AI use at UConn Health is flagging abnormalities in patients. It’s helping doctors save lives

Hartford Courant – Thursday, August 14, 2025
By Livi Stanford

Artificial intelligence reviewing the CT scan of a UConn Health patient in her 20s battling pneumonia identified a small abnormal area, resulting in an uncommon early cancer diagnosis.

The cross country and track and field athlete said the early diagnosis allowed her to continue to be active in her sport.

Dr. Omar Ibrahim, service chief for the Neag Cancer Center and director of interventional pulmonary at UConn Health, said typically no one is looking for cancer in patients in their 20s and 30s and that the new software using AI can help catch suspicious findings and diagnose patients early.

“It has a huge ability to catch so much more in a short amount of time and filter information for us at a capacity that would require multiple people to do,” he said. “Every few days we are able to help a dozen or so patients that have a risk finding that needs to be investigated as opposed to not knowing that it even existed because it was not picked up.”

UConn Health officials said the use of AI in radiology and pulmonary medicine is helping doctors to triage patients that are in need, highlighting the cases that are most urgent. But doctors said the technology does not replace radiologists and doctors who interpret the scans and provide the best care plan for patients. Overall, doctors said the advancement in AI in medicine is an important tool in improving patient care.

Even so, they acknowledge the challenges that arise from securing patient data using AI. There is also concern from other health care workers about AI replacing their jobs, something doctors are skeptical will happen. Instead, they believe the responsibilities in the health care system would shift for workers tackling high level tasks rather than data collection.

The National Institutes of Health reported that “AI is aimed to complement, rather than replace, doctors and health care providers.”

Detecting abnormalities

In the last few years, UConn Health began incorporating AI into medicine.

UConn Health incorporates 12 AI modules looking for potential life-threatening findings on radiology examinations.

“These AI modules flag imaging studies the radiologists should read emergently due to suspected acute findings,” said Dr. Michael Baldwin, UConn Health attending radiologist.

For example, Baldwin said an AI module looking at intracranial bleeds, which refers to bleeding in the brain, is over 95% sensitive and over 94% specific.

“If you have a family member arriving in the ER with symptoms of chest pain, they want to be treated as soon as possible,” he said. “It is nice to know that there is a system in the background that is flagging patients that are critically ill. These modules are very sensitive and they look for any potential abnormalities. It is not always a positive finding. Any potential problem goes to the radiologist and we are able to say whether the finding is true or not.”

Ibrahim added that beyond the diagnosis stage from a surgical or procedural aspect, doctors are a long way off for the trust of automated systems to do procedures or surgeries.

“At some point it will be able to incorporate multiple forms of data that will help providers,” he said. “It will never replace providers. I don’t think anyone would be happy knowing that they are being seen by an AI robot so to speak. What they want is a provider in the room as there is a human aspect to medicine.”

Ibrahim said every CT scan that is done at UConn Health is reviewed through the AI module.

The module also reviews lung nodules, which are small abnormal growths detected.

“The vast majority of lung nodules are benign but our best chance to treat lung cancer is to catch it in its infancy,” he said. “Once we get the information we review it. The vast majority of findings is not significant but if you were to review 100 nodules and we found that three were clinically significant and likely to be cancer that is three people out of 100 that we diagnosed early. The hope is to get them the most appropriate therapy as soon as possible.”

Ibrahim said he has also implemented an AI scribe into his practice, which listens to his conversations and integrates them into a note format.

“It allows me to spend more time with the patient and less time writing my notes,” he said.

Baldwin added that AI in its preliminary stage of diagnosing scans is not always accurate and could tell you you have an intracranial bleed when it is not the case. This is where radiologists’ verification of the scan and interpretation of the data is critical, he added.

“Even when these modules are flagged in studies we need to make sure that we have a physician there to render an opinion on that finding and get the message to the clinician that is caring for that patient,” he said.

By far, Ibrahim said the biggest challenge with AI software is assuring that it is secure, avoiding the hacking of any patient data.

At UConn Health, Ibrahim said the data is secure as the “cybersecurity team makes sure that everything is encrypted.”

Baldwin said that it is critically important at this stage that a human is involved in weighing in on the findings that are generated through AI.

“Overall, our care is improved with the use of AI and in the next five years it will be fascinating to see where it goes and how much better it gets,” he said.

Access this article at its original source.