Real-world use of AI algorithm for collapsed lung cuts radiologist reporting times by 46%
Real-world use of a commercial AI algorithm for collapsed lung can cut radiologists’ reporting times by 46%, according to research published Oct. 29 in Academic Radiology [1].
Prompt diagnosis of such pneumothorax is crucial, with delays linked to longer hospital lengths of stay and greater disease progression. Researchers with University Hospitals Cleveland Medical Center sought to test artificial intelligence software to aid physicians in making these judgments.
Applying the algorithm retrospectively to nearly 27,400 chest X-rays produced promising results. Median reporting times for exams with confirmed pneumothorax fell from 186 minutes down to 100, “enabling swifter clinical response to a critical but treatable condition.”
“There is…great value in developing methods for radiologists to efficiently identify patients with [pneumothorax] and quickly communicate these findings to the care provider,” Joshua G. Hunter, with the Case Western Reserve University School of Medicine in Cleveland, and co-authors wrote Tuesday. “Artificial intelligence has the potential to play a pivotal role by detecting PTx on [chest X-ray] and generating alerts to augment the reading radiologist’s ability to prioritize studies with critical findings and, in turn, significantly improve turnaround time, which is the time between image acquisition and the radiologist’s final report.”
The analysis included front, single-view chest X-rays of adults imaged consecutively at a single institution between 2020 and 2021. A total of 12,728 radiographs were acquired within the AI-integrated system (Critical Care Suite from GE HealthCare), while 14,669 were obtained outside of the AI system and used as the control group. The software creates alerts for collapsed lung within the institution’s picturing archiving and communication system.
AI achieved area under the receiver operator characteristic curve of 0.78, the authors found. Sensitivity was 0.60 and specificity was 0.97. When selecting solely for moderate- or large-sized pneumothorax, AUC (0.93), sensitivity (0.89) and specificity (0.96) all improved.
Hunter and colleagues also experimented with a new routing system for on-call hours after 5 p.m. Traditionally during this shift, radiology reporting urgency is driven by the referring physician’s ordering priority. STAT cases are reviewed by the on-call radiologist within two hours, and routine ones are routed to the PACs worklist for next-day interpretation.
With the new approach, routine priority chest X-rays with findings suspicious of collapsed lung were routed to the on-call list, despite not being ordered as STAT. This served to accelerate interpretation by on-call residents, the authors noted.
“As with most academic radiology departments in the United States, these after-hour exams are generally read by radiologists in-training with little or no supervision by the attending radiologist until the next day,” the authors wrote. “Therefore, the majority of diagnostic decisions during these on-call hours are made by radiologists-in-training who are under stress from high imaging volumes and staffing constraints. As such, assistance in triaging of studies during on-call hours may be especially beneficial.”
The new workflow resulted in the detection of 962 cases of AI-flagged pneumothorax on routine-priority radiographs during on-call hours. Of these, 85 were moderate or large PTx and 31 were new clinically actionable cases, due to either not being detected previously or having increased in size.
“Given the oftentimes heavy workload on radiologists-in-training during on-call hours, the benefits of AI augmentation in this setting are particularly noteworthy,” the authors wrote. “This benefit was especially evident in the 31 CXRs with newly clinically actionable PTx where the AI tool alerted the on-call radiologist-in-training of a ‘suspicious finding’ in a routine priority CXR, which resulted in rerouting of the study for priority reading overnight to enable timely clinical intervention.”
Read more, including potential study limitations, at the link below.
Marty Stempniak has covered healthcare since 2012, with his byline appearing in the American Hospital Association's member magazine, Modern Healthcare and McKnight's. Prior to that, he wrote about village government and local business for his hometown newspaper in Oak Park, Illinois. He won a Peter Lisagor and Gold EXCEL awards in 2017 for his coverage of the opioid epidemic.