Eneko Uruñuela
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Eneko Uruñuela has a total of 270 co-authors. Here are the top co-authors by number of shared publications:
- César Caballero-Gaudes
IKERBASQUE, Basque Foundation for Science, Bilbao, Spain
13 shared publications - Stefano Moia
Maastricht University
6 shared publications - Javier Gonzalez-Castillo
Section on Functional Imaging Methods, Laboratory of Brain and Cognition, National Institute of Mental Health, National Institutes of Health, Bethesda, MD, USA
4 shared publications - Gang Chen
Scientific and Statistical Computing Core, National Institute of Mental Health, Bethesda, MD, USA
3 shared publications - Maite Termenon
UPV/EHU
3 shared publications - Molly G. Bright
Department of Biomedical Engineering, McCormick School of Engineering and Applied Sciences, Northwestern University, Evanston, IL, United States
3 shared publications
Latest Publications
- Open Access
- 27 October 2026
- Published by MIT Press in Imaging Neuroscience
A Multi-Echo-Informed Multimodal Spatio-Temporal CNN for ICA Component Classification in fMRI Denoising
Multi-echo functional MRI (fMRI) improves signal quality by exploiting the fact that true blood-oxygenation-level-dependent (BOLD) signals decay with T2* across echo times while many artifacts do not. By decomposing multi-echo fMRI data using independent component analysis (ICA), each component’s signal decay across echo times can be used to classify it as BOLD-like or non-BOLD-like. The tedana software classifies ICA components using metrics that capture this differential decay via rule-based decision trees. However, approximately 66% of components in our cohort were not classified with high confidence using tedana, leaving residual artifacts that limit downstream analysis sensitivity. To address this gap, we developed a multimodal convolutional neural network (CNN) that jointly learns from the spatial maps and time series of ICA components of optimally combined resting-state multi-echo data, using training labels derived from these multi-echo decay metrics, to more accurately classify these components. Trained on 12,750 high-confidence components from 229 subjects with perfect class balance, the model achieved 98.6% accuracy on held-out data and flagged as artifacts 10,704 components (33.3% of the non-high-confidence set) that tedana had accepted. Critically, denoising with CNN classifications significantly improved network-based contrast-to-noise ratio (CNR) by 6.20% (p=0.011), with six of seven resting-state networks showing gains. Furthermore, CNN-based denoising increased network modularity (Louvain Q +6.8%, p=0.0006) and showed numerically higher CNR test-retest reliability (ICC: 0.664 vs. 0.622). Ablation studies showed that the CNN’s CNR advantage arises from retaining components that tedana rejects, while the additional components it rejects can be removed without loss of network-level signal. On two independent external datasets, the CNN generalized without retraining: CNR gains were directionally consistent but non-significant on a resting-state cohort (+2.6%, p=0.36), whereas on a more dissimilar task-fMRI cohort (four echoes, different scanner, TR = 1.615 s) it significantly improved network CNR over tedana (+16.7%, p<0.001) while performing comparably to tedana in tSNR and task activation, which was enhanced over the non-denoised baseline. SmoothGrad analysis showed broad cortical saliency consistent with resting-state networks, strongest at inferior boundaries, for accepted components, and lower-magnitude edge saliency for rejected ones, confirming neurobiologically meaningful feature learning. These findings demonstrate that learned spatiotemporal features provide superior multi-echo ICA classification compared to traditional metric-based decision trees, offering an interpretable, data-driven solution for multi-echo fMRI denoising.
