In a breakthrough for the healthcare industry, researchers have unveiled Meditron, a suite of open-source large language models (LLMs), specifically designed to assist medical professionals.
Jointly developed by researchers from École Polytechnique Fédérale de Lausanne (EPFL) and Yale School of Medicine, and supported by the International Committee of the Red Cross (ICRC), the LLM is built upon the Meta Llama2 platform and trained on “carefully curated, high-quality medical data sources,” Meta said in a blog post.
Meditron has also been supplied and refined with “continual input from clinicians and experts in humanitarian response,” the blog post said. This medical LLM is expected to assist healthcare professionals with “clinical decision-making and diagnosis.”
Closing the gap in medical AI adoption
“Foundation models have become modern-day intellectual and cultural assets,” Yale professor Mary-Anne Hartley, who is co-leading the project, said in the blog. “When applied to the medical domain, they have the potential to provide life-saving advice and guidance. Yet the lowest-resource settings have the most to gain and remain the least represented.”
Meditron tackles this challenge head-on. Built upon Meta’s Llama 2, Meditron is fine-tuned on a massive dataset of curated medical information including clinical guidelines, medical journals, and real-world data from humanitarian organizations like ICRC.
This training ensures the information Meditron provides aligns with evidence-based practices and meets professional standards, the blog explained. “The Meditron suite has the potential to serve crucial needs in a variety of settings, including emergency scenarios requiring fast and accurate medical response and assisting healthcare workers in diagnosing and treating patients in underserved areas.”
“Meditron represents a significant leap forward in democratizing access to powerful AI tools for healthcare,” said Pradeepta Mishra, co-founder and chief architect of data privacy firm Data Safeguard. “Language model trained on general text data can be fine-tuned for specific medical tasks like medical question answering, clinical documentation, or patient diagnosis.”
Early success and open access
According to Meta, Meditron has been downloaded more than 30,000 times since its release and is “filling an important gap in innovation in low-resource medical settings.” However, the researchers have not stopped innovating and it has already been updated with the latest features of Llama 3.
“Following last week’s release of Meta Llama 3, the team fine-tuned the new 8B model within 24 hours to deliver Llama-3[8B]-MeditronV1.0, which outperforms all state-of-the-art open models within its parameter class on standard benchmarks such as MedQA and MedMCQA,” the blog claimed.
The open access, perhaps, is the most significant aspect of Meditron, believed Hartley. The entire suite – data, model weights, and comprehensive documentation – is freely available. Hartley hoped this could “empower innovation in resource-constrained settings to better ensure representation and create equitable access to medical knowledge. Low-resource settings should not be forced to ‘reinvent the wheel’ in order to have their populations and needs represented in this critical technology.”
Moving beyond benchmarks
While Meditron currently leads the pack of open-source LLMs for medicine on standard benchmarks, the researchers at these universities acknowledged these may not reflect the real-world clinical challenges.
To address this issue, the researchers have launched the Meditron MOOVE (Massive Online Open Validation and Evaluation) initiative, inviting healthcare professionals across the globe to evaluate Meditron’s performance in real-world scenarios, particularly in low-resource environments, the blog added.
“That these time-constrained professionals are volunteering their time in our open-source community to independently validate Meditron is a recognition of its value,” Hartley said. “We are in a unique position to take all this feedback and incorporate it in a new model. We hope funders will recognize the social and commercial value of investing in our academic open-source initiative.”
“Ensuring the accuracy, reliability, and explainability of AI tools like Meditron in real-world clinical settings presents several key technical challenges that need to be addressed,” cautioned Mishra.
Copyright © 2024 IDG Communications, Inc.
