Abstract
<title>Abstract</title> <p>This paper addresses two challenges in multimodal medical image fusion: heterogeneous feature distribution discrepancies and adaptive cross-modal information integration. To tackle these, we propose MESFusion, a modality-specific enhancement and statistically guided adaptive fusion framework. The Modality-Aware Enhancement (MAE) module enhances complementary information from anatomical and functional modalities by using multi-scale gradient extraction and energy statistical modeling. For deep feature representation, the Spatial Gated Mamba (SGMam) module is introduced to capture long-range spatial dependencies and global contextual information. Additionally, the Statistical-Guided Adaptive Fusion (SDAF) module dynamically adjusts fusion behavior according to the first- and second-order statistical characteristics of feature channels. This enables effective integration of heterogeneous modal information. Extensive ablation studies and comparative experiments were conducted on three publicly available multimodal medical image fusion datasets, including CT-MRI, PET-MRI, and SPECT-MRI. n addition, an external PET-CT dataset (NAF-PROSTATE) was employed to evaluate the cross-dataset generalization capability of the proposed method. Experimental results demonstrate that MESFusion achieves competitive and well-balanced performance across multiple quantitative metrics while producing visually coherent fusion results with favorable structural preservation and information representation. These findings indicate the effectiveness and robustness of the proposed framework for multimodal medical image fusion. Our code will release at: https://github.com/bluemountain2025/MESFusion.</p>