Cardiac 3D Mechanical and Electrical Signal Reconstruction via Defocused Speckle Imaging.

Abstract

Defocused speckle imaging (DSI) enables non-contact measurement of mechanocardiography (MCG) signals, such as seismocardiography (SCG) and gyrocardiography (GCG). As cardiac function is inherently governed by the coupling between mechanical motion and electrical excitation, this intrinsic relationship suggests that mechanical signals captured by DSI may further support the inference of electrical activities, i.e., the electrocardiogram (ECG). Therefore, we propose CardioNet, a physics-driven framework that leverages speckle motion signals together with physical model-based priors to reconstruct cardiac mechanical and electrical activities simultaneously. By coupling the MCG representations with the synchronized ECG, CardioNet captures the intrinsic electromechanical relationship of the heart, enabling non-contact and comprehensive cardiac monitoring. CardioNet was evaluated in a cross-subject study involving 20 laboratory subjects and 30 clinical patients, yielding Pearson correlation coefficients of 0.718/0.911 (SCG/ECG) in laboratory settings and 0.725/0.850 in clinical settings. Furthermore, it accurately captures fine-grained biomarkers, achieving mean timing errors of only 1.36/1.39 ms (lab) and 2.20/2.27 ms (clinic) for detecting aortic valve opening events in MCG signals and R-peaks in ECG signals, with similarly low timing errors for other hemodynamics-related markers. While promising for rhythm analysis, capturing subtle morphologies and spatial dynamics remains challenging. Future work will target 12-lead spatial reconstruction and broader pathological cohorts. To facilitate reproducibility, our dataset has been released at https://github.com/contactless-healthcare/CardioNet.