- 27 October 2026
- Published by MIT Press in Imaging Neuroscience
Open AccessA Multi-Echo-Informed Multimodal Spatio-Temporal CNN for ICA Component Classification in fMRI Denoising
Multi-echo functional MRI (fMRI) improves signal quality by exploiting the fact that true blood-oxygenation-level-dependent (BOLD) signals decay with T2* across echo times while many artifacts do not. By decomposing multi-echo fMRI data using independent component analysis (ICA), each component’s signal decay across echo times can be used to classify it as BOLD-like or non-BOLD-like. The tedana software classifies ICA components using metrics that capture this differential decay via rule-based decision trees. However, approximately 66% of components in our cohort were not classified with high confidence using tedana, leaving residual artifacts that limit downstream analysis sensitivity. To address this gap, we developed a multimodal convolutional neural network (CNN) that jointly learns from the spatial maps and time series of ICA components of optimally combined resting-state multi-echo data, using training labels derived from these multi-echo decay metrics, to more accurately classify these components. Trained on 12,750 high-confidence components from 229 subjects with perfect class balance, the model achieved 98.6% accuracy on held-out data and flagged as artifacts 10,704 components (33.3% of the non-high-confidence set) that tedana had accepted. Critically, denoising with CNN classifications significantly improved network-based contrast-to-noise ratio (CNR) by 6.20% (p=0.011), with six of seven resting-state networks showing gains. Furthermore, CNN-based denoising increased network modularity (Louvain Q +6.8%, p=0.0006) and showed numerically higher CNR test-retest reliability (ICC: 0.664 vs. 0.622). Ablation studies showed that the CNN’s CNR advantage arises from retaining components that tedana rejects, while the additional components it rejects can be removed without loss of network-level signal. On two independent external datasets, the CNN generalized without retraining: CNR gains were directionally consistent but non-significant on a resting-state cohort (+2.6%, p=0.36), whereas on a more dissimilar task-fMRI cohort (four echoes, different scanner, TR = 1.615 s) it significantly improved network CNR over tedana (+16.7%, p<0.001) while performing comparably to tedana in tSNR and task activation, which was enhanced over the non-denoised baseline. SmoothGrad analysis showed broad cortical saliency consistent with resting-state networks, strongest at inferior boundaries, for accepted components, and lower-magnitude edge saliency for rejected ones, confirming neurobiologically meaningful feature learning. These findings demonstrate that learned spatiotemporal features provide superior multi-echo ICA classification compared to traditional metric-based decision trees, offering an interpretable, data-driven solution for multi-echo fMRI denoising.
- Open Access
- 7 July 2025
- Published by MIT Press in Imaging Neuroscience
- Vol. 3
Uncovering the Neural Correlates of the Urge-to-Blink: A Study Utilising Subjective Urge Ratings and Paradigm Free Mapping
Neuroimaging plays a significant role in understanding the neurophysiology of Tourette syndrome (TS), in particular the main symptom, tics, and the urges associated with them. Premonitory urge is thought to be a negative reinforcer of tic expression in TS. Tic expression during neuroimaging is most often required as an overt marker of increased urge-to-tic, which can lead to considerable head movement, and thus data loss. This study aims to identify the brain regions involved in urge in healthy subjects using multi-echo functional MRI and a timing-free approach to localise the BOLD response associated with the urge-to-act without information of when these events occur. Blink suppression is an analogous behaviour that can be expressed overtly in the MRI scanner which gives rise to an urge like those described by individuals with TS. We examined the urge-to-blink in 20 healthy volunteers with an experimental paradigm including two conditions, “Okay to blink” and “Suppress blinking”, to identify brain regions involved in blink suppression. Multi-echo functional MRI data was analysed using a novel approach to investigate the BOLD signal correlated with the build-up of the urge-to-blink that participants continuously reported using a rollerball device. In addition, we used the method of multi-echo paradigm free mapping (MESPFM) to identify these regions without prior specification of task timings. Subjective urge scores were correlated with activity in the right posterior and ventral-anterior insula as well as the mid-cingulate and occipital cortices. Whereas blink suppression was associated with activation in the dorsolateral prefrontal cortex, cerebellum, right dorsal-anterior insula, mid-cingulate cortex and thalamus. These findings illustrate that different insula subregions contribute to the urge-for-action and suppression networks. The MESPFM approach showed co-activation of the right insula and cingulate cortex. The MESPFM activation maps showed the highest overlap with activation associated with blink suppression, as identified using general linear model analysis, demonstrating that activity associated with suppression can be determined without prior knowledge of task timings.
