Using ECG-derived respiration for explaining BOLD-fMRI fluctuations during rest and respiratory modulations

Using ECG-derived respiration for explaining BOLD-fMRI fluctuations during rest and respiratory modulations | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Using ECG-derived respiration for explaining BOLD-fMRI fluctuations during rest and respiratory modulations Inês Esteves, Ana R. Fouto, Amparo Ruiz-Tagle, Gina Caetano, Patrícia Figueiredo This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7046365/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 11 Nov, 2025 Read the published version in Scientific Reports → Version 1 posted 11 You are reading this latest preprint version Abstract Recording physiological signals during fMRI is valuable for multiple purposes but often requires additional setup, increasing complexity and participant discomfort. This is particularly challenging in simultaneous EEG-fMRI studies, which typically already include electrocardiogram (ECG) recordings. Here, we aim to leverage the known modulation of ECG by respiration to obtain an ECG-derived respiration (EDR) signal without extra equipment. We acquired EEG-fMRI data from 15 healthy subjects during resting state and two respiratory challenges (slow-paced breathing and breath-holding), with simultaneous ECG and respiratory recordings. Multiple methods were used to extract EDR signals, and the results were evaluated by comparing them with recorded respiration and assessing the quality of physiological regressors for denoising and cerebrovascular reactivity estimation. Amplitude-based EDR methods showed lower correlations with respiration, likely due to ECG distortion in the MRI. Nevertheless, coherence analysis showed that EDR preserved the relevant spectral content. EDR-based regressors were similar to those obtained from measured respiration. Notably, a method based on heart rate variability performed best overall, yielding physiological noise correction and reactivity estimates comparable to those using recorded respiration. Our results demonstrate that meaningful respiratory information can be extracted from ECG within the MRI environment, benefiting EEG-fMRI studies when respiration cannot be reliably recorded. Biological sciences/Neuroscience/Neuro vascular interactions Biological sciences/Physiology/Neurophysiology Physical sciences/Engineering/Biomedical engineering functional magnetic resonance imaging (fMRI) respiration electrocardiogram (ECG) ECG-derived respiration (EDR) resting state respiratory fluctuation Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 1. Introduction Functional magnetic resonance imaging (fMRI) indirectly assesses brain activity by measuring changes in the blood-oxygen level dependent (BOLD) signal, allowing the mapping of brain regions involved in task-induced activation or functionally connected. Since the BOLD contrast relies on hemodynamic changes, which are also influenced by physiological processes such as cardiac pulsatility and respiration, these may represent confounds when interpreting the BOLD signal 1 – 4 . Physiological signal acquisition during fMRI may serve multiple purposes: modeling and correction of non-neuronal BOLD sources, study of physiological modulations of brain activity, or simply monitoring physiological states throughout experiments 5 . Although there are some fMRI data-driven methods used for physiological denoising (e.g. CompCor, Behzadi et al. (2007); FMRIB’s ICA-based X-noiseifier (FIX), Salimi-Khorshidi et al. (2014); ICA-based Automatic Removal Of Motion Artifacts (ICA-AROMA), Pruim et al. (2015)), most methods rely on measuring cardiac and respiratory signals. Critically, this requires additional setup 5 , such as respiratory belts and photoplethysmography, which increases experimental complexity, prolongs preparation times and heightens subject discomfort, representing an additional burden. Respiration is known to modulate the ECG, affecting the morphology of the heartbeats 9 , altering the electrical impedance of the thoracic cavity due to filling/emptying of the lungs 10 , and affecting the heart rate 11 . Consequently, it is possible to obtain an ECG-derived respiration (EDR) signal 12 . Methods for EDR estimation have been widely explored in the context of ambulatory care 13 , 14 , but only one previous EEG-fMRI study reported the use of EDRs in the MRI environment 15 . In the case of EEG-fMRI, placing a single-lead ECG on the back is a common procedure, for a posteriori correction of the pulse artifact induced on the EEG. Therefore, for EEG-fMRI setups, the ECG is usually recorded and may be used to retrieve respiration, in case the dedicated equipment is not available, or the recorded respiration is affected by technical issues. The study by Abreu et al. (2016) compared the use of a subject-specific physiological model derived from several EDRs with an image-based physiological model (based on cerebrospinal fluid (CSF) and white matter (WM) average time courses) and with FIX 7 . The data corrected using the physiological model achieved the best performance in terms of sensitivity and specificity to map epileptic networks and the default mode network. Unfortunately, the MRI environment introduces distortions in the ECG wave morphology as a consequence of both magnetic induction due to movements inside the main magnetic field as well as the effects of radiofrequency pulses 16 . Because the respiratory data were not recorded in that study to provide a ground-truth for comparison with the EDR, it remains to be validated whether EDRs in the MRI environment can be robustly obtained. In this work, we aim to fill this gap in the literature by evaluating EDR signals obtained during EEG-fMRI acquisitions. For this purpose, we collected EEG-fMRI data including simultaneous ECG and respiration recordings, from healthy participants during resting state and while they performed two respiratory challenges aimed at modulating respiration patterns: a slow-paced breathing task and a breath-hold task. We considered these tasks to test the ability of EDR to capture such respiratory variations, which are commonly utilized to assess cerebrovascular reactivity (CVR) 17 . We compared EDR signals with the measured respiratory signals, by assessing their similarity in time and frequency (correlation and coherence) and their ability to predict physiological contributions in fMRI signals, both for physiological denoising and CVR mapping. 2. Materials and Methods 2.1 Data acquisition Data were acquired in the context of a larger research project on brain imaging in migraine (MigN2Treat), taking place at Hospital da Luz, Lisboa. The methods and results corresponding to the other MRI protocols are described elsewhere 18 – 24 . The research protocol and statistical analysis were not preregistered. All methods were performed in accordance with the relevant guidelines and regulations. The study was approved by the Hospital da Luz Ethics Committee and all participants provided written informed consent according to the Declaration of Helsinki 7th revision. 2.1.1 Sample Data was acquired from 15 female healthy subjects (age: median = 30, IQR = 10.5). The participants were healthy, without any diagnosed condition that significantly impaired an active and productive life or reduced their life expectancy to be below 5 years. None of them was undergoing treatment with psychoactive drugs, including anxiolytics, antidepressants, anti-epileptics, or any migraine prophylactics. Furthermore, being pregnant or having contraindications for MRI scanning (e.g. claustrophobia, pacemaker, incompatible implants) were also exclusion criteria. 2.1.2 MRI data The MRI data acquisition was performed with a 3T Siemens Vida system, using a 64-channel RF coil. Functional T2*-weighted images were obtained using 2D-Echo Planar Imaging (EPI), TR/TE = 1260/30ms, flip angle = 70º, in-plane generalized autocalibrating partially parallel acquisition (GRAPPA) acceleration factor of 2, simultaneous multi-slice (SMS) with a factor of 3, 60 slices, with 2.2 mm isotropic resolution. Structural T1-weighted images were acquired with a magnetization-prepared rapid gradient echo (MPRAGE) sequence, with TR = 2300 ms, TE = 2.98 ms, inversion time (TI) = 900 ms, and 1 mm isotropic resolution. Field map magnitude and phase images were obtained using a double-echo gradient echo sequence (TR = 400.0 ms, TE = 4.92/7.38 ms, voxel size: 3.4 × 3.4 × 3, flip angle 60◦). 2.1.3 Physiological data The respiratory signal was continuously recorded with the integrated BioMatrix Sensors from Siemens, through Siemens Physiological Monitoring Unit (PMU), with a sampling frequency of 400Hz. The electrocardiogram (ECG) was acquired simultaneously with fMRI as part of the MR-compatible BrainAmp MR EEG system (Brain Products), with a sampling frequency of 5000Hz, using an Ag/AgCl ring-type electrode placed on the back as caudally as possible, on the left of the spine. The impedance was maintained below 25 kOhm. The SyncBox was used to ensure synchronization between the MRI scanner clock (10 MHz) and the BrainAmp MR system, to improve the quality of the gradient artifact (GA) correction. 2.1.4 Tasks EEG-fMRI data was acquired during resting state (RS) and two respiratory tasks involving respiratory modulations, slow-paced breathing (SPB) and breath-holding (BH). Due to technical issues, some respiratory recordings had missing values at the start or end. These signals were excluded from the analysis and were therefore not included in our sample sizes, except for RS. Consequently, not all tasks were analyzed for every subject, with only a subgroup of participants considered for each task. For RS, we retained the maximum amount of available data that still provided a reasonable duration. Although the original task lasted 7 minutes, only 4.82 minutes (229 fMRI volumes) per subject were usable for analysis. During RS (N = 10), subjects were instructed to keep their eyes open, looking at a black screen, without thinking of anything in particular. The SPB task (N = 9) consisted of an initial fixation cross (5 s), a 1-min free breathing period, a 2-min 0.1 Hz paced respiration with visual written instructions for inhaling (5 s) and exhaling (5 s), and another 1-min free breathing period (194 fMRI volumes). The BH task (N = 12) comprised 4 cycles of post-exhalation breath-hold (15 s) followed by free breathing and naturally paced breathing, with a total of approximately 4.47 min (213 fMRI volumes). The natural breathing frequency of each subject was computed beforehand based on a calibration recording performed outside the scanner, during a period longer than 1 minute while the subject was not performing any particular task. The median breathing period was 4 s for all task sub-samples, ranging from 3–4 s for RS, and 3–5 s for BH and SPB sub-samples. For each subject, all tasks were performed on the same day. The timings of each task are depicted in Fig. 1 . 2.2 Data analysis The fMRI data was preprocessed using FMRIB Software Library (FSL) tools 25 . MATLAB version 2016b was used for respiration and ECG data preprocessing as well as further analysis and visualization. The data analysis pipeline is shown in Fig. 2 . 2.2.1 Data preprocessing 2.2.1.1 fMRI EPI distortion correction was performed using FMRIB's Utility for Geometrically Unwarping EPIs (FUGUE), to correct for geometric distortions and signal loss, using a fieldmap acquisition to characterize B0 field inhomogeneities, containing magnitude and phase information. The Brain Extraction Tool (BET) was used to remove non-brain tissue. Volume realignment was performed by applying FMRIB's Linear Image Registration Tool (FLIRT) with respect to the middle volume, using 6 rigid body motion parameters (MP). Motion outliers (MO) were obtained using FSL Motion Outliers tool, with the dvars option, which considers the root mean square intensity difference of volume N to volume N + 1, thresholded at the 75th percentile + 1.5 times the interquartile range (Power et al. 2012). High-pass temporal filtering with a cut-off frequency of 0.01 Hz was applied to remove slow drift fluctuations and spatial smoothing was performed using SUSAN tool (S.M. Smith and J.M. Brady, 1997), employing a Gaussian kernel with a full width half maximum of 3.3 mm for RS and 3.5 mm for SPB and BH. The code used for fMRI preprocessing can be accessed here: https://github.com/martaxavier/fMRI-Preprocessing . The structural image was used to obtain a gray matter (GM) mask for the functional images. For that, following non-brain tissue removal, tissue segmentation was performed with FMRIB's Automated Segmentation Tool (FAST) to obtain GM partial volume effect (PVE) images, which were then thresholded, binarized and transformed to the subject’s functional space using a registration matrix to the structural image obtained by applying FSL’s FLIRT with 12 degrees of freedom. 