AI in pediatric oncological surgery: Uncharted territories?

This thesis investigates the potential of deep learning (DL) to support clinical care in pediatric abdominal tumors, with a focus on MRI-based surgical planning. First, the current landscape of DL in pediatric oncology was assessed through reviews of imaging and genomic applications, revealing a promising but still immature field. Next, automated tumor segmentation was evaluated in Wilms tumor and neuroblastoma, demonstrating both the potential and limitations of current methods. Following this, key methodological factors influencing performance in Wilms tumor segmentation, including MRI input, dataset size, and tumor characteristics, were explored. Finally, automated Wilms tumor segmentation was prospectively tested within a clinical workflow, demonstrating its feasibility for creating 3D models and supporting future translation of DL into pediatric surgical oncology.