4 increasingly common uses for imaging AI in the ED
As FDA-approved AI software continues to proliferate in radiology—well more than 150 products to date and rising—a trio of Yale radiologists has compiled a status report focused on AI applications available to, specifically, emergency radiology.
Khalid Al-Dasuqi, MD, Michele Johnson, MD, and Joseph Cavallo, MD, MBA, had their paper published online May 29 in Clinical Imaging [1].
Despite the rareness of direct reimbursement for radiology, the authors note, a recent ACR survey found 30% of responding radiologists using AI and another 25% planning to do so. Some 95% of the current users, who were not limited to emergency practice, indicated they were satisfied with the technology’s performance.
In a section on diagnostic AI apps currently used in emergency settings, Al-Dasuqi and colleagues flesh out facts on the four most in-demand diagnostic use cases:
1. Stroke and large vessel occlusion (LVO)
Noting that stroke is the most common target of AI applications in the emergent setting, the authors count 16 FDA-approved AI products for helping to diagnose and triage acute ischemic stroke. They report:
Several studies published over the past year have demonstrated the high accuracy of LVO detection in the emergent setting by commercially available AI products and the associated reduction in time to mechanical thrombectomy. Additionally, the AI-based tools for [Alberta Stroke Programme Early CT Score] on non-contrast head CTs have demonstrated similar accuracy as well as superior precision and reproducibility compared to experienced neuroradiologists.”
2. Intracranial hemorrhage (ICH)
The second most common target of AI-based neuroradiology applications, ICH is a worthwhile indication due to its potential for high mortality and demand for immediate medical and/or surgical care, the authors state. As of early February there were 13 FDA-approved AI tools for ICH detection.
The present applications, like the vast majority of current AI, perform best when complementing the radiologist's role, rather than [operating] in place of it. Prior literature specific to ICH detection revealed a 12.2% increase in ICH detection when radiologist interpretation is augmented by an AI solution, with minimal increase in overcall rate.”
3. Spine injury
Just one FDA-approved AI product is on offer for this indication, although other commercial packages are available for helping diagnose vertebral compression fractures on CT imaging of the chest, abdomen or pelvis, the authors point out.
AI detection of cervical spine fractures has demonstrated variability in accuracy within the published literature. However, it is important to note that study design and population prevalence can have significant influence on results. This variability in performance is not unique to select products and is an important consideration in the deployment and evaluation of these tools at individual institutions.”
4. Pulmonary embolism (PE)
Three FDA-approved AI-based products are on the market for detecting and triaging PE on CT angiograms of the pulmonary arteries.
Performance analyses of the AI solutions demonstrate high diagnostic accuracy for the detection of PE with sensitivity and specificity over 85%. In addition to automated PE detection, two vendors have developed platforms for PE response team activation alerts, which help not only flag cases that are positive for PE, but also enable team members to access relevant patient data, including laboratory test results, imaging, and reports to expedite care.”
The new overview also covers radiographic diagnoses, miscellaneous diagnostic tools, non-interpretive imaging applications, workflow-based applications and other aspects of imaging AI in the ED.
Covering financial considerations, Al-Dasuqi and co-authors spotlight CMS’s 2020 approval of a New Technology Add-On Payment (NTAP) for the detection of large vessel occlusion in acute stroke patients by which the agency reimburses up to $1,300 per use on qualifying patients.
“Between the fourth quarter of 2020 and the second quarter of 2021, the number of hospitals billing for this code has increased from 63 to 130, and the total number of codes billed increased from 1,797 to 4,811, according to vendor database tracking,” Al-Dasuqi et al. write. “There is optimism for additional reimbursement in the future, and there has already been approval of Category III CPT codes for additional AI products, including one for cervical spine fractures.”
More Coverage of Emergency Imaging:
Cross-sectional imaging ordered downstream for just 15% of emergency POCUS patients
Some 78% of parents OK with AI reading their child’s chest X-rays
Imaging shows COVID vaccines effective at warding off pulmonary embolism
Disparities evident as CT stroke imaging rises sharply over 7-year period
Artificial intelligence shows promise predicting patients’ need for CT after traumatic brain injury
Reference:
- Khalid Al-Dasuqi, Michele H. Johnson, Joseph Cavallo, “Use of artificial intelligence in emergency radiology: An overview of current applications, challenges, and opportunities.” Clinical Imaging, May 29, 2022. DOI: https://doi.org/10.1016/j.clinimag.2022.05.010
Dave P. has worked in journalism, marketing and public relations for more than 30 years, frequently concentrating on hospitals, healthcare technology and Catholic communications. He has also specialized in fundraising communications, ghostwriting for CEOs of local, national and global charities, nonprofits and foundations.