arXiv:2606.14072v1 Announce Type: new 
Abstract: Accurate pediatric brain tumor segmentation remains challenging due to limited annotated data, heterogeneous imaging phenotypes, diffuse tumor boundaries, and class imbalance across tumor subregions. Here, we present a two-stage deep learning framework for improving multi-modal pediatric brain MRI segmentation and clinical interpretation. First, we evaluate 3D Res U-Net and Swin-UNETR baselines on BraTS-PEDs MRI scans, using four co-registered modalities to predict tumor core, whole tumor, and enhancing tumor regions. Second, we introduce diffusion-based refinement models conditioned on coarse Swin-UNETR predictions, including a 3D DDPM refiner and MedSegDiff. Conditioning substantially improves diffusion stability and performance, particularly for enhancing tumor boundary segmentation. Conditioned MedSegDiff achieves the strongest boundary agreement with the lowest HD95. Finally, predicted tumor volumes and representative segmentation overlays are integrated with a multimodal language model to generate structured radiology-style reports. Together, our results suggest that coarse-to-refined diffusion segmentation can improve pediatric tumor boundary delineation and support end-to-end interpretable AI-assisted neuro-oncology workflows.

تقدم دراسة جديدة إطار عمل للتعلم العميق من مرحلتين يهدف إلى تحسين تقسيم أورام الدماغ لدى الأطفال وتفسير التصوير بالرنين المغناطيسي سريريًا. يقيم الإطار نماذج 3D Res U-Net وSwin-UNETR على مسح MRI BraTS-PEDs، ويقدم نماذج تحسين قائمة على الانتشار لتحسين تقسيم حدود الأورام.

Un nuevo estudio presenta un marco de aprendizaje profundo en dos etapas destinado a mejorar la segmentación de tumores cerebrales pediátricos y la interpretación clínica de las resonancias magnéticas. El marco evalúa los modelos 3D Res U-Net y Swin-UNETR en escaneos de MRI BraTS-PEDs, introduciendo modelos de refinamiento basados en difusión para mejorar la segmentación de los bordes de los tumores.

Une nouvelle étude présente un cadre d'apprentissage profond en deux étapes visant à améliorer la segmentation des tumeurs cérébrales pédiatriques et l'interprétation clinique des IRM. Le cadre évalue les modèles 3D Res U-Net et Swin-UNETR sur des IRM BraTS-PEDs, introduisant des modèles de raffinement basés sur la diffusion pour améliorer la segmentation des contours des tumeurs.

A new study presents a two-stage deep learning framework aimed at enhancing pediatric brain tumor segmentation and clinical interpretation of MRI scans. The framework evaluates 3D Res U-Net and Swin-UNETR models on BraTS-PEDs MRI scans, introducing diffusion-based refinement models to improve tumor boundary segmentation.