Google introduced Health AI Developer Foundations (HAI-DEF), a public resource for healthcare developers that provides open templates to help them create healthcare applications, initially focused on dermatology, radiology and pathology .
Open-weight AI models are a type of black-box AI technology that allows developers to apply and refine a model for specific tasks, allowing them to adapt and build upon previous work.
HAI-DEF is publicly available for developers who create and implement AI models. It includes open templates, documentation to help you through the various stages of development, and Colab educational notebooks, which allow developers to write and run code.
The models initially available aim to support the development of medical imaging applications, such as those related to chest x-rays, skin images or digital pathology.
The CXR Foundation model for chest x-rays has been trained on over 800,000 x-rays. It allows users to perform efficient data classification and classify specific conditions, among other tasks.
The Derm Foundation for skin images can be used for effective classification of data, including dermatitis, melanoma or psoriasis, and to understand which part of the body is involved.
Path Foundation for Digital Pathology is an integration model that can be used for applications such as “classifying or identifying tumors, classifying tissues or stain types, and determining image quality . Embeds can also be used for similar image search tasks, to find areas within or across slides that look similar,” the tech giant wrote in a statement. blog post.
Google says it introduced HAI-DEF because diverse datasets including various patient populations, protocols or data acquisition devices must be available for models to include environments different from the data they were trained on. The company says the templates also make it easier for developers to take their ideas from concept to prototype.
“For healthcare to continue to realize its potential, it needs innovation from a diverse set of contributors across a multitude of use cases, interfaces and business models,” the healthcare giant wrote. technology.
HAI-DEF allows developers to download and run the models in their environment locally or via the cloud, use them for research or commercial project applications, and fine-tune them for better performance.
Models are available through Google’s Vertex AI Model Garden and Hugging Face.
“HAI-DEF is just one of the ways we are enabling the broader ecosystem to thrive for health, complementing the Open Health Stack and Population Dynamics Foundation model. We are excited to continue investing in this space, including adding more models to HAIDEF and expanding the reach of our laptops. We look forward to seeing the community build on these resources to realize the potential of AI to transform healthcare and life sciences,” Google wrote.
THE BIGGEST TREND
Earlier this year, during Google Check Up EventThe company announced that it is expanding its MedLM models to include multimodal modalities, starting with MedLM for chest radiography, available in an experimental preview on Google Cloud. The goal of the model was to enable result classification, semantic search and more to improve the efficiency of radiologists’ workflows.
Dr Ivor HornDirector of Health Equity and Product Inclusion at Google, also announced the release of a dermatology-focused dataset called the Skin Condition Image Network (SCIN), which includes the skin tones of a diverse group of people with different levels of pathologies.
In October, Google announced it licenses its AI model to detect diabetic retinopathy to healthcare providers and technology partners in Thailand and India, two countries that the tech giant says have a shortage of eye specialists .
In June, the company announced the creation of Tx-LLM, an LLM for drug discovery and therapeutic development, refined from PaLM-2, the company’s generative AI technology that uses Google LLMs to answer medical questions.
The tech giant’s big medical language model Med-PaLM 2 was released last year and was found to generate more comprehensive answers to medical questions than its original version Med-PaLM.