Towards Reliable Pediatric Brain Tumor Segmentation: Task-Specific nnU-Net Enhancements

2025-11-01

AI Summary

Addressing key challenges in pediatric brain tumor segmentation from multiparametric MRI—including limited annotated data, substantial anatomical variability, and cross-center imaging heterogeneity—this work proposes a customized nnU-Net extension. The method introduces a widened residual encoder integrated with Squeeze-and-Excitation (SE) attention, replaces standard convolutions with 3D depthwise separable convolutions to enhance computational efficiency and feature discriminability, incorporates a lesion-specific regularization term, and adopts small-scale Gaussian weight initialization to improve training stability. Additionally, refined post-processing is applied. Evaluated on the BraTS 2025 Task-6 validation set, the approach achieves state-of-the-art performance: lesion-level Dice scores of 0.759 (CC), 0.967 (ED), 0.826 (ET), 0.910 (NET), 0.928 (TC), and 0.928 (WT). These results demonstrate significant improvements in robustness and clinical applicability for automated pediatric brain tumor segmentation.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Calibration & Uncertainty Quantification
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Accurate segmentation of pediatric brain tumors in multi-parametric magnetic resonance imaging (mpMRI) is critical for diagnosis, treatment planning, and monitoring, yet faces unique challenges due to limited data, high anatomical variability, and heterogeneous imaging across institutions. In this work, we present an advanced nnU-Net framework tailored for BraTS 2025 Task-6 (PED), the largest public dataset of pre-treatment pediatric high-grade gliomas. Our contributions include: (1) a widened residual encoder with squeeze-and-excitation (SE) attention; (2) 3D depthwise separable convolutions; (3) a specificity-driven regularization term; and (4) small-scale Gaussian weight initialization. We further refine predictions with two postprocessing steps. Our models achieved first place on the Task-6 validation leaderboard, attaining lesion-wise Dice scores of 0.759 (CC), 0.967 (ED), 0.826 (ET), 0.910 (NET), 0.928 (TC) and 0.928 (WT).

Problem

• Improving pediatric brain tumor segmentation accuracy
• Addressing data limitations and anatomical variability challenges
• Enhancing nnU-Net for pediatric high-grade glioma MRI

Innovation

• Widened residual encoder with squeeze-and-excitation attention
• 3D depthwise separable convolutions for efficient processing
• Specificity-driven regularization term with Gaussian initialization