There are a few (not all) areas where we’ll probably see AI activity going forward.
1) Single biomarker monitoring – One of the lowest hanging fruit for AI is monitoring one specific biomarker that has a known relationship to a specific diseases (e.g. heart rate and arrhythmia). It’s an (easier) place to demonstrate value and enter the market quickly.
However, the really interesting part comes when you start seeing areas where the biomarker-diseases interaction are NOT super well understood. For example, in our new AI in diabetes brief we talked about Cardiogram looking at heart rate to potentially provide a screening for prediabetes.
Once you use AI to better monitor one biomarker-disease relationship, it becomes easier to expand into other relationships with the data you’re now collecting and can study. The key is actually getting tools into the hands of patients as quickly as possible.
2) Image-based diagnostics and monitoring – Computers are getting really good at seeing things thanks to breakthroughs in computer vision and more image datasets to train on. Computers can detect things faster and differences that are more imperceptible (e.g. at a microscopic level) compared to humans.
Images of the eye has been a particularly hot area for this kind of research. We looked deep into the eyes in the first part of our new body series. Check out our analysis of the eye space here.
There’s still a lot of open opportunity around image analysis. Think about how many places in healthcare involve a human being watching something and trying to look for subtle changes.
Analyzing MRI images is one example that comes to mind quickly. In our Google in Healthcare report we talked about machine vision tools potentially being used to look at video footage for labs, for example.
New opportunities are opening here thanks to image generating tools scaling down and getting into the hands of more people. The Butterfly Network raised $250M a month ago to do portable ultrasounds. Athelas is aiming to bring blood slide analysis to the home (tbd on FDA approval).
Athelas is also starting with a single biomarker-disease with neutrophils/white blood cells and cancer, but there’s a lot of expansion opportunity in blood monitoring once it has a consistent stream of image data to train on.
3) Time sensitive notifications – There are conditions where time is of the essence (sepsis, stroke, acute kidney injury, a lot within neonatology, etc.). In these conditions, using AI tools to quickly detect an issue (even with low specificity) is useful in preventing expensive and life threatening events.
Tech giants are starting to enter this area like using the Apple Watch for fall detection and Google using cameras to detect face discoloration for stroke. There opportunities for AI to be applied to these other time sensitive issues. For example, AliveCor is studying whether it can detect STEMI (ST-Elevation Myocardial Infarction) and prevent cardiac arrest/muscle damage.
An area I’m still skeptical that we’ll see deployment of AI in the near term is back-office functions, mostly because the data is a hot mess and we’re still in the data structuring + basic statistical analysis/decision tree phase. For expert intelligence clients, we looked at companies working to tackle the administrative costs in healthcare. See them here.
Humana’s offering a look into its tech playbook at TRANSFORM.
Ramu Kannan, SVP of Technology Advancement, will talk about the disruptive forces in healthcare, and Humana’s plans to leverage blockchain, AI and other emerging tech.
Hear more from Ramu, Paul Bleicher of UnitedHealth’s OptumLabs and myself at TRANSFORM. Newsletter readers get a special discount and special access to a breakfast briefing using the codedigitalhealth (save $750).