The News: Researchers from Google report that their latest model Med-PaLM 2, a multimodal biomedical AI, has achieved a remarkable 86.5% accuracy on US Medical License Exam (USMLE)-style questions, representing a substantial 19% improvement over the previous state-of-the-art results with Med-PaLM. In addition, clinicians preferred Med-PaLM 2 reports over radiologists’ reports in up to 40.5% of cases in a direct comparison of 246 retrospective chest X-rays. Read the announcement on the Google Research website.
Google Research Introduces Med-PaLM 2 To Improve Health Outcomes
Analyst Take: Researchers from Google have reported that their latest model Med-PaLM 2, a multimodal biomedical AI, has achieved a remarkable 86.5% accuracy on USMLE-style questions, representing a substantial 19% improvement over their previous state-of-the-art results with Med-PaLM. Med-PaLM 2 was developed focusing on challenges in the medical field, such as early disease detection, medical image analysis, and personalized treatment recommendations. The integration of AI with biomedicine provides the opportunity for healthcare experts to improve patient outcomes, allowing healthcare experts the ability to quickly and accurately analyze extensive data, such as medical images, electronic health records (EHRs), and scientific literature, unlocking greater efficiency and precision.
What is unique about Med-PaLM 2 is its ability to process and analyze multimodal data including medical images, such as X-rays, MRIs, and CT scans, along with clinical data, patient histories, genetic information, and medical literature. This ability enables Med-PaLM 2 to offer more comprehensive and accurate insights into complex medical conditions than a healthcare expert is able to do. To develop Med-PaLM 2, the team of researchers assembled MultiMedBench, which is an extensive multimodal medical dataset, encompassing 14 tasks across text, medical imaging, and genomics. For example, MultiMedBench has more than 1 million examples that are tailored for tasks ranging from question answering, report generation, classification, and other clinically significant objectives. MultiMedBench played a pivotal role in training and assessing the biomedical capabilities of Med-PaLM 2.
Med-PaLM 2 Benefits
- Multimodal Data Analysis: Med-PaLM 2 can process and analyze multimodal data, allowing it to uncover patterns that are difficult or time-consuming for healthcare experts, improving diagnostic accuracy and providing more personalized patient treatment recommendations.
- Early Disease Detection: Med-PaLM 2’s algorithms enable healthcare experts to identify early signs of diseases, providing patients with timely interventions and improved prognosis.
- Precision Medicine: Med-PaLM 2 enables healthcare experts to develop personalized treatment plans tailored to each individual patient, which ensures more effective and targeted therapies.
- Speed and Efficiency: Med-PaLM 2 can analyze large datasets in a fraction of the time it would take healthcare experts, allowing them to make more informed decisions quickly.
Through language-based instructions and prompts, Med-PaLM 2 applies its knowledge on new medical tasks and concepts and demonstrate multimodal reasoning without specific training. For example, Med-PaLM 2 displayed accurate identification and description of tuberculosis in chest X-rays, even though it had never been exposed to images of the disease before. In addition, clinicians preferred Med-PaLM 2 reports over radiologists’ reports in up to 40.5% of cases in a direct comparison of 246 retrospective chest X-rays.
Google will be rolling out Med-PaLM 2 to a select group of Google Cloud customers for limited testing purposes to explore potential use cases and gather feedback while ensuring that the technology will be employed safely, responsibly, and in meaningful ways.
Disclosure: The Futurum Group is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of The Futurum Group as a whole.
Other Insights from The Futurum Group:
Amazon Introduces AWS HealthScribe to Enhance Patient Medical Care
Google Cloud Next: A Deep Dive Into AI and Modern Infrastructure
Google Cloud Unveils Cutting-Edge AI Tools for Accelerating Drug Discovery and Precision Medicine
Author Information
Clint brings over 20 years of market research and consulting experience, focused on emerging technology markets. He was co-founder and CEO of Dash Network, an integrated research and digital media firm focused on the CX market, which was acquired by The Futurum Group in 2022. He previously founded Tractica with a focus on human interaction with technology, including coverage of AI, user interface technologies, advanced computing, and other emerging sectors. Acquired by Informa Group, Clint served as Chief Research Officer for Informa’s research division, Omdia, with management and content strategy responsibility, formed by the combination of Tractica, Ovum, IHS Markit Technology, and Heavy Reading.
Clint was previously the founder and President of Pike Research, a leading market intelligence firm focused on the global clean technology industry, which was acquired by Navigant Consulting where he was Managing Director of the Navigant Research business.
Prior to Pike Research, Clint was Chief Research Officer at ABI Research, a New York-based industry analyst firm concentrating on the impact of emerging technologies on global consumer and business markets.
Clint holds a Master of Business Administration in Telecommunications Management from the University of Dallas and a Bachelor of Arts in History from Washington & Lee University.
