Direct Estimation of Pediatric Heart Rate Variability from BOLD-fMRI: A Machine Learning Approach Using Dynamic Connectivity

TL;DR

This study introduces a machine learning method using 1D-CNN and GRU to directly estimate pediatric heart rate variability from BOLD-fMRI data, eliminating the need for peripheral devices and improving robustness.

Contribution

The paper presents a novel hybrid deep learning model that accurately reconstructs HRV from fMRI data without external sensors, enhancing pediatric neuroimaging analysis.

Findings

Achieved 8% improvement in HRV estimation accuracy

Eliminated need for peripheral photoplethysmography devices

Enhanced robustness of pediatric fMRI studies

Abstract

In many pediatric fMRI studies, cardiac signals are often missing or of poor quality. A tool to extract Heart Rate Variation (HRV) waveforms directly from fMRI data, without the need for peripheral recording devices, would be highly beneficial. We developed a machine learning framework to accurately reconstruct HRV for pediatric applications. A hybrid model combining one-dimensional Convolutional Neural Networks (1D-CNN) and Gated Recurrent Units (GRU) analyzed BOLD signals from 628 ROIs, integrating past and future data. The model achieved an 8% improvement in HRV accuracy, as evidenced by enhanced performance metrics. This approach eliminates the need for peripheral photoplethysmography devices, reduces costs, and simplifies procedures in pediatric fMRI. Additionally, it improves the robustness of pediatric fMRI studies, which are more sensitive to physiological and developmental variations than those in adults.