Memory-Efficient SEEG Data Processing Pipeline for Large Datasets

Stereoencephalography (SEEG) is a powerful technique for intracranial recording of brain activity, crucial for localizing epileptic foci and studying neural dynamics. However, interpreting SEEG data is computationally demanding due to the high spatial and temporal resolution of recordings, significant data volume, electrode placement variability, intrinsic noise, and complex, non-stationary signals. This paper proposes a memory-efficient SEEG data processing pipeline designed to manage large datasets effectively while preserving critical signal information. The preprocessing pipeline includes loading and inspecting raw data, applying zero-phase FIR band-pass filters (1-50 Hz) to eliminate noise without distorting phase relationships, and segmenting data for frequency-specific analysis. Frequency domain analysis is conducted using Fast Fourier Transform (FFT), Short-Time Fourier Transform (STFT), Morlet wavelet transforms, and Welch’s method for Power Spectral Density (PSD) estimation. These methods enable robust exploration of frequency dynamics, capturing both transient and stable oscillatory brain activity. The presented pipeline maintains computational efficiency through optimized windowing parameters and filtering strategies, ensuring high-quality data interpretation without extensive resource demands. While primarily linear and limited by fixed parameters and potential redundancy in time-frequency overlaps, the approach successfully addresses common challenges in SEEG interpretation, including artifact reduction, spectral clarity, and reproducibility. This work offers a practical framework for large-scale SEEG analysis, facilitating clinical decision-making and advanced neurophysiological research.