Google digs deeper into healthcare AI: 5 notes

Google has launched Health AI Developer Foundations, a suite of open-weight AI models designed to simplify and accelerate the development of healthcare applications, particularly in medical imaging.

"AI has the potential to transform healthcare by improving diagnostic accuracy, expanding access to care, and reducing administrative burdens," said Google in its announcement. "With HAI-DEF, we aim to lower the barriers to entry for health AI development and empower a diverse set of contributors to innovate."

Five things to know:

1. The initial release includes tools for radiology, dermatology, and pathology, aiming to democratize AI development and improve access to cutting-edge technology.

2. The new platform offers developers a starting point to build AI applications with minimal additional data and compute requirements. Each model produces embeddings that efficiently represent medical images, allowing for tasks like semantic image search, zero-shot classification, and condition-specific classification.

3. Developers can download and run these models locally or in the cloud, fine-tune them to meet specific needs, and incorporate them into broader healthcare AI architectures. The specialized models for medical imaging include:

CXR Foundation: Pre-trained on over 800,000 chest X-rays, this model leverages advanced architectures and supervised contrastive training to enable efficient classification and text-image interaction.
Derm Foundation: This model aids in tasks like identifying dermatological conditions, determining body part involvement, and assessing image quality. It enables developers to address clinical challenges such as diagnosing melanoma or psoriasis.
Path Foundation: Specializing in hematoxylin and eosin stained pathology slides, this model supports applications like tumor grading, tissue classification, and identifying similar image regions across slides.

4. HAI-DEF builds upon Google's Medical AI Research Foundations repository and complements its Open Health Stack and Population Dynamics Foundation Model, further expanding the tools available for healthcare developers.

5. The open-weight models are a response to feedback from researchers and healthcare institutions who expressed a need for downloadable models that can be deployed within institutional boundaries.

Copyright © 2024 Becker's Healthcare. All Rights Reserved. Privacy Policy. Cookie Policy. Linking and Reprinting Policy.

 

Articles We Think You'll Like

 

Featured Whitepapers

Featured Webinars