How Artificial Intelligence Can Help With Efficiency in Healthcare

By Matt Phillion

An aging population, a shortage of clinicians, and an abundance of data—treating patients grows more and more complicated all the time. Leveraging available and emerging technology to maximize efficiency, however, offers a chance to improve care in innovative ways.

“The population is aging, and more and more people are suffering from cardiac issues. Expertise is expensive, and there is limited access to those experts,” says Jia Li, co-founder of Cardiologs, a medical technology company developing medical-grade artificial intelligence (AI) and cloud technology to improve cardiac diagnoses. “If we want accurate diagnoses and to deliver the proper therapies to the right patient, this shortage of expertise is a problem.”

The shortage affects not just cardiology, but the entire spectrum of care—and it extends beyond places traditionally thought of as having limited healthcare options. From the U.S. and Europe to Africa and Asia, the universal need for more highly skilled physicians and other clinicians places these experts, and their time, at a premium.

There’s also the matter of cost. “[Cardiac] diagnoses are extremely labor-intensive,” says Li. “Before delivering therapy, physicians need to interpret and analyze the data gathered on the patient.” Electrocardiogram (ECG) data, for example, can take up to 90 minutes for a physician to interpret. And with today’s technology, Li notes, we are creating more and more data to be analyzed and interpreted.

“Our healthcare system, our doctors cannot analyze all the data available,” he says. “I don’t think it’s just a problem of there being not enough people with the expertise. While the science and the knowledge is developed, we have a need for more and more specialists in one domain. The problem is that a human, even an expert in their domain, has a limited capacity to digest all the data and information presented, and it requires years of training to reach that expert level.”

Beyond this, specializing in one domain may not be enough—the task of analysis becomes more difficult, even for experts, as the amount of data grows.

Leveraging emerging technology

The answer, Li says, lies in adapting to available technology to improve the efficiency of that analysis. For example, he specifically works with developing machine learning to address ECG processing problems. Machine learning is the study of computer algorithms that improve automatically through experience, and is part of the general concept of AI. These algorithms build a model from “training data” to make decisions or predictions.

But, one might ask, won’t letting AI do part of the analysis mean we’ll need or rely on humans less? Not so, Li explains.

“I don’t think AI is going to take the job from humans,” he says. “Humans don’t have anything to fear at this stage. We are far from what we see in science fiction in terms of AI capability. We’re not moving in a direction in the industry in which AIs can replace human beings.”

The current context, Li adds, is that AI would work alongside humans to deliver better care. “It’s extremely important for us to find the correct interface between the two,” he says. So what would that interface look like?

In terms of communication, the first thing an AI would do is help physicians make their diagnosis. This is already happening in CT scans, MRIs, and ECGs. “This allows the doctor to spend less time on the diagnosis, and focus instead on the therapy and the time they spend with the patient,” says Li.

AI wouldn’t remove the doctor from the diagnosis completely, Li says. Some diagnoses are extremely complicated, but others can be labor-intensive and repetitive—and an AI can shoulder that latter work, freeing up physicians to handle the more complex tasks.

Involving AI in diagnosing patients could also help create more accurate data. Li notes that in ECGs, physicians have developed rules for biomarkers for certain conditions. In the future, AI could help analyze the large amount of data collected to identify biomarkers that are not necessarily visible to the human eye. Experts could then study the results the AI provides, get a better understanding of a biomarker, and better correlate how it is related to the condition involved.

Concerns about AI

While fears about AI taking over human jobs are unfounded at this point, there are some limitations to AI that are worth noting so that the humans using this technology can work through them.

One important topic in this area, Li notes, is bias. “There’s a lot of debate about this,” he says. “Studies have shown that there might be bias introduced either from the data set used, or the way the human developers model their data that could introduce this bias into the machine learning or AI model.”

A typical example of bias outside of healthcare is facial recognition. It’s known that racial bias can show up in this technology, rendering it far less accurate when scanning images of darker-skinned women, for example.

While the origin of bias is not always 100% understood depending on the data the AI is examining, healthcare teams developing AI are aware that it can crop up.

“This is a hot topic, and people are trying to address this concern by making more robust data sets, more robust evaluation,” says Li. “From our end, we also make sure to include variability and large sample sizes in the demographics of our ECG training data, so as to account for potential bias linked to ethnicity.”

Another concern is being able to explain what the machine learning model has learned. The process of machine learning can be difficult to explain, and experts on the healthcare side might be confused by the conclusions the AI comes up with.

“This is why dedicated researchers and experts try to improve transparency and to develop tools to explain what happens inside the machine learning model,” Li says. Cardiologs, for example, tries to build a comprehensive visual interface for physicians to interact with the AI.

Helping physicians connect the dots comes down to understanding how the AI does its job. For example, most of the time AI will report the result of the specific analysis it was asked to undergo. It will say whether a given condition exists or does not exist. It will look for what it’s told to look for.

“The AI will tell you it sees no dog, it sees no cat,” Li says. But a human can ask: What about another animal?

Future benefits

The potential benefits of AI extend beyond helping with efficiency and freeing humans from labor-intensive analysis, Li says; it could also help with the spread of expertise across the industry.

“In the future, if machine learning could provide a deep dive into specific areas, we might need more general practitioners or technologists, cross-domain experts to work with that data,” Li says. “The AI would enhance or augment doctors on a very specific topic.”

For example, a neurologist working with a stroke patient may need to work with the cardiology department to look for atrial fibrillation if they’re unable to read the results of a test themselves. But neurologists can sometimes have trouble synching up with the expertise they need because of limited access to specialists. In this case, an AI could help them analyze the data, speeding up diagnosis and treatment.

Li imagines the industry might end up needing more generalists, pilots who can oversee the big picture provided by AI analysis.

He also talks about AI’s potential in remote monitoring of patients. Healthcare has already seen an increase in remote monitoring over the past 10 years, driven further by the COVID-19 pandemic. The ubiquitous nature of mobile devices helps make recording digital signals remotely far more accessible than before. There are 10 times as many Apple Watch® wearables worldwide than there are Holter recorders for monitoring ECG signals, for example. This has the potential to generate more and more data—the sheer volume of which, Li points out, would call for AI assistance.

“Doctors are already saying they don’t have the capacity for this much data—and that means it may not have as much value,” Li says. “AI will be able to help doctors triage this amount of data.”

At the end of the day, the concept of using AI in healthcare is meant to expand how patients receive care, rather than take away from the value of human expertise.

“The goal of AI in the medical domain is similar to all medical devices. We want to bring benefits to the patients,” Li says. “We want to enhance and help doctors.”

Matt Phillion is a freelance writer covering healthcare, cybersecurity, and more. He can be reached at matthew.phillion@gmail.com.