基于机器学习的呼吸波动估计 - 基于BOLD fMRI与头部运动参数的健康成人研究

Aug 6, 2026

Upload by Aurora

Functional MRI studies often lack clean respiratory recordings or suffer from low-quality respiratory signals, limiting reliable reconstruction of respiratory variation (RV) waveforms. This study tests the hypothesis that head motion parameters carry informative cues about respiratory patterns and events. A convolutional neural network is proposed to reconstruct RV waveforms from head motion parameters together with BOLD signals. Results show that combining motion parameters and BOLD signals improves RV estimation, potentially reducing fMRI cost and simplifying participant setup by avoiding respiratory bellows.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages6

Reading Time15 min

AccessFree to download

Introduction

Background and motivation

Method

Dataset and preprocessing

Model design (1D-CNN) and RV definition

Results

RV waveform estimation performance

Impact

Cost reduction and participant burden

Machine Learning-based Estimation of Respiratory Fluctuations in a Healthy Adult Population using BOLD fMRI and Head Motion Parameters* Abdoljalil Addeh 1-3,5, Fernando Vega 1-3,5, Rebecca J Williams6 G. Bruce Pike,3-5 M. Ethan MacDonald 1-3,5 1 Department of Biomedical Engineering, Schulich School of Engineering, University of Calgary, Canada 2 Department of Electrical & Software Engineering, Schulich School of Engineering, University of Calgary, Canada 3 Department of Radiology, Cumming School of Medicine, University of Calgary, Canada 4 Department of Clinical Neurosciences, Cumming School of Medicine, University of Calgary, Canada 5 Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary, Canada 6 Faculty of Health, Charles Darwin University, Australia Synopsis: Motivation: In many fMRI studies, respiratory signals are often missing or of poor quality. Therefore, it could be highly beneficial to have a tool to extract respiratory variation (RV) waveforms directly from fMRI data without the need for peripheral recording devices. Goal: Investigate the hypothesis that head motion parameters contain valuable information regarding respiratory patter, which can help machine learning algorithms estimate the RV waveform. Approach: This study proposes a CNN model for reconstruction of RV waveforms using head motion parameters and BOLD signals. Results: This study showed that combining head motion parameters with BOLD signals enhances RV waveform estimation. Impact It is expected that application of the proposed method will lower the cost of fMRI studies, reduce complexity, and decrease the burden on participants as they will not be required to wear a respiratory bellows. * 2024 International Society of Magnetic Resonance in Medicine. Singapore, May 4-9. Abstract Number 3276 2024 ISMRM A. Addeh, F. Vega, R. Williams, et al. Introduction Acquisition of clean external respiratory data during functional magnetic resonance imaging (fMRI) to remove the effect of low-frequency respiratory variation is not always possible [1] . Several machine learning-based approaches have been developed recently that utilize the information contained within blood oxygen level-dependent (BOLD) signals to estimate respiratory variations (RV) waveforms [2-4] . These methods showed promising results when tested on the Human Connectome Project in Young Adults (HCP-YA) dataset but faced limitations in complete RV signal reconstruction due to edge-effects. Respiration can also influence head motion in fMRI. Recent studies have shown that respiration generates real and pseudomotion of the head at the respiratory rate ( ∼0.3 Hz for healthy adults) [5] . Thus, there may be a potential use for head motion parameters in RV estimation using machine learning approaches. The aim of this study is to investigate the hypothesis that head motion parameters contain valuable information regarding the different respiratory events, which can help machine learning algorithms estimate the RV waveform. Method We utilized 900 resting-state fMRI scans from the HCP-YA dataset. Fig.1 demonstrates how current respiration affects subsequent BOLD signals and head motion using a gray plot of BOLD signals [6] . In addition, current breathing depth and rate are dependent upon previous breaths [7] . Fig. 2 displays the power spectra derived from motion parameters. Notable oscillations around 0.3 Hz and 0.12 Hz, especially in the phase encoding direction, are evident. These oscillations are attributed to both the physical movement ofthe head due to breathing and the pseudomotion artifact, which arises from changes in lung volume and subsequent shift in the B0 field [5] . Figure 1 . Illustration of respiratory events impact on BOLD signals and subject’s head motion parameters. Vertical black bands after each single isolated deep breath (shown by a green arrow) or burst of deep breaths (shown by yellow arrows) reflect the BOLD s

FAQs

为什么在fMRI研究中难以获得高质量呼吸信号？

许多fMRI研究无法始终采集到干净的外部呼吸数据，或呼吸信号质量不足，导致呼吸变异（RV）波形重建受限。

研究的核心假设是什么？

头部运动参数包含与不同呼吸事件相关的信息，可为机器学习算法估计RV波形提供有效线索。

模型如何利用BOLD信号与头部运动参数来重建RV？

研究提出在时间维度上使用三个1D-CNN，将BOLD时间序列与头部运动参数作为输入重建RV波形；并在滑动窗口内以呼吸波形的标准差定义RV。

Aug 6, 2026

Upload by Aurora