- 7 July 2025
- Published by MIT Press in Imaging Neuroscience
- Vol. 3
Open AccessUncovering the Neural Correlates of the Urge-to-Blink: A Study Utilising Subjective Urge Ratings and Paradigm Free Mapping
Neuroimaging plays a significant role in understanding the neurophysiology of Tourette syndrome (TS), in particular the main symptom, tics, and the urges associated with them. Premonitory urge is thought to be a negative reinforcer of tic expression in TS. Tic expression during neuroimaging is most often required as an overt marker of increased urge-to-tic, which can lead to considerable head movement, and thus data loss. This study aims to identify the brain regions involved in urge in healthy subjects using multi-echo functional MRI and a timing-free approach to localise the BOLD response associated with the urge-to-act without information of when these events occur. Blink suppression is an analogous behaviour that can be expressed overtly in the MRI scanner which gives rise to an urge like those described by individuals with TS. We examined the urge-to-blink in 20 healthy volunteers with an experimental paradigm including two conditions, “Okay to blink” and “Suppress blinking”, to identify brain regions involved in blink suppression. Multi-echo functional MRI data was analysed using a novel approach to investigate the BOLD signal correlated with the build-up of the urge-to-blink that participants continuously reported using a rollerball device. In addition, we used the method of multi-echo paradigm free mapping (MESPFM) to identify these regions without prior specification of task timings. Subjective urge scores were correlated with activity in the right posterior and ventral-anterior insula as well as the mid-cingulate and occipital cortices. Whereas blink suppression was associated with activation in the dorsolateral prefrontal cortex, cerebellum, right dorsal-anterior insula, mid-cingulate cortex and thalamus. These findings illustrate that different insula subregions contribute to the urge-for-action and suppression networks. The MESPFM approach showed co-activation of the right insula and cingulate cortex. The MESPFM activation maps showed the highest overlap with activation associated with blink suppression, as identified using general linear model analysis, demonstrating that activity associated with suppression can be determined without prior knowledge of task timings.
- Open Access
Uncovering the Neural Correlates of the Urge-to-Blink: A Study Utilising Subjective Urge Ratings and Paradigm Free Mapping
Neuroimaging plays a significant role in understanding the neurophysiology of Tourette syndrome (TS), in particular the main symptom, tics, and the urges associated with them. Premonitory urge is thought to be a negative reinforcer of tic expression in TS. Tic expression during neuroimaging is most often required as an overt marker of increased urge-to-tic, which can lead to considerable head movement, and thus data loss. This study aims to identify the brain regions involved in urge in healthy subjects using multi-echo functional MRI and a timing- free approach to localise the BOLD response associated with the urge-to-act without information of when these events occur. Blink suppression is an analogous behaviour that can be expressed overtly in the MRI scanner which gives rise to an urge like those described by individuals with TS. We examined the urge-to-blink in 20 healthy volunteers with an experimental paradigm including two conditions, “Okay to blink” and “Suppress blinking”, to identify brain regions involved in blink suppression. Multi-echo functional MRI data was analysed using a novel approach to investigate the BOLD signal correlated with the build-up of the urge-to-blink that participants continuously reported using a rollerball device. In addition, we used the method of multi-echo paradigm free mapping (MESPFM) to identify these regions without prior specification of task timings. Subjective urge scores were correlated with activity in the right posterior and ventral-anterior insula as well as the mid-cingulate and occipital cortices. Furthermore, blink suppression was associated with activation in the dorsolateral prefrontal cortex, cerebellum, right dorsal-anterior insula, mid-cingulate cortex and thalamus. These findings illustrate that different insula subregions contribute to the urge-for-action and suppression networks. The MESPFM approach showed co-activation of the right insula and cingulate cortex. The MESPFM activation maps showed the highest overlap with activation associated with blink suppression, as identified using general linear model analysis, demonstrating that activity associated with suppression can be determined without prior knowledge of task timings.
