Google Shares Health AI, Large Language Model Research Updates
Google announced research updates on its medical large language model and recent partnerships to apply artificial intelligence to cancer, tuberculosis, and maternal health.
Source: Getty Images
By - At Google’s annual health event, The Check Up, earlier this week, the tech company provided updates on its health artificial intelligence (AI) research, including improvements in its medical large language model (LLM) and partnerships to leverage AI for cancer treatments, ultrasounds for maternal health and breast cancer detection, and tuberculosis (TB) screenings.
The biggest news centers on Google’s medical LLM, known as Med-PaLM, as the hype around ChatGPT, and its performance on questions from the United States Medical Licensing Examination (USMLE), continues to grow.
LLMs, at their core, are tools designed to perform a wide range of natural language processing tasks, such as summarizing, predicting, and generating text. Google introduced its LLM, Pathways Language Model (PaLM), last year. Since then, the company has developed Med-PaLM, a version of PaLM tuned for the medical domain, and its updated iteration, Med-PaLM 2.
The original Med-PaLM achieved a passing score on USMLE-style multiple choice and open-ended questions, in addition to providing a rationale for its responses. Med-PaLM 2 has surpassed its predecessor, recently scoring 85 percent and performing at an 'expert' doctor level, the announcement states.
The announcement also notes that the tool has outperformed similar AI models, but Google states that more improvements to Med-PaLM 2 are needed before it can be applied in real-world settings. The model was reportedly tested against 14 criteria — including scientific factuality, precision, medical consensus, reasoning, bias, and harm — and subsequently evaluated by experts, both clinicians and non-clinicians, from various fields around the world.
However, these evaluations highlighted gaps in the model’s ability to effectively and appropriately answer medical questions, which Google aims to address through further research and testing.
Google also announced a formal agreement with Mayo Clinic to improve cancer radiotherapy through AI, which builds on previous work the two have undertaken in recent years.
According to the announcement, radiotherapy is a common but time-intensive cancer treatment because of the 'contouring' process, which requires clinicians to draw lines on CT scans to help separate cancer from nearby areas of healthy tissue that can be damaged during treatment.
The partnership is investigating whether AI can help streamline this process. The two organizations will publish research on their AI-based radiotherapy model soon, the announcement states.
Similarly, Google shared that it would partner with Jacaranda Health, a Kenya-based nonprofit focused on improving health outcomes for mothers and babies in government hospitals, to explore the potential role of AI-assisted ultrasound to improve image interpretation and enhance maternal and infant outcomes.
The company has also launched a partnership with Taiwan-based Chang Gung Memorial Hospital (CGMH) to conduct exploratory research into using ultrasound for breast cancer detection. Mammograms are typically used for breast cancer screening, but high costs and a lack of screening programs make mammography inaccessible to many in CGMH’s patient population.
To address this, the partnership is investigating the utility of an AI-based ultrasound approach for early breast cancer detection.
Finally, Google will be working with Right to Care, a not-for-profit with significant experience in tuberculosis care within Africa, to advance the accessibility of AI-powered, chest X-ray-based TB screenings in Sub-Saharan Africa.
These efforts are part of a larger undertaking by Google to promote health AI and other digital health solutions.
In November, Google Cloud, in collaboration with Hackensack Meridian Health, Lifepoint Health, and other organizations, launched three new Healthcare Data Engine (HDE) accelerators to improve patient care and health system efficiency.
The accelerators are designed to improve interoperability and data unification across health systems and help support digital transformation. They are focused on leveraging social determinants of health (SDOH) and advancing health equity, enhancing operations to bolster patient experience and flow, and advancing value-based care and interoperability to improve the quality of care.
