Artificial Intelligence Applications in Pediatric Oncology Imaging: A Systematic Review of Diagnostic, Prognostic, and Therapeutic Evaluation Tools

Abstract

Artificial intelligence (AI) has emerged as a transformative tool in pediatric oncology imaging, enhancing diagnostic accuracy, prognostic evaluation, and treatment monitoring. This systematic review synthesizes evidence from 22 studies to evaluate AI applications—including machine learning (ML) and deep learning (DL)—in tumor classification, segmentation, radiogenomics, and treatment response assessment. Key findings reveal that convolutional neural networks (CNNs) and radiomics pipelines achieve expert-level performance in classifying pediatric brain tumors (e.g., medulloblastoma, pilocytic astrocytoma) with AUCs >0.95 and Dice scores up to 0.96 for segmentation tasks. AI also shows promise in predicting molecular markers (e.g., MYCN, BRAF) and automating longitudinal tumor volume measurements using frameworks like RAPNO. However, challenges persist, such as data scarcity due to the rarity of pediatric cancers, heterogeneity in imaging protocols, and limited external validation. Ethical concerns regarding data privacy and model interpretability further hinder clinical adoption. Multi-institutional collaborations (e.g., Children’s Brain Tumor Network) and explainable AI (XAI) tools (e.g., Grad-CAM) are proposed to address these limitations. Future research should prioritize large-scale, prospective studies, standardized reporting frameworks (e.g., TRIPOD-AI), and integration of AI into clinical workflows. While AI demonstrates significant potential to revolutionize pediatric oncology imaging, overcoming current barriers is essential for robust, generalizable, and ethically sound implementations.

Ähnliche Arbeiten