Multimodality Multimask and Multitask Auto-Segmentation Network for Organs-at-Risk in Head and Neck Radiation Therapy.

Purpose: Accurate segmentation of head and neck organs-at-risk remains a critical challenge in radiation therapy planning, where current single-modality approaches often fail to address the inherent complexity of soft-tissue differentiation and interpatient anatomic variations. This study aims to develop a clinically robust auto-segmentation framework that synergistically integrates multimodal imaging features while optimizing computational efficiency. Methods and Materials: We present multimodality multimask and multitask auto-segmentation network (M3-Net), a triple-interlocked deep learning architecture featuring: (1) cross-modality fusion modules with attention-guided feature recalibration between computed tomography density maps and magnetic resonance imaging soft-tissue contrast; (2) a hierarchical multimask generator producing organ-specific, regional, and global masks through parallel encoding pathways; and (3) a dual-task learning mechanism combining segmentation with deformable image registration to establish voxel-level modality correspondence. The model was trained on 200 retrospective cases (160/20/20 split) with expert-reviewed contours from a tertiary cancer center, supplemented by 10 prospective cases for clinical validation. Results: M3-Net demonstrated significant improvements across 3 key dimensions: Efficiency: reduced inference time by 63.6% (548 ± 23 seconds vs 198 ± 15 seconds; Conclusions: M3-Net establishes new state-of-the-art performance for head and neck organs-at-risk segmentation, by simultaneously addressing accuracy-efficiency tradeoffs and modality discordance. The clinically validated workflow reduces contouring time by 75% while maintaining dosimetrically significant precision, enabling rapid adoption in adaptive radiation therapy protocols.