Automated analytical techniques for magnetocardiography (MCG) are crucial for the diagnosis and prediction of cardiovascular diseases. However, labeled data sparsity and susceptibility to interference remain significant challenges in MCG research. Moreover, the complex temporal and spatial features of MCG signals pose additional challenges for deep learning models. Learning generic and meaningful representations from MCG signals via self-supervised pre-training is an effective method to alleviate these issues. To enhance feature extraction performance of the pre-trained model, we propose a novel self-supervised learning framework named CrossLGNet. First, we generated two distinct augmented views of the MCG signals using temporal-and channel-domain augmentation and CrossLGNet was used to learn the local and global feature representations of these augmented views. The local prediction module performs the local cross-view prediction task by extracting contextual information from past and future local views, we introduced a cross-spatio-temporal attention fusion module into local prediction to facilitate the capability of the model to extract temporal and spatial features. The global comparison module learns the overall discriminative representation of the different augmented views in parallel. We comprehensively evaluated CrossLGNet based on multiple downstream tasks to demonstrate its advantages in model performance, label efficiency, robustness, and computational complexity. Notably, CrossLGNet achieved superior performance with limited labeled data and noise interference, which was crucial to improving the clinical application value of MCG and also highlighted the potential of self-supervised learning in the automatic analysis of MCG signals.

Surface electromyography (sEMG) signals are characterized by low amplitude, nonlinearity, non-stationarity, and susceptibility to various types of noise during acquisition. Although signal decomposition-based preprocessing methods are widely used, the performance of existing algorithms heavily depends on mode selection strategies, leading to suboptimal denoising outcomes under low signal-to-noise ratio(SNR) conditions. These limitations significantly restrict the broader application of sEMG. In this paper, a hybrid denoising framework integrating variational mode decomposition (VMD) and refined composite multiscale dispersion entropy (RCMDE) is proposed. Based on the variational mode functions(VMFs) derived from VMD, an RCMDE-weighted VMF screening mechanism is established to effectively identify noise-dominant components and suppress spectral leakage. Subsequently, local mean decomposition(LMD) and wavelet thresholding(WT) are combined to achieve superior denoising performance. Simulations on synthetic sEMG and real sEMG signals, along with gesture classification experiments, demonstrate that the proposed algorithm outperforms conventional VMD and other mode selection methods, such as those based on correlation coefficient or multiscale dispersion entropy, in terms of SNR and root mean square error(RMSE), particularly under low-SNR scenarios. This method contributes to improving gesture classification accuracy and provides an efficient preprocessing solution for sEMG signals, which lays a foundation for its applications in medical diagnosis and human-computer interaction.

The availability and distribution of isolated X-ray pulsars suitable for navigation are limited in space. In contrast, there is a considerable number of binary pulsars available for navigation, albeit with the added complexity of accounting for the orbit of the binary pulsar system. Directly utilizing binary pulsars for spacecraft navigation has significant systematic biases. Therefore, to establish a measurement model applicable to both binary pulsars and isolated pulsars, effectively suppressing time-varying systematic biases and enhancing navigation accuracy, this paper introduces a pulsar navigation method based on the phase and Doppler frequency shift of binary pulsars. Initially, we formulated a navigation measurement model considering systematic biases, leveraging the phase and Doppler frequency shift of binary pulsars. Subsequently, a detailed analysis of systematic biases was conducted. Recognizing the characteristics of systematic biases, we further established a binary pulsar navigation measurement model with sequential difference. For deep space spacecraft, the time-varying system bias amplitude is large, but it changes slowly during the filtering period. Therefore, the proposed binary pulsar navigation measurement model with sequential difference effectively suppresses the majority of systematic bias effects. The effectiveness of the proposed method is demonstrated through 100 Monte Carlo simulation trials based on the Tianwen-1 Mars mission trajectory. The results show that the proposed method achieves a position error of approximately 824 m and a velocity estimation accuracy better than 1 m/s. Further analysis indicates that the remaining errors are primarily influenced by measurement noise and process noise, suggesting that reducing noise levels or incorporating more accurate models could further improve navigation performance.

