It is becoming harder to find enough radiologists to manually review the rising number of medical scansCredit: Phil Boorman/Getty
Radiology has been a fundamental platform of healthcare for more than a century. From basic X-rays to more sophisticated CT scans, the technology offers pictures from inside the body that can help diagnose everything from tumours to blood clots.
Timely analysis of these images is critical for early interventions and to put patients on the right treatment path for better care outcomes. But radiology faces a crisis: a global shortage of the skilled experts needed to interpret the images has increased waiting times and reduced access to necessary care. Delays in diagnosis can have life-threatening consequences. And as the population of the United States and other countries age, there are an increasing number of these images that need to be reviewed.
Physicians at Atlantic Health System in New Jersey are showing that artificial intelligence (AI) can help. Their patients are already benefitting from the speed and efficiency that machine learning techniques offer.
In 2020, Atlantic Health introduced an AI algorithm that can analyse CT scans for the signs of pulmonary embolism — a potentially life-threatening condition when blood flow is blocked to part of the lung. Humans are still very much part of the loop. AI screens positive scans to be urgently reviewed by radiologists, leading to earlier diagnosis and faster treatment for the people who most need it. In this way, the technology supports the core mission of Atlantic Health, to maximize patient safety and quality of care.
“AI takes two to three minutes to screen an image,” says Devon Klein, a diagnostic radiologist at Atlantic Health. “Different clinicians may interact with AI in different ways, but I use it as a triage tool to help me prioritize the next case I should read.”
Faster results
Used originally in the emergency department setting, the algorithm meant that patients attending Atlantic Health hospitals could receive a faster diagnosis of pulmonary embolism — about a third quicker than relying on radiologists alone.
Inspired by this success, the system extended its use to the outpatient setting, where embolisms are seen less often, and so delays in receiving treatment are more likely.
Screening for incidental pulmonary embolism in outpatient imaging centres in this way is part of what Atlantic Health calls its “AI ambulatory safety net”. A recent review of the programme showed that 0.3% of almost 12,000 CT scans were positive for pulmonary embolism. The wait time for these patients to get a diagnosis and start a treatment plan was under 90 minutes, compared to the average outpatient wait of around 15 hours.
This ten-fold reduction in wait time saves lives, says Klein.
“You don't expect pulmonary embolism patients to come to the outpatient centre,” he says. “But even if it happens only five or ten times a year, we want to treat those cases as fast as possible.”
He adds: “We are now able to catch those patients, contact the referring physician and coordinate emergency care before they leave the department, which is something very novel.”
The success at using AI algorithms to find embolisms in radiology images, as well as a different condition called intracranial bleeds, helped Atlantic Health last year to win a CIO 100 award for ‘AI-assisted radiology tool powers’. These annual prizes celebrate teams at 100 organizations who use IT in innovative ways to deliver business value.
Solutions for success
Atlantic Health is now working to expand the successful use of AI to many other areas of care.
“We are looking at AI technologies for clinical screening purposes in oncology, for early detection of breast, lung, pancreatic and colon cancers”, says Suja Mathew, Atlantic Health’s executive vice president and chief clinical officer.
Organized into cross-functional teams that merge clinicians and colleagues with expertise in IT, law, finance and administration, this allows Atlantic Health’s experts to view clinical problems from different perspectives. “We want technologies not just to fix an isolated issue, but to help us provide a comprehensive care delivery experience that will measurably improve clinical outcome.”
One significant challenge for healthcare providers looking to incorporate AI technologies into their systems is the ever-increasing number of tools and technologies gaining FDA approval. So how can they choose the right one?
Atlantic Health argues it has a head start. Through its early decision to adopt and incorporate new technologies into its 400-plus sites, including seven hospitals, the system has developed a robust internal governance structure. This includes ways to monitor AI applications in healthcare and measure the outcomes.
This structure ensures that the system will implement AI only when a genuine problem exists that the technology can help with, which stops it falling into the trap of adopting technologies just because they are available. A trap that Sunil Dadlani, executive vice president, and chief information and digital transformation officer, calls “shiny toy syndrome”.
“Clinician input is paramount through every step of the process and dictates what we do,” Mathew adds. “Clinicians understand the technology, they understand potential biases, and they can use AI tools alongside their clinical judgment.”
Data quality
The performance of all AI tools, in healthcare and beyond, rests on the quality of the data sets used to train the algorithms. If a tool has been tried and tested on mainly white Europeans, for example, it may not be so reliable when used with a more heterogeneous population in New Jersey. Atlantic Health officials acknowledge this challenge and are very aware of it.
“We have made huge strides in terms of data quality,” says Dadlani. The system gathers relevant information from dozens of sources alongside its own electronic health records. These include social vulnerability and population data. “Bringing together these different data sources creates a more holistic picture relevant to our patient population,” he says.
Better data doesn’t just improve the experience for patients. It can help physicians too. Atlantic Health is using new ambient voice technologies that records conversations in the clinic so that doctors don’t have to spend so long writing notes by hand. This gives them more time, to see more patients and to work more efficiently. That’s a crucial step. One reason for the lack of trained radiologists is burn out. AI won’t replace doctors but will enhance human intelligence, enabling quicker, accurate and timely decisions, and can also help keep them healthier and happier. And that’s a good outcome for everyone.