Can AI Triage for CT Detection of Intracranial Hemorrhages Have an Impact for Pediatric Patients?

Do pediatric patients get lost in the shuffle in the evaluation and potential implementation of adjunctive AI triage tools for acute presentations in emergency department settings?

In a recent interview with Diagnostic Imaging, Marla Sammer, MD, MHA, FAAP, and Joseph Cavallo, MD, MBA, discussed the overall dearth of AI applications for pediatric patients and potential ramifications.

“The main reason to look at this is right now there's overall very little commercially available AI that's also available for children, or even exclusively for children. It's mostly for adult patients. Then the bigger even more pointed part of that is that there are no triage tools that are for pediatric patients,” noted Dr. Sammer, the vice chair of AI and innovation in the Department of Radiology at Texas Children’s Hospital in Houston.¹

“I think the reason why triage tools jump out a bit is because if you are preferentially only identifying urgent pathology in the adult population, you could be unknowingly deprioritizing pediatric patients with the same acute pathology, right? We're getting alerted if it's happening to an adult, but we're not getting alerted if it's happening to somebody under the age of 18,” pointed out Dr. Cavallo, the vice chair of clinical affairs for Yale Radiology and an assistant professor at the Yale University School of Medicine.

Accordingly, Dr. Cavallo, Dr. Sammer and their colleagues recently evaluated data from 1,996 non-contrast head computed tomography (CT) scans to assess the impact of a CT-based AI software (Briefcase — Intracranial Hemorrhage Triage, Aidoc) for improving detection of intracranial hemorrhage (ICH) in pediatric patients six years of age and older.² The AI software is cleared by the FDA for ICH detection in adults, according to the study.

For the detection of ICH in this pediatric cohort, the retrospective multicenter study revealed a 94.2 percent sensitivity rate, a 94.7 percent specificity rate, a 62.5 percent positive predictive value (PPV) and a 99.4 percent negative predictive value (NPV).²

However, Dr. Sammer emphasized the importance of looking beyond these measurements when assessing the value of AI tools.

“For patients, what I would say is really important and for the practice, does it actually improve their care? … Are there more emergent findings that are getting detected faster? That probably is a lot more important. … Does using the AI tool in your actual practice speed things up? Does it slow things down? On an even more individualized basis, (if we expand the AI use to patients to six years old and up), how does that actually impact the patients who are not getting triaged?,” posited Dr. Sammer, the chair of the Pediatric AI Working Group with the Informatics Commission of the American College of Radiology.

References

1. Sammer MBK, Akbari YS, Barth RA, et al. Use of artificial intelligence in radiology: impact on pediatric patients, a white paper from the ACR Pediatric AI Workgroup. J Am Coll Radiol. 2023;20(8):730-737.
2. Cavallo J, Sher A, Chen D, Avondo J, Sammer M. Performance of an adult-trained AI tool for intracranial hemorrhage detection on head CT in children aged 6-17 years. Pediatr Radiol. 2026;56(3):649-657.

(Editor’s note: For related content, see “FDA Clears AI-Powered CT Triage Software for Intracranial Hemorrhage,” “Pertinent Insights on Evaluating the Value of AI Models in Radiology, Part 1” and “Can AI Assessment of Longitudinal MRI Scans Improve Prediction for Pediatric Glioma Recurrence?”)