LaserNet: Contactless Electrocardiogram Monitoring Using Structured-Light Defocused Speckle Imaging

In high-field magnetic resonance imaging (MRI), conventional electrocardiogram (ECG) monitoring is challenged by the need for skin-contact electrodes and its susceptibility to magnetohydrodynamic effects and scanner-related electromagnetic interference, which can compromise signal stability and patient comfort. To address this issue, we propose a contactless ECG waveform reconstruction framework based on structured-light-assisted defocused speckle imaging (SL-DSI). Specifically, we collect structured-light video data from 66 participants under both no-occlusion and ultra-flexible-coil conditions, from which camera-based seismocardiography (Camera-SCG) and high-frequency camera-based phonocardiography (Camera-PCG) signals are derived to capture complementary mechanical information. On this basis, we develop LaserNet, an end-to-end mechanical-to-electrical signal translation network that combines parallel multi-scale convolutions for local cardiac motion encoding, Mamba blocks for long-range temporal modeling, and a dilated causal convolution decoder for ECG waveform reconstruction. Compared with representative mainstream methods, the proposed method achieves a median temporal error below 16 ms for key cardiac events, together with a Pearson correlation coefficient of 90.6% and a root-mean-square error of 0.080 mV for waveform reconstruction. These results demonstrate the potential of the proposed system for contactless, continuous, and accurate ECG monitoring in high-field MRI environments.