Non-Invasive Triboelectric Pulse Sensor and Regression-Type MLP-Mixer for Extracting Cardiovascular Physiological Information

This study constructs fabric electrodes on the surface of polyester fiber cloth using copper-nickel alloy and prepares SnCl<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">4</inf>-modified PVA fiber membranes through electrospinning technology, forming a single-electrode mode flexible triboelectric pulse sensor. Relying on the flexible characteristics of the electrodes and the high triboelectric performance of the fiber membranes, this sensor can effectively detect pulse fluctuations. Some traditional neural networks struggle to extract and analyze the physiological information contained in the pulse with high precision, and the training process of these networks often takes too long. To achieve better testing results, this article conducts a regression analysis of pulse signals based on the MLP-Mixer (Multi-Layer Perceptron Mixer) neural network, ultimately outputting predictions for four aspects: heart rate (HR), the time difference between P1 and P3 (PPT), reflection index (RI), and the systolic upstroke time (UT). Experimental results indicate that the proposed method has a faster convergence speed and accuracy compared to most related methods.