Google announces funding for AI-enabled digital health projects

Google announced it is funding 15 AI-powered projects, including digital health initiatives to improve provider experience and patient access to care, via its commitment to advancing the United Nations Sustainable Development Goals. 

Each project received $3 million in technical assistance, cash support and Google Cloud credits. A handful of projects got Google.org Fellowships, where a team of Google employees works with an organization pro bono full-time for up to six months. 

Of the 15 AI projects funded, the following eight digital health endeavors were awarded funding:

RAD-AID provides low-source hospitals with an AI-enabled platform that helps triage patients, primarily regarding respiratory disease and breast cancer. The platform also helps interpret X-rays and scans and provide test results. 

Wuqu’ Kawoq and safe+natal are collaborating to develop a machine learning-enabled toolkit to help midwives in rural areas of Guatemala detect neonatal complications in real-time, such as poor fetal growth and fetal stress during delivery. The toolkit will consist of an ultrasound and blood pressure monitor connected to one’s smartphone. 

MATCH (Music Attuned Technology – Care via eHealth) is a project built out of the University of Melbourne and CSIRO that combines music and wearable sensor technology to decrease agitation in patients with dementia. Google’s grant will help the team develop the sensor technology and AI-enabled adaptive music system.  

Makerere AI Lab will develop a 3D printed adapter that processes images using AI and is compatible with a phone or microscope. The goal is to help providers in Uganda diagnose patients with illnesses, such as tuberculosis, malaria and cancer, in low- and middle-income countries where lab technicians are scarce.   

IDinsight with Reach Digital Health developed a natural language-enabled question-answering service for expectant mothers in South Africa, which provides answers to inquiries and vital health information.  

Causal Foundry seeks to develop a smartphone-based tool that utilizes machine learning to help community health providers in Sub-Saharan Africa manage patient information and behavior changes related to pregnancy and childbirth.

Jacaranda Health delivers an SMS-based digital health platform that answers questions for expectant mothers in Sub-Saharan Africa. The platform provides behavioral nudges and includes a natural language-powered helpdesk that helps triage patients and connects them to human agents. The funding will be used to refine the machine learning model within the platform.  

The University of Surrey and Signapse will use generative AI to translate online and offline text in real time for deaf people in the U.S. and U.K. and provide photorealistic videos in sign language, permitting more accessible access to healthcare and other information.  

THE LARGER TREND

Google has its own machine learning technology, dubbed Med-PaLM 2, aimed at improving healthcare information access. Med-PaLM 2 utilizes the tech company’s large language model to answer medical questions. 

In March, Med-PaLM 2 was tested on U.S. Medical Licensing Examination-style questions and performed at an “expert” test-taker level with 85%+ accuracy. It also received a passing score on the MedMCQA dataset, a multiple-choice dataset designed to address real-world medical entrance exam questions. 

One month later, Google announced it would make Med-PaLM 2 available to select Google Cloud customers to explore use cases, share feedback and for limited testing. 

The company also announced a new AI-enabled Claims Acceleration Suite, created to help with the process of prior authorization and claims processing of health insurance. The Suite converts unstructured data (datasets not organized in a pre-defined manner) into structured data (datasets highly organized and easily decipherable).

In July, a study performed by Google researchers and published in Nature revealed that Med-PaLM provided long-form answers aligned with scientific consensus on 92.6% of questions submitted, which aligns with clinician-generated answers at 92.9%.