2.2.1.2 Respiration Respiratory data (Resp) was matched with the fMRI acquisition, downsampled to 250 Hz and bandpass filtered from 0.01 to 1 Hz using a Butterworth filter of order 2. Peak detection was performed to find maximum and minimum values of inspiration and expiration, using a threshold based on the 80th percentile of the data for RS and BH, and on the 40th percentile for the SPB, since the amplitude of the signal during slow breathing was larger. This procedure was followed by manual verification and the threshold was readjusted to the 75th percentile for one subject. 2.2.1.3 ECG The ECG was corrected for the GA using the average artifact subtraction (AAS) method 26 , implemented in the FMRIB plugin of the EEGLAB toolbox, using windows with 30 volumes and without performing adaptive noise cancelation, to avoid distorting the ECG waveform. The signal was then downsampled to 250Hz and ECG R-peaks were automatically detected using a long short term memory (LSTM) network trained on ECG data acquired in the MR environment ( https://github.com/LaSEEB/deepQRS ) and manually inspected and corrected using the interactive tool ( https://github.com/LaSEEB/interactiveQRS ). Furthermore, we corrected for missing and false heart beats, using the approach proposed by de Chazal et al. (2003), which had also been employed by Varon et al. (2020) prior to EDR computation. This procedure resulted in the detection and correction of a missing peak for one subject, for the BH task. Finally, the signal was bandpass filtered 0.05-40 Hz using a FIR filter with a Hamming window. The lower edge of the band was chosen considering the preservation of respiratory frequencies. 2.2.2 ECG-derived respiration (EDR) extraction categories EDRs were extracted using a publicly available repository ( https://github.com/rmabreu/Respiratory_Signal_Estimation ) which comprises methods from 7 categories: ECG envelope (ENV), heart-rate variability (HRV), amplitude modulation (AM), QRS-area modulation (QRS-AM), principal component analysis (PCA), kernel PCA (kPCA), and empirical mode decomposition (EMD), previously described in the work by Abreu et al. (2017). Each method leverages different ECG features with some focusing on amplitude (e.g., ENV, AM, QRS-AM), others on frequency components (e.g., HRV, EMD), and some using statistical decomposition techniques (e.g., PCA, kPCA). Several EDRs were obtained for AM, PCA and kPCA (e.g. variations relying on different ECG features, such as using either the whole beat or only the QRS or T wave portions), resulting in 51 EDRs per task for each subject. Some of the methods induced a steep rise/decrease of the EDR signal at the start or end, due to insufficient information for an accurate respiratory estimate. To correct abnormal fluctuations at the signal edges, we applied nearest-value interpolation if the absolute gradient in the first or last second exceeded 80% of the maximum absolute gradient of the entire signal. Peak detection of minimum and maximum values was performed using the same procedure that had been employed for respiratory signals. 2.2.3 Similarity between respiratory signals We assessed the similarity between Resp and EDR signals in both the time and frequency domains, using correlation and coherence, respectively. To compute the similarity metrics for each participant and task, both signals were first standardized (mean of zero and standard deviation of one) and downsampled to 10 Hz 29 , 30 . Correlation was computed as the maximum cross correlation coefficient over a lag of +/-5s, optimizing for phase delays, and with no lag, to mimic a real situation with unknown optimal lags. Coherence was based on the magnitude squared coherence, estimated using Welch's method to obtain the spectra (1024-point Fast Fourier Transform; Hamming window; 8 segments with equal length and 50% of overlap). The mean magnitude squared coherence was computed for a range around the fundamental respiratory frequency (fR), considered to be the frequency with highest power within 0.15–0.45 Hz for RS and BH, and within 0.08–0.45 Hz for SPB. The range was determined by the frequencies around fR with more than half of the peak power 29 , 30 . For each category, we identified the EDR method yielding the highest correlation with Resp (using the optimal lag) for each task. The EDR method selected for each category was then identified based on a majority vote across the three tasks. For AM, a version without baseline removal was chosen by consensus, being the highest across all tasks. Both for PCA and kPCA, methods using the first principal component and the whole ECG beat were chosen, having achieved the highest correlation for SPB and BH, respectively. For kPCA, the method also included a polynomial kernel of order 2. To determine whether the correlations were significantly greater than zero, we conducted a Wilcoxon signed-rank test for each EDR category and task, applying the Benjamini-Hochberg procedure to control the false discovery rate (FDR) across multiple comparions. To evaluate differences between categories within each task, we used a Kruskal-Wallis test followed by post hoc pairwise comparisons using the Dunn test with Sidák correction to control the family wise error rate (FWER). The significance level was set as \(\:\alpha\:\) = 0.05. 2.2.4 Physiological signal modeling in fMRI To explore the impact on the physiological noise correction, nuisance regressors regarding respiration and cardiac pulsatility were obtained (with no lag): retrospective image correction (RETROICOR) terms, respiratory volume per time (RVT) and cardiac rate (CR). For RETROICOR, the phases of respiratory and cardiac cycles at each fMRI volume acquisition were estimated. A Fourier series up to the second order was built from respiratory and cardiac terms, accounting for quasi periodic signal fluctuations 4 . RVT considers non-periodic signal fluctuations due to changes in the depth and rate of breathing 1 . The RVT timecourse was obtained from the respiratory peaks and troughs using the amplitude (difference between consecutive maximum and minimum) divided by the time period between them. The result was interpolated to obtain a uniform sampling frequency corresponding to that of the fMRI, normalized and convolved with the respiratory response function as described by Birn et al. (2008). Similarly, for CR, the time difference between consecutive R peaks, which had been previously detected, was used to obtain the heart rate. The heart rate was interpolated, normalized and convolved with the cardiac response function as described in 3 . The respiratory regressors were estimated from EDRs and Resp, and their similarity was assessed using the correlation coefficient, in this case without considering a lag. The GM BOLD percent signal change (PSC) was calculated by first averaging the BOLD time series across all voxels within the GM mask to obtain a mean time course. The mean of this time course was then subtracted from each time point, and the result was divided by the same mean before being multiplied by 100. Three general linear models (GLM) were fitted to the GM BOLD PSC: Basic (6 MP, variable number of MO and, only for SPB, a task block convolved with the double-gamma hemodynamic response function (HRF)); Physio-rRetr (RETROICOR cardiac terms; CR; RETROICOR respiratory terms); and Physio-RVT (RETROICOR cardiac terms; CR; RVT). The two GLM’s including respiratory regressors were estimated using either EDR or Resp for subsequent comparison. The variance explained (%VE) for each GLM was computed as 100 x adjusted R 2 (R 2 adj ), and the %VE by the physiological terms was then obtained as (VE Physio -VE Basic ) for Physio-rRetr and Physio-RVT. As previously, we conducted a Wilcoxon signed-rank test for each EDR category and task to assess whether the variance explained was significantly greater than zero, applying the Benjamini-Hochberg procedure to control the FDR. In this case, respiration was also included in the comparison alongside the EDR categories. To evaluate differences among categories and compared them with Resp within each task, we used a Kruskal-Wallis test followed by post-hoc pairwise comparisons corrected using the Dunn test with Sidák correction, to control FWER. The significance level was set as \(\:\alpha\:\) = 0.05. 2.2.5 Mapping fMRI changes associated with RVT The BOLD changes associated with RVT were mapped by fitting a GLM for each task, using FEAT (FMRI Expert Analysis Tool) v6.00. For RS and SPB, we incorporated RVT and CR in the GLM, along with their respective derivatives as well as MP and MO, as nuisance regressors. For SPB, we also added the task paradigm convolved with a double-gamma HRF to the GLM as the regressor of interest. Two models were created for each subject, one using RVT derived from Resp and another using RVT derived from the EDR method selected in the previous analysis. A cluster threshold of z-stat > 2.3 and a cluster significance level p-value < 0.05 were applied to generate thresholded z-stat maps for RVT. For BH, we included RVT in the GLM, as well as its temporal derivative, as a regressor of interest, since it may be used as a proxy of PetCO2 for CVR mapping 31 . We also included MP and MO in the GLM as nuisance regressors. In this case, the clustering threshold was increased to z-stat > 3.1 to enhance the significance of the results. For the three tasks, we examined the maps of BOLD changes associated with each regressor, assessing the number of voxels that survived the threshold as well as the average PSC across the whole brain, GM and WM. Differences between Resp and HRV were tested using a Wilcoxon rank-sum test, followed by the Benjamini-Hochberg procedure to control the FDR. The significance level was set as \(\:\alpha\:\) = 0.05. 3. Results 3.1 Similarity metrics Results are presented only for the EDR methods yielding the highest correlation coefficients with Resp for each category. The results for all methods are presented in the Supplementary Material. The similarity metrics computed between the EDR signals and Resp across all subjects and tasks are presented in Fig. 3 . We found that, using optimal lags, the correlation between Resp and EDRs is significantly larger than zero for all categories, for RS, SPB and BH. Among the methods, EMD consistently outperforms others, while AM performs significantly worse than at least one other method for all tasks. Notably, HRV is significantly outperformed by other methods for RS and BH, but for SPB it shows high values, significantly exceeding AM. At zero lag, the differences in correlation between EDR methods become more pronounced. PCA and EMD consistently show stronger correlations, while Env and HRV perform consistently worse than at least two other methods. In terms of coherence, amplitude-based methods consistently show the poorest performance across all tasks. For RS, the methods that are not based on the amplitude perform similarly among them, with only EMD outperforming QRS-AM (amplitude-based). For SPB, the Env and HRV methods significantly outperform the lowest-performing method, AM. For BH, more subjects exhibit low coherence values across all methods, with no significant differences among them. In line with the previous results, the example in Fig. 4 shows that the EDR methods differ in their ability to accommodate breathing changes. EDR waveforms of some of the methods resemble respiration, though with amplitude differences and a time shift. Nevertheless, in general, their power spectra overlap the respiratory frequency band. Furthermore, the differences obtained for correlations considering the optimal lag relative to lag zero may be explained by the large scattering of optimal time lags, which are also mostly not centered around zero, as shown in Fig. 5 . 3.2 Physiological signal modelling in fMRI Figure 6 depicts the average GM fMRI signal % VE by respiratory regressors (rRetr and RVT) computed based on Resp and the different EDR methods. The performance of EDRs for physiological regression of the BOLD signal differs for rRetr and RVT and is dependent on the task. Resp-based regressors only outperform the AM method, for the model including RVT for the BH task. In general, although smaller, the % VE by Resp is equivalent to the one explained by several EDR methods, with HRV method showing the best performance. Overall, results suggest that HRV is the most consistent method, particularly in the case of Physio-rRetr models. In general, the amplitude-based methods show the worst performance. For these reasons, the following analyses were performed only with HRV. Figure 7 shows that the Basic and both Physio models (using Resp and HRV, the best performing EDR method) can fit the data, following it more closely in the case of respiratory tasks, as expected. 3.3 Mapping fMRI changes associated with RVT The results obtained for the mapping of BOLD changes associated with RVT, computed from Resp and EDR, for RS, SPB and BH, are presented in Figs. 8 , 9 , and 10 , respectively. For RS and SPB, no significant changes were found between Resp and EDR for either the number of voxels showing significant BOLD changes and the average PSC (Figs. 8 and 9 , panel B). For BH, fewer voxels were associated with the EDR-derived RVT compared with Resp for the whole brain, with gray and white matter following the same pattern. No significant differences were found between Resp and HRV (Wilcoxon rank-sum test, corrected p-values). Nonetheless, the average PSC seems to be equivalent for respiration and HRV, considering either the whole brain or gray/white matter. 