Hemodynamic Response Function Model Research Articles

Improved Activation and Hemodynamic Response Function of Olfactory fMRI Using Simultaneous Multislice with Reduced TR Acquisition.

Objectives The respiration could decrease the time synchronization between odor stimulation and data acquisition, consequently deteriorating the functional activation and hemodynamic response function (HRF) in olfactory functional magnetic resonance imaging (fMRI) with a conventional repetition time (TR). In this study, we aimed to investigate whether simultaneous multislice (SMS) technology with reduced TR could improve the blood oxygen level-dependent (BOLD) activation and optimize HRF modeling in olfactory fMRI. Methods Sixteen young healthy subjects with normal olfaction underwent olfactory fMRI on a 3T MRI scanner using a 64 channel head coil. FMRI data were acquired using SMS acceleration at three different TRs: 3000 ms, 1000 ms, and 500 ms. Both metrics of BOLD activation (activated voxels, mean, and maximum t-scores) and the HRF modeling (response height and time to peak) were calculated in the bilateral amygdalae, hippocampi, and insulae. Results The 500 ms and 1000 ms TRs both significantly improved the number of activated voxels, mean, and maximum t-score in the amygdalae and insulae, compared with a 3000 ms TR (all P < 0.05). But the increase of these metrics in the hippocampi did not reach a statistical significance (all P > 0.05). No significant difference in any BOLD activation metrics between TR 500 ms and 1000 ms was observed in all regions of interest (ROIs) (all P > 0.05). The HRF curves showed that higher response height and shorter time to peak in all ROIs were obtained at 500 ms and 1000 ms TRs compared to 3000 ms TR. TR 500 ms had a more significant effect on response height than TR 1000 ms in the amygdalae (P = 0.017), and there was no significant difference in time to peak between TR 500 ms and 1000 ms in all ROIs (all P > 0.05). Conclusions The fast image acquisition technique of SMS with reduced TR could significantly improve the functional activation and HRF curve in olfactory fMRI.

Hemodynamic Response Function in Brain White Matter in a Resting State.

The hemodynamic response function (HRF) characterizes temporal variations of blood oxygenation level-dependent (BOLD) signals. Although a variety of HRF models have been proposed for gray matter responses to functional demands, few studies have investigated HRF profiles in white matter particularly under resting conditions. In the present work we quantified the nature of the HRFs that are embedded in resting state BOLD signals in white matter, and which modulate the temporal fluctuations of baseline signals. We demonstrate that resting state HRFs in white matter could be derived by referencing to intrinsic avalanches in gray matter activities, and the derived white matter HRFs had reduced peak amplitudes and delayed peak times as compared with those in gray matter. Distributions of the time delays and correlation profiles in white matter depend on gray matter activities as well as white matter tract distributions, indicating that resting state BOLD signals in white matter encode neural activities associated with those of gray matter. This is the first investigation of derivations and characterizations of resting state HRFs in white matter and their relations to gray matter activities. Findings from this work have important implications for analysis of BOLD signals in the brain.

Hemodynamic Response Function Modeling to Determine the Areas with High Blood Supply in Block-Design fMRI Experiments

Background: Determining the activated brain areas due to different activities is one of the most common targets in functional magnetic resonance imaging (fMRI) data analysis, which could be carried out by hemodynamic response function (HRF) evaluation. The HR functions reflect changes of cerebral blood flow (CBF) in response to neural activity. Objectives: In this study, five models of HRF estimation were evaluated based on a simulated dataset. Models with higher accuracy were used to determine HRF parameters of the block-design fMRI data. Methods: The fMRI data were acquired in a 3 Tesla scanner. For block-design fMR imaging, CO2 gas was administered using a face-mask under physiological monitoring. Three patients with brain tumors were scanned. The fMRI data analysis was performed using the SPM 12 MATLAB toolbox. Akaike’s information criterion (AIC), Schwarz’ Bayesian (SBC), and mean square error (MSE) criteria were used to select the best HRF estimation model. Results: In simulation studies, the estimated HRFs by the canonical HRF plus its temporal derivative (TD), finite impulse response (FIR), and inverse logistic (IL) models were almost equal to the standard HRF. Mean square error, AIC, and SBC indices were ignorable for TD, FIR, and IL models (MSE/AIC/SBC magnitudes for TD, FIR, and IL models were 0.052/-1235.1/-1223.9, 0.055/-1206.4/-1194.9, and 0.068/-1091.5/-1049.2, respectively), which indicates that these models could accurately estimate HRF in block design fMRI studies. Conclusions: The HRF models could non-invasively evaluate the change of MR signal intensity under cerebrovascular reactivity (CVR) conditions and they might be helpful to investigate changes in human cerebral blood flow.

