BQM-Transformer Multimodal Dynamic Physiological Fusion with Quantum Attention

Autism spectrum disorder (ASD) early intervention is constrained by existing models' inability to disentangle dynamic physiological signals from static features. This study proposes a Bio-Quantum Memory Transformer (BQM-Transformer) that integrates multimodal time-physiological features (EEG, heart rate, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{SpO}_{2}$</tex>) through three novel components: Bio-inspired Dynamic Positional Encoding (BDPE), Quantum-enhanced Attention Mechanism (QAM), and Memory-Augmented Feed-Forward Network (MA-FFN). BDPE dynamically modulates temporal representations using real-time heart rate and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{SpO}_{2}$</tex> fluctuations, resolving static encoding biases in transient physiological events. QAM leverages EEG signals to generate quantum-state vectors, enhancing sensitivity to ASD-specific cognitive fluctuations. MA-FFN incorporates LSTM gating to mitigate long-term dependency forgetting. Experimental results demonstrate state-of-the-art performance: BQM-Transformer achieves 95.83% accuracy in ASD interest prediction, 4.16% improvement over the benchmark Transformer model, And reveals a strong correlation between post-stimulus attention fluctuations and heart rate surges. Ablation studies validate the biological plausibility of each module. This work advances ASD research by providing an interpretable, high-precision framework for non-verbal intent understanding and personalized intervention.