4. Discussion We demonstrated the feasibility of extracting meaningful respiratory signals from the ECG recorded in the MRI environment for the first time. Using different methods, we compared the extracted EDR signals with the measured respiration signal, both during resting state and two respiration-modulation tasks. Overall, the HRV method, which relies on heart rate changes modulated by respiration, exhibited the best performance across tasks, explaining the BOLD-signal variance as well as measured respiration. Our results indicate that EDR holds great potential for physiological noise correction or CVR imaging, in cases when respiration cannot be recorded, or the signal is corrupted, particularly in the context of EEG-fMRI. Similarity between EDR and measured respiration signals In general, all EDR methods yielded similar respiratory traces to the ground-truth, when using the optimal lag, with EMD showing the highest similarity. Consistently with our results, Abreu et al. (2017) also found this method to be the most accurate for EDR estimation in the MRI environment, in terms of both temporal dynamics and spectral content 28 . Moreover, our results demonstrated high coherence values in addition to the strong correlations. Nevertheless, by comparing the EDR signals with the ground-truth respiration signals in our study, we found that correlations are generally low at a lag of 0 s, with the optimal time lag varying considerably between subjects. In fact, none of the methods could be linked to a fixed lag based on our time lag study, which would have enabled further research to use them in situations where the respiratory signal is unavailable for retrieval. Furthermore, it may be dependent on the respiratory frequency itself, which will in principle be unknown as well. This EDR time lag dependence and variability represents a limitation when a ground-truth respiration is not available. In contrast, coherence values are unaffected by time lags, and they were high across all tasks. The HRV method exhibited the highest coherence, despite low correlation. At the other end, amplitude-based methods had the lowest coherence values, indicating their inability to effectively capture the frequency content of the measured respiration. Physiological signal modelling in fMRI analysis Physiological noise correction is often necessary to remove non-neuronal contributions to the BOLD signal. Because frequency features play an important role in estimating physiological regressors, the high coherence of HRV signals allowed this method to provide the best models for all tasks. Additionally, the voxelwise analysis showed that HRV-based EDR also yielded BOLD signal changes associated with RVT that were similar to those obtained using the measured respiration signals, for both resting state and slow-paced breathing. These results indicate that the HRV EDR approach may be useful for the purpose of physiological noise correction when respiration is lacking. For CVR imaging based on the breath-holding task, EDR-based models yielded similar BOLD percent signal change values compared with respiration, although with a considerably lower number of voxels activated across the whole brain and the gray matter. These results indicate that, despite being less sensitive, EDR may provide reliable CVR estimates. Relation with the EDR literature Although EDR methods have not previously been evaluated against the respiration ground truth using ECG signals recorded in the MR environment, a substantial literature exists for recordings conducted outside of it. Of special relevance to our work, a few studies have investigated the performance of EDR methods in conditions with varying breathing patterns. Varon et al. (2020) systematically compared several types of methods for estimating EDR, assessing respiratory wave morphology, respiratory rate and cardiorespiratory information 14 . They applied the EDR methods to three datasets with a variety of physiological conditions, including relaxation, stress and healthy/abnormal sleep patterns. Their findings showed that simple methods based on morphological changes of the ECG caused by respiration, particularly those relying on the QRS complex, outperformed the other ones. The fact that our findings are not aligned with this study, may be explained by the distortion of the ECG in the MRI environment. Even after MR-artifact correction and further preprocessing, residual artifacts inevitably persist, and the typical ECG signal morphology is not preserved. As a result, amplitude-based methods are expected to perform poorly in our case. Additionally, the placement of the ECG electrode in the EEG-fMRI setup (on the subject's back, as recommended for artifact minimization) differs from the standard single-lead ECG montage used in most EDR studies. This may further affect the waveform morphology, making direct comparisons with the literature challenging. Machine learning, including deep learning models, have also been utilized to infer respiratory information from fMRI spatiotemporal patterns 32 , 33 . However, these models are more complex and require larger datasets, which makes them more difficult to apply to the typically small EEG-fMRI datasets. Furthermore, they are more challenging to implement in real time when needed. Limitations One of the greatest limitations of our study is the need for optimizing the lag introduced by the EDR estimation, which may hinder the use of EDR signals alone for assessing respiratory patterns at specific time points. Another limitation is the strong dependence of the EDR on the quality of the ECG data, as any degradation in ECG signal quality can significantly affect the accuracy of EDR estimations. Furthermore, it can be difficult to distinguish between cardiac and respiratory contributions when analyzing fMRI data, primarily due to the inherent nature of the EDR signal, which integrates both physiological processes. Similarly, it may be that the good performance of the HRV method is also linked to the ability to explain an autonomic system component, as this system is closely related to heart rate variability in general and may contribute to the BOLD signal. Conclusion Our study demonstrates the feasibility of extracting respiratory signals from ECG in the MRI environment and highlights the potential of this approach even in the presence of respiratory modulations, making it useful for monitoring and the computation of physiological regressors in fMRI analysis. The HRV method showed the best performance across tasks, indicating the potential of using EDRs as a physiological regressor in EEG-fMRI studies where direct respiration data is unavailable or corrupted. Importantly, we demonstrated the performance of EDR methods, not only during resting state, but also during respiratory manipulations of slow-paced breathing and breath-holding, broadening the range of potential applications. Declarations Funding Declaration This work was supported by LARSyS FCT funding [DOI: 10.54499/LA/P/0083/2020, 10.54499/UIDP/50009/2020, and 10.54499/UIDB/50009/2020], PRR project Center for Responsible AI [grant C645008882-00000055] and FCT [grants PD/BD/150356/2019, PTDC/EMD-EMD/29675/2017, LISBOA-01-0145-FEDER-029675]. Author Contribution IE: Methodology, Software, Formal analysis, Investigation, Data Curation, Writing – original draft, Writing – review and editing, Visualization; ARF: Investigation, Resources, Data Curation; ART: Investigation, Resources, Data Curation; GC: Investigation, Resources, Writing – review and editing; PF: Methodology, Conceptualization, Resources, Writing – review and editing, Supervision, Project administration, Funding acquisition. All authors have approved the final manuscript. Data availability Data supporting this study will be made available by the corresponding author upon reasonable request. References Birn, R. M., Diamond, J. B., Smith, M. A. & Bandettini, P. A. Separating respiratory-variation-related fluctuations from neuronal-activity-related fluctuations in fMRI. NeuroImage 31 , 1536–1548 (2006). Birn, R. M., Smith, M. A., Jones, T. B. & Bandettini, P. A. The respiration response function: The temporal dynamics of fMRI signal fluctuations related to changes in respiration. NeuroImage 40 , 644–654 (2008). Chang, C., Cunningham, J. P. & Glover, G. H. Influence of heart rate on the BOLD signal: The cardiac response function. NeuroImage 44 , 857–869 (2009). Glover, G. H., Li, T.-Q. & Ress, D. Image‐based method for retrospective correction of physiological motion effects in fMRI: RETROICOR. Magn. Reson. Med. Off. J. Int. Soc. Magn. Reson. Med. 44 , 162–167 (2000). Bulte, D. & Wartolowska, K. Monitoring cardiac and respiratory physiology during FMRI. NeuroImage 154 , 81–91 (2017). Behzadi, Y., Restom, K., Liau, J. & Liu, T. T. A Component Based Noise Correction Method (CompCor) for BOLD and Perfusion Based fMRI. NeuroImage 37 , 90–101 (2007). Salimi-Khorshidi, G. et al. Automatic denoising of functional MRI data: Combining independent component analysis and hierarchical fusion of classifiers. NeuroImage 90 , 449–468 (2014). Pruim, R. H. R. et al. ICA-AROMA: A robust ICA-based strategy for removing motion artifacts from fMRI data. NeuroImage 112 , 267–277 (2015). Przystup, P., Poliński, A. & Wtorek, J. QRS Morphology-Based EDR Signal—Factors Determining its Properties. IEEE Access 10 , 34665–34676 (2022). Moody, G. B., Mark, R. G., Zoccola, A. & Mantero, S. Derivation of Respiratory Signals from Multi-lead ECGs. Comput. Cardiol. 12 , 113–116 (1985). Yasuma, F. & Hayano, J. Respiratory Sinus Arrhythmia. Chest 125 , 683–690 (2004). Dong, K. et al. An integrated framework for evaluation on typical ECG-derived respiration waveform extraction and respiration. Comput. Biol. Med. 135 , 104593 (2021). Boyle, J., Bidargaddi, N., Sarela, A. & Karunanithi, M. Automatic Detection of Respiration Rate From Ambulatory Single-Lead ECG. IEEE Trans. Inf. Technol. Biomed. 13 , 890–896 (2009). Varon, C. et al. A Comparative Study of ECG-derived Respiration in Ambulatory Monitoring using the Single-lead ECG. Sci. Rep. 10 , 5704 (2020). Abreu, R. et al. Ballistocardiogram artifact correction taking into account physiological signal preservation in simultaneous EEG-fMRI. NeuroImage 135 , 45–63 (2016). Niendorf, T., Winter, L. & Frauenrath, T. Electrocardiogram in an MRI environment: Clinical Needs, Practical Considerations, Safety Implications, Technical Solutions and Future Directions. in Advances in Electrocardiograms: Methods and Analysis (BoD – Books on Demand, 2012). Pinto, J., Bright, M. G., Bulte, D. P. & Figueiredo, P. Cerebrovascular Reactivity Mapping Without Gas Challenges: A Methodological Guide. Front. Physiol. 11 , 608475 (2021). Domingos, C. et al. Impact of susceptibility-induced distortion correction on perfusion imaging by pCASL with a segmented 3D GRASE readout. Magn. Reson. Imaging 102 , 141–150 (2023). Esteves, I. et al. Uncovering longitudinal changes in the brain functional connectome along the migraine cycle: a multilevel clinical connectome fingerprinting framework. J. Headache Pain 26 , 29 (2025). Fouto, A. R. et al. Impact of truncating diffusion MRI scans on diffusional kurtosis imaging. Magn. Reson. Mater. Phys. Biol. Med. (2024) doi:10.1007/s10334-024-01153-y. Fouto, A. R. et al. Alterations of white matter microstructure in migraine patients vary in the peri-ictal phases. eNeuro ENEURO.0300-24.2024 (2024) doi:10.1523/ENEURO.0300-24.2024. Guadilla, I. et al. White matter alterations in episodic migraine without aura patients assessed with diffusion MRI: effect of free water correction. J. Headache Pain 26 , 31 (2025). Matoso, A. et al. Involvement of the cerebellum in structural connectivity enhancement in episodic migraine. J. Headache Pain 25 , 154 (2024). Ruiz-Tagle, A. et al. Preserved working memory performance along with subcortical modulation during peri-ictal phases in spontaneous migraine attacks. Headache J. Head Face Pain 65 , 407–419 (2025). Jenkinson, M., Beckmann, C. F., Behrens, T. E. J., Woolrich, M. W. & Smith, S. M. FSL. NeuroImage 62 , 782–790 (2012). Niazy, R. K., Beckmann, C. F., Iannetti, G. D., Brady, J. M. & Smith, S. M. Removal of FMRI environment artifacts from EEG data using optimal basis sets. NeuroImage 28 , 720–737 (2005). de Chazal, P. et al. Automated processing of the single-lead electrocardiogram for the detection of obstructive sleep apnoea. IEEE Trans. Biomed. Eng. 50 , 686–696 (2003). Abreu, R., Nunes, S., Leal, A. & Figueiredo, P. Physiological noise correction using ECG-derived respiratory signals for enhanced mapping of spontaneous neuronal activity with simultaneous EEG-fMRI. NeuroImage 154 , 115–127 (2017). Varon, C. & Van Huffel, S. ECG-derived respiration for ambulatory monitoring. in 2015 Computing in Cardiology Conference (CinC) 169–172 (IEEE, Nice, France, 2015). doi:10.1109/CIC.2015.7408613. Widjaja, D., Varon, C., Dorado, A., Suykens, J. A. K. & Van Huffel, S. Application of Kernel Principal Component Analysis for Single-Lead-ECG-Derived Respiration. IEEE Trans. Biomed. Eng. 59 , 1169–1176 (2012). Zvolanek, K. M. et al. Comparing end-tidal CO2, respiration volume per time (RVT), and average gray matter signal for mapping cerebrovascular reactivity amplitude and delay with breath-hold task BOLD fMRI. NeuroImage 272 , 120038 (2023). Bayrak, R. G., Salas, J. A., Huo, Y. & Chang, C. A Deep Pattern Recognition Approach for Inferring Respiratory Volume Fluctuations from fMRI Data. in Medical Image Computing and Computer Assisted Intervention – MICCAI 2020 (eds. Martel, A. L. et al.) 428–436 (Springer International Publishing, Cham, 2020). doi:10.1007/978-3-030-59728-3_42. Addeh, A. et al. Direct machine learning reconstruction of respiratory variation waveforms from resting state fMRI data in a pediatric population. NeuroImage 269 , 119904 (2023). Additional Declarations No competing interests reported. Supplementary Files manuscriptSREDRsupplementary.docx Cite Share Download PDF Status: Published Journal Publication published 11 Nov, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 04 Sep, 2025 Reviews received at journal 12 Aug, 2025 Reviews received at journal 28 Jul, 2025 Reviewers agreed at journal 28 Jul, 2025 Reviewers agreed at journal 23 Jul, 2025 Reviewers agreed at journal 23 Jul, 2025 Reviewers invited by journal 23 Jul, 2025 Editor assigned by journal 19 Jul, 2025 Editor invited by journal 10 Jul, 2025 Submission checks completed at journal 07 Jul, 2025 First submitted to journal 07 Jul, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7046365","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":489820161,"identity":"fe7c8610-f3be-4392-980e-70b12c6f0138","order_by":0,"name":"Inês Esteves","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABEUlEQVRIie3QMWrDMBSAYQlBsyT1+oxccgWXgqEkJVdJEDiLS9cMwSgENAW8+igdHR64i6BXaMkFVLw0S2hst5SAlUKnUvQPDzR8PEmEuFx/sIuTkyHsOJksqPyZQD1o3hD6RdjZbQ1h/U9CzpFLvkZDFmnqgcBqvBjNJMxW2/3jeEh4VnReLChjIBrBz+OYJ3peE4kDHV/LALvfAklEqCog1Dpi9wpbQhVSCcJCHipDDylM9HNV3R5ast0rnNhJQoBKBmFvQziVLSkGzTobiSOYlujnTyryN+X8RvVfm7cIFWDYRYa52BmzTD1vzXbmfTm6ynoC344/dpfx1Uvnmrrpyd7vC1iBtV8Ql8vl+pd9ABAJX+Tp1Om9AAAAAElFTkSuQmCC","orcid":"","institution":"ISR-Lisboa/LARSyS, Instituto Superior Técnico – Universidade de Lisboa","correspondingAuthor":true,"prefix":"","firstName":"Inês","middleName":"","lastName":"Esteves","suffix":""},{"id":489820162,"identity":"a9b11192-b7b6-46c1-8de5-9223e73f8e1e","order_by":1,"name":"Ana R. Fouto","email":"","orcid":"","institution":"ISR-Lisboa/LARSyS, Instituto Superior Técnico – Universidade de Lisboa","correspondingAuthor":false,"prefix":"","firstName":"Ana","middleName":"R.","lastName":"Fouto","suffix":""},{"id":489820163,"identity":"b7758f3a-2f60-48e7-8ddf-5e0f0de199e4","order_by":2,"name":"Amparo Ruiz-Tagle","email":"","orcid":"","institution":"ISR-Lisboa/LARSyS, Instituto Superior Técnico – Universidade de Lisboa","correspondingAuthor":false,"prefix":"","firstName":"Amparo","middleName":"","lastName":"Ruiz-Tagle","suffix":""},{"id":489820164,"identity":"2ff1b193-c3da-4e4b-b33c-6f7a9bcef482","order_by":3,"name":"Gina Caetano","email":"","orcid":"","institution":"ISR-Lisboa/LARSyS, Instituto Superior Técnico – Universidade de Lisboa","correspondingAuthor":false,"prefix":"","firstName":"Gina","middleName":"","lastName":"Caetano","suffix":""},{"id":489820165,"identity":"420c2688-142c-4f06-ba5d-e0901f49f0b9","order_by":4,"name":"Patrícia Figueiredo","email":"","orcid":"","institution":"ISR-Lisboa/LARSyS, Instituto Superior Técnico – Universidade de Lisboa","correspondingAuthor":false,"prefix":"","firstName":"Patrícia","middleName":"","lastName":"Figueiredo","suffix":""}],"badges":[],"createdAt":"2025-07-04 11:38:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7046365/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7046365/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-23131-7","type":"published","date":"2025-11-11T15:57:09+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":87576431,"identity":"e3f759a8-0eb1-4efd-a3aa-139d3de4f470","added_by":"auto","created_at":"2025-07-25 11:42:18","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":72264,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic representation of task timings for resting state (RS) and the two respiratory modulations, slow- paced breathing (SPB) and breath-holding (BH). Each task is divided into distinct intervals marked along the horizontal axis indicating the corresponding breathing pattern.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7046365/v1/02b02204f5e489ef9c35d013.png"},{"id":87578102,"identity":"d508b4c2-7e03-4f49-af78-7ebb938af701","added_by":"auto","created_at":"2025-07-25 11:58:18","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":239360,"visible":true,"origin":"","legend":"\u003cp\u003eData analysis pipeline. \u003cstrong\u003eAcronyms:\u003c/strong\u003e BOLD = blood oxygen level dependent; CR = cardiac rate; dCR = cardiac rate derivative; dRVT = respiratory volume per time derivative; GLM = general linear model; GM = gray matter; MP = motion parameters; MO = motion outliers; RVT = respiratory volume per time; VE = variance explained.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7046365/v1/1c14abff009110bae4c1fb5f.png"},{"id":87576445,"identity":"5a027858-56ee-4864-927a-f2eff0c2c62f","added_by":"auto","created_at":"2025-07-25 11:42:18","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":372779,"visible":true,"origin":"","legend":"\u003cp\u003eDistributions across subjects of the similarity metrics between each EDR method and Resp (x-axis, colours): Correlation (Optimal lag: maximum cross-correlation over a lag = +/-5s; No lag: correlation without at lag = 0 s); and Coherence. Square brackets with * outside the plots denote significant differences between methods (Kruskal-Wallis test, corrected p-values); * inside the plots denote values significantly larger than zero for each method (Wilcoxon signed-rank test, corrected p-values). \u003cstrong\u003eAcronyms:\u003c/strong\u003eEnv = ECG Envelope, HRV = Heart Rate Variability, AM = Amplitude Modulation, QRS-AM = QRS-area Modulation, PCA = Principal Component Analysis, kPCA = Kernel Principal Component Analysis, EMD = Empirical Mode Decomposition.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7046365/v1/45fd2b72f99fd210a1027d5e.png"},{"id":87576447,"identity":"fb0c22c1-1de2-47dc-9374-d8b6e46ff8c5","added_by":"auto","created_at":"2025-07-25 11:42:18","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":205568,"visible":true,"origin":"","legend":"\u003cp\u003eRespiratory signals (black) and corresponding EDRs shifted by the optimal lag (colour), for the three tasks, in an illustrative subject: (Top) Signal amplitudes during a representative 30 s period of the task; and (Bottom) Power spectra for the whole signal (shaded area corresponds to the full width at half maximum of the power). The strong respiratory modulations in the SPB and BH tasks can be appreciated in the substantially different amplitude and spectral profiles. \u003cstrong\u003eAcronyms:\u003c/strong\u003e Env = ECG Envelope, HRV = Heart Rate Variability, AM = Amplitude Modulation, QRS-AM = QRS-area Modulation, PCA = Principal Component Analysis, kPCA = Kernel Principal Component Analysis, EMD = Empirical Mode Decomposition.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7046365/v1/1b0e5d549db512a755809304.png"},{"id":87576456,"identity":"5554ea34-7337-472d-9e8a-0ac0a3be7774","added_by":"auto","created_at":"2025-07-25 11:42:18","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":100170,"visible":true,"origin":"","legend":"\u003cp\u003eDistributions across subjects of the time lag yielding the maximum correlation over a range of +/-5s between each EDR method and Resp (x-axis, colours), for each task. The scattering of optimal time lags varies across methods and is larger for the respiratory challenges, particularly SPB. \u003cstrong\u003eAcronyms:\u003c/strong\u003e Env = ECG Envelope, HRV = Heart Rate Variability, AM = Amplitude Modulation, QRS-AM = QRS-area Modulation, PCA = Principal Component Analysis, kPCA = Kernel Principal Component Analysis, EMD = Empirical Mode Decomposition.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7046365/v1/2827cf129e18fac28e7a9e15.png"},{"id":87576438,"identity":"af2eae4f-fcc4-4545-aa20-bd8380c4fafb","added_by":"auto","created_at":"2025-07-25 11:42:18","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":205905,"visible":true,"origin":"","legend":"\u003cp\u003eDistributions across subjects of the GM average fMRI signal Variance Explained (%VE) by the respiratory regressors relative to a basic model (VE\u003csub\u003ePhysio\u003c/sub\u003e-VE\u003csub\u003eBasic\u003c/sub\u003e), for Resp (gray box plot) and each EDR method (x-axis, colours): (Top) Physio - rRetr: including the RETROICOR respiratory terms; and (Bottom) Physio – RVT: includes RVT. Square brackets with * outside the plots denote significant differences between methods (Kruskal-Wallis test, corrected p-values); * inside the plots denote values significantly larger than zero for each method (Wilcoxon signed-rank test, corrected p-values). \u003cstrong\u003eAcronyms:\u003c/strong\u003e Env = ECG Envelope, HRV = Heart Rate Variability, AM = Amplitude Modulation, QRS-AM = QRS-area Modulation, PCA = Principal Component Analysis, kPCA = Kernel Principal Component Analysis, EMD = Empirical Mode Decomposition.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7046365/v1/d6034578a4f5acb53648a4f5.png"},{"id":87576443,"identity":"a6546c2c-222a-4af9-a1ab-e5170237c997","added_by":"auto","created_at":"2025-07-25 11:42:18","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":349865,"visible":true,"origin":"","legend":"\u003cp\u003eGM average BOLD percent signal change in each task, for an illustrative subject: original preprocessed BOLD signal (GM psc) and respective model fit using only basic regressors (Basic fit) and cardiac regressors adding either RETROICOR respiratory terms (Physio-rRetr fit) or RVT (Physio – RVT fit), obtained from Resp and HRV.\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7046365/v1/51ee1639d8d8a86ddd0206fb.png"},{"id":87577111,"identity":"5136a131-54f3-4d76-8171-54846a317a5b","added_by":"auto","created_at":"2025-07-25 11:50:19","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":331113,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePhysiological noise:\u003c/strong\u003e mapping of fMRI changes associated with RVT for RS. A. Illustrative subject: map of significant BOLD modulation with RVT (z-stat map, cluster threshold 2.3; p-value \u0026lt; 0.05; and B. Group results: distributions across subjects of volume (nb voxels, top) and average amplitude (% signal change, bottom) of significant BOLD modulation with RVT.\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7046365/v1/9f3758a203139bf582121739.png"},{"id":87576440,"identity":"bb494954-0fce-4fb2-968c-2bc033c7ed49","added_by":"auto","created_at":"2025-07-25 11:42:18","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":306360,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePhysiological noise:\u003c/strong\u003e mapping of fMRI changes associated with RVT for SPB. A. Illustrative subject: map of significant BOLD modulation with RVT (z-stat map, cluster threshold 2.3; p-value \u0026lt; 0.05) and B. Group results: distributions across subjects of volume (nb voxels, top) and average amplitude (% signal change, bottom) of significant BOLD modulation with RVT.\u003c/p\u003e","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7046365/v1/fc0a2b9a1fe4aac6261c2c51.png"},{"id":87576439,"identity":"cee7c4bd-9e96-416c-9f8b-6a7cb51956c9","added_by":"auto","created_at":"2025-07-25 11:42:18","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":499751,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCVR imaging:\u003c/strong\u003e mapping of fMRI changes associated with RVT for BH; A. Illustrative subject: map of significant BOLD modulation with RVT (z-stat map, cluster threshold 3.1; p-value \u0026lt; 0.05 and B. Group results: distribution across subjects of volume (nb voxels) and average amplitude (% signal change) of BOLD modulation with RVT.\u003c/p\u003e","description":"","filename":"floatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7046365/v1/47c6cb10553a2697594b50d1.png"},{"id":96104975,"identity":"e9e4ba29-ab72-403b-be61-a2ebc54a9cc1","added_by":"auto","created_at":"2025-11-17 16:05:36","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3095518,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7046365/v1/f0c4ab1d-ade1-4dd8-9833-6c358283e9a0.pdf"},{"id":87576432,"identity":"4201f4b1-ffec-468a-a4f7-97459b6670eb","added_by":"auto","created_at":"2025-07-25 11:42:18","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":542827,"visible":true,"origin":"","legend":"","description":"","filename":"manuscriptSREDRsupplementary.docx","url":"https://assets-eu.researchsquare.com/files/rs-7046365/v1/b587a39f5eef9c564955311d.