level of medical autonomy. This pressing need is driving NASA to develop “Earth-independent” healthcare solutions for its crews, a critical step towards enabling sustained human presence far beyond our planet.
One pioneering initiative in this endeavor is the Crew Medical Officer Digital Assistant (CMO-DA), a proof-of-concept AI medical assistant being co-developed by NASA and Google. Designed to empower astronauts to diagnose and treat health issues when direct communication with Earth is unavailable or a medical doctor is not present, CMO-DA represents a significant leap forward in on-orbit medical care. This advanced tool leverages a multimodal interface, integrating speech, text, and images to facilitate comprehensive medical assessments, and operates within Google Cloud’s Vertex AI environment, a robust platform for artificial intelligence development.
The collaboration is structured under a fixed-price agreement with Google’s Public Sector unit, which encompasses the necessary cloud services, infrastructure for application development, and the crucial training of the AI models. While Google provides the underlying Vertex AI platform and access to various models, NASA retains ownership of the application’s source code and has played an active role in fine-tuning the AI models to meet the specific demands of space medicine. David Cruley, a customer engineer at Google’s Public Sector business unit, highlighted that this collaborative approach ensures the tool is tailored precisely to NASA’s unique requirements.
To rigorously test CMO-DA’s capabilities, the system was put through three distinct medical scenarios: an ankle injury, flank pain, and ear pain. A panel of three physicians, including an astronaut, meticulously evaluated the assistant’s performance across several key metrics, including initial patient assessment, history-taking, clinical reasoning, and the proposed treatment plan. The results demonstrated a high degree of diagnostic accuracy: CMO-DA achieved an impressive 88% likelihood of correctness for the ankle injury evaluation and treatment plan, 80% for ear pain, and 74% for flank pain, underscoring its potential as a reliable medical aid.
The development roadmap for CMO-DA is intentionally incremental. NASA scientists plan to integrate additional data sources, such as inputs from medical devices, to further enhance the system’s diagnostic capabilities. A critical future objective is to train the model to be “situationally aware,” meaning it will be attuned to unique space medicine conditions, particularly the physiological effects of microgravity on the human body.
While the primary focus remains on space applications, the implications of CMO-DA could extend beyond the cosmos. Cruley hinted at the possibility of pursuing regulatory clearance for Earth-based applications if the model proves its efficacy in orbit, suggesting that such a medical assistant could eventually find its way into terrestrial doctor’s offices. The lessons learned from developing and deploying this tool for astronaut health, he noted, hold significant potential for broader applicability in other areas of healthcare, promising advancements that could benefit patients far beyond the confines of a spacecraft.