Using AI to help San Antonio radiologists reduce X-ray reading errors

Dr. Rajeev Suri, seated in front of a desktop computer, searched for signs of lung cancer on a digital X-ray emblazoned with red, orange and yellow clusters.

The colors weren’t highlighting cancerous masses, however. They were marking where on the image Suri was looking, as gauged by artificial intelligence software. It uses infrared light and sensors to monitor his eye movements and pupil dilation.

Suri, 54, a professor of radiology at UT Health San Antonio’s Long School of Medicine, is participating in a study launched this month to collect data from 60 radiologists as they read X-rays at University Hospital.

By feeding large amounts of data into the AI system, researchers hope to train the software to mimic human reading patterns to provide a machine version of a second opinion.

The radiologists at the testing site — backed by engineers at the Southwest Research Institute and a team from the University of Texas at San Antonio’s engineering and psychology departments — are laying the groundwork for the growing use of AI in San Antonio’s biotechnology industry.

“San Antonio has become a hub for biotech, and AI is the new kid on the block,” said Suri, who’s also chief of staff of University Hospital’s radiology department. “It makes sense for this to align.”

Among computer scientists and engineers, the term AI broadly refers to software that can adapt and learn how to perform tasks like humans. AI is what you rely on when you’re using the map app on your phone to find the fastest route to your destination. Social media platforms, such as Facebook and Twitter, use it to fill your feeds with posts and news links you’re likely to be interested in.

Douglas Brooks, manager at SwRI’s Applied Sensing Department, focuses on the most popular AI subfield — machine learning, which allows computer algorithms to take in data and train itself to find patterns. It fuels Netflix movie recommendations, language translation apps, Facebook Messenger and online chatbots.

Dr. Rajeev Suri, interim chair of UT Health San Antonio’s radiology department, participates in an eye-tracking study at University Hospital in the South Texas Medical Center, Wednesday afternoon, April 13, 2022. The study uses an artificial intelligence software to collect eye-tracking data from 60 radiologists, including professional and resident doctors, as they look over hundreds of chest X-rays.

Sam Owens, Staff Photographer / San Antonio Express-News

“Deep learning” is a more complex form of machine learning, one that relies less on humans to improve its performance. Algorithms use artificial neural networks that mirror human brains to learn connections in large data sets. They can be trained to identify faces in photographs, predict cancer risk from mammograms and teach autonomous vehicles to avoid crashing into people crossing a street.

“In machine learning, a human plays a major role in creating algorithms and updating it and making mathematical modifications so the machine can do its job,” said Brooks, 40. “In deep learning, we give it data and labels, and it figures out the thing on its own.”

One of Brooks’ SwRI colleagues, principal engineer David Chambers, helped develop the deep-learning algorithm for the radiology study.

“SwRI can use the data to train a machine-learning model that performs some of the visual tasks of a radiologist,” said Chambers, 35. “This capability could be used to train and improve other physicians by giving them a second opinion that factors in how the radiologist scanned the image.”

San Antonio is an ideal AI testing ground, the engineers say, because it’s home to a large health care industry, with a web of clinics and civilian and military hospitals with connections throughout Texas.

“AI in medical is a huge deal here,” Brooks said. “Imagine more accurately diagnosing patients and being able to save lives.”

UT Health San Antonio ’s Dr. Kal Clark talks about how AI software collects eye-tracking data.

UT Health San Antonio ’s Dr. Kal Clark talks about how AI software collects eye-tracking data.

Sam Owens, Staff Photographer / San Antonio Express-News

Money for AI research

Researchers are also finding financial support in San Antonio.

The San Antonio Medical Foundation, the nonprofit that helped bring a University of Texas medical school to San Antonio and launch the South Texas Medical Center, donated $187,000 for the radiology study, which runs through August.

Researchers are looking to reduce the number of false-positive readings. Every time a problem is found on an image, there’s a 33 percent chance the radiologist made an error. Most mistakes are minor and can be corrected quickly. But in some cases, mistakes can mean missing early indications of cancer or other diseases.

Dr. Kal Clark, vice chairman of radiology informatics at UT Health, who is overseeing the X-ray readings for the study, said radiologists and AI can work better together than either software or humans alone.

“To err is human,” said Clark, 36. “We’re better at interacting with people than we are at being a detector. We’re going to use AI to augment our abilities to detect so we could do more of the art of medicine.”

Jim Reed, president of the medical foundation, also believes the study will demonstrate that radiologists can partner with AI systems to avoid errors.

“The artificial intelligence would improve the radiologists identifying cancers by 16 to 20 percent, based on the database they would gain through the artificial intelligence,” he said, citing the proposal for the study.

The medical foundation has previously backed AI research in biotech.

Last year, it gave a $200,000 grant to SwRI, UT Health and University Health to develop AI algorithms for cancer detection.

Also in 2021, the nonprofit awarded $200,000 to a team from SwRI, UT Health, UTSA and Texas Biomedical Research Institute to develop AI software to identify molecules that could block the coronavirus from infiltrating human cells.

Reed said his organization plans to fund more AI studies.

“AI is a growing thing,” he said. “I think we’ll see more grant applications with the artificial intelligence, particularly as everybody is moving in that direction.”

Dr. Rajeev Suri, interim chair of UT Health San Antonio’s radiology department, participates in an eye-tracking study as a radiology resident watches at University Hospital in the South Texas Medical Center, Wednesday afternoon, April 13, 2022. The study uses an artificial intelligence software to collect eye-tracking data from 60 radiologists, including professional and resident doctors, as they look over hundreds of chest X-rays.

Dr. Rajeev Suri, interim chair of UT Health San Antonio’s radiology department, participates in an eye-tracking study as a radiology resident watches at University Hospital in the South Texas Medical Center, Wednesday afternoon, April 13, 2022. The study uses an artificial intelligence software to collect eye-tracking data from 60 radiologists, including professional and resident doctors, as they look over hundreds of chest X-rays.

Sam Owens, Staff Photographer / San Antonio Express-News

Fewer radiologists?

Radiology is one of the hotbeds for AI research. And as deep-learning algorithms improve, demand for radiologists could decline.

In 2016, Geoffrey Hinton, a computer science professor at the University of Toronto known as the “Godfather of AI”— told an AI conference that “people should stop training radiologists” because they would become obsolete within five years as a result of advances in deep learning.

SwRI’s Chambers rejects dire predictions like Hinton’s. “I think he’s wrong,” he said. “Certainly, the space is changing — but slowly.”

At University Hospital, Suri said he wasn’t afraid of radiologists losing their jobs to AI.

“No, we’re helping develop it,” he said, laughing.

Still, future reductions in hiring are likely.

“I would be scared if I wasn’t embracing technology,” UT Health’s Clark said. “If AI can help us, we only need to hire eight radiologists instead of 10 radiologists. Now, the patients are taken care of just the same, maybe even better, with eight radiologists with AI than 10 radiologists without AI.”

Researchers say AI could take over time-consuming tasks, such as data analysis, to support a radiologist’s findings. But radiologists will remain important because AI can’t yet be trusted to make a diagnosis alone.

“AI can be wrong, too,” Chambers said. “That certainly is scary. That keeps me up.”

[email protected]

https://www.expressnews.com/business/article/AI-Artificial-intelligence-radiologists-Xray-17089163.php

Related Posts