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Using ECG-derived respiration for explaining BOLD-fMRI fluctuations during rest and respiratory modulations","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eFunctional magnetic resonance imaging (fMRI) indirectly assesses brain activity by measuring changes in the blood-oxygen level dependent (BOLD) signal, allowing the mapping of brain regions involved in task-induced activation or functionally connected. Since the BOLD contrast relies on hemodynamic changes, which are also influenced by physiological processes such as cardiac pulsatility and respiration, these may represent confounds when interpreting the BOLD signal \u003csup\u003e\u003cspan additionalcitationids=\"CR2 CR3\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Physiological signal acquisition during fMRI may serve multiple purposes: modeling and correction of non-neuronal BOLD sources, study of physiological modulations of brain activity, or simply monitoring physiological states throughout experiments \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. Although there are some fMRI data-driven methods used for physiological denoising (e.g. CompCor, Behzadi et al. (2007); FMRIB\u0026rsquo;s ICA-based X-noiseifier (FIX), Salimi-Khorshidi et al. (2014); ICA-based Automatic Removal Of Motion Artifacts (ICA-AROMA), Pruim et al. (2015)), most methods rely on measuring cardiac and respiratory signals. Critically, this requires additional setup \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e, such as respiratory belts and photoplethysmography, which increases experimental complexity, prolongs preparation times and heightens subject discomfort, representing an additional burden.\u003c/p\u003e\u003cp\u003eRespiration is known to modulate the ECG, affecting the morphology of the heartbeats \u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, altering the electrical impedance of the thoracic cavity due to filling/emptying of the lungs \u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, and affecting the heart rate \u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. Consequently, it is possible to obtain an ECG-derived respiration (EDR) signal \u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Methods for EDR estimation have been widely explored in the context of ambulatory care \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e, but only one previous EEG-fMRI study reported the use of EDRs in the MRI environment \u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. In the case of EEG-fMRI, placing a single-lead ECG on the back is a common procedure, for \u003cem\u003ea posteriori\u003c/em\u003e correction of the pulse artifact induced on the EEG. Therefore, for EEG-fMRI setups, the ECG is usually recorded and may be used to retrieve respiration, in case the dedicated equipment is not available, or the recorded respiration is affected by technical issues. The study by Abreu et al. (2016) compared the use of a subject-specific physiological model derived from several EDRs with an image-based physiological model (based on cerebrospinal fluid (CSF) and white matter (WM) average time courses) and with FIX \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. The data corrected using the physiological model achieved the best performance in terms of sensitivity and specificity to map epileptic networks and the default mode network. Unfortunately, the MRI environment introduces distortions in the ECG wave morphology as a consequence of both magnetic induction due to movements inside the main magnetic field as well as the effects of radiofrequency pulses \u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. Because the respiratory data were not recorded in that study to provide a ground-truth for comparison with the EDR, it remains to be validated whether EDRs in the MRI environment can be robustly obtained.\u003c/p\u003e\u003cp\u003eIn this work, we aim to fill this gap in the literature by evaluating EDR signals obtained during EEG-fMRI acquisitions. For this purpose, we collected EEG-fMRI data including simultaneous ECG and respiration recordings, from healthy participants during resting state and while they performed two respiratory challenges aimed at modulating respiration patterns: a slow-paced breathing task and a breath-hold task. We considered these tasks to test the ability of EDR to capture such respiratory variations, which are commonly utilized to assess cerebrovascular reactivity (CVR) \u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. We compared EDR signals with the measured respiratory signals, by assessing their similarity in time and frequency (correlation and coherence) and their ability to predict physiological contributions in fMRI signals, both for physiological denoising and CVR mapping.\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Data acquisition\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eData were acquired in the context of a larger research project on brain imaging in migraine (MigN2Treat), taking place at Hospital da Luz, Lisboa. The methods and results corresponding to the other MRI protocols are described elsewhere \u003csup\u003e\u003cspan additionalcitationids=\"CR19 CR20 CR21 CR22 CR23\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. The research protocol and statistical analysis were not preregistered. All methods were performed in accordance with the relevant guidelines and regulations. The study was approved by the Hospital da Luz Ethics Committee and all participants provided written informed consent according to the Declaration of Helsinki 7th revision.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cdiv id=\"Sec4\" class=\"Section3\"\u003e\u003ch2\u003e2.1.1 Sample\u003c/h2\u003e\u003cp\u003eData was acquired from 15 female healthy subjects (age: median\u0026thinsp;=\u0026thinsp;30, IQR\u0026thinsp;=\u0026thinsp;10.5). The participants were healthy, without any diagnosed condition that significantly impaired an active and productive life or reduced their life expectancy to be below 5 years. None of them was undergoing treatment with psychoactive drugs, including anxiolytics, antidepressants, anti-epileptics, or any migraine prophylactics. Furthermore, being pregnant or having contraindications for MRI scanning (e.g. claustrophobia, pacemaker, incompatible implants) were also exclusion criteria.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section3\"\u003e\u003ch2\u003e2.1.2 MRI data\u003c/h2\u003e\u003cp\u003eThe MRI data acquisition was performed with a 3T Siemens Vida system, using a 64-channel RF coil. Functional T2*-weighted images were obtained using 2D-Echo Planar Imaging (EPI), TR/TE\u0026thinsp;=\u0026thinsp;1260/30ms, flip angle\u0026thinsp;=\u0026thinsp;70\u0026ordm;, in-plane generalized autocalibrating partially parallel acquisition (GRAPPA) acceleration factor of 2, simultaneous multi-slice (SMS) with a factor of 3, 60 slices, with 2.2 mm isotropic resolution. Structural T1-weighted images were acquired with a magnetization-prepared rapid gradient echo (MPRAGE) sequence, with TR\u0026thinsp;=\u0026thinsp;2300 ms, TE\u0026thinsp;=\u0026thinsp;2.98 ms, inversion time (TI)\u0026thinsp;=\u0026thinsp;900 ms, and 1 mm isotropic resolution. Field map magnitude and phase images were obtained using a double-echo gradient echo sequence (TR\u0026thinsp;=\u0026thinsp;400.0 ms, TE\u0026thinsp;=\u0026thinsp;4.92/7.38 ms, voxel size: 3.4 \u0026times; 3.4 \u0026times; 3, flip angle 60◦).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section3\"\u003e\u003ch2\u003e2.1.3 Physiological data\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe respiratory signal was continuously recorded with the integrated BioMatrix Sensors from Siemens, through Siemens Physiological Monitoring Unit (PMU), with a sampling frequency of 400Hz. The electrocardiogram (ECG) was acquired simultaneously with fMRI as part of the MR-compatible BrainAmp MR EEG system (Brain Products), with a sampling frequency of 5000Hz, using an Ag/AgCl ring-type electrode placed on the back as caudally as possible, on the left of the spine. The impedance was maintained below 25 kOhm. The SyncBox was used to ensure synchronization between the MRI scanner clock (10 MHz) and the BrainAmp MR system, to improve the quality of the gradient artifact (GA) correction.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section3\"\u003e\u003ch2\u003e2.1.4 Tasks\u003c/h2\u003e\u003cp\u003eEEG-fMRI data was acquired during resting state (RS) and two respiratory tasks involving respiratory modulations, slow-paced breathing (SPB) and breath-holding (BH). Due to technical issues, some respiratory recordings had missing values at the start or end. These signals were excluded from the analysis and were therefore not included in our sample sizes, except for RS. Consequently, not all tasks were analyzed for every subject, with only a subgroup of participants considered for each task. For RS, we retained the maximum amount of available data that still provided a reasonable duration. Although the original task lasted 7 minutes, only 4.82 minutes (229 fMRI volumes) per subject were usable for analysis. During RS (N\u0026thinsp;=\u0026thinsp;10), subjects were instructed to keep their eyes open, looking at a black screen, without thinking of anything in particular. The SPB task (N\u0026thinsp;=\u0026thinsp;9) consisted of an initial fixation cross (5 s), a 1-min free breathing period, a 2-min 0.1 Hz paced respiration with visual written instructions for inhaling (5 s) and exhaling (5 s), and another 1-min free breathing period (194 fMRI volumes). The BH task (N\u0026thinsp;=\u0026thinsp;12) comprised 4 cycles of post-exhalation breath-hold (15 s) followed by free breathing and naturally paced breathing, with a total of approximately 4.47 min (213 fMRI volumes). The natural breathing frequency of each subject was computed beforehand based on a calibration recording performed outside the scanner, during a period longer than 1 minute while the subject was not performing any particular task. The median breathing period was 4 s for all task sub-samples, ranging from 3\u0026ndash;4 s for RS, and 3\u0026ndash;5 s for BH and SPB sub-samples. For each subject, all tasks were performed on the same day. The timings of each task are depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Data analysis\u003c/h2\u003e\u003cp\u003eThe fMRI data was preprocessed using FMRIB Software Library (FSL) tools \u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. MATLAB version 2016b was used for respiration and ECG data preprocessing as well as further analysis and visualization. The data analysis pipeline is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cdiv id=\"Sec9\" class=\"Section3\"\u003e\u003ch2\u003e2.2.1 Data preprocessing\u003c/h2\u003e\u003cdiv id=\"Sec10\" class=\"Section4\"\u003e\u003ch2\u003e2.2.1.1 fMRI\u003c/h2\u003e\u003cp\u003eEPI distortion correction was performed using FMRIB's Utility for Geometrically Unwarping EPIs (FUGUE), to correct for geometric distortions and signal loss, using a fieldmap acquisition to characterize B0 field inhomogeneities, containing magnitude and phase information. The Brain Extraction Tool (BET) was used to remove non-brain tissue. Volume realignment was performed by applying FMRIB's Linear Image Registration Tool (FLIRT) with respect to the middle volume, using 6 rigid body motion parameters (MP). Motion outliers (MO) were obtained using FSL Motion Outliers tool, with the dvars option, which considers the root mean square intensity difference of volume N to volume N\u0026thinsp;+\u0026thinsp;1, thresholded at the 75th percentile\u0026thinsp;+\u0026thinsp;1.5 times the interquartile range (Power et al. 2012). High-pass temporal filtering with a cut-off frequency of 0.01 Hz was applied to remove slow drift fluctuations and spatial smoothing was performed using SUSAN tool (S.M. Smith and J.M. Brady, 1997), employing a Gaussian kernel with a full width half maximum of 3.3 mm for RS and 3.5 mm for SPB and BH. The code used for fMRI preprocessing can be accessed here: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/martaxavier/fMRI-Preprocessing\u003c/span\u003e\u003cspan address=\"https://github.com/martaxavier/fMRI-Preprocessing\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e\u003cp\u003eThe structural image was used to obtain a gray matter (GM) mask for the functional images. For that, following non-brain tissue removal, tissue segmentation was performed with FMRIB's Automated Segmentation Tool (FAST) to obtain GM partial volume effect (PVE) images, which were then thresholded, binarized and transformed to the subject\u0026rsquo;s functional space using a registration matrix to the structural image obtained by applying FSL\u0026rsquo;s FLIRT with 12 degrees of freedom.