Understanding artificial intelligence for radiologists

Automation often breeds worry, as the people working in an area being automated sometimes fear that technology will put them out of a job. Yet, Montefiore Nyack Hospital — a 250-bed community hospital located 30 miles north of New York City — found that implementing artificial intelligence (AI) in the radiology department didn’t lead to panic, but instead was viewed as a much-appreciated, reliable safety net.

Montefiore Nyack Hospital has a long history of using technology to improve efficiency, reduce waste and ensure quality. With its AI initiative, the organization wanted to speed turnaround for reading radiologic images, so that patients with critical conditions could receive life-saving treatments faster. The organization also hoped that quicker turnaround times could yield shorter lengths of stay and higher patient satisfaction.

The Aidoc’s AI solution, integrated into the Change Healthcare Workflow Intelligence, uses three FDA-cleared algorithms that address common emergency room trauma cases to read images and prioritize them, allowing radiologists to review the most clinically urgent studies first.

Before implementing AI, the hospital used metadata from the clinical exam to rank studies, considering a patient’s age, condition and treatment location. However, with the AI solution, the hospital can now prioritize studies using the actual image, making sure that the most urgent cases are triaged and moved to the top of the list.

Although some of the hospital’s radiologists were initially skeptical of the technology, it quickly became clear that AI was not there to take over their jobs.

The AI solution has a highly specific purview. It closely examines images for one of three findings — brain bleeds, pulmonary embolisms or cervical spine fractures — and indicates whether the finding is positive. It is not designed to determine what’s causing the finding or what other factors may be involved.

For example, the AI may point out there is bleeding on a patient’s brain, but it doesn’t say why there is bleeding, such as because of an intracranial hemorrhage, subarachnoid hemorrhage or subdural blood. It is up to the radiologist to integrate all the available data points, examine the image in context and come up with a possible recommendation.

While the hospital has seen a 17% improvement in turnaround time, the radiologists are more impressed with some of the technology’s other benefits.

First and foremost, the AI solution serves as a dependable backup. It’s widely accepted that double-reading radiology images increases quality.

The problem with this approach to date has been that it is not cost-effective for hospitals to have all their images double-read: the time and physician resources needed is cost-prohibitive. However, when one of those radiologists is a computer, an organization can cost-effectively augment the human reading, improving accuracy and ultimately providing better outcomes for patients. When computers and humans serve as double checks for each other, the results are increased accuracy and better outcomes for the patient.

The advantages of the double read became apparent to Montefiore Nyack’s radiologists shortly after the technology was up and running. On its first day, the AI solution found a bleed in a patient’s brain that one of the radiologists had missed. It was nearly imperceptible, but it completely changed the interpretation of the study. News of the incident spread throughout the department. Radiologists went from wondering if the technology would provide benefit to actively asking for an AI reading prior to dictating a case.

The hospital has also found the technology reduces radiologists’ stress levels. Reading images is a complex task that requires focus and precision. When a radiologist is reviewing hundreds of images during an on-call shift, it is easy to overlook subtle issues, particularly if the radiologist is the only one on call and is reading images outside of his or her specialty.

Since Montefiore Nyack Hospital is a community hospital, it doesn’t always have every sub-specialist on call. In those instances, providers often read images from areas in which they are not fellowship trained. This practice, along with the large workload, can cause stress — especially if there are time pressures. Having a reliable second look puts radiologists’ minds at ease. The hospital has received anecdotal feedback from its radiologists that the AI solution takes some of the pressure off. Over time, curtailing stress in this manner may reduce the likelihood of physician burnout — something with which the entire medical field struggles.

The idea that AI will replace radiologists is a myth. In fact, quite the opposite is true: the technology makes these hard-working physicians more efficient and increases care quality in ways that have not been possible before. Instead of viewing the technology as a threat, radiologists can think of it as a partner, helping to improve patient care, mitigate risk and curb physician burnout.