Objective Emerging clinical research evidence has substantiated magnetocardiography (MCG) as a promising non-invasive diagnostic tool for cardiovascular assessment. This meta-analysis was designed to comprehensively evaluate the current diagnostic efficacy of MCG for coronary artery diseases (CAD) and myocardial ischemia. Methods PubMed, Web of Science, Embase, and the Cochrane Library databases were searched for English-language studies on MCG diagnosing CAD or myocardial ischemia published from the time of library construction to Feb 4, 2026. Bivariate random-effects models were used to calculate pooled estimates for overall and subgroup analyses. Results Thirty-six studies from eight countries published between Mar 2003 and Feb 2026 were analyzed, with a total of 7055 subjects. MCG had an overall sensitivity of 85% (95% CI, 82–87%), specificity of 80% (95% CI, 75–84%), and area under the curve (AUC) of summary receiver operating characteristic of 0.90 (95% CI, 0.87–0.92). The pooled sensitivity for CAD and myocardial ischemia were similar (84% and 82%), while the pooled specificity for CAD was higher (79% vs. 62%). Of these, for non-ST-segment elevation acute coronary, the pooled sensitivity and specificity were 86% and 68%, respectively. ST-segment fluctuation score was the most predominant parameter (AUC: 0.92), but its pooled sensitivity was low at 79% (95% CI, 67–87%). MCG based on superconducting quantum interference device outperformed optically pumped magnetometer with pooled specificity of 82% and 47%, respectively. Conclusion MCG shows high sensitivity in the diagnosis of CAD and myocardial ischemia, but the specificity is heterogeneous. It is needed to continuously evaluate the applicability of MCG in clinical settings.

Background:Non-ST-segment elevation acute coronary syndrome (NSTE-ACS) is a leading cause of acute chest pain in clinical practice. Magnetocardiography (MCG) is a non-invasive and rapid functional imaging technique with high sensitivity to early, subtle electrophysiological changes associated with myocardial ischemia. Aims:To develop and validate a machine learning (ML)-based diagnostic model for NSTE-ACS using MCG-derived features. Study Design:Retrospective cohort study. Methods:Patients presenting with acute chest pain and admitted between September 2023 and May 2024 were consecutively enrolled. Pretreatment cardiac magnetic signals were collected using a 36-channel optically pumped magnetometer-based MCG system. A total of 13 feature categories (188 parameters) were extracted from the ST segment and T wave. Three feature selection methods [Boruta, least absolute shrinkage and selection operator (LASSO), and maximum relevance minimum redundancy], along with hyperparameter tuning and unbiased performance estimation for five ML algorithms, were implemented within a nested cross-validation (CV) framework. Model performance was assessed using the area under the curve (AUC). The optimal model was further validated in an independent test set. SHapley Additive exPlanations (SHAP) were used to interpret the final model. Results:A total of 578 patients were included (366 with NSTE-ACS and 212 without NSTE-ACS). The support vector machine (SVM) model, based on nine LASSO-selected features, demonstrated the best performance, achieving an AUC of 0.91 ± 0.01 in nested CV. In the independent test set, the model achieved an AUC of 0.89 (95% confidence interval: 0.81-0.95), with an accuracy of 0.84, sensitivity of 0.89, and specificity of 0.77. Exploratory subgroup analyses showed consistent performance across age, sex, body mass index, and comorbidity groups. SHAP analysis identified the minimum magnetic field strength at the T-peak time (T_min_mag) as the most influential predictor. Conclusion:The SVM-based MCG model showed strong potential as an auxiliary tool for identifying NSTE-ACS. Its application may improve chest pain management and reduce misdiagnoses.

To address the issue in traditional SINS/DVS integrated navigation methods that is required a spacecraft to simultaneously carry three sensors to observe three navigation stars for comprehensive velocity error correction,an inertial/time-segmented astronomical Doppler velocity integrated navigation method based on sensor maneuvering(SINS/TS-DVS)was proposed.By periodically changing the optical axis pointing through sensor maneuvering,DVS measurements in different directions were acquired.This method ensured navigation accuracy while reducing the burden on the spacecraft.Simulation results demonstrate that the position errors of the SINS/TS-DVS integrated navigation in longitude,latitude,and altitude directions were significantly improved compared to the traditional Strapdown Inertial Navigation System(SINS)and the SINS/DVS method observing only one star.It can effectively enhance spacecraft positioning accuracy.