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section4\"\u003e\u003ch2\u003e2.2.1.2 Respiration\u003c/h2\u003e\u003cp\u003eRespiratory data (Resp) was matched with the fMRI acquisition, downsampled to 250 Hz and bandpass filtered from 0.01 to 1 Hz using a Butterworth filter of order 2. Peak detection was performed to find maximum and minimum values of inspiration and expiration, using a threshold based on the 80th percentile of the data for RS and BH, and on the 40th percentile for the SPB, since the amplitude of the signal during slow breathing was larger. This procedure was followed by manual verification and the threshold was readjusted to the 75th percentile for one subject.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section4\"\u003e\u003ch2\u003e2.2.1.3 ECG\u003c/h2\u003e\u003cp\u003eThe ECG was corrected for the GA using the average artifact subtraction (AAS) method \u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e, implemented in the FMRIB plugin of the EEGLAB toolbox, using windows with 30 volumes and without performing adaptive noise cancelation, to avoid distorting the ECG waveform. The signal was then downsampled to 250Hz and ECG R-peaks were automatically detected using a long short term memory (LSTM) network trained on ECG data acquired in the MR environment (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/LaSEEB/deepQRS\u003c/span\u003e\u003cspan address=\"https://github.com/LaSEEB/deepQRS\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and manually inspected and corrected using the interactive tool (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/LaSEEB/interactiveQRS\u003c/span\u003e\u003cspan address=\"https://github.com/LaSEEB/interactiveQRS\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Furthermore, we corrected for missing and false heart beats, using the approach proposed by de Chazal et al. (2003), which had also been employed by Varon et al. (2020) prior to EDR computation. This procedure resulted in the detection and correction of a missing peak for one subject, for the BH task. Finally, the signal was bandpass filtered 0.05-40 Hz using a FIR filter with a Hamming window. The lower edge of the band was chosen considering the preservation of respiratory frequencies.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section3\"\u003e\u003ch2\u003e2.2.2 ECG-derived respiration (EDR) extraction categories\u003c/h2\u003e\u003cp\u003eEDRs were extracted using a publicly available repository (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/rmabreu/Respiratory_Signal_Estimation\u003c/span\u003e\u003cspan address=\"https://github.com/rmabreu/Respiratory_Signal_Estimation\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) which comprises methods from 7 categories: ECG envelope (ENV), heart-rate variability (HRV), amplitude modulation (AM), QRS-area modulation (QRS-AM), principal component analysis (PCA), kernel PCA (kPCA), and empirical mode decomposition (EMD), previously described in the work by Abreu et al. (2017). Each method leverages different ECG features with some focusing on amplitude (e.g., ENV, AM, QRS-AM), others on frequency components (e.g., HRV, EMD), and some using statistical decomposition techniques (e.g., PCA, kPCA). Several EDRs were obtained for AM, PCA and kPCA (e.g. variations relying on different ECG features, such as using either the whole beat or only the QRS or T wave portions), resulting in 51 EDRs per task for each subject. Some of the methods induced a steep rise/decrease of the EDR signal at the start or end, due to insufficient information for an accurate respiratory estimate. To correct abnormal fluctuations at the signal edges, we applied nearest-value interpolation if the absolute gradient in the first or last second exceeded 80% of the maximum absolute gradient of the entire signal. Peak detection of minimum and maximum values was performed using the same procedure that had been employed for respiratory signals.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section3\"\u003e\u003ch2\u003e2.2.3 Similarity between respiratory signals\u003c/h2\u003e\u003cp\u003eWe assessed the similarity between Resp and EDR signals in both the time and frequency domains, using correlation and coherence, respectively. To compute the similarity metrics for each participant and task, both signals were first standardized (mean of zero and standard deviation of one) and downsampled to 10 Hz \u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e,\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. Correlation was computed as the maximum cross correlation coefficient over a lag of +/-5s, optimizing for phase delays, and with no lag, to mimic a real situation with unknown optimal lags. Coherence was based on the magnitude squared coherence, estimated using Welch's method to obtain the spectra (1024-point Fast Fourier Transform; Hamming window; 8 segments with equal length and 50% of overlap). The mean magnitude squared coherence was computed for a range around the fundamental respiratory frequency (fR), considered to be the frequency with highest power within 0.15\u0026ndash;0.45 Hz for RS and BH, and within 0.08\u0026ndash;0.45 Hz for SPB. The range was determined by the frequencies around fR with more than half of the peak power \u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e,\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eFor each category, we identified the EDR method yielding the highest correlation with Resp (using the optimal lag) for each task. The EDR method selected for each category was then identified based on a majority vote across the three tasks. For AM, a version without baseline removal was chosen by consensus, being the highest across all tasks. Both for PCA and kPCA, methods using the first principal component and the whole ECG beat were chosen, having achieved the highest correlation for SPB and BH, respectively. For kPCA, the method also included a polynomial kernel of order 2.\u003c/p\u003e\u003cp\u003eTo determine whether the correlations were significantly greater than zero, we conducted a Wilcoxon signed-rank test for each EDR category and task, applying the Benjamini-Hochberg procedure to control the false discovery rate (FDR) across multiple comparions. To evaluate differences between categories within each task, we used a Kruskal-Wallis test followed by post hoc pairwise comparisons using the Dunn test with Sid\u0026aacute;k correction to control the family wise error rate (FWER). The significance level was set as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\alpha\\:\\)\u003c/span\u003e\u003c/span\u003e = 0.05.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section3\"\u003e\u003ch2\u003e2.2.4 Physiological signal modeling in fMRI\u003c/h2\u003e\u003cp\u003eTo explore the impact on the physiological noise correction, nuisance regressors regarding respiration and cardiac pulsatility were obtained (with no lag): retrospective image correction (RETROICOR) terms, respiratory volume per time (RVT) and cardiac rate (CR). For RETROICOR, the phases of respiratory and cardiac cycles at each fMRI volume acquisition were estimated. A Fourier series up to the second order was built from respiratory and cardiac terms, accounting for quasi periodic signal fluctuations \u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. RVT considers non-periodic signal fluctuations due to changes in the depth and rate of breathing \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. The RVT timecourse was obtained from the respiratory peaks and troughs using the amplitude (difference between consecutive maximum and minimum) divided by the time period between them. The result was interpolated to obtain a uniform sampling frequency corresponding to that of the fMRI, normalized and convolved with the respiratory response function as described by Birn et al. (2008). Similarly, for CR, the time difference between consecutive R peaks, which had been previously detected, was used to obtain the heart rate. The heart rate was interpolated, normalized and convolved with the cardiac response function as described in \u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. The respiratory regressors were estimated from EDRs and Resp, and their similarity was assessed using the correlation coefficient, in this case without considering a lag.\u003c/p\u003e\u003cp\u003eThe GM BOLD percent signal change (PSC) was calculated by first averaging the BOLD time series across all voxels within the GM mask to obtain a mean time course. The mean of this time course was then subtracted from each time point, and the result was divided by the same mean before being multiplied by 100. Three general linear models (GLM) were fitted to the GM BOLD PSC: Basic (6 MP, variable number of MO and, only for SPB, a task block convolved with the double-gamma hemodynamic response function (HRF)); Physio-rRetr (RETROICOR cardiac terms; CR; RETROICOR respiratory terms); and Physio-RVT (RETROICOR cardiac terms; CR; RVT). The two GLM\u0026rsquo;s including respiratory regressors were estimated using either EDR or Resp for subsequent comparison. The variance explained (%VE) for each GLM was computed as 100 x adjusted R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e (R\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e\u003csub\u003eadj\u003c/sub\u003e), and the %VE by the physiological terms was then obtained as (VE\u003csub\u003ePhysio\u003c/sub\u003e-VE\u003csub\u003eBasic\u003c/sub\u003e) for Physio-rRetr and Physio-RVT.\u003c/p\u003e\u003cp\u003eAs previously, we conducted a Wilcoxon signed-rank test for each EDR category and task to assess whether the variance explained was significantly greater than zero, applying the Benjamini-Hochberg procedure to control the FDR. In this case, respiration was also included in the comparison alongside the EDR categories. To evaluate differences among categories and compared them with Resp within each task, we used a Kruskal-Wallis test followed by post-hoc pairwise comparisons corrected using the Dunn test with Sid\u0026aacute;k correction, to control FWER. The significance level was set as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\alpha\\:\\)\u003c/span\u003e\u003c/span\u003e = 0.05.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec16\" class=\"Section3\"\u003e\u003ch2\u003e2.2.5 Mapping fMRI changes associated with RVT\u003c/h2\u003e\u003cp\u003eThe BOLD changes associated with RVT were mapped by fitting a GLM for each task, using FEAT (FMRI Expert Analysis Tool) v6.00. For RS and SPB, we incorporated RVT and CR in the GLM, along with their respective derivatives as well as MP and MO, as nuisance regressors. For SPB, we also added the task paradigm convolved with a double-gamma HRF to the GLM as the regressor of interest. Two models were created for each subject, one using RVT derived from Resp and another using RVT derived from the EDR method selected in the previous analysis. A cluster threshold of z-stat\u0026thinsp;\u0026gt;\u0026thinsp;2.3 and a cluster significance level p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 were applied to generate thresholded z-stat maps for RVT. For BH, we included RVT in the GLM, as well as its temporal derivative, as a regressor of interest, since it may be used as a proxy of PetCO2 for CVR mapping \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. We also included MP and MO in the GLM as nuisance regressors. In this case, the clustering threshold was increased to z-stat\u0026thinsp;\u0026gt;\u0026thinsp;3.1 to enhance the significance of the results. For the three tasks, we examined the maps of BOLD changes associated with each regressor, assessing the number of voxels that survived the threshold as well as the average PSC across the whole brain, GM and WM. Differences between Resp and HRV were tested using a Wilcoxon rank-sum test, followed by the Benjamini-Hochberg procedure to control the FDR. The significance level was set as \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\alpha\\:\\)\u003c/span\u003e\u003c/span\u003e = 0.05.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\n \u003ch2\u003e3.1 Similarity metrics\u003c/h2\u003e\n \u003cp\u003eResults are presented only for the EDR methods yielding the highest correlation coefficients with Resp for each category. The results for all methods are presented in the Supplementary Material. The similarity metrics computed between the EDR signals and Resp across all subjects and tasks are presented in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e. We found that, using optimal lags, the correlation between Resp and EDRs is significantly larger than zero for all categories, for RS, SPB and BH. Among the methods, EMD consistently outperforms others, while AM performs significantly worse than at least one other method for all tasks. Notably, HRV is significantly outperformed by other methods for RS and BH, but for SPB it shows high values, significantly exceeding AM. At zero lag, the differences in correlation between EDR methods become more pronounced. PCA and EMD consistently show stronger correlations, while Env and HRV perform consistently worse than at least two other methods. In terms of coherence, amplitude-based methods consistently show the poorest performance across all tasks. For RS, the methods that are not based on the amplitude perform similarly among them, with only EMD outperforming QRS-AM (amplitude-based). For SPB, the Env and HRV methods significantly outperform the lowest-performing method, AM. For BH, more subjects exhibit low coherence values across all methods, with no significant differences among them.