Cost Effectiveness Analysis of the Use of Machine Learning-based Software for the Diagnosis of Neonatal Cranial Deformities

Introduction: Prior retrospective studies have shown that image-fed AI models can provide cheap, scalable point-of-care (POC) diagnostic analyses of cranial deformities that approach CT accuracy and may reduce the need for radiography during the initial evaluation of neonatal head shape abnormalities. Our group has had AI-analyses of neonatal deformity photographs available at POC since 2020. These AI reports provide standard craniometrics and relative risk estimations for craniosynostosis. We examine here the relationship between the availability of AI-augmented cranial deformity analysis and provider diagnostic ordering patterns.

Methods: A retrospective chart review of pediatric patients referred for concern for “plagiocephaly”, “brachycephaly”, “abnormal head shape”, or “craniosynostosis” was conducted. Patient demographics, final encounter diagnosis, review of an AI-analysis, and provider orders were documented. Patients were separated based on whether an AI cranial deformity analysis was generated during the index evaluation. Pearson’s Chi-square test was used to evaluate rates of provider ordering between the AI-augmented and conventional evaluation groups.

Results: 500 patient charts were reviewed (381 conventional; 119 AI). 98.8% of final diagnoses were deformational, and there were 6 cases (1.2%) of craniosynostosis. 73.2% of patients had multiple encounters. The conventional evaluation group ordered radiographic studies 12.1% of the time, while the AI-augmented group ordered radiographic studies 1.7% of the time (p = 0.015). Among MDs, this was a 94.3% reduction in radiography usage and a 100% reduction among advanced practice providers.

Conclusion : The availability of AI-augmented analyses of standardized neonatal head images was associated in this single-center retrospective review with a 10-fold reduction in radiography utilization. As neurosurgeons and pediatricians continue to work to limit neonatal radiation exposure and contain healthcare costs, this data highlights a potentially cheap and easily scalable diagnostic approach that is associated with fewer unnecessary radiographic diagnostic tests for neonatal cranial deformities.