Open AccessUncovering the Neural Correlates of the Urge-to-Blink: A Study Utilising Subjective Urge Ratings and Paradigm Free Mapping
Neuroimaging plays a significant role in understanding the neurophysiology of Tourette syndrome (TS), in particular the main symptom, tics, and the urges associated with them. Premonitory urge is thought to be a negative reinforcer of tic expression in TS. Tic expression during neuroimaging is most often required as an overt marker of increased urge-to-tic, which can lead to considerable head movement, and thus data loss. This study aims to identify the brain regions involved in urge in healthy subjects using multi-echo functional MRI and a timing- free approach to localise the BOLD response associated with the urge-to-act without information of when these events occur. Blink suppression is an analogous behaviour that can be expressed overtly in the MRI scanner which gives rise to an urge like those described by individuals with TS. We examined the urge-to-blink in 20 healthy volunteers with an experimental paradigm including two conditions, “Okay to blink” and “Suppress blinking”, to identify brain regions involved in blink suppression. Multi-echo functional MRI data was analysed using a novel approach to investigate the BOLD signal correlated with the build-up of the urge-to-blink that participants continuously reported using a rollerball device. In addition, we used the method of multi-echo paradigm free mapping (MESPFM) to identify these regions without prior specification of task timings. Subjective urge scores were correlated with activity in the right posterior and ventral-anterior insula as well as the mid-cingulate and occipital cortices. Furthermore, blink suppression was associated with activation in the dorsolateral prefrontal cortex, cerebellum, right dorsal-anterior insula, mid-cingulate cortex and thalamus. These findings illustrate that different insula subregions contribute to the urge-for-action and suppression networks. The MESPFM approach showed co-activation of the right insula and cingulate cortex. The MESPFM activation maps showed the highest overlap with activation associated with blink suppression, as identified using general linear model analysis, demonstrating that activity associated with suppression can be determined without prior knowledge of task timings.
- Open Access
Individual variability in the relationship between physiological and resting-state fMRI metrics
Cerebrovascular Reactivity (CVR), the brain’s vascular response to a vasodilatory stimulus, can be measured using fMRI during breathing challenges that modulate arterial CO 2 levels. CVR is an important indicator of cerebrovascular health, although its estimation can be challenging due to the extra experimental setup and/or the subject compliance required. To overcome these limitations, summary metrics based on resting state fluctuations (RSF), such as the (fractional) amplitude of low frequency fluctuations (f/ALFF) and the resting state fluctuation amplitude (RSFA), have been proposed as alternative estimates of CVR, as they are frequently associated with vascular and physiological factors. Previous studies have reported a significant relationship between CVR estimates obtained by means of respiratory paradigms and RSF metrics. However, the total sample sizes, considering both subjects and sessions, are typically small, and not all studies agree on the degree of the relationship. Furthermore, to our knowledge, these studies have only reported cross-sectional analyses, whereas intra-subject longitudinal relationships between CVR estimates and RSF metrics are unknown. Leveraging a unique dense sampling dataset in which resting state and breath-hold multi-echo fMRI were collected in 7 subjects with 10 sessions each, we provide evidence of high individual variability in the inter-session (i.e. intra-subject) relationship between RSF metrics and CVR. These results indicate that RSF metrics might not be a suitable proxy of CVR in clinical settings or for BOLD signal calibration as they may not properly account for intra-individual physiological variations in BOLD fMRI data.
