One of the biggest problems with AI solutions in healthcare right now is getting quality data that you can use to train your AI models. When you train an AI model using generic data, it is like a child trying to teach another child. For the AI models to be effective, they require the highest quality data including meta data that helps the model to understand the context of the data. This is particularly true in healthcare which has extremely complex data models, superfluous data, and unique terminology.
The good news for healthcare IT companies is that there are companies who can help them clean up and annotate their data so that it can be used effectively in tools like ambient clinical voice and other healthcare AI solutions. One of those companies that provides human-driven data annotation services is called Xelex.ai. I wanted to learn more about the services they offered, so I sat down with Mark Christensen, CEO at Xelex.ai.
In the video interview below, Christensen shares how his company took their years of clinical documentation experience acquired through their WebChartMD platform and pivoted to now provide training data services to health IT organizations. A few years back he realized that AI solutions like ambient clinical voice were going to need not just data, but data that had been effectively annotated by experts in order to effectively train these AI models. You couldn’t just throw more data at the problem. You needed the right kind of data.
I asked Christensen whether they provide health data themselves for training or if they work to clean up and improve other people’s data (Spoiler Alert: They do the latter, but can help with creating synthetic data). He also shared the complexity involved in creating good training data and why it often takes a human to ensure the data can actually be used by the AI. In fact, Xelex.ai will even use clinicians to annotate their data for some use cases that require the expertise of a clinician to really achieve the clinically valid data that’s being requested by health IT companies.
It turns out, there’s a lot more to data annotation than meets the eye. Experts can check for factual correctness, but they are also needed for the more subtle and nuanced tasks like accurately interpreting ambiguous content, creating natural-sounding clinical summaries, and appropriately grouping content that may have been shared at random times throughout an office visit. Plus, Christensen shared that not everyone is good at annotating data. In fact, one of the keys to good annotation of health data is ensuring the right oversight of the process, the technical tools involved in the annotation, and having metrics about each specific annotator so you can train them on any errors they’re making. It isn’t enough to just have metrics for the overall project. Christensen also mentioned how important it is for their process to be nimble so that it can change and adapt to the clients often changing requirements.
If you’re interested in the process of creating high quality healthcare training data and the unique approach that Xelex.ai takes to create this data, you’ll enjoy this interview.
Learn more about Xelex.AI: https://xelex.ai/
Learn more about WebChartMD: https://www.webchartmd.org/
Listen and subscribe to the Healthcare IT Today Interviews Podcast to hear all the latest insights from experts in healthcare IT.
And for an exclusive look at our top stories, subscribe to our newsletter and YouTube.
Tell us what you think. Contact us here or on Twitter at @hcitoday. And if you’re interested in advertising with us, check out our various advertising packages and request our Media Kit.
WebChart MD is a proud sponsor of Healthcare Scene.
