The Clinical Impact Score: A Novel Framework for Guiding Artificial Intelligence Development in Radiology for Pediatric Populations

Purpose: Artificial Intelligence (AI) has the potential to revolutionize radiology, but few algorithms have been developed for use in pediatric patients. We propose the Clinical Impact Score (CIS), a novel framework that could help guide AI growth in pediatric radiology, an area where progress has lagged.

Methods/Materials: Imaging examinations performed on pediatric patients (age < 18 years) at our institution from 2019-2021 were analyzed. To create the CIS framework, four separate variables were determined for each of 18 locally defined critical findings: two objective (A) number of exams performed during the study period with potential critical findings, (B) median turnaround time; and two qualitative (C) likelihood of diagnostic error, (D) likelihood of patient harm if finding is missed. These variables were selected based on their potential to impact diagnostic accuracy and patient outcomes. Qualitative variables were determined by radiologist survey. Operational (A*B) and Criticality Scores (C*D) were determined along with the Clinical Impact Score (CIS, A*B*C*D) for each critical finding. Quadrant analysis and CIS magnitude were used to prioritize exam types by their potential for AI development.

Results: Of the 18 critical findings, line malposition, significant foreign body, and tension pneumothorax had the highest CIS, while unreduced intussusception had the lowest. The finding of a significant foreign body was ranked with the highest Operational Score, followed by line malposition and tension pneumothorax. Suspected child abuse and increased intracranial pressure were equally ranked with the highest Criticality Scores.

Conclusions: The Clinical Impact Score ranks critical findings in pediatric radiology, indicating high-impact use cases for AI development and highlight areas needing urgent attention. Development of AI algorithms using this framework optimizes impact on patient outcomes in pediatric radiology.