Real World AI Use Cases: Digital Healthcare

No matter how many theoretical or high-level discussions we have about AI implementation, nothing beats the illustrative power of use cases. They’re great for getting ideas about what’s possible with your own customer work, and it never hurts to check in on the latest trends in a particular space.

?

So, when I had the opportunity to talk to software development and data analytics expert Richard Kedziora , I wanted to pick his brain about AI’s impact on digital healthcare. He’s the founder of Estenda Solutions, a custom software data analytics turned AI company that helps other entities create and bring products to market.

?

Curious about what AI is bringing to the digital healthcare table? Here’s what Kedziora shared on his bonus episode of The Business of Tech .

?

Estenda’s expertise

?

Digital health is now a booming industry, but Kedziora worked in it long before it was even called digital health. He co-founded Estenda Solutions about 20 years ago to help everyone from Fortune 50s to large medical device manufacturers to startups develop the solutions and products they’re taking to market. The team’s focus has always been improving health and wellness through innovation and data analytics, which now includes AI.

?

For example, Kedziora has been involved with the Indian Health Services department for decades through a partnership with the Joslin Diabetes Center’s diabetic retinopathy surveillance program. Estenda’s work involves over 100 cameras deployed throughout the country that take images of the backs of people’s eyes for the detection of diabetic retinopathy. In the startup realm, he’s helped companies fuse data with wearable information to create digital weight loss solutions.

?

And, as many experts who stop by the show can attest, AI has been around for quite a while now:

?

“It's not something that just jumped on the scene. We've been involved in various AI systems over the years, developing medication recommendation systems that were expert system-based back in the early 2000s, and now today, implementing Gen AI types of solutions to address the challenges in healthcare,” he said.

?

Kedziora’s take on the evolution of AI

?

Still, there's something about this moment in time that makes AI feel more cutting-edge than ever. What does Kedziora think is different about AI now versus AI-related work from 20 years ago?

?

For one, he believes that the average non-technical person now being able to take advantage of AI – and even drive impact with it – is a major difference. Take the medication recommendation system he mentioned; it required a large amount of data, had an extensive expert-level system behind the scenes, and had a Y2K-style interface. Today, to create that system, APIs and generative AI could craft the same rules and present it in a far more user-friendly fashion.

?

Along those lines, the shift away from having to learn SQL is another major difference he’s observed:

?

“It's relatively easy to learn in the grand scheme of things, but it's something you have to learn and understand and how to use it effectively. Where now, we're developing systems that put the gen AI on top of that data. So the average person doesn't have to learn or understand tools and technology and can start querying and looking at that data,” he said.

?

This development has changed the game so much that now, according to Kedziora, people might not even have to learn Excel to query patient population data, for example.

?

That reminded me of my personal experience dealing with the woes of report building. I asked Kedziora if AI-powered solutions could replace that type of work, and he agreed: natural language prompts could absolutely replace current report space norms. Imagine typing prompts like ‘find me all the patients that have diabetes or find all the patients that have congestive heart disease’ — that’s where he believes AI could continue making waves in digital health.

?

?

Prerequisites for taking advantage of AI

?

But how can we lead clients to that point? I asked Kedziora for his take on what’s necessary to help customers prepare their data for AI solutions.

?

If you’ve tracked what previous AI experts had to say on this front, you already know his answer: start with the data strategy.

?

“You can't just have a mass of information that is then understood by an AI. You need to provide that structure or methodology to be able to put that data in a format that is usable by the AI. Whether it is unstructured data out of a PDF document or your electronic medical record system, it has to have an understanding of what that information is and what is available,” he said.

?

Of course, safety also warrants heavy consideration. He believes that all of the cybersecurity privacy guidelines you already use very much apply to AI-related data. For health tech folks, that includes regulations like HIPAA.

?

Later, when I asked Kedziora what AI customers ask about the most, he brought us back to this prerequisite stage. Understanding data is the #1 thing people need to get started with AI-powered solutions, so there’s certainly an opportunity here.

?

Frameworks for accelerating AI adoption

?

Next comes implementation. How does Kedziora recommend safely accelerating this process for customers?

?

His answer is simple: checklists. More specifically, checklists inspired by standards like ISO, OWASP, and NIST.

?

He suggests thinking of this process as you would a zero-trust architecture – as you get logged into the system, you have zero access to anything until you slowly apply each principle to it. So, think of these projects as needing ground-up development, and if you can check one box today, remember to apply it at a later stage.

?

As you go, don’t forget about the human training component.

?

“Technology is important, but it's very much about people and processes as well. When you are implementing a cybersecurity system or protecting your data, looking at privacy, you can't forget the people and the processes in place to control that data,” he said.

?

AI’s role in Estenda’s development process

?

I was also curious how Kedziora uses AI in his own development process, outside of hands-on customer work. What practical applications is he finding the most useful?

?

Big picture, he believes AI will soon be unavoidable as a developer. Day to day, he loves using natural language prompts to speed up his usual work. For example, instead of hand-crafting a demo solution, he used ChatGPT to get Python up and running on his system and asked it to help figure out a new API he found.

?

“It's not going to get it right all of the time. But it's going to give you that jumpstart to understanding it. And then, you have to understand the framework. What else is going on within your production application to make sure it makes sense?” he said.

?

From an operational and higher level product perspective, he also uses natural language prompts to ask questions like, what have I forgotten (from his aforementioned checklist)? What else do I need to consider to be able to help jumpstart those thought processes? When he conducts user interviews, he might also ask ChatGPT for help coming up with questions.

?

While he could easily do all these tasks himself, to Kedziora, it’s simply about efficiency.

?

?

?

What use cases related to AI implementation do you hope to hear next? As always, my inbox is open for questions, stories, insights, or whatever else is on your mind.