Last year, Google decided to explore the use of large language models (LLMs) for healthcare, resulting in the creation of Med-PaLM, an open-source large language model designed for medical purposes.
The model achieved an 85% score on USMLE MedQA, which is comparable to an expert doctor and surpassed similar AI models such as GPT-4.
Just like Med-PaLM, several LLMs positively impact clinicians, patients, health systems, and the broader health and life sciences ecosystem. As per a Microsoft study, 79% of healthcare organisations reported using AI technology currently.
The use of such models in healthcare is only expected to grow due to the ongoing investments in artificial intelligence and the benefits they provide.
LLMs in Medical Research
Recently, Stanford University Researchers used an LLM to find a potential new heart disease treatment. Using MeshGraphNet, an architecture based on graph neural networks (GNNs), the team created a one-dimensional Reduced Order Model (1D ROM) to simulate blood flow.
MeshGraphnet provides various code optimisations, including data parallelism, model parallelism, gradient checkpointing, cuGraphs, and multi-GPU and multi-node training, all of which are useful for constructing GNNs for cardiovascular simulations.
Llama in Medicine
Researchers at the Yale School of Medicine and the School of Computer and Communication Sciences at the Swiss science and technology institute EPFL used Llama to bring medical know-how into low-resource environments.
One such example is Meditron, a large medical multimodal foundation model suite created using LLMs. Meditron assists with queries on medical diagnosis and management through a natural language interface.
This tool could be particularly beneficial in underserved areas and emergency response scenarios, where access to healthcare professionals may be limited.
According to a preprint in Nature, Meditron has been trained in medical information, including biomedical literature and practice guidelines. It’s also been trained to interpret medical imaging, including X-ray, CT, and MRI scans.
Bolstering Clinical Trials
Quantiphi, an AI-first digital engineering company, uses NVIDIA NIM to develop generative AI solutions for clinical research and development. These solutions, powered by LLMs, are designed to generate new insights and ideas, thereby accelerating the pace of medical advancements and improving patient care.
Likewise, ConcertAI is advancing a broad set of translational and clinical development solutions within its CARA AI platform. The Llama 3 NIM has been incorporated to provide population-scale patient matching for clinical trials, study automation, and research.
Data Research
Mendel AI is developing clinically focused AI solutions to understand the nuances of medical data at scale and provide actionable insights. It has deployed a fine-tuned Llama 3 NIM for its Hypercube copilot, offering a 36% performance improvement.
Mendel is also investigating possible applications for Llama 3 NIM, such as converting natural language into clinical questions and extracting clinical data from patient records.
Advancing Digital Biology
The Techbio pharmaceutical companies and life sciences platform providers use NVIDIA NIM for generative biology, chemistry, and molecular prediction.
This involves using LLMs to generate new biological, chemical, and molecular structures or predictions, thereby accelerating the pace of drug discovery and development.
Transcripta Bio, a company dedicated to drug discovery has a Rosetta Stone to systematically decode the rules by which drugs affect the expression of genes within the human body. Its proprietary AI modelling tool Conductor AI discovers and predicts the effects of new drugs at transcriptome scale.
It also uses Llama 3 to speed up intelligent drug discovery.
BioNeMo is a generative AI platform for drug discovery that simplifies and accelerates the training of models using your own data and scaling the deployment of models for drug discovery applications. BioNeMo offers the quickest path to both AI model development and deployment.
Then there is AtlasAI drug discovery accelerator, powered by the BioNeMo, NeMo and Llama 3 NIM microservices. AtlasAI is being developed by Deloitte.
Medical Knowledge and Medical Core Competencies
One way to enhance the medical reasoning and comprehension of LLMs is through a process called ‘fine-tuning’. This involves providing additional training with questions in the style of medical licensing examinations and example answers selected by clinical experts.
This process can help LLMs to better understand and respond to medical queries, thereby improving their performance in healthcare applications.
Examples of such tools are First Derm, a teledermoscopy application for diagnosing skin conditions, enabling dermatologists to assess and provide guidance remotely, and Pahola, a digital chatbot for guiding alcohol consumption.
Chatdoctor, created using an extensive dataset comprising 100,000 patient-doctor dialogues extracted from a widely utilised online medical consultation platform, could be proficient in comprehending patient inquiries and offering precise advice.
They used the 7B version of the LLaMA model.
