Acquisition and Joint Processing Pipeline of Intracranial Local Field Potentials and Magnetoencephalography in Treatment-Resistant Depression Undergoing Deep Brain Stimulation: From Raw Signals to Frequency Domain Feature Extraction

Dataset Overview and Processing PipelineThe dataset originates from a clinical study on Deep Brain Stimulation (DBS) for the treatment of depression. Data were acquired synchronously via two modalities: 1) Intracranial Local Field Potential (LFP) recordings, and 2) Magnetoencephalography (MEG) recordings.LFP Data:Following DBS implantation, data were collected monthly via electrodes implanted in the nucleus accumbens and the anterior limb of the internal capsule. During each recording session, the stimulator was temporarily turned off, and patients remained seated with eyes open, refraining from speech or movement. Immediately following each recording, patients were assessed using the Hamilton Depression Rating Scale (HAMD). Data were acquired from six bipolar contact pairs across both hemispheres (left and right), with a sampling rate of 1000 Hz. Each recording lasted approximately 3 minutes per contact pair.MEG Data:MEG data were acquired using a magnetoencephalography system prior to DBS surgery and at 1, 3, and 6 months postoperatively. During recording, patients lay supine in a shielded room with their eyes closed, and data were continuously acquired for 10 minutes at a sampling rate of 2000 Hz.Data Processing StepsData processing proceeded along parallel pipelines for LFP and MEG:LFP Processing:​ Raw signals were first subjected to band-pass filtering (1–100 Hz) and notch filtering to remove power-line interference at 40 Hz and 80 Hz. All filtering was performed using zero-phase filtering to prevent phase distortion. Subsequently, Power Spectral Density (PSD) was calculated using Welch’s method (Hanning window, 1-second window length, 50% overlap, 2048 FFT points) and decomposed into six standard frequency bands: delta, theta, alpha, beta, low-gamma, and high-gamma.MEG Processing:​ Preprocessing included visual inspection and removal of artifacts related to head movement, blinking, and muscle activity, as well as the exclusion of noisy channels. This was followed by band-pass filtering (1–100 Hz) and 50 Hz notch filtering. Source localization comprised three stages: First, an individual Boundary Element Method (BEM) head model was constructed based on individual MRI scans. Second, the inverse problem in the cortical source space was solved using Dynamic Statistical Parametric Mapping (dSPM) and Weighted Minimum Norm Estimation (wMNE). Finally, individual source estimates were mapped to a standard brain template, and cortical activity was estimated based on the Desikan-Killiany atlas. Power spectral density and Granger Causality connectivity were calculated for the same six frequency bands used in the LFP analysis.Equipment and SoftwareAcquisition Devices:​ Intracranial LFP recordings were controlled by SceneRay software (SceneRay Co., Ltd.) via the DBS device; MEG data were acquired using a 306-channel Elekta Neuromag TRIUX system.Processing Tools:​ MRI segmentation and co-registration were performed using SPM12; MEG source localization and time-frequency analysis were primarily conducted using the Brainstorm toolbox; the overall processing workflow was integrated via modified and customized Python (v3.8) scripts.