title:: Generative AI for Healthcare
publisher:: Amazon Web Services
people:: Razvan Ionasec, Christoph Russ, Regina Hackenberg, James Wiggins, Ujjwal Ratan, Khan Siddiqui, Jory Tremblay, Prabhu Arumugam, Wieland Sommer, Jeroen van der Laak, Mahesh Pancholi, Jonathan Larbey, Ruben Amarasingham, Mattia Capulli, Andrea Riposati, Alberto Rizzoli
organization:: Amazon Web Services (AWS), RadboudUMC, Aiosyn, Hoppr, Genomics England, Smart Reporting, UK Biobank, T-Pro, Pieces Technologies, Dante Genomics, V7, University Hospital Essen, NVIDIA, Heidelberg University
domain:: Healthcare, Artificial Intelligence, Machine Learning, Natural Language Processing, Computer Vision, Genomics, Pathology, Radiology
link:: https://pages.awscloud.com/rs/112-TZM-766/images/AWS-GenAI-for-HCLS-Whitepaper_062024.pdf?version=0
Summary
This white paper discusses the emerging field of generative AI and its applications in healthcare. It covers the history of AI/ML, the new era of generative AI enabled by foundation models (FMs) and large language models (LLMs), and the key industry players building FMs and applications. Use cases explored include clinical documentation, data exploration, patient interfaces, research, and multi-modal models spanning imaging, pathology, genomics, and more.
The paper highlights AWS’s infrastructure, tools and services for training/deploying FMs and generative AI applications responsibly. It features insights from leaders at AWS, healthcare organizations, startups and academia working on generative AI for areas like radiology, pathology, genomics, clinical workflows and documentation.
Key topics include the technology stack (hardware, platforms, APIs), emerging models and innovators, business impact, guidance for getting started, architectural patterns, open source, future research directions, and responsible AI considerations like accuracy, bias, privacy, and transparency.
Data Points
- Generative AI represents a paradigm shift, with emerging FMs/LLMs surpassing human performance across cognitive tasks
- Estimated to unlock $60-110B annual value in healthcare through productivity/innovation
- AWS offers custom AI chips (Inferentia, Trainium), SageMaker tools, and infrastructure at scale for training/deploying FMs
- Bedrock service provides access to commercial and open source FMs like LLaMA, with secure fine-tuning
- Interconnected with AWS data/analytics services for healthcare workloads (HealthLake, HealthImaging, HealthOmics)
- Other key offerings: Q developer tools, HealthScribe clinical documentation, generative BI
- Leaders innovating in radiology FMs (Hoppr), pathology (RadboudUMC, Aiosyn), multi-modal models (Genomics England), healthcare data (UK Biobank)
- Commercial builders of apps: Smart Reporting (structured reporting), T-Pro (clinical documentation), Pieces Technologies (clinical workflows)
- Startups working on genomics (Dante Genomics), multi-modal data (V7)
- Key people: Razvan Ionasec, Khan Siddiqui, Prabhu Arumugam, Jeroen van der Laak, Wieland Sommer, Jonathan Larbey, Ruben Amarasingham, Mattia Capulli, Andrea Riposati, Alberto Rizzoli
- Future areas: Performance/cost optimization, multi-modal reasoning, data-centric FMs
- Responsible AI priorities: Fairness, explainability, privacy, robustness, transparency