Open AccessIndividual variability in the relationship between physiological and resting-state fMRI metrics
Cerebrovascular Reactivity (CVR), the brain’s vascular response to a vasodilatory stimulus, can be measured using fMRI during breathing challenges that modulate arterial CO 2 levels. CVR is an important indicator of cerebrovascular health, although its estimation can be challenging due to the extra experimental setup and/or the subject compliance required. To overcome these limitations, summary metrics based on resting state fluctuations (RSF), such as the (fractional) amplitude of low frequency fluctuations (f/ALFF) and the resting state fluctuation amplitude (RSFA), have been proposed as alternative estimates of CVR, as they are frequently associated with vascular and physiological factors. Previous studies have reported a significant relationship between CVR estimates obtained by means of respiratory paradigms and RSF metrics. However, the total sample sizes, considering both subjects and sessions, are typically small, and not all studies agree on the degree of the relationship. Furthermore, to our knowledge, these studies have only reported cross-sectional analyses, whereas intra-subject longitudinal relationships between CVR estimates and RSF metrics are unknown. Leveraging a unique dense sampling dataset in which resting state and breath-hold multi-echo fMRI were collected in 7 subjects with 10 sessions each, we provide evidence of high individual variability in the inter-session (i.e. intra-subject) relationship between RSF metrics and CVR. These results indicate that RSF metrics might not be a suitable proxy of CVR in clinical settings or for BOLD signal calibration as they may not properly account for intra-individual physiological variations in BOLD fMRI data.
- Open Access
- 13 March 2024
- Published by The Open Journal in Journal of Open Source Software
- Vol. 9 (95) , 6326
PyProximal - scalable convex optimization in Python
A broad class of problems in scientific disciplines ranging from image processing and astrophysics, to geophysics and medical imaging call for the optimization of convex, non-smooth objective functions. Whereas practitioners are usually familiar with gradient-based algorithms, commonly used to solve unconstrained, smooth optimization problems, proximal algorithms can be viewed as analogous tools for non-smooth and possibly constrained versions of such problems. These algorithms sit at a higher level of abstraction than gradient-based algorithms and require a basic operation to be performed at each iteration: the evaluation of the so-called proximal operator of the functional to be optimized. PyProximal is a Python-based library aimed at democratizing the application of convex optimization to scientific problems; it provides the required building blocks (i.e., proximal operators and algorithms) to define and solve complex, convex objective functions in a high-level, abstract fashion, shielding users away from any unneeded mathematical and implementation details.
- 13 March 2024
- Published by The Open Journal in Journal of Open Source Software
- Vol. 9 (95) , 6326
Open AccessPyProximal - scalable convex optimization in Python
A broad class of problems in scientific disciplines ranging from image processing and astrophysics, to geophysics and medical imaging call for the optimization of convex, non-smooth objective functions. Whereas practitioners are usually familiar with gradient-based algorithms, commonly used to solve unconstrained, smooth optimization problems, proximal algorithms can be viewed as analogous tools for non-smooth and possibly constrained versions of such problems. These algorithms sit at a higher level of abstraction than gradient-based algorithms and require a basic operation to be performed at each iteration: the evaluation of the so-called proximal operator of the functional to be optimized. PyProximal is a Python-based library aimed at democratizing the application of convex optimization to scientific problems; it provides the required building blocks (i.e., proximal operators and algorithms) to define and solve complex, convex objective functions in a high-level, abstract fashion, shielding users away from any unneeded mathematical and implementation details.
