AI imaging in pediatric oncology

Abstract

This thesis examined how artificial intelligence can extract clinically meaningful information from routine brain MRI scans of children with brain tumors. It demonstrated that these scans contain overlooked signals that extend beyond tumor size and can be quantified consistently with automated methods. The work introduced models that measure the temporalis muscle, a marker of physical resilience, and showed that reduced muscle thickness is associated with adverse outcomes. It also developed tools that quantify tumor growth trajectories and software that streamlines the review of automated segmentations in clinical settings. A second research line addressed the brain age and limited longitudinal data in pediatrics by estimating brain maturity from diffusion MRI and generating realistic synthetic follow-up scans to simulate developmental change. Collectively, the thesis showed that carefully validated AI systems can convert standard imaging into reproducible biomarkers that support earlier recognition of clinical risk, stronger evidence for decision making, and more transparent interpretation of pediatric neuro oncology data.

Ähnliche Arbeiten