ACS Nano. 2026 Aug 25;20(33):23259-23269. doi: 10.1021/acsnano.6c04605.
ABSTRACT
The pulse wave is a vital physiological signal reflecting cardiovascular function. It carries a wealth of critical health information, playing a significant role in the early diagnosis and continuous monitoring of cardiovascular diseases. Although considerable progress has been made in pulse monitoring technologies in recent years, several key technical challenges remain unresolved: (i) signal acquisition is susceptible to motion artifacts, and (ii) high-precision monitoring equipment relies on professional medical instruments, resulting in elevated costs and hindering the realization of autonomous, long-term monitoring. Based on the traditional 2D convolutional neural network (CNN), the pulse wave sequence needs to be converted into a 2D image. The conversion process leads to the loss of critical potential time-domain features due to time-frequency transformation and significantly increases external system complexity and computational overhead. This work proposes a one-dimensional intelligent convolutional monitoring system based on a single convolutional transistor. The system adopts a hybrid architecture that integrates self-powered pulse sensing with externally biased convolutional-transistor computing. The system directly acquires human pulse signals through a high-sensitivity flexible sensor and employs an approximate one-dimensional convolutional neural network for in situ feature extraction and analysis, reducing the total area of devices by over 80%. Validated on a public physiological waveform data set, the system achieves 98.2% accuracy in pulse-related abnormal waveform classification. This study provides a highly integrated and low-power solution for wearable health monitoring devices, demonstrating significant potential for applications in the early screening of cardiovascular diseases.
PMID:42674437 | DOI:10.1021/acsnano.6c04605