- Open Access
Bayesian Modelling Approaches for Breath-Hold Induced Cerebrovascular Reactivity
Cerebrovascular reactivity (CVR) reflects the ability of blood vessels to dilate and constrict in response to a vasoactive stimulus and is an important indicator of cerebrovascular health. CVR can be mapped non-invasively with functional magnetic resonance imaging (fMRI) based on blood oxygen level-dependent (BOLD) contrast in combination with a breath-hold (BH) task. There are several ways to analyse this type of data and retrieve individual CVR amplitude and timing information. The most common approach involves employing a time-shifted general linear model with the measured end-tidal carbon dioxide signal as a regressor of interest. In this work, we introduce a novel method for CVR mapping based on a variational Bayesian approach. We analysed BOLD fMRI data from six participants that performed a BH task in ten different sessions each, and computed the corresponding CVR amplitude and delay maps for each session/subject. No statistically significant differences were observed between the modelling approaches in the CVR delay and amplitude maps in grey matter. Notably, the largest difference between methods was apparent in the case of low CVR amplitude, attributed to how each method addressed noisy voxels, particularly in white matter and cerebral spinal fluid. Both approaches showed highly reproducible CVR amplitude maps where between-subject variability was significantly larger than between-session variability. Furthermore, our results illustrated that the Bayesian approach is more computationally efficient, and future implementations could incorporate more complex noise models, non-linear fitting, and physiologically meaningful information into the model in the form of priors. This work demonstrates the utility of variational Bayesian modelling for CVR mapping and highlights its potential for characterising BOLD fMRI dynamics in the study of cerebrovascular health and its application to clinical settings.
Open AccessBayesian Modelling Approaches for Breath-Hold Induced Cerebrovascular Reactivity
Cerebrovascular reactivity (CVR) reflects the ability of blood vessels to dilate and constrict in response to a vasoactive stimulus and is an important indicator of cerebrovascular health. CVR can be mapped non-invasively with functional magnetic resonance imaging (fMRI) based on blood oxygen level-dependent (BOLD) contrast in combination with a breath-hold (BH) task. There are several ways to analyse this type of data and retrieve individual CVR amplitude and timing information. The most common approach involves employing a time-shifted general linear model with the measured end-tidal carbon dioxide signal as a regressor of interest. In this work, we introduce a novel method for CVR mapping based on a variational Bayesian approach. We analysed BOLD fMRI data from six participants that performed a BH task in ten different sessions each, and computed the corresponding CVR amplitude and delay maps for each session/subject. No statistically significant differences were observed between the modelling approaches in the CVR delay and amplitude maps in grey matter. Notably, the largest difference between methods was apparent in the case of low CVR amplitude, attributed to how each method addressed noisy voxels, particularly in white matter and cerebral spinal fluid. Both approaches showed highly reproducible CVR amplitude maps where between-subject variability was significantly larger than between-session variability. Furthermore, our results illustrated that the Bayesian approach is more computationally efficient, and future implementations could incorporate more complex noise models, non-linear fitting, and physiologically meaningful information into the model in the form of priors. This work demonstrates the utility of variational Bayesian modelling for CVR mapping and highlights its potential for characterising BOLD fMRI dynamics in the study of cerebrovascular health and its application to clinical settings.
- 6 November 2023
- Published by Elsevier in Medical Image Analysis
- Vol. 91, 103010
Whole-brain multivariate hemodynamic deconvolution for functional MRI with stability selection
- 6 November 2023
- Published by Elsevier in Medical Image Analysis
- Vol. 91, 103010
Whole-brain multivariate hemodynamic deconvolution for functional MRI with stability selection
- Open Access
- 28 August 2023
- Published by Organization for Human Brain Mapping in Aperture Neuro
- Vol. 3, 1-25
Hemodynamic Deconvolution Demystified: Sparsity-Driven Regularization at Work
Deconvolution of the hemodynamic response is an important step to access short timescales of brain activity recorded by functional magnetic resonance imaging (fMRI). Albeit conventional deconvolution algorithms have been around for a long time (e.g., Wiener deconvolution), recent state-of-the-art methods based on sparsity-pursuing regularization are attracting increasing interest to investigate brain dynamics and connectivity with fMRI. This technical note revisits the main concepts underlying two main methods, paradigm free mapping and total activation, in the most accessible way. Despite their apparent differences in the formulation, these methods are theoretically equivalent as they represent the synthesis and analysis sides of the same problem, respectively. We demonstrate this equivalence in practice with their best-available implementations using both simulations, with different signal-to-noise ratios, and experimental fMRI data acquired during a motor task and resting state. We evaluate the parameter settings that lead to equivalent results and showcase the potential of these algorithms compared to other common approaches. This note is useful for practitioners interested in gaining a better understanding of state-of-the-art hemodynamic deconvolution and aims to answer questions that practitioners often have regarding the differences between the two methods.