Estimation of Hemodynamic Response Function in the Brain and Brain Tumors: Comparison of Inverse Logistic and Canonical Hemodynamic Response Function Models

Estimation of Hemodynamic Response Function in the Brain and Brain Tumors: Comparison of Inverse Logistic and Canonical Hemodynamic Response Function Models

P.2.047 - Validation of the canonical hemodynamic response function model used in fMRI studies

Introduction There is need to validate the fMRI analysis methods using large-scale datasets [1]. It is known that the hemodynamic response function (HRF) varies across brain regions and subjects [2]. Different HRF models are currently available in fMRI analysis packages [3]. The most popular HRF model is the canonical HRF: sum of two fixed gamma curves. We compare some of the implemented HRF models in terms of sensitivity-specificity efficiency. Aims We aim to identify an optimal HRF model for healthy subjects and to check if this model is flexible enough for unusual subject populations like impaired consciousness patients or the elderly. For such patient populations the hemodynamic response might differ and the canonical model might be too fixed to detect experimentally induced neural activation. Methods We compare HRF models in the most popular fMRI analysis softwares: AFNI, FSL and SPM, regarding: (1) positive rates (at least one significant voxel in the entire brain) (2) spatial distribution of the significant voxels and (3) numbers of significant voxels. For our analyses we employ three different baselines: (1) simulated data, for which we developed an R package: https://github.com/wiktorolszowy/HRF_simulation_for_fMRI_experiments, (2) resting state data with assumed dummy design and (3) task based data with assumed wrong design. Results The following table shows mean numbers of significant voxels for 70 impaired consciousness subjects performing a visual task. The visual stimulus was presented for 16s after 16s of rest. This design was repeated 10 times. The results refer to different assumed experimental designs and different HRF models. Conclusions The FIR model as implemented in FSL returned suspiciously high numbers of significant voxels which suggests that the degrees of freedom might be mis-specified. FIR behaved in a similar way when using simulated data and resting state data and for different subject populations (not shown in the above table). The effect of adding the first derivative to the gamma models was negligible. This suggests that the canonical HRF model is inflexible. Interestingly, significance was detected more often for assumed low experimental frequencies. This might suggest that the employed degrees of freedom do not model the correlations between periods when the stimulus was shown and the rest periods.

Detecting white matter activity using conventional 3 Tesla fMRI: An evaluation of standard field strength and hemodynamic response function

Detection of functional magnetic resonance imaging (fMRI) activation in white matter has been increasingly reported despite historically being controversial. Much of the development work to-date has used high-field 4 T MRI and specialized pulse sequences. In the current study, we utilized conventional 3 T MRI and a commonly applied gradient-echo-planar imaging sequence to evaluate white matter (WM) fMRI sensitivity within a common framework. Functional WM activity was replicated in target regions of interest within the corpus callosum, at the group and individual levels. As expected there was a reduction in overall WM activation sensitivity. Individual analyses revealed that 8 of the 13 individuals showed white matter activation, showing a lower percentage of individuals with WM activation detected. Importantly, WM activation results were sensitive to analyses that applied alternate hemodynamic response functions, with an increase in the group level cluster when hemodynamic response function (HRF) onset slope was reduced. The findings supported the growing evidence that WM activation is detectable, with activation levels are closer to thresholds used for routine 3T MRI studies. Optimization factors, such as the HRF model, appear to be important to further enhance the characterization of WM activity in fMRI.

