Scalable and Robust Tensor Decomposition of Spontaneous Stereotactic EEG Data

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Understanding spontaneous brain activity dynamics is crucial for neuroscience.
• Resting-state brain signals offer insights into neural network function.
• Identifying these networks requires advanced analytical techniques.

Purpose of the Study:

• To introduce a novel tensor-based model for identifying brain networks from resting-state signals.
• To develop a scalable and robust framework for tensor decomposition.
• To compare the proposed method against conventional algorithms.

Main Methods:

• Developed a rank-recursive scalable and robust sequential canonical polyadic decomposition (SRSCPD) framework.
• Utilized a warm-start strategy for enhanced robustness and scalability.
• Applied SRSCPD to simulated data and in-vivo stereotactic EEG (sEEG) data from epilepsy patients.

Main Results:

• SRSCPD consistently outperformed the multi-start alternating least square (ALS) algorithm in simulations across various ranks and SNRs, with reduced computational cost.
• Identified consistent default mode and motor networks in epilepsy patients' resting-state sEEG data.
• Achieved high within-subject consistency of identified networks across multiple recording sessions.

Conclusions:

• SRSCPD successfully identified consistent brain networks and their dynamic behaviors from resting-state sEEG data.
• The SRSCPD framework is scalable for large datasets, making it a valuable tool for analyzing long neural recordings.
• This method offers a promising approach for brain network identification in neuroscience research.