Alzheimer's disease recognition based on multimodal image fusion

In response to the current challenges of low diagnostic accuracy for Alzheimer's disease (AD) and the weak ability of single-modal imaging to extract lesion feature information, a deep learning-based multimodal image fusion method for AD classification has been proposed. First, a novel residual network architecture is used to extract lesion features from three-dimensional images. Then, the improved residual network is employed as a feature extractor to separately extract image features from magnetic resonance imaging (MRI) and positron emission tomography (PET) scans. Afterward, the features from both modalities are fused and subsequently classified. Finally, to enable the fusion network to better capture the spatial relationships between dimensions and channels in three-dimensional medical images, a coordinate attention mechanism is introduced into the network structure. Experimental results show that the improved fusion network achieved an accuracy of 91.07% in the AD/mild cognitive impairment (MCI)/normal cognitive (CN) classification task. This represents a 7.14% improvement over the basic residual network, a 21.43% improvement over single-modal MRI-based methods, and a 12.5% improvement over single-modal PET-based methods. The improved fusion network demonstrates superior classification performance compared to the basic residual network in AD/CN, AD/MCI, and CN/MCI classification tasks, proving its effectiveness.