Parallel Factorization to Implement Group Analysis in Brain Networks Estimation

Area of Science:

• Neuroscience
• Network Analysis
• Signal Processing

Background:

• Group analysis of complex functional brain networks remains a challenge.
• Extracting a reliable grand average connectivity matrix is crucial for understanding network dynamics.

Purpose of the Study:

• To investigate the efficacy of Parallel Factor Analysis (PARAFAC) for extracting grand average connectivity matrices.
• To evaluate the impact of data characteristics and algorithm parameters on PARAFAC performance.

Main Methods:

• PARAFAC was applied to both simulated and real electroencephalography (EEG) datasets.
• Simulated data were manipulated for network dimension, sample size, and noise levels.
• The PARAFAC approach was solved using varying numbers of rank-one tensors (PAR-FACT).

Main Results:

• PARAFAC demonstrated potential for grand average connectivity extraction in both synthetic and real data.
• Optimal performance (low False Positive Rate, False Negative Rate, high Area Under the Curve) was associated with larger sample sizes and lower noise.
• Increasing the number of rank-one tensors (PAR-FACT) in the PARAFAC solution negatively impacted estimation accuracy.

Conclusions:

• PARAFAC is a promising tool for group analysis of functional brain networks.
• Data quality (sample size, noise) and algorithm parameterization (PAR-FACT) are critical for accurate grand average estimation.