Prior-guided multi-expert consensus fusion for multi-center thyroid nodule classification

Abstract

Thyroid nodule classification in ultrasound imaging is challenged by entangled visual patterns and distribution shifts across multi-center data due to variations in devices, protocols, and gain settings. We propose a Prior-Guided Multi-Expert Consensus Fusion Network (PMCF) for robust cross-center classification. The Prior-Guided Routing Mechanism leverages structure-aware spatial priors from fixed operators and learnable filters to modulate deep features across multiple expert branches, encouraging complementary spatial responses and implicit feature disentanglement. The Consensus Fusion Mechanism models inter-expert agreement together with adaptive gating information to aggregate expert outputs, enhancing robustness under domain shifts. Experiments on three multi-center thyroid ultrasound datasets demonstrate that PMCF outperforms state-of-the-art methods in classification accuracy and generalization. Ablation studies confirm that prior-guided feature modulation improves representation diversity, while consensus-based fusion enhances prediction stability, highlighting the effectiveness of combining structured priors with expert collaboration for multi-center diagnosis. Code is available at https://github.com/guangguangLi/MultiExpert.