Artificial intelligence for pediatric neuroimaging.

Authors

Abstract

Artificial intelligence (AI) is revolutionizing health care, particularly in radiology for which large retrospective electronic datasets are naturally suited to training large machine and deep learning models. To date, the vast majority of United States Food and Drug Administration (FDA)-cleared AI software tools are geared toward radiology applications, with most in neuroradiology and very few in pediatric radiology. Pediatric innovations have historically lagged behind adults due to children's limited exam tolerance, small and rare datasets, age-related variability in normal and disease processes, and ethical/legal concerns for vulnerable populations. The chasm between research and clinical applications is further hampered by a smaller commercial market share in pediatrics. Nevertheless, understanding of AI successes and failures in adult neuroradiology can help inform progress in pediatric neuroradiology. Furthermore, the rise of generative AI can help overcome current limitations and enable complex multimodal pattern recognition for a broader variety of use cases. Human expert oversight will help mitigate AI risks including data and algorithmic bias, black-box errors, and regulatory gaps. In this review, we will discuss AI technical principles and pitfalls, relevant clinical tools, promising research advances, and future directions for pediatric neuroimaging. Key AI applications to be discussed include quality/safety, fetal, neonatal, hydrocephalus, malformations, tumors, epilepsy, demyelination, stroke, and trauma.

Tags

Topics