Google Research at The Check Up: from healthcare innovation to real-world care settings
Google Research argues that AI is entering a new phase in healthcare: moving beyond isolated tools toward personalized care, clinical collaboration, public-health planning, and scientific discovery. The post highlights research partnerships, open models, and real-world deployments designed to make healthcare more accurate, accessible, and proactive. Google emphasizes that these advances must be developed responsibly through clinical validation, peer review, and collaboration with healthcare institutions. ## AI for Personalized Healthcare - A Fitbit collaboration studied how AI could support preventative care across the United States. - The research found that a Personal Health Agent (PHA) modeled on a collaborative health team could provide more effective long-term support than single-purpose fitness or tracking apps. - The PHA combines: - Data analysis - Medical and domain expertise - Health coaching - Large multimodal models can transform wearable data into personalized guidance about sleep, fitness, and overall health. ## AI as a Clinical Collaborator - Google’s breast cancer research with Imperial College London and the UK’s NHS used diverse datasets and expert-validated ground truth data. - The experimental system identified 25% of “interval cancers”—cancers missed during screening and later detected after symptoms appeared. - Integrated into clinical workflows, the system could reduce radiologists’ workload while maintaining safe detection performance. - Google’s diabetic retinopathy screening model has been deployed through partnerships with medical institutions in India, Thailand, and Australia. - It has supported more than one million screenings. - Patients can receive results in roughly two minutes. - AMIE, a multi-agent medical AI system, can reason across medical histories, laboratory results, and medical images to identify overlooked patterns. - Google is testing AMIE with Beth Israel Deaconess Medical Center to assist with pre-visit history-taking and flag urgent symptoms. - An IRB-approved national study with Included Health will evaluate AI-supported telehealth care. ## Open Models for Healthcare Developers - Google’s Health AI Developer Foundations (HAI-DEF) provides free open-weight models and open-source tools for building healthcare applications. - MedGemma supports: - Medical text and image interpretation - High-dimensional 3D imaging - Medical-specific speech recognition - The All India Institute of Medical Sciences is using MedGemma for outpatient triage and dermatology screening. - Singapore’s Ministry of Health is adapting the model for locally relevant primary- and specialty-care applications. - The MedGemma Impact Challenge received more than 850 submissions aimed at turning AI research into practical, human-centered healthcare tools. ## AI for Public Health - Google Earth AI combines geospatial models and datasets to study connections between environmental conditions, population behavior, and health outcomes. - Researchers at Mount Sinai and Boston Children’s Hospital/Harvard used Google data and surveys to estimate childhood MMR vaccination coverage at ZIP-code resolution. - The resulting “super-resolution” maps identified pockets of under-vaccination that corresponded with recent measles outbreaks. - Such analysis could help public-health officials target outreach and prevention efforts more effectively. ## AI for Biomedical Discovery - Co-Scientist and Gemini Deep Think are being used to generate scientific hypotheses and support research across fields including single-cell analysis, public health, and neuroscience. - Google is also exploring evolutionary coding agents that run scientific-computing experiments in parallel. - DeepSomatic, a genomic analysis tool, is designed to improve the detection of cancer-related genetic mutations across multiple cancer types. Google’s broader recommendation is to treat AI as a validated collaborator and infrastructure layer rather than a replacement for clinicians or researchers. Continued clinical testing, expert oversight, transparent publication, and open developer access will be essential to translating these systems into safe, practical benefits.
Read original(opens in new tab)