Contrastive Learning Neural Network with Multimodal Physiological Signal Fusion for Early Detection of Cognitive Impairment

With the rapid advancement of artificial intelligence in healthcare, early detection of cognitive impairment based on physiological signals has emerged as a significant research focus. Existing methods often struggle with challenges such as strong modality heterogeneity, data scarcity, and difficulty in feature alignment when processing multimodal inputs. To address these issues, this paper proposes a contrastive learning neural network framework (MCL-Net) that fuses multimodal physiological signals, including EEG, ECG, and EDA. The model employs modality-specific encoders to extract deep temporal representations and introduces a cross-modal contrastive mechanism to enhance semantic consistency and discriminative capacity. Trained in an end-to-end manner, MCL-Net jointly optimizes self-supervised contrastive loss and classification loss, thereby improving generalization under limited data conditions. Experimental evaluations conducted on a synthesized multimodal physiological dataset demonstrate that MCL-Net outperforms several baseline models in terms of accuracy, F1-score, and recall, while maintaining high stability even with reduced training data. Ablation studies further confirm the effectiveness of the contrastive module and modality fusion strategy. This work contributes a novel approach at the intersection of self-supervised learning, multimodal modeling, and physiological signal processing, providing both theoretical and practical support for the development of intelligent cognitive impairment screening systems and deep representation learning from multimodal physiological data.