- 28 August 2023
- Published by Organization for Human Brain Mapping in Aperture Neuro
- Vol. 3, 1-25
Open AccessHemodynamic Deconvolution Demystified: Sparsity-Driven Regularization at Work
Deconvolution of the hemodynamic response is an important step to access short timescales of brain activity recorded by functional magnetic resonance imaging (fMRI). Albeit conventional deconvolution algorithms have been around for a long time (e.g., Wiener deconvolution), recent state-of-the-art methods based on sparsity-pursuing regularization are attracting increasing interest to investigate brain dynamics and connectivity with fMRI. This technical note revisits the main concepts underlying two main methods, paradigm free mapping and total activation, in the most accessible way. Despite their apparent differences in the formulation, these methods are theoretically equivalent as they represent the synthesis and analysis sides of the same problem, respectively. We demonstrate this equivalence in practice with their best-available implementations using both simulations, with different signal-to-noise ratios, and experimental fMRI data acquired during a motor task and resting state. We evaluate the parameter settings that lead to equivalent results and showcase the potential of these algorithms compared to other common approaches. This note is useful for practitioners interested in gaining a better understanding of state-of-the-art hemodynamic deconvolution and aims to answer questions that practitioners often have regarding the differences between the two methods.
- Open Access
Whole-brain multivariate hemodynamic deconvolution for multi-echo fMRI with stability selection
Conventionally, analysis of functional MRI (fMRI) data relies on available information about the experimental paradigm to establish hypothesized models of brain activity. However, this information can be inaccurate, incomplete or unavailable in multiple scenarios such as resting-state, naturalistic paradigms or clinical conditions. In these cases, blind estimates of neuronal-related activity can be obtained with paradigm-free analysis methods such as hemodynamic deconvolution. Yet, current formulations of the hemodynamic deconvolution problem have three important limitations: 1) their efficacy strongly depends on the appropriate selection of regularization parameters, 2) being univariate, they do not take advantage of the information present across the brain, and 3) they do not provide any measure of statistical certainty associated with each detected event. Here we propose a novel approach that addresses all these limitations. Specifically, we introduce MvME-SPFM (multivariate multi-echo sparse paradigm free mapping), a novel hemodynamic deconvolution algorithm that operates at the whole brain level and adds spatial information via a mixed-norm regularization term over all voxels. Additionally, MvME-SPFM employs a stability selection procedure that removes the need to select regularization parameters and also lets us obtain an estimate of the true probability of having a neuronal-related BOLD event at each voxel and time-point based on the area under the curve (AUC) of the stability paths. Besides, the formulation is tailored for multi-echo fMRI acquisitions, which allows us to better isolate fluctuations of BOLD origin on the basis of their linear dependence with Echo Time (TE) and to assign physiologically interpretable units (i.e., changes in the apparent transverse relaxation ) to the resulting deconvolved events. We demonstrate that this algorithm outperforms existing state-of-the-art deconvolution approaches, and shows higher spatial and temporal agreement with the activation maps and BOLD signals obtained with a standard model-based linear regression approach, even at the level of individual neuronal events. Consequently, the proposed algorithm provides more reliable estimates of neuronal-related activity, here in terms of , for the study of the dynamics of brain activity when no information about the timings of the BOLD events is available. This algorithm will be made publicly available as part of the splora Python package.
