Applications of artificial intelligence in pediatric surgical pathology: A systematic review.

BACKGROUND

Artificial Intelligence (AI) techniques can transform and enhance diagnosis and treatment response predictions in pediatric surgical pathology. AI offers the potential to reduce the workload of pathologists by automating routine and labor-intensive tasks. In this systematic review, we investigate current applications of computational pathology in pediatric surgical conditions.

METHODS

Nine databases were searched from inception until January 2025 to retrieve articles looking at the use of machine learning, AI, or virtual reality in the pathological diagnosis of pediatric surgical conditions, without language restrictions. PRISMA standards were followed, and abstract screening was performed by two reviewers, with conflicts resolved by the senior author. Original studies and reviews exploring computational pathology for diagnosing, grading, or predicting outcomes in pediatric surgical diseases were included.

RESULTS

The authors screened 3363 articles, with 34 meeting the inclusion criteria. AI applications primarily involved convolutional neural networks (24, 70.6 %), trained on whole-slide images. The most frequently studied diseases were childhood cancers (18, 52.9 %) and Hirschsprung's disease (8, 23.5 %), with diagnostic support being the most common objective (23, 67.6 %). Nearly half (15, 44.1 %) included model explainability tools, while performance metrics were heterogeneous, most often reporting accuracy (24, 70.6 %) and AUROC (21, 61.8 %). Twenty-two studies (64.7 %) had a high risk of bias, mainly due to small cohorts, poorly defined predictors, and outcome-informed assessments.

CONCLUSION

Our results highlight promising avenues of AI application in pediatric surgical pathology and identify the current gaps in model validation, performance, and clinical implementation.