A major bottleneck for radiology report turnaround speed is the amount of time a study sits idle in the workflow before it’s read, and each passing moment can have an impact on patient care. But imaging and informatics experts found success using a deep learning tool to help break up this congestion.
Dallas-based University of Texas Southwestern Medical Center researchers used a commercially available algorithm to flag abnormal noncontrast CT exams for intracranial hemorrhage and reprioritize routine and urgent exams in providers’ worklists.
While the pop-up notifications had no impact on wait times, actively reprioritizing worklists with AI had an outsized effect, reducing wait times from 15.75 minutes per exam to 12.01.
The tool, which was tested in their busy neuroradiology department, may lead to positive gains in productivity, efficiency, and patient care, researchers explained in Radiology: Artificial Intelligence.
“We observed that more than 90% of the report turnaround time can be contributed to wait time for examinations with the lowest priority (routine) and more than 60% for STAT examinations (highest priority) on average,” the authors reported. “Because report turnaround time is an important component of the imaging-to-treatment time, efficiencies introduced to expedite the interpretation of noncontrast-enhanced CT have the potential to improve patient outcomes,” they added.
O’Neill et al. first shared their results late last year, but the findings recently pushed two Wisconsin imaging experts to ask a simple question: Should AI tell radiologists which study to read next?
The pair noted that O’Neill and colleagues showed an “optimal” example of using technology in radiology practice, but also pointed out their research only included unenhanced CT, whereas many reading lists contain a variety of exams. It also omitted exams completed outside normal hours.
Stacy D. O’Connor, MD, MPH, and Manav Bhalla, MD, both with the Medical College of Wisconsin’s Department of Radiology, posed further questions: Should a head CT with one positive finding like ICH be prioritized above or below a chest x-ray with two positives such as pneumothorax and pneumonia? And what if the brain bleed is larger today than yesterday’s CT?
For these situations, and AI tools going forward, variables such as patient type (emergency, outpatient, etc.) and time elapsed since image acquisition, will need to be incorporated to balance AI results.
“As reliable AI tools are developed that can reprioritize examinations with urgent findings appropriately, methods to adjudicate priority adjustments from an ensemble of algorithms will be required, especially for work lists that contain examinations from multiple modalities for multiple body parts,” O’Connor and Bhalla concluded.
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.