Open AccessWhole-brain multivariate hemodynamic deconvolution for multi-echo fMRI with stability selection
Conventionally, analysis of functional MRI (fMRI) data relies on available information about the experimental paradigm to establish hypothesized models of brain activity. However, this information can be inaccurate, incomplete or unavailable in multiple scenarios such as resting-state, naturalistic paradigms or clinical conditions. In these cases, blind estimates of neuronal-related activity can be obtained with paradigm-free analysis methods such as hemodynamic deconvolution. Yet, current formulations of the hemodynamic deconvolution problem have three important limitations: 1) their efficacy strongly depends on the appropriate selection of regularization parameters, 2) being univariate, they do not take advantage of the information present across the brain, and 3) they do not provide any measure of statistical certainty associated with each detected event. Here we propose a novel approach that addresses all these limitations. Specifically, we introduce MvME-SPFM (multivariate multi-echo sparse paradigm free mapping), a novel hemodynamic deconvolution algorithm that operates at the whole brain level and adds spatial information via a mixed-norm regularization term over all voxels. Additionally, MvME-SPFM employs a stability selection procedure that removes the need to select regularization parameters and also lets us obtain an estimate of the true probability of having a neuronal-related BOLD event at each voxel and time-point based on the area under the curve (AUC) of the stability paths. Besides, the formulation is tailored for multi-echo fMRI acquisitions, which allows us to better isolate fluctuations of BOLD origin on the basis of their linear dependence with Echo Time (TE) and to assign physiologically interpretable units (i.e., changes in the apparent transverse relaxation ) to the resulting deconvolved events. We demonstrate that this algorithm outperforms existing state-of-the-art deconvolution approaches, and shows higher spatial and temporal agreement with the activation maps and BOLD signals obtained with a standard model-based linear regression approach, even at the level of individual neuronal events. Consequently, the proposed algorithm provides more reliable estimates of neuronal-related activity, here in terms of , for the study of the dynamics of brain activity when no information about the timings of the BOLD events is available. This algorithm will be made publicly available as part of the splora Python package.
- Open Access
Hemodynamic Deconvolution Demystified: Sparsity-Driven Regularization at Work
Deconvolution of the hemodynamic response is an important step to access short timescales of brain activity recorded by functional magnetic resonance imaging (fMRI). Albeit conventional deconvolution algorithms have been around for a long time (e.g., Wiener deconvolution), recent state-of-the-art methods based on sparsity-pursuing regularization are attracting increasing interest to investigate brain dynamics and connectivity with fMRI. This technical note revisits the main concepts underlying two main methods, Paradigm Free Mapping and Total Activation, in the most accessible way. Despite their apparent differences in the formulation, these methods are theoretically equivalent as they represent the synthesis and analysis sides of the same problem, respectively. We demonstrate this equivalence in practice with their best-available implementations using both simulations, with different signal-to-noise ratios, and experimental fMRI data acquired during a motor task and resting-state. We evaluate the parameter settings that lead to equivalent results, and showcase the potential of these algorithms compared to other common approaches. This note is useful for practitioners interested in gaining a better understanding of state-of-the-art hemodynamic deconvolution, and aims to answer questions that practitioners often have regarding the differences between the two methods.
Open AccessHemodynamic Deconvolution Demystified: Sparsity-Driven Regularization at Work
Deconvolution of the hemodynamic response is an important step to access short timescales of brain activity recorded by functional magnetic resonance imaging (fMRI). Albeit conventional deconvolution algorithms have been around for a long time (e.g., Wiener deconvolution), recent state-of-the-art methods based on sparsity-pursuing regularization are attracting increasing interest to investigate brain dynamics and connectivity with fMRI. This technical note revisits the main concepts underlying two main methods, Paradigm Free Mapping and Total Activation, in the most accessible way. Despite their apparent differences in the formulation, these methods are theoretically equivalent as they represent the synthesis and analysis sides of the same problem, respectively. We demonstrate this equivalence in practice with their best-available implementations using both simulations, with different signal-to-noise ratios, and experimental fMRI data acquired during a motor task and resting-state. We evaluate the parameter settings that lead to equivalent results, and showcase the potential of these algorithms compared to other common approaches. This note is useful for practitioners interested in gaining a better understanding of state-of-the-art hemodynamic deconvolution, and aims to answer questions that practitioners often have regarding the differences between the two methods.