Artificial intelligence in the diagnosis of necrotising enterocolitis from abdominal radiographs: a systematic review and meta-analysis

Context: Necrotising enterocolitis (NEC) is a leading cause of morbidity and mortality in premature and very low birth weight neonates. Early diagnosis is challenging as clinical and laboratory features are non-specific. Artificial intelligence (AI) offers a potential means to improve diagnostic consistency and timeliness. Objective: To systematically review and meta-analyse the diagnostic accuracy of AI-based models for identifying NEC from abdominal radiographs. Evidence acquisition: A systematic search of MEDLINE, Embase, CINAHL, IEEE Xplore, and the Cochrane Library was conducted for studies published between January 1, 2015, and July 20, 2025. Studies evaluating AI, machine learning, or deep learning models applied to abdominal radiographs for NEC diagnosis or stratification, and reporting diagnostic performance metrics, were included. Study selection, data extraction, and risk-of-bias assessment (modified PROBAST) were performed independently by two reviewers. Diagnostic accuracy was pooled using hierarchical summary receiver operating characteristic models. Results: Ten retrospective studies met inclusion criteria; six were eligible for meta-analysis. Most employed convolutional neural networks, with limited external validation. The pooled sensitivity for AI-based diagnosis of NEC was 0.78 (95% CI 0.67-0.85) and pooled specificity was 0.84 (95% CI 0.73-0.91), with substantial heterogeneity. Positive and negative likelihood ratios were 4.84 and 0.27, respectively, indicating moderate diagnostic value. Explainability analyses commonly highlighted clinically relevant bowel features. Conclusion: AI-based interpretation of abdominal radiographs demonstrates moderate accuracy for NEC diagnosis and may serve as a decision-support adjunct rather than a standalone test. Clinicians should view these tools as complementary aids within existing diagnostic frameworks, pending prospective validation of AI models and standardised implementation into existing workflows. Systematic Review Registration: https://www.crd.york.ac.uk/PROSPERO/view/CRD420251090229.