Artificial intelligence (AI) can detect symptomatic pulmonary emboli (iPE) in chest computed tomography scans performed for other indications, according to a new study published in American Journal of Oncology.
In a retrospective review of conventional contrast-enhanced chest computed tomography, the authors found that a commercial AI algorithm had a high negative predictive value for iPE. In addition, the AI picked up pulmonary emboli that the radiologists had missed – but the radiologists also picked up some of the pulmonary emboli that the AI had missed.
“Sometimes it’s hard to see these incidental findings in tests that aren’t optimized for PE,” said Paul H. Yi, MD, associate professor of diagnostic radiology and nuclear medicine at the University of Maryland School of Medicine and rector of the university. Intelligent Medical Imaging Center, in an interview with Medscape Medical News. Yi was not involved in the study.
“This AI works for this, and that purpose could be really helpful, because we don’t always benefit from CTPA.” [CT pulmonary angiography],” He said.
Echoing one of the authors’ conclusions, Yi added that AI might help radiologists by giving them “a second reading or a second opinion, sort of looking over our shoulders.”
said lead author Kiran Batra, MD Medscape That in addition to being that second reader, AI can flag certain studies for priority reading, helping radiologists sort through their ever-increasing workloads.
“I think it would be a symbiosis and teamwork between the two,” said Batra, an assistant professor of radiology, UT Southwestern Medical Center.
AI driving test
The authors performed a retrospective study of 3,003 contrast-enhanced chest CT scans that did not use pulmonary angiography protocols.
These tests 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 CTs:
An FDA-approved commercial AI algorithm (Aidoc) was applied to the images with the goal of iPE detection. This algorithm was trained on conventional chest CT scans. It was applied before the current study, and the radiologists caring for patients were unable to come up with the results.
A Natural Language Processing (NLP) algorithm (RepScheme) was applied to clinical radiologists’ readings of scans to see which ones were mentioned.
If either algorithm flags iPE, two radiologists independently separate relevant scans to determine if iPE is present, with a third radiologist available to resolve the discrepancies.
In addition, a radiologist examined the NLP results and corrected any misclassification mention of iPE.
Way to help rule out PE
The average age of the patients was 53.6 years and just over half were women. More than 70% of CT scans were done because of cancer.
After judgment, about 40 iPEs were detected. AI found four iPEs that doctors didn’t miss, while doctors discovered seven that AI missed.
For AI versus clinical reports, the performance was as follows:
Sensitivity: 82.5% vs. 90.0%, probability = .37
Specificity: 92.7% vs. 99.8%, Probability = 0.045
Positive predictive value: 86.8% vs. 97.3%, P = 0.03
Negative predictive value: 99.8% vs. 99.9%, P = 0.36
“If I’m reading the scan as a radiologist, and I don’t find PE, I have to look at the AI to see if it has found PE, because it has a high negative predictive value,” Batra said. . “If AI does not find PE, and I do not find PE, chances are [the patient] Not having it is too high.”
Limitations included a low incidence of iPE, which limits the power of the study. Manual review was applied only to scans that were positive by AI or NLP; Thus, had iPEs been incorrectly missed by both techniques, the authors would have missed them as well. The authors noted that generalizability is limited, as protocols and patient groups vary.
The role of artificial intelligence in vascular radiology
PE can present indefinitely and be easy to miss. It affects between 71 to 117 per 100,000 people in the United States annually, according to the authors, and is particularly threatening in patients with cancer, which can portend a worse prognosis.
AI is good at capturing PE on PE- CTs protocol, also called CTPA. These tracks adjust the timing of the contrast dose to highlight the pulmonary arteries.
But it wasn’t previously clear how well the technology could pick up iPE from chest CT scans that were done for other indications, such as cancer or lung disease.
Amid reports of radiologist fatigue, a worldwide radiologist shortage, and an increased demand for imaging, artificial intelligence may play an important role. But according to me, ray AI is still in its infancy.
“We have a long way to go,” he said. “I think there are early victories [in] Things like sorting and trying to get a high negative predictive value. But we’re really far from replicating what a radiologist does.”
However, Yi added, there are a lot of nuances in radiology, and studies like this one that validate these products clinically will be needed.
“This is a third party, unfunded, and unbiased assessment of [the AI algorithm]He said, This is wonderful. “It seems to work as they claim.”
The study was unfunded. Batra and colleagues did not disclose any relevant financial relationships. Yi is a consultant at Bunkerhill Health.
AJR am J Roentgenol. Published online July 13, 2022. Abstract
Jenny Blair is a physician, journalist, writer and editor in Vermont.
For more news, follow Medscape on Facebook, Twitterand Instagram and YouTube.
#Artificial #intelligence #excludes #accidental #chest