Distributed fMRI–Physiology Coupling and Age Prediction

This repository contains the analysis code supporting the study:

Distributed fMRI patterns coupled to low-frequency cardiorespiratory dynamics provide markers of aging
https://www.biorxiv.org/content/10.64898/2025.12.15.694468v1 (Wang et al. 2025)

---

Overview

This codebase provides an end-to-end pipeline for quantifying fMRI–physiology coupling and using these features to predict chronological age.

It includes two major components:

1. MATLAB: Physio–fMRI Variability Extraction

The MATLAB code implements:

• Generation of synthetic datasets for demonstration and testing
• Preprocessing of fMRI and physiological signals (respiration and heart rate)
• Fitting physiological regressors to fMRI time series
• Extraction of physio–fMRI variability measures at the segment level, including:
  • RV-coupled fMRI variability
  • HR-coupled fMRI variability
  • Joint RVHR-coupled fMRI variability
  • Raw and regression-based fALFF/ALFF features

These scripts produce feature matrices saved as .mat files, which are used as inputs to the prediction models.

---

2. Python: Subject-Level Age Prediction (SVR)

The Python pipeline performs:

• Loading of extracted MATLAB features
• Subject-level cross-validation using predefined or automatically generated splits
• Support Vector Regression (SVR) modeling with configurable kernels
• Segment-level prediction followed by subject-level averaging
• Performance evaluation using:
  • Pearson correlation (PCC)
  • Mean Absolute Error (MAE)

Python Requirements

• Python 3.9.7
• Packages:
  • numpy==1.22.4
  • scipy==1.8.0
  • pandas==1.4.2
  • scikit-learn==1.6.1

Installation

pip install -r requirements.txt

---

Purpose

This repository is intended to:

• Support reproducibility of the published analyses
• Provide a reference implementation for physio–fMRI coupling methods
• Serve as a template for extending these models to other datasets or phenotypes

---