Functional Regression with Reconstructed Functions from Hybrid Principal Components Analysis: An EEG–fMRI Application

Published in Statistics, Optimization & Information Computing, 2022

Title

Functional Regression with Reconstructed Functions from Hybrid Principal Components Analysis: An EEG–fMRI Application

Authors

Mohammad Fayaz, Alireza Abadi, Soheila Khodakarim

Journal Information

Journal: Statistics, Optimization & Information Computing
Year: 2022
Volume: 10
Issue: 3
Pages: 890–903

DOI: [https://doi.org/10.19139/soic-2310-5070-1437]


Keywords

  • Functional Data Analysis
  • Hybrid Principal Components Analysis (HPCA)
  • Functional Regression
  • EEG
  • fMRI
  • Feature Extraction
  • Prediction Accuracy

Abstract

Objective:
This study develops a unified framework combining hybrid principal component analysis (HPCA) with functional regression for the analysis of high-dimensional EEG–fMRI data. The proposed approach reconstructs functional signals using HPCA as a feature extraction step and evaluates predictive performance using functional regression models.

Methods:
The hybrid data structure is decomposed into eigencomponents using HPCA. Reconstructed signals are partitioned into training and testing datasets. Both function-on-function and scalar-on-function regression models are applied to EEG–fMRI data collected from auditory and visual oddball paradigms. Model performance is evaluated using mean squared prediction error (MSPE), and the optimal number of HPCA components is selected based on predictive accuracy.

Results:
Simulation studies demonstrate that reconstruction using the first HPCA component yields the lowest MSPE compared to raw and fully reconstructed representations. In real EEG–fMRI applications, the first HPCA component consistently provides superior predictive performance across auditory and visual tasks, as well as across multiple brain regions and stimulus conditions.

Conclusion:
The results indicate that HPCA-based reconstruction significantly improves predictive performance in functional regression models for EEG–fMRI data. The proposed framework outperforms traditional functional PCA-based approaches in capturing complex neurophysiological patterns.


Full Text

View Full Text (PDF)


Supplementary Material

The supplementary material provides an extended analysis of four EEG–fMRI datasets obtained from 17 participants.

Dataset Structure

  • Auditory oddball task — standard stimuli
  • Auditory oddball task — target stimuli
  • Visual oddball task — standard stimuli
  • Visual oddball task — target stimuli

In this framework, EEG signals are treated as response functions, while fMRI region-of-interest (ROI) curves based on Brodmann areas are used as functional covariates.

EEG Region Grouping

  • Whole brain
  • Frontal
  • Central
  • Parietal
  • Occipital
  • Right temporal
  • Left temporal

Analytical Pipeline

  1. Decomposition of EEG signals into functional, longitudinal, and regional components using HPCA
  2. Reconstruction of EEG curves using the first to KLM HPCA components
  3. Function-on-function regression with multiple ROI covariates (train/test split)
  4. Evaluation using correlation and mean squared prediction error (MSPE)
  5. Selection of optimal HPCA components based on MSPE, correlation, and computational efficiency

Supplementary File

Supplementary Report


Preprint (arXiv)


Conference Presentation

Selected results from this study were presented as a poster at:

  • 13th International Conference of the ERCIM WG on Computational and Methodological Statistics (CMStatistics 2020) and 14th International Conference on Computational and Financial Econometrics (CFE 2020, Virtual), 19–21 December 2020, King’s College London, London, UK. (Virtual Presentation)

Conference link: https://www.cmstatistics.org/CMStatistics2020/index.php
Book of Abstracts: https://mohammad-fayaz.github.io/publications_files/2022/Y_2022_1_BoACFECMStatistics2020.pdf
Poster: https://mohammad-fayaz.github.io/publications_files/2022/Y_2022_1_PosterCFECMStatistics2020.png


Citation

Fayaz M, Abadi A, Khodakarim S. Functional regression with reconstructed functions from hybrid principal components analysis: an EEG–fMRI application. Statistics, Optimization & Information Computing. 2022;10(3):890–903.

Recommended citation: Fayaz M, Abadi A, Khodakarim S. Functional regression with reconstructed functions from hybrid principal components analysis: an EEG–fMRI application. Statistics, Optimization & Information Computing. 2022;10(3):890–903.
Download Paper