Artificial intelligence rules out incidental PE in chest CT

According to a new study published in American Journal of X-rays.

In a retrospective review of conventional contrast-enhanced chest CTs, the authors found that a commercial AI algorithm had a high negative predictive value for iPE. In addition, the AI ​​detected pulmonary embolism that radiologists had overlooked, but the radiologists also picked up some pulmonary embolism that the AI ​​had overlooked.

“Sometimes these occasional PEs are harder to see on the exams that aren’t optimized for PE,” says Paul H. Yi, MD, assistant professor of diagnostic radiology and nuclear medicine at the University of Maryland School of Medicine and director of the university. Medical Intelligent Imaging Center, in an interview with Medscape medical news. Yi was not involved in the investigation.

“This AI works for this purpose, and this purpose could be really helpful because we don’t always have the advantage of a CTPA [CT pulmonary angiography],” he said.

Echoing one of the authors’ conclusions, Yi added that AI can help radiologists by giving them a “second reading or a second opinion, looking over our shoulder a bit.”

Lead author Kiran Batra, MD, shared: Medscape that AI is not only that second reader, but can also flag certain studies for a priority study, allowing radiologists to assess the ever-increasing workload.

“I think it’s going to be a symbiosis and teamwork between the two,” said Batra, an assistant professor of radiology, at UT Southwestern Medical Center.

Test-driving AI

The authors conducted a retrospective study of 3003 consecutive contrast-enhanced CT scans of the chest without using pulmonary angiography protocols.

These were conducted on 2,555 adults between September 2019 and February 2020 at Parkland Health in Dallas, Texas.

The authors examined the results of two algorithms previously applied to the CTs:

  • An FDA-approved commercial AI algorithm (Aidoc) was applied to the images to detect iPE. This algorithm has been trained on conventional chest CTs. It had been applied before the current study and radiologists caring for the patients had no access to the results.

  • A natural language processing (NLP) algorithm (RepScheme) was applied to the clinical radiologists’ scans to see which one mentioned iPE.

If either algorithm flagged an iPE, two radiologists independently reviewed the relevant scans to determine if iPE was present, with a third radiologist available to resolve discrepancies.

In addition, a radiologist examined NLP results and corrected any misclassification of iPE.

A way to help rule out PE

The mean age of the patients was 53.6 years and slightly more than half were female. More than 70% of CTs were done because of cancer.

After arbitration, about 40 iPEs were detected. AI found four iPEs that clinicians had missed, while clinicians saw seven that AI had overlooked.

For AI vs. clinical reports, the performance was as follows:

  • Sensitivity: 82.5% vs 90.0%, P = .37

  • Specificity: 92.7% vs 99.8%, P = .045

  • Positive predictive value: 86.8% vs. 97.3%, P = 0.03

  • Negative predictive value: 99.8% vs. 99.9%, P = 0.36

“As a radiologist, if I read a scan, and I don’t find a PE, I should look at the AI ​​to see if it found a PE or not, because it has a high negative predictive value,” Batra said. “If the AI ​​didn’t find a PE, and I didn’t find a PE, then the chances are [the patient] don’t have are quite high.”

Limitations included low iPE incidence, which limits study capacity. Manual grading was applied only to scans positive by AI or NLP; so if iPEs had been mistakenly missed by either technique, the authors would have missed them as well. And the authors pointed out that generalizability is limited because protocols and patient populations vary.

The role of AI in vascular radiology

PE can present non-specifically and be notoriously easy to miss. It affects between 71 and 117 per 100,000 people in the US each year, and it primarily threatens cancer patients, for whom it may herald a worse prognosis.

AI is good at picking up PE on PE protocol CTs, also known as CTPA. Those CTs time the contrast bolus to highlight the pulmonary arteries.

But it had previously been less clear how well the technology would pick up iPE from chest CT scans performed for other indications, such as cancer or lung disease.

Amid reports of burnout among radiologists, a global shortage of radiologists and an increased demand for imaging, AI can play an important role. But AI for radiology is still in its infancy, according to Yi.

“There is still a long way to go,” he said. “I think there are early wins [in] things like triage and trying to have a high negative predictive value. But we’re really far from replicating what a radiologist does.”

That said, Yi added, there is a lot of nuance in radiology, and there will be a need for studies like this that clinically validate these products.

“This is an unfunded, unbiased evaluation of [the AI algorithm]and that’s pretty cool,” he said. “It seems to work as they claim.”

The study was unfunded. Batra and co-authors have not disclosed any relevant financial relationships. Yi is a consultant for Bunkerhill Health.

AJR Am J Roentgenol. Published online July 13, 2022. Summary

Jenny Blair, MD, is a journalist, writer, and editor based in Vermont.

For more news, follow Medscape on Facebook, TwitterInstagram and YouTube.

Leave a Comment