Key Takeaways from The Check-up with Google Health 2024

The Check Up with Google Health was streamed live on March 19, 2024. It highlighted their achievements, current collaborations and ongoing research to use AI as a tool to improve healthcare provision. Ensuring privacy and health equity was at the core of all their demonstrations. Here are some of the key ideas from the talk -

1. “Health moves at the speed of trust.” - Dr. Karen DeSalvo. Building trust in AI as a tool and, building that trust specifically in the AI models used in healthcare workflows is key to universal adoption of AI. That means ensuring that the models are safe, secure, equitable.

2. MedPaLM, a large language model was introduced by Google in 2023. It is designed to provide high-quality answers to medical questions. This year, Google released MedLM, a group of models specifically fine-tuned for incorporation into healthcare workflows. A key goal of MedLM is multi-modality. MedLM for chest x-ray is now available on Google Cloud in experimental preview, which helps interpret and generate reports on chest X-rays.

3. MedLM is being utilized to enhance specific workflows within hospital practice. Google, in collaboration with HCA healthcare, is using MedLM to improve nurse-to-nurse handoffs during shift changes. This process involves the use of AI to summarize information and produce a comprehensive handoff report. The objective is to boost the performance of the care team, not to replace the clinician.

4. Research is ongoing to train the MedLM model and expand its capability to process multi-modal information like images (both 2D and 3D), written text and speech. AMIE or Articulate Medical Intelligence Explorer is another model being trained to ask contextually relevant questions to arrive at a diagnosis, while incorporating caregiver qualities like empathy and respect for individuals.

5. Google in collaboration with Stanford Medicine, released a dataset called SCIN - Skin Condition Image Network. It is an open source dataset with over 10,000 images of skin, nail and hair concerns with a diversity in condition severity, skin tone, age and gender.

6. Google talks about using the HEAL Framework - Health Equity Assessment of Machine Learning Performance in their research. It acts as another layer of evaluation to help identify the areas in which medical AI models need to be trained to reduce bias.

7. The event had a segment where they showcased Google’s collaboration with Apollo Health in India, where they are using AI to evaluate chest-Xrays and screen for the presence of TB. Watch the clip here - https://www.youtube.com/watch?v=CzgMUVPduZA
   

8. Google showcased its efforts to make healthcare information more accessible to people. This includes features like -

   1. Using Google lens to search for information using pictures,

   2. Displaying WHO verified information on knowledge panels,

   3. Using graphics and images on knowledge panels for health questions

   4. Introduction of Aloud for Youtube Health which is an AI powered dubbing tool to allow for easy translation of medical information.

9. And lastly, the event showcased the launch of Fitbit Labs which will allow premium users to have personalized, tailored insights into their health using AI.

I believe that the ideas presented here are just the start. If built and implemented correctly, AI has the potential to streamline, revolutionize, and increase the accessibility of healthcare for everyone.