\u003c/p\u003e\n \u003cp\u003eIn line with the previous results, the example in Fig. \u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e shows that the EDR methods differ in their ability to accommodate breathing changes. EDR waveforms of some of the methods resemble respiration, though with amplitude differences and a time shift. Nevertheless, in general, their power spectra overlap the respiratory frequency band.\u003c/p\u003e\n \u003cp\u003eFurthermore, the differences obtained for correlations considering the optimal lag relative to lag zero may be explained by the large scattering of optimal time lags, which are also mostly not centered around zero, as shown in Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\n \u003ch2\u003e3.2 Physiological signal modelling in fMRI\u003c/h2\u003e\n \u003cp\u003eFigure 6 depicts the average GM fMRI signal % VE by respiratory regressors (rRetr and RVT) computed based on Resp and the different EDR methods. The performance of EDRs for physiological regression of the BOLD signal differs for rRetr and RVT and is dependent on the task. Resp-based regressors only outperform the AM method, for the model including RVT for the BH task. In general, although smaller, the % VE by Resp is equivalent to the one explained by several EDR methods, with HRV method showing the best performance. Overall, results suggest that HRV is the most consistent method, particularly in the case of Physio-rRetr models. In general, the amplitude-based methods show the worst performance. For these reasons, the following analyses were performed only with HRV.\u0026nbsp;\u003c/p\u003e\n \u003cp\u003eFigure 7 shows that the Basic and both Physio models (using Resp and HRV, the best performing EDR method) can fit the data, following it more closely in the case of respiratory tasks, as expected.\u0026nbsp;\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\n \u003ch2\u003e3.3 Mapping fMRI changes associated with RVT\u003c/h2\u003e\n \u003cp\u003eThe results obtained for the mapping of BOLD changes associated with RVT, computed from Resp and EDR, for RS, SPB and BH, are presented in Figs. \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e, \u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e, and \u003cspan class=\"InternalRef\"\u003e10\u003c/span\u003e, respectively. For RS and SPB, no significant changes were found between Resp and EDR for either the number of voxels showing significant BOLD changes and the average PSC (Figs. \u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e and \u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e, panel B). For BH, fewer voxels were associated with the EDR-derived RVT compared with Resp for the whole brain, with gray and white matter following the same pattern. No significant differences were found between Resp and HRV (Wilcoxon rank-sum test, corrected p-values). Nonetheless, the average PSC seems to be equivalent for respiration and HRV, considering either the whole brain or gray/white matter.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eWe demonstrated the feasibility of extracting meaningful respiratory signals from the ECG recorded in the MRI environment for the first time. Using different methods, we compared the extracted EDR signals with the measured respiration signal, both during resting state and two respiration-modulation tasks. Overall, the HRV method, which relies on heart rate changes modulated by respiration, exhibited the best performance across tasks, explaining the BOLD-signal variance as well as measured respiration. Our results indicate that EDR holds great potential for physiological noise correction or CVR imaging, in cases when respiration cannot be recorded, or the signal is corrupted, particularly in the context of EEG-fMRI.\u003c/p\u003e\u003cp\u003e\u003cem\u003eSimilarity between EDR and measured respiration signals\u003c/em\u003e\u003c/p\u003e\u003cp\u003eIn general, all EDR methods yielded similar respiratory traces to the ground-truth, when using the optimal lag, with EMD showing the highest similarity. Consistently with our results, Abreu et al. (2017) also found this method to be the most accurate for EDR estimation in the MRI environment, in terms of both temporal dynamics and spectral content \u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e. Moreover, our results demonstrated high coherence values in addition to the strong correlations. Nevertheless, by comparing the EDR signals with the ground-truth respiration signals in our study, we found that correlations are generally low at a lag of 0 s, with the optimal time lag varying considerably between subjects. In fact, none of the methods could be linked to a fixed lag based on our time lag study, which would have enabled further research to use them in situations where the respiratory signal is unavailable for retrieval. Furthermore, it may be dependent on the respiratory frequency itself, which will in principle be unknown as well. This EDR time lag dependence and variability represents a limitation when a ground-truth respiration is not available. In contrast, coherence values are unaffected by time lags, and they were high across all tasks. The HRV method exhibited the highest coherence, despite low correlation. At the other end, amplitude-based methods had the lowest coherence values, indicating their inability to effectively capture the frequency content of the measured respiration.\u003c/p\u003e\u003cp\u003e\u003cem\u003ePhysiological signal modelling in fMRI analysis\u003c/em\u003e\u003c/p\u003e\u003cp\u003ePhysiological noise correction is often necessary to remove non-neuronal contributions to the BOLD signal. Because frequency features play an important role in estimating physiological regressors, the high coherence of HRV signals allowed this method to provide the best models for all tasks. Additionally, the voxelwise analysis showed that HRV-based EDR also yielded BOLD signal changes associated with RVT that were similar to those obtained using the measured respiration signals, for both resting state and slow-paced breathing. These results indicate that the HRV EDR approach may be useful for the purpose of physiological noise correction when respiration is lacking. For CVR imaging based on the breath-holding task, EDR-based models yielded similar BOLD percent signal change values compared with respiration, although with a considerably lower number of voxels activated across the whole brain and the gray matter. These results indicate that, despite being less sensitive, EDR may provide reliable CVR estimates.\u003c/p\u003e\u003cp\u003e\u003cem\u003eRelation with the EDR literature\u003c/em\u003e\u003c/p\u003e\u003cp\u003eAlthough EDR methods have not previously been evaluated against the respiration ground truth using ECG signals recorded in the MR environment, a substantial literature exists for recordings conducted outside of it. Of special relevance to our work, a few studies have investigated the performance of EDR methods in conditions with varying breathing patterns. Varon et al. (2020) systematically compared several types of methods for estimating EDR, assessing respiratory wave morphology, respiratory rate and cardiorespiratory information \u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. They applied the EDR methods to three datasets with a variety of physiological conditions, including relaxation, stress and healthy/abnormal sleep patterns. Their findings showed that simple methods based on morphological changes of the ECG caused by respiration, particularly those relying on the QRS complex, outperformed the other ones. The fact that our findings are not aligned with this study, may be explained by the distortion of the ECG in the MRI environment. Even after MR-artifact correction and further preprocessing, residual artifacts inevitably persist, and the typical ECG signal morphology is not preserved. As a result, amplitude-based methods are expected to perform poorly in our case. Additionally, the placement of the ECG electrode in the EEG-fMRI setup (on the subject's back, as recommended for artifact minimization) differs from the standard single-lead ECG montage used in most EDR studies. This may further affect the waveform morphology, making direct comparisons with the literature challenging.\u003c/p\u003e\u003cp\u003eMachine learning, including deep learning models, have also been utilized to infer respiratory information from fMRI spatiotemporal patterns \u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e,\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. However, these models are more complex and require larger datasets, which makes them more difficult to apply to the typically small EEG-fMRI datasets. Furthermore, they are more challenging to implement in real time when needed.\u003c/p\u003e\u003cp\u003e\u003cem\u003eLimitations\u003c/em\u003e\u003c/p\u003e\u003cp\u003eOne of the greatest limitations of our study is the need for optimizing the lag introduced by the EDR estimation, which may hinder the use of EDR signals alone for assessing respiratory patterns at specific time points. Another limitation is the strong dependence of the EDR on the quality of the ECG data, as any degradation in ECG signal quality can significantly affect the accuracy of EDR estimations. Furthermore, it can be difficult to distinguish between cardiac and respiratory contributions when analyzing fMRI data, primarily due to the inherent nature of the EDR signal, which integrates both physiological processes. Similarly, it may be that the good performance of the HRV method is also linked to the ability to explain an autonomic system component, as this system is closely related to heart rate variability in general and may contribute to the BOLD signal.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eOur study demonstrates the feasibility of extracting respiratory signals from ECG in the MRI environment and highlights the potential of this approach even in the presence of respiratory modulations, making it useful for monitoring and the computation of physiological regressors in fMRI analysis. The HRV method showed the best performance across tasks, indicating the potential of using EDRs as a physiological regressor in EEG-fMRI studies where direct respiration data is unavailable or corrupted. Importantly, we demonstrated the performance of EDR methods, not only during resting state, but also during respiratory manipulations of slow-paced breathing and breath-holding, broadening the range of potential applications.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding Declaration\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by LARSyS FCT funding [DOI: 10.54499/LA/P/0083/2020, 10.54499/UIDP/50009/2020, and 10.54499/UIDB/50009/2020], PRR project Center for Responsible AI [grant C645008882-00000055] and FCT [grants PD/BD/150356/2019, PTDC/EMD-EMD/29675/2017, LISBOA-01-0145-FEDER-029675].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contribution\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIE: Methodology, Software, Formal analysis, Investigation, Data Curation, Writing \u0026ndash; original draft, Writing \u0026ndash; review and editing, Visualization; ARF: Investigation, Resources, Data Curation; ART: Investigation, Resources, Data Curation; GC: Investigation, Resources, Writing \u0026ndash; review and editing; PF: Methodology, Conceptualization, Resources, Writing \u0026ndash; review and editing, Supervision, Project administration, Funding acquisition. All authors have approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData supporting this study will be made available by the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBirn, R. M., Diamond, J. B., Smith, M. A. \u0026amp; Bandettini, P. A. Separating respiratory-variation-related fluctuations from neuronal-activity-related fluctuations in fMRI. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 1536\u0026ndash;1548 (2006).\u003c/li\u003e\n\u003cli\u003eBirn, R. M., Smith, M. A., Jones, T. B. \u0026amp; Bandettini, P. A. The respiration response function: The temporal dynamics of fMRI signal fluctuations related to changes in respiration. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e40\u003c/strong\u003e, 644\u0026ndash;654 (2008).\u003c/li\u003e\n\u003cli\u003eChang, C., Cunningham, J. P. \u0026amp; Glover, G. H. Influence of heart rate on the BOLD signal: The cardiac response function. