Expanding the Clinical Application of OPM-MEG: An Effective Method for Automatically Suppressing Metal Artifacts from Dental Braces

Background

Magnetoencephalography (MEG) is a technique that uses multi-channel magnetic field measurement sensors to reconstruct the neural current distribution and functional networks of the brain. Compared to electroencephalography (EEG), MEG has significant advantages in source spatial resolution. Additionally, its magnetic signals are not disturbed by the conductivity of the skull and scalp tissues, making it superior in temporal resolution compared to functional magnetic resonance imaging (fMRI). Therefore, MEG holds significant importance in studying brain function and cognition, clinical applications for epilepsy, and neuropsychiatric research.

Currently, MEG measurements primarily rely on two types of devices: commercial superconducting quantum interference devices (SQUID) and wearable optically pumped magnetometer (OPM) devices. However, SQUID devices are costly and bulky, and they require liquid helium for cooling, making their operational costs extremely high. In contrast, OPM devices are portable, wearable, and can be placed close to the scalp, meeting the needs of future clinical applications and neuroscience research.

However, in actual clinical settings, many patients have metal implants due to treatment needs, such as vagus nerve stimulators for epilepsy treatment, pacemakers for heart disease treatment, and magnetic materials for dental treatments. These metal materials can generate strong interference during MEG recording, making effective suppression of these artifacts an important challenge.

This study provides a comprehensive analysis of the temporal, frequency, and time-frequency characteristics of metal artifacts for the first time, focusing on metal braces. By modulating metal artifacts through breathing and head movements, we identified their sub-Gaussian distribution and high absolute power ratio in the 0.5-8Hz frequency band. Existing metal artifact suppression algorithms are limited in traditional SQUID-MEG devices. In response to these issues, this study proposes a second-order blind source separation (SOBI) algorithm to separate OPM-MEG measurement signal components through multiple time delays, and independently developed an automatic artifact component screening method.

Paper Source

This paper was written by Ruonan Wang, Kaiwen Fu, Ruochen Zhao, Dawei Wang, Zhimin Yang, Wei Bin, Yang Gao, and Xiaolin Ning. The affiliated institutions include the School of Precision Instruments and Optoelectronic Engineering at Beihang University, the Department of Radiology at Qilu Hospital of Shandong University, and the Second Affiliated Hospital of Guangzhou University of Chinese Medicine. The paper was published in the 2024 editions of “NeuroImage” and “ScienceDirect.”

Research Process

Detailed Research Process

1. Artifact Characteristic Analysis: a. Experimental Subjects: Four subjects with dental braces were selected to measure metal artifacts induced by breathing and head movements. b. Testing Equipment: The second-generation optically pumped magnetometer device (QZFM Gen-2 OPM) was used, with 33 data acquisition channels. c. Characteristic Analysis: Analyze the temporal, frequency, and time-frequency characteristics of the obtained data, determining that the main frequency band of metal artifacts is between 0.5-8Hz with a sub-Gaussian distribution.

2. Algorithm Implementation: a. Second-order Blind Source Separation Method (SOBI): Introduce multiple time delay parameters, preprocess the signal by subtracting the mean and spatial decorrelation, calculate the lagged correlation matrix, and apply the SOBI method for signal separation. b. Automatic Artifact Identification: Utilize the high absolute power ratio of metal artifacts in the low-frequency band, root mean square (RMS), and mutual information methods to achieve automatic identification and removal of artifact components.

3. Simulation Experiments: a. Resting State Simulation Data: Form simulated MEG data with artifact interference through linear regression, compare SOBI, FastICA, Infomax, and AMUSE algorithms. b. Evoked State Simulation Experiment: Simulate event-related signals, superimpose metal artifact interference data, and calculate the root mean square error (RMSE) and signal-to-noise ratio (SNR) of each algorithm to evaluate their performance.

4. Actual Testing: a. Auditory Evoked Experiment: Set up pure-tone stimulation experiments, process MEG data containing metal artifacts using the four algorithms mentioned above, and calculate the signal-to-noise ratio and source localization accuracy.

Data Analysis and Results

1. Resting State Experiment Analysis:

   • Evaluation Metrics: Use normalized mean square error (NMSE) and Delta band normalized power error (▽δ) to evaluate the artifact suppression effect.
   • Results: In all subject data, the proposed method showed the best performance in NMSE and ▽δ metrics, significantly suppressing metal artifact interference and reconstructing the source signal amplitude most accurately.

2. Evoked State Experiment Results:

   • Evaluation Metrics: Use root mean square error (RMSE) and signal-to-noise ratio (SNR) to evaluate the artifact suppression effect and calculate the accuracy of the simulated evoked signal.
   • Results: The proposed SOBI separation identification method demonstrated excellent performance in suppressing artifacts, with minimal distortion in the amplitude of the reconstructed evoked state simulated waveform and the highest SNR.

3. Actual OPM-MEG Auditory Evoked Experiment Results:

   • Artifact Identification and Suppression: The proposed artifact identification method accurately separated and identified metal artifacts, confirming the effectiveness of the algorithm through mutual information and spectral analysis.
   • Experimental Results: The proposed method significantly improved the SNR of event-related signals, especially showing prominent effects in the N100 component’s response.

Conclusion and Prospects

This study utilized optically pumped magnetometer devices to comprehensively analyze the characteristics of dental brace metal artifacts for the first time. The proposed SOBI separation identification algorithm effectively suppressed and removed metal artifacts through automatic identification of artifact components, enhancing the clinical application of OPM-MEG devices.

Research Highlights and Significance

• Innovativeness: For the first time, this study deeply analyzed metal artifacts in optically pumped magnetometer devices and proposed a metal artifact suppression method based on second-order blind source separation and automatic identification.
• Practicality: The method significantly suppresses artifacts caused by metals such as dental braces, improving data quality and the application prospects of OPM-MEG in clinical settings.
• Future Prospects: With the miniaturization of optically pumped magnetometer devices and an increase in sensor numbers, future integration with spatial filtering algorithms or deep learning techniques is expected to further enhance artifact identification and suppression effectiveness.