Evidential regression-based blood glucose detection using waveform features

<div>As one of the necessary diabetes control and treatment methods, the photoacoustic blood glucose detection technology has great potential due to its deep detection depth and low interference from stray light. Previous research mainly focused on improving the detection capabilities of hardware systems and ignored the exploration of the physical meaning of the signal itself. We analyzed the characteristics of the signal amplitude decay in the photoacoustic signal and employed the forced damping vibration equation to model the signal waveform. A new waveform feature was constructed to describe the amplitude attenuation rate. Moreover, facing low accuracy of blood glucose prediction in the case of small data, we proposed a stable and effective blood glucose detection combining time-frequency feature and waveform features with evidential regression. Finally, in human tissue and glucose solution experiments, the minimum error is achieved 1.02±0.71 mg/dL and 13.28±10.33 mg/dL, respectively.</div><div><br></div>