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e44\u003c/strong\u003e, 857\u0026ndash;869 (2009).\u003c/li\u003e\n\u003cli\u003eGlover, G. H., Li, T.-Q. \u0026amp; Ress, D. Image‐based method for retrospective correction of physiological motion effects in fMRI: RETROICOR. \u003cem\u003eMagn. Reson. Med. Off. J. Int. Soc. Magn. Reson. Med.\u003c/em\u003e \u003cstrong\u003e44\u003c/strong\u003e, 162\u0026ndash;167 (2000).\u003c/li\u003e\n\u003cli\u003eBulte, D. \u0026amp; Wartolowska, K. Monitoring cardiac and respiratory physiology during FMRI. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e154\u003c/strong\u003e, 81\u0026ndash;91 (2017).\u003c/li\u003e\n\u003cli\u003eBehzadi, Y., Restom, K., Liau, J. \u0026amp; Liu, T. T. A Component Based Noise Correction Method (CompCor) for BOLD and Perfusion Based fMRI. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e37\u003c/strong\u003e, 90\u0026ndash;101 (2007).\u003c/li\u003e\n\u003cli\u003eSalimi-Khorshidi, G. \u003cem\u003eet al.\u003c/em\u003e Automatic denoising of functional MRI data: Combining independent component analysis and hierarchical fusion of classifiers. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e90\u003c/strong\u003e, 449\u0026ndash;468 (2014).\u003c/li\u003e\n\u003cli\u003ePruim, R. H. R. \u003cem\u003eet al.\u003c/em\u003e ICA-AROMA: A robust ICA-based strategy for removing motion artifacts from fMRI data. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e112\u003c/strong\u003e, 267\u0026ndash;277 (2015).\u003c/li\u003e\n\u003cli\u003ePrzystup, P., Poliński, A. \u0026amp; Wtorek, J. QRS Morphology-Based EDR Signal\u0026mdash;Factors Determining its Properties. \u003cem\u003eIEEE Access\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 34665\u0026ndash;34676 (2022).\u003c/li\u003e\n\u003cli\u003eMoody, G. B., Mark, R. G., Zoccola, A. \u0026amp; Mantero, S. Derivation of Respiratory Signals from Multi-lead ECGs. \u003cem\u003eComput. Cardiol.\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 113\u0026ndash;116 (1985).\u003c/li\u003e\n\u003cli\u003eYasuma, F. \u0026amp; Hayano, J. Respiratory Sinus Arrhythmia. \u003cem\u003eChest\u003c/em\u003e \u003cstrong\u003e125\u003c/strong\u003e, 683\u0026ndash;690 (2004).\u003c/li\u003e\n\u003cli\u003eDong, K. \u003cem\u003eet al.\u003c/em\u003e An integrated framework for evaluation on typical ECG-derived respiration waveform extraction and respiration. \u003cem\u003eComput. Biol. Med.\u003c/em\u003e \u003cstrong\u003e135\u003c/strong\u003e, 104593 (2021).\u003c/li\u003e\n\u003cli\u003eBoyle, J., Bidargaddi, N., Sarela, A. \u0026amp; Karunanithi, M. Automatic Detection of Respiration Rate From Ambulatory Single-Lead ECG. \u003cem\u003eIEEE Trans. Inf. Technol. Biomed.\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 890\u0026ndash;896 (2009).\u003c/li\u003e\n\u003cli\u003eVaron, C. \u003cem\u003eet al.\u003c/em\u003e A Comparative Study of ECG-derived Respiration in Ambulatory Monitoring using the Single-lead ECG. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 5704 (2020).\u003c/li\u003e\n\u003cli\u003eAbreu, R. \u003cem\u003eet al.\u003c/em\u003e Ballistocardiogram artifact correction taking into account physiological signal preservation in simultaneous EEG-fMRI. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e135\u003c/strong\u003e, 45\u0026ndash;63 (2016).\u003c/li\u003e\n\u003cli\u003eNiendorf, T., Winter, L. \u0026amp; Frauenrath, T. Electrocardiogram in an MRI environment: Clinical Needs, Practical Considerations, Safety Implications, Technical Solutions and Future Directions. in \u003cem\u003eAdvances in Electrocardiograms: Methods and Analysis\u003c/em\u003e (BoD \u0026ndash; Books on Demand, 2012).\u003c/li\u003e\n\u003cli\u003ePinto, J., Bright, M. G., Bulte, D. P. \u0026amp; Figueiredo, P. Cerebrovascular Reactivity Mapping Without Gas Challenges: A Methodological Guide. \u003cem\u003eFront. Physiol.\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 608475 (2021).\u003c/li\u003e\n\u003cli\u003eDomingos, C. \u003cem\u003eet al.\u003c/em\u003e Impact of susceptibility-induced distortion correction on perfusion imaging by pCASL with a segmented 3D GRASE readout. \u003cem\u003eMagn. Reson. Imaging\u003c/em\u003e \u003cstrong\u003e102\u003c/strong\u003e, 141\u0026ndash;150 (2023).\u003c/li\u003e\n\u003cli\u003eEsteves, I. \u003cem\u003eet al.\u003c/em\u003e Uncovering longitudinal changes in the brain functional connectome along the migraine cycle: a multilevel clinical connectome fingerprinting framework. \u003cem\u003eJ. Headache Pain\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 29 (2025).\u003c/li\u003e\n\u003cli\u003eFouto, A. R. \u003cem\u003eet al.\u003c/em\u003e Impact of truncating diffusion MRI scans on diffusional kurtosis imaging. \u003cem\u003eMagn. Reson. Mater. Phys. Biol. Med.\u003c/em\u003e (2024) doi:10.1007/s10334-024-01153-y.\u003c/li\u003e\n\u003cli\u003eFouto, A. R. \u003cem\u003eet al.\u003c/em\u003e Alterations of white matter microstructure in migraine patients vary in the peri-ictal phases. \u003cem\u003eeNeuro\u003c/em\u003e ENEURO.0300-24.2024 (2024) doi:10.1523/ENEURO.0300-24.2024.\u003c/li\u003e\n\u003cli\u003eGuadilla, I. \u003cem\u003eet al.\u003c/em\u003e White matter alterations in episodic migraine without aura patients assessed with diffusion MRI: effect of free water correction. \u003cem\u003eJ. Headache Pain\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 31 (2025).\u003c/li\u003e\n\u003cli\u003eMatoso, A. \u003cem\u003eet al.\u003c/em\u003e Involvement of the cerebellum in structural connectivity enhancement in episodic migraine. \u003cem\u003eJ. Headache Pain\u003c/em\u003e \u003cstrong\u003e25\u003c/strong\u003e, 154 (2024).\u003c/li\u003e\n\u003cli\u003eRuiz-Tagle, A. \u003cem\u003eet al.\u003c/em\u003e Preserved working memory performance along with subcortical modulation during peri-ictal phases in spontaneous migraine attacks. \u003cem\u003eHeadache J. Head Face Pain\u003c/em\u003e \u003cstrong\u003e65\u003c/strong\u003e, 407\u0026ndash;419 (2025).\u003c/li\u003e\n\u003cli\u003eJenkinson, M., Beckmann, C. F., Behrens, T. E. J., Woolrich, M. W. \u0026amp; Smith, S. M. FSL. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e62\u003c/strong\u003e, 782\u0026ndash;790 (2012).\u003c/li\u003e\n\u003cli\u003eNiazy, R. K., Beckmann, C. F., Iannetti, G. D., Brady, J. M. \u0026amp; Smith, S. M. Removal of FMRI environment artifacts from EEG data using optimal basis sets. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e28\u003c/strong\u003e, 720\u0026ndash;737 (2005).\u003c/li\u003e\n\u003cli\u003ede Chazal, P. \u003cem\u003eet al.\u003c/em\u003e Automated processing of the single-lead electrocardiogram for the detection of obstructive sleep apnoea. \u003cem\u003eIEEE Trans. Biomed. Eng.\u003c/em\u003e \u003cstrong\u003e50\u003c/strong\u003e, 686\u0026ndash;696 (2003).\u003c/li\u003e\n\u003cli\u003eAbreu, R., Nunes, S., Leal, A. \u0026amp; Figueiredo, P. Physiological noise correction using ECG-derived respiratory signals for enhanced mapping of spontaneous neuronal activity with simultaneous EEG-fMRI. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e154\u003c/strong\u003e, 115\u0026ndash;127 (2017).\u003c/li\u003e\n\u003cli\u003eVaron, C. \u0026amp; Van Huffel, S. ECG-derived respiration for ambulatory monitoring. in \u003cem\u003e2015 Computing in Cardiology Conference (CinC)\u003c/em\u003e 169\u0026ndash;172 (IEEE, Nice, France, 2015). doi:10.1109/CIC.2015.7408613.\u003c/li\u003e\n\u003cli\u003eWidjaja, D., Varon, C., Dorado, A., Suykens, J. A. K. \u0026amp; Van Huffel, S. Application of Kernel Principal Component Analysis for Single-Lead-ECG-Derived Respiration. \u003cem\u003eIEEE Trans. Biomed. Eng.\u003c/em\u003e \u003cstrong\u003e59\u003c/strong\u003e, 1169\u0026ndash;1176 (2012).\u003c/li\u003e\n\u003cli\u003eZvolanek, K. M. \u003cem\u003eet al.\u003c/em\u003e Comparing end-tidal CO2, respiration volume per time (RVT), and average gray matter signal for mapping cerebrovascular reactivity amplitude and delay with breath-hold task BOLD fMRI. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e272\u003c/strong\u003e, 120038 (2023).\u003c/li\u003e\n\u003cli\u003eBayrak, R. G., Salas, J. A., Huo, Y. \u0026amp; Chang, C. A Deep Pattern Recognition Approach for Inferring Respiratory Volume Fluctuations from fMRI Data. in \u003cem\u003eMedical Image Computing and Computer Assisted Intervention \u0026ndash; MICCAI 2020\u003c/em\u003e (eds. Martel, A. L. et al.) 428\u0026ndash;436 (Springer International Publishing, Cham, 2020). doi:10.1007/978-3-030-59728-3_42.\u003c/li\u003e\n\u003cli\u003eAddeh, A. \u003cem\u003eet al.\u003c/em\u003e Direct machine learning reconstruction of respiratory variation waveforms from resting state fMRI data in a pediatric population. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cstrong\u003e269\u003c/strong\u003e, 119904 (2023).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"functional magnetic resonance imaging (fMRI), respiration, electrocardiogram (ECG), ECG-derived respiration (EDR), resting state, respiratory fluctuation","lastPublishedDoi":"10.21203/rs.3.rs-7046365/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7046365/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eRecording physiological signals during fMRI is valuable for multiple purposes but often requires additional setup, increasing complexity and participant discomfort. This is particularly challenging in simultaneous EEG-fMRI studies, which typically already include electrocardiogram (ECG) recordings. Here, we aim to leverage the known modulation of ECG by respiration to obtain an ECG-derived respiration (EDR) signal without extra equipment.\u003c/p\u003e\u003cp\u003eWe acquired EEG-fMRI data from 15 healthy subjects during resting state and two respiratory challenges (slow-paced breathing and breath-holding), with simultaneous ECG and respiratory recordings. Multiple methods were used to extract EDR signals, and the results were evaluated by comparing them with recorded respiration and assessing the quality of physiological regressors for denoising and cerebrovascular reactivity estimation.\u003c/p\u003e\u003cp\u003eAmplitude-based EDR methods showed lower correlations with respiration, likely due to ECG distortion in the MRI. Nevertheless, coherence analysis showed that EDR preserved the relevant spectral content. EDR-based regressors were similar to those obtained from measured respiration. Notably, a method based on heart rate variability performed best overall, yielding physiological noise correction and reactivity estimates comparable to those using recorded respiration.\u003c/p\u003e\u003cp\u003eOur results demonstrate that meaningful respiratory information can be extracted from ECG within the MRI environment, benefiting EEG-fMRI studies when respiration cannot be reliably recorded.\u003c/p\u003e","manuscriptTitle":"Using ECG-derived respiration for explaining BOLD-fMRI fluctuations during rest and respiratory modulations","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-25 11:42:13","doi":"10.21203/rs.3.rs-7046365/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-09-04T17:04:32+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-12T08:47:08+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-28T12:32:23+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"200450280314442705158810048577543210518","date":"2025-07-28T04:24:56+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"173656906655693508807533389257983434649","date":"2025-07-23T14:10:38+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"69647695400659819960676350501681983653","date":"2025-07-23T09:41:43+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-07-23T04:14:10+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-07-20T02:36:44+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-07-10T13:11:01+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-07-07T15:10:06+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-07-07T15:06:52+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"928b85cb-d31f-42bc-8e0d-3efdd6ede536","owner":[],"postedDate":"July 25th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":52004062,"name":"Biological sciences/Neuroscience/Neuro vascular interactions"},{"id":52004063,"name":"Biological sciences/Physiology/Neurophysiology"},{"id":52004064,"name":"Physical sciences/Engineering/Biomedical engineering"}],"tags":[],"updatedAt":"2025-11-17T15:59:55+00:00","versionOfRecord":{"articleIdentity":"rs-7046365","link":"https://doi.org/10.1038/s41598-025-23131-7","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-11-11 15:57:09","publishedOnDateReadable":"November 11th, 2025"},"versionCreatedAt":"2025-07-25 11:42:13","video":"","vorDoi":"10.1038/s41598-025-23131-7","vorDoiUrl":"https://doi.org/10.1038/s41598-025-23131-7","workflowStages":[]},"version":"v1","identity":"rs-7046365","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7046365","identity":"rs-7046365","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}