machine learning models for pediatric intussusception diagnosis and prognosis

Source article: Unfolding new horizons: Machine learning applications for pediatric intussusception - A systematic review and meta-analysis

Background Pediatric intussusception is among the most common surgical emergencies in early childhood. Ultrasonography is the diagnostic standard but remains operator-dependent, and no validated machine learning (ML) tools exist for triage or prognostic stratification. We systematically evaluated ML models for diagnosis and prognosis of pediatric intussusception. Methods This PROSPERO-registered systematic review and meta-analysis (CRD420251073590) included studies evaluating ML models for pediatric intussuscept…

By the Truvace deskOctober 6, 20262 min read1 sourcePeer-reviewedTRV-2026-1297· cite & verify

Trace impact readingPositive state

P 75

Both sides are scored from claims and sources, not community votes.

G 80

In brief

This PROSPERO-registered systematic review and meta-analysis evaluated 11 studies comprising 37 model variants and 36,863 patients assessing machine learning for pediatric intussusception. Internally validated ultrasound models achieved pooled sensitivity 0.914 and specificity 0.980, with externally validated models showing sensitivity 0.946 and specificity 0.958, while abdominal radiograph triage models performed lower externally.

The findings matter because ultrasonography is operator-dependent and no validated ML triage tools currently exist, yet the review shows AI assistance can improve junior reader specificity and reduce examination time without loss of accuracy. Uncertainty remains due to detected publication bias, absence of clinical endpoint data, wide prognostic performance range, and need for multicenter prospective validation outside Asia before implementation.

Main points

Systematic review and meta-analysis included 11 studies, 37 model variants, 36,863 patients, PROSPERO-registered CRD420251073590.
Internally validated US models k=4 pooled sensitivity 0.914 and specificity 0.980; externally validated US models k=3 sensitivity 0.946 and specificity 0.958.
AI assistance improved junior readers' specificity by 15.9 percentage points overall, up to 37.6 points for most junior, and cut ultrasound time by 61% without loss of accuracy.
Abdominal radiograph triage models performed lower externally with sensitivity 0.800 and specificity 0.748, positioned as screening adjunct.

The gain

Systematic review of 11 studies (36,863 patients) found ultrasound-based ML models achieved high pooled sensitivity and specificity with external validation, and AI assistance improved junior readers' specificity and cut examination time by 61% without loss of accuracy.

The problem

Review detected possible publication bias, found no studies reporting clinical endpoint data, and found abdominal radiograph triage models had substantially lower external accuracy, limiting readiness for clinical implementation.

The rundown

The review searched for ML models for pediatric intussusception diagnosis or prognosis reporting sensitivity, specificity or AUC, assessed bias with PROBAST, and applied a bivariate random-effects model to internally and externally validated datasets with subgroup analyses by input modality.

Results showed US-based models maintained high accuracy externally, while AXR models were lower and suited only to raising suspicion, with prognostic AUCs ranging from 0.555 to 0.962, and authors concluded AXR-based models should be positioned as an adjunct that raises suspicion, never as a gate that determines which child receives an ultrasound.

What this doesn’t fix

Evidence is limited to retrospective diagnostic accuracy studies with detected publication bias, no reported clinical endpoint data, and lack of multicenter prospective validation outside Asia, requiring regulatory development before implementation.