For years, our team at Google Research has worked to find solutions to seemingly impossible challenges, like quantum computing and flood forecasting. And we’ve also focused on doing this in healthcare — from helping radiologists to detect breast cancer, to supporting diabetic retinopathy screening — a domain where we believe AI can catalyze the creation of more personalized, accessible and effective solutions.
Collaborating closely with healthcare providers, researchers and industry partners, we’ve published research, created open-source tools, and built AI systems that have the potential to positively impact health outcomes for people globally. Here’s a look at how AI is transforming healthcare, how we are bringing research to reality, and some of our learnings as we build next-generation solutions.
Exploring how generative AI can transform healthcare
Last year, we sought out to build a large language model (LLM) capable of reaching a passing score in the U.S. Medical License Exam-style questions. We built on Google’s LLM technology PaLM to fine-tune it for medicine and created Med-PaLM, which achieved state-of-the-art performance in medical question answering tasks. Our research, however, didn’t stop there. While Med-PaLM and Med-PaLM 2 allowed us to solve a long-standing grand challenge in AI, our goal is to positively impact healthcare and do that in a bold and responsible way.
Through further research and model development, we’ve continued to evaluate generative AI’s promise in medicine, identifying how we can improve the safety, accuracy and equity of these tools. With our latest multimodal explorations in medicine, we showcased ELIXR, a novel approach to grafting LLMs with medical imaging models to create specialized multimodal models. Additionally, we’re having clinicians evaluate Med-PaLM’s performance and sharing our results publicly. Now, we’re working with partners through Google Cloud to explore how these medically fine-tuned models can transform healthcare.
Partnering with healthcare organizations to bring research to reality
From showing that AI can help radiologists identify signs of breast cancer in mammograms, to using AI to contour organs for radiotherapy, our published research has illustrated a future in which AI can help advance the field of medicine. However, without real clinical validation and application, it remains just that — research.
Over the past several years, we’ve learned the importance of partnering with healthcare organizations to validate our research in clinical settings. We’re working with Jacaranda Health in Kenya to improve our fetal ultrasound AI models, with Osaka University to improve our dermatology classifiers, and with a group of healthcare organizations to understand where and how Med-PaLM 2 technology is most helpful. Working with best-in-class research partners help us all understand how AI can support real clinical workflows.
Taking this a step further, we know that partners are key to scaling this work and improving the health of millions globally. In genomics specifically, we’re partnering with PacBio, a developer of genome sequencing instruments, to improve genomic analysis using DeepVariant and DeepConsensus.
Sharing our learnings in building and deploying health solutions
By building AI-powered solutions that integrate into health systems and improve health outcomes, we’re often dealing with novel, complex challenges. Healthcare organizations are burdened by different tech tools, so AI systems need to quickly add value rather than contribute to the tech burden. We’ve learned a lot in our journey toward impact. From understanding the need to build more equitable datasets, to identifying the right way to embed in clinical workflows, our goal is to learn as much as possible and help others understand how to deploy health AI solutions effectively and safely.
Beyond sharing our learnings, it’s important to make our tools publicly available to health researchers and innovators. Through open sourcing technologies like Open Health Stack and CXR Foundation, we’re empowering others to build next-generation digital health solutions that can save time and improve health outcomes.
Through both foundational and applied research in health AI, we hope to help people envision a future in healthcare that is more accessible, accurate, and equitable.