Molecular-Information-Guided Framework for Head and Neck Tumor and Lymph Node Segmentation in PET/CT Images

Jung Jee Yoon, Min Jeong Cho, Jae Sung Lee

Keywords: HECKTOR2025, MICCAI2025, Segmentation, Challenge, Deep Learning, 3D CT, 3D PET

Abstract: Accurate segmentation of Head and Neck (H\&N) tumor in PET/CT images is essential for diagnosis and treatment planning, yet remains challenging due to heterogeneous lesion morphology and variable physiological uptake. In this paper, we propose 3D U-Net-based architecture with a PET-guided Multimodal Spatial Attention Module (PSAM). PSAM consumes the PET image to generate a spatial attention map that gates encoder skip features. Furthermore, we use Squeeze-and-Excitation (SE) Normalization, which dynamically recalibrates channel responses and improves multimodal fusion. Notably, we introduce a novel molecular-information-guided preprocessing pipeline, which minimizes the size of input data. This design enables modality-aware modeling and aims for robust, generalizable segmentation of both primary Gross Tumor Volume (GTVp) and nodal Gross Tumor Volumes (GTVn) across multi-center PET/CT data. We obtain mean Dice of 0.6073 with class-wise Dice 0.7133 (GTVp) and 0.5013 (GTVn) on the validation set.

Submission Number: 9