谷歌研究进展：从医疗创新到实际护理场景的应用
内容总结：
谷歌AI医疗新突破：从癌症早筛到个性化健康管理，人工智能重塑医疗未来
在近日举办的“The Check Up”年度活动上，谷歌研究院展示了其在人工智能（AI）医疗领域的一系列前沿进展。这些技术正从实验室加速走向全球临床实践，致力于实现“让每个人享有高质量医疗”的愿景。
AI成为临床医生的“超级协作者”
谷歌重点介绍了其与全球顶尖医疗机构合作取得的成果。其中，与英国帝国理工学院及国家医疗服务体系（NHS）联合开展的研究显示，其AI系统能够帮助医生多发现约25%的传统筛查易漏诊的“间期乳腺癌”，有望在保证安全的前提下减轻放射科医生工作量，让其更专注于患者照护。此外，谷歌与印度、泰国等地眼科机构合作推广的糖尿病视网膜病变AI筛查模型，已提供超百万次筛查，能在两分钟内给出诊断，帮助预防这一致盲性疾病。
个性化健康管理迎来“智能体”时代
谷歌揭示了下一代预防性医疗的形态。通过与Fitbit合作的研究，谷歌发现，模仿完整医疗团队（集数据科学家、领域专家和健康教练于一体）的“个人健康智能体”（PHA），比仅记录步数或卡路里的单一功能应用更能有效支持长期健康。该智能体能整合可穿戴设备数据，通过多模态大模型为用户睡眠、健康及健身提供个性化指导。
赋能全球开发者，构建医疗AI生态
为加速医疗创新普惠，谷歌推出了“健康AI开发者基础模型”（HAI-DEF），免费提供开源模型与工具。其中，医学多模态模型MedGemma已成为全球开发者的重要“启动平台”。例如，印度新德里全印医学科学研究所已将其用于门诊分诊和皮肤病筛查；新加坡卫生部正基于它构建本土化的多模态模型，以普及健康信息服务。
从公共卫生到科学发现，AI应用边界持续拓展
在公共卫生领域，谷歌正利用其地理空间AI模型分析人口行为与环境因素的复杂交互，助力 proactive 的社区健康干预。例如，通过高分辨率数据分析麻疹疫苗接种覆盖率，能精准定位接种不足的社区，为防控提供关键洞察。
在生物医学研究方面，谷歌开发的DeepSomatic基因组分析工具，能更精准地识别与癌症相关的基因突变，已在多种癌症测试中发现了既往尖端工具遗漏的关键变异，为癌症研究、诊断和治疗带来新潜力。
谷歌研究院表示，当前正处于一场深刻的医疗变革之中。通过多模态开源模型和多智能体系统的研究，其目标是让医疗更精准、更个性化，让疾病更易被检测和治疗，并增强公共卫生系统的韧性。公司强调，在推进突破性研究的同时，始终坚持最高的科学严谨标准，并与患者、医疗专业人员及研究人员紧密合作，负责任地将技术从实验室推向临床。
中文翻译：
谷歌研究在“健康检查”峰会：从医疗创新到真实临床场景
2026年3月17日
谷歌研究副总裁 Avinatan Hassidim 与产品副总裁 Katherine Chou
人工智能在帮助数十亿人活得更长久、更健康方面可以发挥关键作用。今天，在“健康检查”峰会上，我们的同事分享了我们将如何进入一个创新新时代，以推动科学和临床研究的普及化。
快速链接
十年来，我们一直致力于推进基础计算机科学研究，以应对现实世界的医疗健康挑战，追求科学和临床发现。尽管当今人工智能研究的发展速度惊人，但我们的责任感始终坚定不移。我们坚持最高的准确性标准，并与全球的医疗服务提供者、科学家、医院、公共卫生官员和学者紧密合作，以确保我们的创新安全且有益。
对我们而言，人工智能在医疗健康领域的未来意味着为每个人提供高质量的医疗服务。以下是我们在今天活动中重点介绍的最新突破概览。
人工智能助力个性化医疗
人工智能使临床医生能够全面评估患者，并深入实现真正的个性化护理。我们与 Fitbit 合作，在全美范围内进行了一项研究，旨在探索下一代由人工智能驱动的预防性护理可能是什么样。我们发现，模拟协作式健康团队的“个人健康助手”，比仅追踪步数或卡路里的单一功能应用，能更有效地支持长期健康。PHA 就像一个由数据科学家、领域专家和健康教练组成的集成团队，提供统一的智能分析和持续支持。我们还发现，通过利用大型多模态模型，我们可以将可穿戴设备的日常数据转化为个性化洞察，为用户提供关于睡眠、健康和健身的指导。
人工智能作为临床医生的协作伙伴
人工智能有潜力提升护理标准，并使临床医生能有更多时间与患者相处。
上周，我们分享了与伦敦帝国理工学院和英国国家医疗服务体系合作进行的两项研究结果，这些结果发表在《自然·癌症》期刊上，揭示了人工智能如何改善乳腺癌检测。我们整理了多样化的全球数据集，并通过专家医生共识开发了高度可靠的基准数据集，使我们的诊断模型能够达到专家级水平。我们的实验性研究人工智能系统识别出了25%先前被漏诊的“间期癌”——这类癌症通常逃过传统筛查，在症状出现后才被发现。当将这一能力整合到现有工作流程中时，该系统展现出安全减轻放射科医生工作负担的潜力，使他们能将更多时间投入到直接的病患护理中。
我们在临床期刊上发表人工智能研究成果，通过同行评审确保透明度、可重复性和稳健性。这些出版物使我们与印度、泰国和澳大利亚的医学研究机构和眼科医院的临床医生建立了全球联系网络。我们共同扩展了针对糖尿病视网膜病变（一种可预防的致盲主要原因）的筛查模型，提供了超过一百万次筛查。通过这些合作，患者最快可在两分钟内获得诊断，这有可能挽救他们的视力。
