Zhanyi Liu

High-sensitivity miniature optically pumped magnetometers (OPMs) enable noninvasive magnetocardiography (MCG) with flexible sensor configurations. Accurate magnetic source imaging (MSI) critically depends on precise registration between MCG measurements and anatomical computed tomography (CT) data, while registration errors can substantially degrade source localization accuracy. In this study, we systematically analyze the data acquisition and coordinate transformation processes to identify major sources of registration error in OPM-MCG systems. We show that existing approaches often overlook sensor pose deviations introduced by substituting the manufactured sensor array with an ideal design model during registration. To address this limitation, we propose an optical image–based sensor array calibration method using the local tangent plane distance (LTPD) measure and a sensor face filtering (SFF) algorithm. The proposed method directly extracts sensor positions and orientations without relying on predefined array parameters, thereby correcting errors caused by sensor assembly inaccuracies and array deformation. Experimental results demonstrate a mean position error of 0.53 mm and an orientation error of 0.84° for the sensitive axis. In addition, we investigate the effects of data volume and registration algorithm selection on transformation accuracy, revealing their critical roles in achieving high-precision registration. The optimized registration pipeline improves both accuracy and computational efficiency, facilitating reliable MSI for both rigid and wearable OPM-MCG systems.

Automated analytical techniques for magnetocardiography (MCG) are essential for diagnosing and predicting cardiovascular diseases. Clinically acquired MCG signals are often contaminated by various types of noise, which negatively impact subsequent signal analysis. However, traditional methods have limitations in denoising long-term MCG signals with complex spatial structures. We propose a high-precision, robust representation learning method based on skip connection multi-scale feature fusion (SkipDAEformer) for effectively removing random mixed noise in MCG signals. SkipDAEformer integrates attention fusion mechanisms into a basic denoising autoencoder to extract and fuse critical temporal and spatial information from each feature map, thus enhancing the model's ability to capture long-range dependencies and spatial features in MCG signals. Meanwhile, we further supplement and refine the semantic information for the feature maps through a global feature fusion method. By fusing multi-scale features from different skip connections, SkipDAEformer can learn more comprehensive representations of MCG signals, enabling the effective separation of clean signals from noise. Experimental results demonstrate that SkipDAEformer outperforms existing methods in denoising performance, channel consistency, feature consistency, and generalization ability and can be extended to a self-supervised learning framework. In actual noise reduction and diagnostic classification tasks, SkipDAEformer shows superior clinical acceptability and diagnostic value, potentially advancing MCG data analysis.

Recent advancements in optically pumped magnetometers (OPMs) have driven the development of magnetocardiography (MCG) systems, which provide high-quality signals and greater flexibility at a lower cost. While the OPM sensor configuration is more flexible, it introduces new challenges to the co-registration of MCG and computed tomography (CT), which directly impacts the spatial accuracy of source localization. In this study, we present an automatic co-registration method consisting of two main components: sensor localization and sensor-to-CT registration. Sensor positions and orientations are determined within 3D images reconstructed through optical scanning based on the bed registration unit. Abdominal surface matching is employed to automatically align anatomical images with sensor positions. The results demonstrate that the mean registration position and orientation errors across all sensors on the reference phantom are 0.49 mm and less than 0.2°, respectively, with a computational runtime of less than 10 seconds. Compared with existing methods involving manual procedures, such as moving digitizers to markers and removing sensors, the proposed co-registration method is more accurate and efficient.