Google LLC today introduced a cloud-based automation toolkit for healthcare organizations and previewed Med-PaLM 2, a neural network capable of answering medical exam questions.

The company debuted the technologies at its annual The Check Up healthcare event.

AI advances

The first major highlight of Google’s healthcare event was the announcement of Med-PaLM 2, a new internally developed artificial intelligence model. It can take medical questions as input and provide detailed responses in a natural language format. According to Google, the AI is also capable of explaining the rationale behind its replies.

The search giant evaluated Med-PaLM 2’s accuracy by having it answer a series of questions in the style of the U.S. Medical Licensing Examination. The AI achieved a 85% score, according to Google, 18% higher than a previous-generation neural network called Med-PaLM. The company says that Med-PaLM 2’s performance “far surpasses” similar AI models from other companies.

Google’s cloud business plans to make Med-PaLM 2 available to a limited number of customers in the coming weeks. The goal, according to the search giant, is to explore how the model could be applied in the medical field.

Google hopes to “understand how Med-PaLM 2 might be used to facilitate rich, informative discussions, answer complex medical questions, and find insights in complicated and unstructured medical texts,” Google Cloud executives Aashima Gupta and Amy Waldron wrote in a blog post. “They might also explore its utility to help draft short- and long-form responses and summarize documentation and insights from internal data sets and bodies of scientific knowledge.”

Med-PaLM 2 is one of several AI models that Google has built to support the work of medical professionals. It’s collaborating with multiple healthcare organizations to advance its research in this area. Alongside the announcement of Med-PaLM 2 today, the company detailed four new healthcare partnerships.

The first partnership is with an “AI-based organization” led by the healthcare nonprofit Right to Care. The collaboration focuses on making AI-powered tuberculosis screenings widely available in Sub-Saharan Africa. According to Google, its partners have committed to donating 100,000 free screenings.

The three other healthcare AI partnerships are with Kenya-based nonprofit Jacaranda Health, Taiwan-based Chang Gung Memorial Hospital and Mayo Clinic. The former two collaborations focus on using machine learning to interpret ultrasound images. Google’s partnership with Mayo Clinic, in turn, seeks to develop an AI model that can help physicians more quickly plan radiotherapy treatments.

Claims Acceleration Suite

Alongside its new partnerships and Med-PaLM 2 model, Google Cloud announced an offering called the Claims Acceleration Suite. It uses AI to reduce administrative work for healthcare organizations. The offering makes use of several existing Google Cloud services, including its Document AI API for extracting information from documents. 

The Claims Acceleration Suite is designed to speed up two common healthcare administration tasks. The first is insurance claims processing, while the other is health insurance prior authorization. On launch, the offering supports only the latter use case. 

Health insurance prior authorization is the process of evaluating whether a treatment plan is medically necessary. The evaluation involves reviewing medical records and other patient data that is often scattered across multiple documents. According to Google, organizing that data for processing can involve a significant amount of manual work.

The Claims Acceleration Suite is designed to speed up the task. It can automatically turn medical data stored in unstructured files, such as PDF documents, into a structured format that lends itself better to processing. Additionally, the offering provides a search tool that medical professionals can use to browse the collected data.

Image: Google