智能体人工智能的兴起，正在将这项技术从工具转变为医疗服务提供者真正的协作伙伴。一个典型的例子是 AMIE，这是由谷歌研究和 Google DeepMind 开发的研究型多智能体系统。我们最新的研究发现，它能够解读并推理病史、实验室结果和复杂的医学影像。它通过同时分析患者健康的完整图谱，识别出可能被忽视的模式。我们目前正在与贝斯以色列女执事医疗中心在临床研究环境中测试 AMIE，探索它如何帮助减轻患者就诊前实时病史采集的负担，同时标记紧急症状。我们还最近与 Included Health 合作，启动了一项首创的、经机构审查委员会批准的大规模全国性研究，以评估人工智能驱动的远程医疗护理。
人工智能作为医疗健康开发者生态系统的基石
我们正在采取措施，赋能全球社区以扩大社会创新。我们的“健康人工智能开发者基础”计划提供免费的开源权重模型和配套工具，帮助开发者构建支持人工智能的下一代医疗健康应用。
作为该计划一部分的 MedGemma，是一套医学文本和图像解读模型，支持高维3D成像和医学专用语音识别。MedGemma 已从一个理论研究模型，转变为全球医疗服务提供者和研究人员的开发启动平台。
最近，新德里的全印度医学科学研究所使用 MedGemma 为其门诊分诊和皮肤病筛查应用提供支持。在新加坡，卫生部正在对 MedGemma 进行微调，以构建适用于初级保健和专业医疗场景的本地化多模态模型，普及健康信息的获取。今年早些时候，我们与 Kaggle 合作推出了 MedGemma 影响力挑战赛，邀请开发者构建以人为本的人工智能应用原型，帮助我们弥合人工智能研究与临床影响之间的差距。我们收到了超过850份参赛作品，并将于下周公布获奖者。
人工智能作为公共卫生的导航仪
我们对医疗健康的愿景涵盖从单个细胞到整个星球层面。我们现在正利用“谷歌地球人工智能”——我们提供深度地球智能的地理空间模型和数据集集合——进行公共卫生研究。人口动态与流动洞察为人口行为与环境因素之间的复杂相互作用提供了宝贵见解。这类智能有助于将数十年的研究转化为针对社区的、有效的、主动的护理方案。
例如，随着麻疹疫情的抬头，西奈山医疗中心和波士顿儿童医院/哈佛大学的研究人员将我们的数据与调查相结合，生成了针对幼儿麻疹-腮腺炎-风疹疫苗接种覆盖率的“超分辨率”估计，精确到邮政编码级别。这揭示了与近期疫情爆发相吻合的低疫苗接种率聚集区，可协助公共卫生团队开展更积极主动的本地外展工作。
人工智能作为生物医学与科学发现的加速器
我们也看到人工智能在整个科学方法中的应用势头强劲，在生物医学和生命科学研究方面已取得有希望的早期成果。“共同科学家”——谷歌研究、谷歌云人工智能和 Google DeepMind 的合作项目——以及 Gemini Deep Think，正在成为假设生成方面有价值的人工智能协作者。我们在人工智能驱动的专家级经验软件方面的研究更进一步，使我们能够将科学计算过程重新构想为由进化编码智能体运行的一系列并行实验。我们已在单细胞分析、公共卫生和神经科学等多个跨学科挑战上测试了这些系统。
基于我们在基因组学创新方面的历史，我们开发了 DeepSomatic，这是一个旨在更准确识别癌症相关基因突变的基因组数据分析研究工具。在对多种癌症类型进行测试时，DeepSomatic 识别出了先前最先进工具遗漏的关键变异，为我们的合作伙伴改进癌症研究、诊断和治疗提供了潜力。
医疗健康的新时代
我们正处于一场深刻的变革之中。我们在多模态开源权重模型和多智能体系统方面的研究，有潜力使医疗健康服务更加精准和个性化，使疾病更易检测、更易治疗，并使公共卫生生态系统更具韧性。
在我们推进突破性研究的同时，我们继续坚持最高的科学严谨性标准，并负责任地将我们的研究从实验室带到临床环境，与患者、医疗专业人员、公共官员、研究人员和科学家紧密合作。我们相信，如果我们基于证据、透明度和可重复性来构建追求真理的模型和系统，我们将看到改变生活和挽救生命的成果。我们满怀期待，希望充分发挥人工智能的优势，帮助世界各地的每个人活得更长久、更健康。
英文来源：
Google Research at The Check Up: from healthcare innovation to real-world care settings
March 17, 2026
Avinatan Hassidim, VP Research, and Katherine Chou, VP Product, Google Research
AI can be instrumental in helping billions of people live longer, healthier lives. Today at The Check Up, our colleagues shared how we’re entering a new era of innovation to democratize scientific and clinical research.
Quick links
