Traumatic brain injury (TBI) in infants under 3 months is rare but potentially severe, and clinical decision-making is complicated by nonspecific symptoms and limited neurologic assessment. This study aimed to evaluate the diagnostic yield of PECARN, CATCH, and CHALICE rules.

This single-center retrospective study included infants aged 0–3 months who underwent cranial CT following blunt head trauma between January 2020 and December 2023. Clinical and radiological data were extracted from medical records, and each case was retrospectively evaluated using the PECARN, CATCH, and CHALICE rules. Diagnostic yield was defined as the proportion of rule-positive infants with CT-confirmed traumatic brain injury. Multivariable logistic regression and dimensionality reduction techniques (t-SNE, UMAP) were applied to explore predictive patterns.

Among 151 infants aged 0–3 months who underwent cranial CT, 20 (13.2 %) had traumatic brain injury (TBI). Diagnostic yield was 14.8 % for PECARN, 18.9 % for CATCH, and 25.6 % for CHALICE. In multivariable analysis, having a Glasgow Coma Scale (GCS) score of 15 (vs. <15) was independently associated with lower odds of TBI (adjusted OR: 0.003, 95 % CI: ∼0.00–0.040), while post-traumatic seizure was associated with higher odds (adjusted OR: 26.17, 95 % CI: 2.12–394.45). Dimensionality reduction techniques (t-SNE, UMAP) revealed partial clustering of TBI-positive cases.

In this CT-based cohort of infants under 3 months, CHALICE achieved the best balance by detecting all TBI cases with fewer scans than PECARN. CATCH missed two TBIs despite modest CT reduction. Multivariable modeling and dimensionality reduction supported integrated, data-driven risk stratification.