基于多头注意力机制的青少年早期抑郁症检测方法研究OA

Objective To propose a neural network-based classification method utilizing a multi-head attention mechanism,so as to realize early detection of depression in adolescents.Methods Firstly,a serial hybrid model was designed by combi-ning a convolutional neural network(CNN)with a long short-term memory(LSTM)network to address the issues of gradient vanishing and feature loss that single-network models encountered when used for processing multimodal long-sequence data;secondly,a multi-head attention mechanism was introduced to construct the CLAL model,and through parallel subspace learning differentiated weights were automatically assigned to modalities such as EEG,ECG and Audio,effectively capturing the subtle manifestations of depression across various physiological signals.To validate the effectiveness of the CLAL model for detecting early-stage depression in adolescents,ablation experiments were conducted using the public dataset from the 9th National Undergraduate Biomedical Engineering Innovation Design Competition in 2024 and comparison analyses were carried out between the model and the transformer-based multimodal spatio-temporal attention transformer approach for depression detection(DepMSTAT)model and the multimodal fusion framework model based on the audio,video and text fusion-three branch network(AVTF-TBN).Results The CLAL model achieved an accuracy of 0.907,a precision of 0.911,a recall of 0.907 and an F1 score of 0.908,all of which outperformed those of the DepMSTAT and AVTF-TBN models;ablation experiment results indicated that under multimodal(EEG+ECG+Audio)experimental conditions the CLAL model had a standard deviation of 0.000 2,demonstrating high stability compared with single-modal models.Conclusion The proposed method shows high accuracy and reliability,offering an effective approach for identifying early signs of depression in adolescents.[Chinese Medical Equipment Journal,2026,47(3):1-8]