For a decade, we’ve been advancing fundamental computer science research to address real-world healthcare challenges, in pursuit of scientific and clinical discovery. While the pace of AI research today is staggering, our sense of responsibility remains steadfast. We uphold the highest standards of accuracy and collaborate closely with healthcare providers, scientists, hospitals, public health officials, and academics around the world to ensure that our innovations are safe and helpful.
For us, the future of AI in healthcare means high-quality care for everyone. Here’s an overview of our latest breakthroughs highlighted at today’s event.
AI towards personalized healthcare
AI is allowing clinicians to assess patients holistically and go deeper into truly personalized care. In collaboration with Fitbit, we conducted a study across the US to identify what the next generation of preventative care could look like with AI. We found that a Personal Health Agent (PHA) that emulates a collaborative health team supports long-term health more effectively than single-task apps that only track steps or calories. PHA acts as an integrated team, composed of a data scientist, a domain expert and a health coach, offering unified intelligence and ongoing support. We also found that by leveraging large multimodal models, we can turn everyday data from wearables into personalized insights that provide guidance for users about their sleep, health and fitness.
AI as a collaborator for clinicians
AI has the potential to improve the standard of care and enable clinicians to spend more time with their patients.
Last week we shared the results of two studies conducted in collaboration with Imperial College London and the UK’s National Health Service, published in Nature Cancer, revealing how AI can improve breast cancer detection. We curated diverse global datasets and developed highly reliable ground truth datasets through expert doctor consensus, enabling our diagnostic models to achieve expert-level performance. Our experimental research AI system identified 25% of “interval cancers” that were previously missed — cases that typically evade traditional screenings and surface after symptoms appear. When integrating this capability into existing workflows, the system demonstrated potential to safely reduce the workload of radiologists, allowing them to dedicate more time to direct patient care.
