Convolutional Neural Network-Self-Attention Mechanism Enhanced Near-Infrared: Non-Invasive Breakthrough for...

Alzheimer's disease (AD) and vascular dementia (VaD) are two common forms of dementia. Differentiating between them is challenging due to the lack of clear clinical and auxiliary test differences. In this study, we developed a novel diagnostic method combining near-infrared spectroscopy with a convolutional neural network and self-attention mechanism (CNN-SAM). The CNN-SAM model, which integrates the self-attention mechanism to highlight important spectral features, outperformed other models with 99.3% accuracy. Data pre-processing, feature extraction, and parameter optimization further enhanced the model's performance. Visualization using the self-attention mechanism revealed key spectral bands at 1364 and 1484 nm as crucial for distinguishing AD and VaD. This approach offers a rapid, non-invasive, and accurate method for the diagnosis of AD and VaD, potentially advancing clinical practice.