ObjectiveTo investigate the neuroelectrophysiological characteristics of children with obstructive sleep apnea (OSA) accompanied by attention deficit hyperactivity disorder (ADHD)-like symptoms using attention network test (ANT) task-state electroencephalography analysis.MethodsAll participants completed the Attention Deficit Hyperactivity Disorder rating scale. Electroencephalography data were collected using 32 leads while the participants performed the ANT, followed by polysomnography.ResultsOf the 87 children, 21 were in the control group, 56 were in the OSA group, and 10 were in the OSA accompanied by ADHD-like symptoms group (OSA-co-ADHD). Each group had similar response time and accuracy for the different ANT conditions. At the O2 electrode, compared with the controls, the latency of the P1 component in the OSA group was shorter (P < 0.05). During the alerting network stage, in the PZ electrode, the beta band energy of the OSA-co-ADHD group was greater than that of the control group (Z   -3.067, P = 0.006). During the executive control network stage, at the CZ electrode, the OSA-co-ADHD group exhibited higher alpha band energy than the OSA group (P < 0.05). During the alerting and orienting network phases, the connections of multiple brain regions changed (all P < 0.05), while the executive control network remained unchanged.ConclusionExcessive activation of the alerting network, reorganization of the orienting network in the brain, and compensation for inhibitory control during the process of the executive control network may be important characteristics of children with OSA accompanied by ADHD-like symptoms.

Optically pumped magnetometer magnetoencephalography (OPM-MEG) enables the study of pediatric neurodevelopment during naturalistic tasks. Here, we investigated functional connectivity (FC) changes in 30 typically developing (TD) children and 10 children with developmental language disorder (DLD), aged 4-7 years, during a rest task ("Inscapes" video) and a story task (short narrative videos). Both groups exhibited similar power modulation (δ enhancement, α, β, and low-γ suppression) and decreased network modularity when transitioning from rest to story processing. However, children with DLD displayed increased high-γ network modularity, indicating a more segregated neural architecture. During the story task, TD children displayed enhanced low-γ FC within a left-hemisphere network with hubs in temporal, parietal, central, and occipital regions, whereas children with DLD showed weaker connectivity. These findings suggest that altered γ-band network organization may be a core neural signature of DLD and highlight the potential of OPM-MEG in pediatric neurodevelopmental disorders.

The strapdown inertial navigation system (SINS) suffers from cumulative error growth, where velocity and acceleration drifts cause position errors to grow quadratically. While multi-sensor fusion using Doppler Velocity Sensors (DVS) can correct these drifts, traditional methods often assume a uniform radial velocity across the solar disk, thereby ignoring the differential surface velocities caused by solar rotation. This mismatch introduces a significant unmodeled system bias with a mean value of approximately 1.45 km/s and an upper bound of up to 2.07 km/s, which is orders of magnitude larger than the standard measurement error (~1 m/s). Such a dominant deterministic error leads to severe filter divergence, undermining the reliability of the entire integrated navigation system. This mismatch introduces unmodeled system bias, degrading filter performance. To address this issue, this paper proposes a solar disk differential velocity method within a multi-source fusion framework. By constructing a velocity measurement model that reflects the Sun's actual geometric and rotational characteristics, this method effectively eliminates the deterministic system bias. Combined with dual-star Doppler measurements, the method enhances the observability of 3D velocity errors. Simulation results show that the proposed method significantly improves navigation accuracy compared to the traditional method. Position errors are reduced from kilometer-level magnitudes to the range of tens to hundreds of meters, with velocity errors remaining within 10-2-10-1 m/s.

Source imaging algorithms have been widely used to localize functional and lesion areas. Brain source reconstruction is limited by complex experimental environments (noise interference, distributed brain activity, acquisition systems, etc.), and range estimation is not accurate. This study proposes a variational sparse source imaging method based on the time basis matrix (VSSI-TBM) algorithm. VSSI-TBM permits the source spatial signal to consist of several temporal basis functions by using low-rank decomposition to extract effective signals. In a compressed space, mixed-norm constraints and a cortical source variation operator ensure spatial sparsity and smoothness. In clinical examinations or research, other a priori information regarding brain activity may be available. VSSI-TBM using lead field guide constraints can further enhance the reconstruction results. The simulation results demonstrate the robust performance of VSSI-TBM in environments with a low signal-to-noise ratio (SNR), large sources (>11 cm2), and multiple sources. Additionally, integrating prior information enhances the imaging performance in complex environments. The algorithm is evaluated using an open-source dataset and an optically pumped magnetometer-based magnetoencephalography (OPM-MEG) system with a noisy 30-channel uniform layout. The results reveal a strong robustness of the spatial range reconstruction. Moreover, the combination of prior information effectively improves the imaging performance of the OPM-MEG system.