We publish our AI results in clinical journals to ensure transparency, reproducibility and robustness through peer review. These publications connected us with a global network of clinicians at medical research institutes and eye hospitals across India, Thailand, and Australia. Together, we scaled our screening model for diabetic retinopathy, a leading cause of blindness that is preventable when detected early, to provide over one million screenings. Through these partnerships, patients receive a diagnosis in as little as two minutes, that could potentially save their sight.
The rise of agentic AI is transforming the technology from a tool into a true collaborator for healthcare providers. A prime example is AMIE, a research multi-agent system developed by Google Research and Google DeepMind. Our latest research has found that it can interpret and reason across medical histories, lab results, and complex medical images. It identifies patterns that might otherwise be overlooked by analyzing the entire map of a patient's health simultaneously. We’re now testing AMIE in clinical research settings with Beth Israel Deaconess Medical Center, exploring how it can help reduce the burden of real-time history-taking before a patient’s visit while flagging urgent symptoms. We also recently partnered with Included Health to launch a first-of-its-kind, IRB-approved national-scale study to evaluate AI-driven telehealth care.
AI as a building block for the healthcare developer ecosystem
We’re taking steps to empower the global community to scale social innovation. Our Health AI Developer Foundations (HAI-DEF) offers free open-weight models and open-source companion tools to help developers build AI-enabled, next generation healthcare applications.
MedGemma, which is part of HAI-DEF, is a set of medical text and image interpretation models that supports high-dimensional 3D imaging and medical-specific speech recognition. MedGemma has shifted from a theoretical research model to a development launch pad for healthcare providers and researchers worldwide.
Recently, All India Institute of Medical Sciences in New Delhi used MedGemma to power applications for outpatient triage and dermatology screening. In Singapore, the Ministry of Health is fine-tuning MedGemma to build a locally tuned multimodal model for primary care and specialty settings, democratizing access to health information. Earlier this year we launched the MedGemma Impact Challenge in collaboration with Kaggle, inviting developers to prototype human-centered AI applications and help us bridge the gap between AI research and clinical impact. We received 850+ submissions and will announce the winners next week.
AI as a navigator for public health
Our vision for healthcare spans from the individual cell all the way to the planetary level. We’re now harnessing Google Earth AI — our collection of geospatial models and datasets, which provide deep planetary intelligence — for public health research. PDFM insights provide valuable insights on complex interactions between population behaviors and environmental factors. This type of intelligence can help turn decades of research into effective, proactive care for communities.
For instance, with the rise of measles outbreaks, researchers at Mount Sinai and Boston Children’s Hospital / Harvard combined our data with surveys to produce "super-resolution" estimates of MMR coverage among young children, down to the ZIP-code level. This revealed clusters of undervaccination that align with recent outbreaks, and could assist public health teams to conduct more proactive local outreach.
AI as an accelerator of biomedical and scientific discovery
We also see incredible momentum in using AI across the scientific method, with promising early results for biomedical and life sciences research. Co-Scientist — a collaboration across Google Research, Cloud AI and Google DeepMind — along with Gemini Deep Think, are becoming valuable AI collaborators for hypothesis generation. Our research in AI-driven expert-level empirical software goes one step further, allowing us to reimagine the process of scientific computing as a series of parallel experiments run by an evolutionary coding agent. We’ve tested these systems on a wide range of multidisciplinary challenges, spanning the fields of single cell analysis, public health and neuroscience.
Building on our history of genomics innovation, we developed DeepSomatic, a genomic data analysis research tool designed to more accurately identify cancer-related genetic mutations. When tested on multiple cancer types, DeepSomatic identified key variants missed by prior state-of-the-art tools, offering the potential for our partners to improve cancer research, diagnosis and treatment.
A new era of healthcare
We are in the midst of a profound transformation. Our research in multimodal open-weight models and multi-agent systems has the potential to make healthcare more accurate and personalized, diseases more detectable and more treatable, and public health ecosystems more resilient.
As we advance our breakthrough research, we continue to adhere to the highest standards of scientific rigor, and to bring our research from the lab to clinical settings responsibly, working closely with patients, medical professionals, public officials, researchers and scientists. We believe if we build models and systems to be truth-seeking, based on evidence, transparency, and reproducibility, we should see both life-changing and life-saving results. We are excited to realize the full benefits of AI to help everyone, everywhere live longer, healthier lives.