Title : AI-driven predictions in paediatric traumatic brain injuries: Enhancing outcomes with data-driven insights
Abstract:
Objectives: To evaluate the application and effectiveness of artificial intelligence (AI) in predicting outcomes in pediatric traumatic brain injuries (TBIs).
Design: A systematic review of studies examining AI-driven models for outcome prediction in pediatric TBI.
Subjects: Studies included pediatric patients (<18 years) with TBI, where AI models were used for predicting outcomes such as recovery, mortality, or long-term impairments.
Methods: A systematic search of PubMed, Embase, and Scopus databases was conducted for studies published in the last 10 years. Inclusion criteria were original peer-reviewed research focusing on pediatric TBIs, utilising AI for outcome predictions, and reporting performance metrics. Articles were screened, and data on AI model type, accuracy (e.g., AUC, sensitivity, specificity), and clinical applicability were extracted.
Results: Of 243 initial articles, 25 met the inclusion criteria. Most studies used machine learning algorithms, with several incorporating imaging and clinical data. Predictive accuracy for long-term outcomes ranged from AUC 0.80 to 0.93. However, limitations included small, heterogeneous datasets and a lack of external validation. Few studies focused specifically on pediatric populations, limiting generalizability.
Conclusions: AI demonstrates potential in predicting outcomes in pediatric TBI but requires further validation. Developing standardised pediatric datasets and improving model transparency could enhance clinical adoption.