Brain's phase-unlocked rhythmic oscillatory activity typically induces synchronized discharges in large-scale neuronal populations. Accurate imaging of oscillatory sources is crucial for brain science and brain disease research. However, traditional brain source imaging methods struggle with the signals for locking phase and time, but handle the time-locked phase-locked signals lacking phase locking with difficulty. To address these challenges, the time-frequency multiple sparse priors Champagne (TF-MSPC) approach is proposed. By leveraging data-driven pre-segmentation (DDP) and incorporating Bayesian learning with a diagonal noise variance structure, TF-MSPC improves coherence matrix estimation accuracy. Additionally, TF-MSPC introduces priors from different modalities through matrix rearrangement encoding, further improving coherence matrix estimation accuracy and constructs a frequency-domain beamformer based on the averaged estimation of cross-trial coherence matrices, thereby overcoming the attenuation of non-phase-locked signals caused by direct time-domain averaging. TF-MSPC outperforms other benchmark algorithms in both numerical simulations and optically pumped magnetometer magnetoencephalography (OPM-MEG)-fMRI bilateral movement experiments. Notably, under low signal-to-noise ratio and multi-source conditions, TF-MSPC effectively reconstructs transient neural activities across various frequency bands and mitigates energy leakage due to correlated sources. These findings underscore the strong potential and practicality of TF-MSPC for reconstructing neural oscillation activities.

To improve the accuracy and stability of X-ray pulsar time-delay estimation for multi-scenario celestial remote sensing and navigation, this paper proposes a time-delay estimation method based on a waterfall-plot multi-criteria framework and develops an end-to-end simulation framework for multi-scenario applications. First, a pulsar profile waterfall-plot model is built, and principal component analysis is performed to characterize candidate periodic structures. The contribution rate of the principal eigenvalue is used to describe the overall significance of the candidate period, and the projection variance of the first principal component is used to measure the prominence of the candidate pattern in the principal subspace. Second, support vector regression is used to fit the peak track of the waterfall plot, and a regression slope is used to describe the geometric stability of the candidate period. These three indicators are fused for pulsar period and time-delay estimation. Tests based on Insight-HXMT satellite observation data show that, compared with the χ2 and Z2 test methods, our method improves time-delay estimation accuracy by 68.68% and 50.43%, respectively. Multi-scenario navigation simulations indicate positioning improvements of approximately 0.83 km, 3.04 km, and 1.05 km in the Earth-orbiting, Earth–Moon transfer, and Mars approach scenarios, respectively. These results suggest that the proposed framework can improve pulsar time-delay estimation and may provide useful measurement support for celestial remote sensing and navigation.

To further improve the accuracy and speed of real-time dynamic estimation of X-ray pulsar periods, this paper proposes a pulsar period estimation model based on the interlayer phase difference (IPD) of the fast folding algorithm (FFA) and the weighted Z2 (WZ) test. This paper adopts a staged estimation strategy and divides the pulsar period estimation into a fast initial estimation stage and a local refinement search stage. First, in the fast initial estimation stage, an FFA IPD model based on the relationship among phase, time, and period is established. The interlayer phase is used to directly perform a single initial estimation of a large range of periods, thereby improving the period estimation speed. Second, in the local refinement search stage, the response coverage index is proposed for the Z2 test function. The WZ test function is constructed to perform a refinement test on the local candidate period to improve the period estimation accuracy. Meanwhile, for the PSR B0531+21 source, we conducted ablation tests, analyzed influencing factors and simulation performance of the proposed method, and validated its practical application performance using Neutron Star Interior Composition Explorer observation data. We also performed generalization performance tests on other sources such as PSR B0540-69 and SMC X-1. The results show that our method has significant advantages compared to several existing estimation methods. Specifically, for the PSR B0531+21 source, compared to the integrated chi 2 test method, our method improves estimation accuracy by 50.21% and reduces computational time by 73.47%.

