Reconstruction of planar Magnetocardiography via spatiotemporal overlapping block sparse Bayesian learning

Magnetocardiography (MCG) is a noninvasive technique for measuring cardiac magnetic fields. The limited number of sensors and occasional signal corruption in individual sensors often results in suboptimal array density in practical MCG systems. This study proposes a spatiotemporal overlapping block-sparse Bayesian learning (STOBSBL) algorithm to reconstruct high-density sensor array signals from low-density measurements. The STOBSBL algorithm incorporates a Gaussian overlapping block sparse structure for spatial prior modeling, coupled with a sliding window mechanism for localized temporal magnetic field reconstruction. Extensive numerical simulations demonstrated that STOBSBL outperformed the benchmark algorithms under diverse conditions (various noise levels and missing sensor counts), thus achieving optimal robustness. Under zero-noise conditions with four missing sensors, STOBSBL achieved exceptional performance metrics: a spatial correlation coefficient of 0.9984 (relative error of 0.0452) and a waveform correlation coefficient of 0.9979 (relative error of 0.0282), thus demonstrating superior reconstruction fidelity. A comparative analysis of sparse basis matrices reveals that the discrete cosine transform (DCT) exhibits superior performance for high-density array reconstructions. Experimental validation using a 32-channel optically pumped magnetometer (OPM)-MCG system further confirmed the efficacy of STOBSBL. This study establishes a novel technical framework for high-precision MCG measurements using sparse sensor arrays.