The development of magnetoencephalography (MEG) based on optically pumped magnetometers (OPMs) has accelerated the evolution of brain imaging toward portable and movable configurations. Source localization using OPM-MEG plays a crucial role in both neuroscience and clinical applications. However, few studies have systematically evaluated localization performance under realistic measurement conditions that account for potential sensor errors. In this study, we simulated empirical OPM-MEG data with a biologically plausible model of ongoing brain activity to investigate the effects of calibration errors, specifically gain error, crosstalk, and angular misalignment, on the localization performance of three OPM array configurations: single-axis, dual-axis, and tri-axis sensors. We also compared four source inversion algorithms: Multiple Sparse Priors (MSP), Empirical Bayesian Beamformer (EBB), Minimum Norm (IID), and LORETA (COH). Our results indicate that gain error had minimal impact on localization accuracy, whereas crosstalk and angular error significantly degraded performance. These findings underscore the importance of OPM array calibration using reference magnetic fields, particularly for dual- and tri-axis configurations. Among the inversion methods tested, MSP demonstrated relative robustness to calibration errors and achieved the best overall performance, making it the most recommended approach. Furthermore, analysis of the relationship between cortical anatomy and localization accuracy revealed that deep neural sources are more susceptible to gain errors and thus require particular attention. To achieve accuracy comparable to an ideal OPM-MEG system, crosstalk should be kept below 2

In X-ray pulsar-based navigation, phase and Doppler frequency estimation based on the maximum likelihood estimation and grid search methods is widely used in practical missions and theoretical analysis. However, due to the non-convexity of the objective function and the inefficiency of the grid search method, a key challenge is how to achieve fast estimation of phase and Doppler frequency while maintaining accuracy. To address this issue, the fast and high-precision estimation of phase and Doppler frequency based on prior information and non-convex optimization is proposed in this paper. First, leveraging the prior state information of the spacecraft, an enhanced on-orbit phase model is established by considering a reference time at any given moment. Then, the statistical properties of the parameters to be estimated are analyzed, and the corresponding prior probability model is constructed, using the Bayesian estimation model as the objective function. Finally, incorporating non-convex optimization theory, the Nesterov-adaptive moment estimation is employed to automatically adjust the step size of the quasi-Newton algorithm. Simulation and experimental results demonstrate that the proposed method achieves rapid convergence with high precision, effectively balancing real-time performance and estimation accuracy compared to traditional phase and Doppler frequency estimation methods. Using the Crab pulsar as the primary case study, when the observation duration is 1800 s and the detector area is 30 cm2, the proposed method reduces the running time by 99.91

Speech perception relies on distributed neuronal populations, yet traditional decoding often utilizes static strategies that overlook inherent temporal dependencies and dynamic regulation. Therefore, we introduce the concept of system identification into multivariate decoding. By modeling brain response characteristics through time-lagged regression between speech stimuli and neural responses, we propose a temporal response function-based representational similarity analysis method (TRF-RSA). This method models the dynamic time-lag mapping from continuous stimulus features to neural responses, effectively separating stimulus-driven coherent activity from high-dimensional noise. More importantly, it elevates the analytical perspective from static comparisons of raw signals to dynamic trajectories in weight space. We conducted an auditory experiment and incorporated high spatiotemporal resolution optically pumped magnetometer magnetoencephalography magnetoencephalography (OPM-MEG) with electroencephalography (EEG). The results showed that TRF-RSA significantly enhanced the pattern similarity between speech sounds and the ability to discriminate between pattern differences. Furthermore, it revealed stronger similarities elicited by biological vocalizations, indicating a preference in the brain for these species-specific sounds. Source localization results not only confirmed the classical speech perception network but also revealed activation in limbic and deep brain regions. By modeling the relationship between stimulus features and neural responses, TRF-RSA dynamically quantified the spatiotemporal patterns of stimulus-driven neural activity, improving the sensitivity of representational pattern decoding during the encoding process. These findings suggest that this method is a sensitive neuroimaging tool that not only advances our understanding of the spatiotemporal dynamics of speech processing but also provides a new reference for population dynamics research.

The inversion of asteroid lightcurve data is formulated as a nonlinear optimization problem in a high-dimensional parameter space, which typically yields a large number of virtual solutions. These virtual solutions are characterized by small photometric chi-square values but large errors. To scientifically evaluate such virtual solutions, we introduce the concept of an insensitive subspace, defined as the subspace within which virtual solutions produce low chi-square values for lightcurve fits while still exhibiting large errors. To efficiently estimate the insensitive subspace, we propose an insensitive subspace search algorithm based on the Kepler optimization algorithm. First, the algorithm obtains an initial set of insensitive deviation vectors through singular value decomposition. Thereafter, vectors corresponding to large singular values are discarded, and those associated with smaller singular values are retained to construct a low-dimensional search space. Subsequently, the algorithm applies KOA to explore additional insensitive deviation vectors. Finally, these vectors are used to construct the insensitive subspace. In addition, we theoretically prove the chi-square constraint theorem for the insensitive subspace. Experimental results demonstrate that, compared to the dimension-ratio-based observability analysis method, our algorithm increases the dimensionality of the insensitive subspace from 4 to 8-9, and in some cases even up to 12, thereby significantly enhancing the ability to characterize observability, with stable performance across different parameter settings and optimization strategies. These findings provide a novel methodological approach for the observability analysis of asteroid physical parameters.

Magnetoencephalography (MEG) offers high temporal and spatial resolution for clinical and neuroscience applications. Traditional sensor registration methods depend on complex point cloud reconstruction, which is error-prone, labor-intensive, and lacks adaptability. Recently, registration methods based on multi-view images have improved efficiency but still face challenges in dynamic responsiveness and speed. We propose a Interpretable Self-Supervised Differentiable Rendering Network (ISDR-Net) that enables monocular dynamic sensor-head pose tracking and registration. By integrating geometry-guided differentiable rendering with a transparent, unrolled optimization process, ISDR-Net ensures interpretability, computational efficiency, and adaptability. A coarse-to-fine optimization strategy rapidly initializes alignment, while adaptive refinement driven by keyframes and motion cues robustly tracks subtle head movements. Experimental results show that ISDR-Net aligns with the sub-millimeter registration accuracy of existing multi-view methods using only monocular single-frame images, with significantly reduced computational cost and stable performance under dynamic conditions. These results highlight the potential of ISDR-Net to support practical deployment of next-generation OPM-MEG systems in naturalistic environments.

In order to improve the performance of the x-ray pulsar timing system and further refine the pulsar time series analysis process, we propose a Hankel-singular value decomposition (SVD) principal modal vector energy-fusion based phase delay estimation method for x-ray pulsar. First, the pulsar profile signal is subjected to Hankel embedding and singular value decomposition. The principal modal vector in the structured subspace is extracted from the one-dimensional profile signal, effectively suppressing redundant components such as observation noise. Secondly, cross-spectral analysis is performed on the standard and folded profiles in the main modal domain. Combining the phase ratio method, the phase difference of each frequency component is analytically solved. By utilizing the frequency-domain phase continuity, the resolution limitation caused by discrete search in the time domain is overcome, achieving subsampling level phase delay estimation below the sampling interval. Finally, combining the frequency energy distribution characteristics of each order, we propose a fusion strategy based on cross-spectral energy constraints, using spectral amplitude to construct weights. The phase shifts of each order are weighted and adjusted to further improve estimation accuracy. Meanwhile, we conduct ablation tests, computational complexity analysis, and influencing factor analysis on the proposed method. Furthermore, we perform ground-based experimental tests using actual observation data from the NICER and XPANV-1 satellites. The results show that compared to cross-correlation estimation and maximum likelihood estimation, our method improves estimation accuracy by 26.18% and 42.72%, respectively, and improves estimation stability by 16.91% and 43.88%, respectively. The proposed method has advantages such as high estimation accuracy, strong estimation stability, and subsampling level phase estimation resolution, providing important support for the development of high-precision pulsar timing.

Magnetoencephalography (MEG) based on optically pumped magnetometers (OPM) offers new opportunities for brain research, yet its recordings are highly susceptible to various noise interferences. Noise sources such as sensor slippage, cable movement, muscle activity, or head movement can cause sudden spikes or fluctuations with significantly higher amplitudes than normal signals in certain channels over specific time periods, manifesting as sparse artifacts in the time domain. These temporally sparse artifacts can significantly distort signal waveforms and severely interfere with subsequent data analysis. To address this issue, this study proposes a novel method called Centered Sliding-Window Neighborhood Subspace Projection (CSNSP) for effectively suppressing temporally sparse artifacts and improving signal quality. This method captures the intrinsic patterns of the MEG signals by identifying the intersection of the common signal subspaces between artifact-free segments and artifact-contaminated segments in the time domain. Based on this intersection, a signal subspace is constructed, and the artifact-contaminated segments are projected onto it to isolate and correct the artifact effects, achieving more accurate signal restoration. The CSNSP method is evaluated with both simulated and real data, and found to be highly effective in removing or attenuating temporally sparse artifacts. The method demonstrates strong robustness, unaffected by either the number of contaminated channels or the severity of the artifacts.