Optimizing Stimulation and Analysis Protocols for Neonatal fMRI

The development of brain function in young infants is poorly understood. The core challenge is that infants have a limited behavioral repertoire through which brain function can be expressed. Neuroimaging with fMRI has great potential as a way of characterizing typical development, and detecting abnormal development early. But, a number of methodological challenges must first be tackled to improve the robustness and sensitivity of neonatal fMRI. A critical one of these, addressed here, is that the hemodynamic response function (HRF) in pre-term and term neonates differs from that in adults, which has a number of implications for fMRI. We created a realistic model of noise in fMRI data, using resting-state fMRI data from infants and adults, and then conducted simulations to assess the effect of HRF of the power of different stimulation protocols and analysis assumptions (HRF modeling). We found that neonatal fMRI is most powerful if block-durations are kept at the lower range of those typically used in adults (full on/off cycle duration 25-30s). Furthermore, we show that it is important to use the age-appropriate HRF during analysis, as mismatches can lead to reduced power or even inverted signal. Where the appropriate HRF is not known (for example due to potential developmental delay), a flexible basis set performs well, and allows accurate post-hoc estimation of the HRF.

On the distinguishability of HRF models in fMRI.

Modeling the Hemodynamic Response Function (HRF) is a critical step in fMRI studies of brain activity, and it is often desirable to estimate HRF parameters with physiological interpretability. A biophysically informed model of the HRF can be described by a non-linear time-invariant dynamic system. However, the identification of this dynamic system may leave much uncertainty on the exact values of the parameters. Moreover, the high noise levels in the data may hinder the model estimation task. In this context, the estimation of the HRF may be seen as a problem of model falsification or invalidation, where we are interested in distinguishing among a set of eligible models of dynamic systems. Here, we propose a systematic tool to determine the distinguishability among a set of physiologically plausible HRF models. The concept of absolutely input-distinguishable systems is introduced and applied to a biophysically informed HRF model, by exploiting the structure of the underlying non-linear dynamic system. A strategy to model uncertainty in the input time-delay and magnitude is developed and its impact on the distinguishability of two physiologically plausible HRF models is assessed, in terms of the maximum noise amplitude above which it is not possible to guarantee the falsification of one model in relation to another. Finally, a methodology is proposed for the choice of the input sequence, or experimental paradigm, that maximizes the distinguishability of the HRF models under investigation. The proposed approach may be used to evaluate the performance of HRF model estimation techniques from fMRI data.

Investigating the neural bases for intra-subject cognitive efficiency changes using functional magnetic resonance imaging.

Several fMRI studies have examined brain regions mediating inter-subject variability in cognitive efficiency, but none have examined regions mediating intra-subject variability in efficiency. Thus, the present study was designed to identify brain regions involved in intra-subject variability in cognitive efficiency via participant-level correlations between trial-level reaction time (RT) and trial-level fMRI BOLD percent signal change on a processing speed task. On each trial, participants indicated whether a digit-symbol probe-pair was present or absent in an array of nine digit-symbol probe-pairs while fMRI data were collected. Deconvolution analyses, using RT time-series models (derived from the proportional scaling of an event-related hemodynamic response function model by trial-level RT), were used to evaluate relationships between trial-level RTs and BOLD percent signal change. Although task-related patterns of activation and deactivation were observed in regions including bilateral occipital, bilateral parietal, portions of the medial wall such as the precuneus, default mode network regions including anterior cingulate, posterior cingulate, bilateral temporal, right cerebellum, and right cuneus, RT-BOLD correlations were observed in a more circumscribed set of regions. Positive RT-BOLD correlations, where fast RTs were associated with lower BOLD percent signal change, were observed in regions including bilateral occipital, bilateral parietal, and the precuneus. RT-BOLD correlations were not observed in the default mode network indicating a smaller set of regions associated with intra-subject variability in cognitive efficiency. The results are discussed in terms of a distributed area of regions that mediate variability in the cognitive efficiency that might underlie processing speed differences between individuals.

Data-driven HRF estimation for encoding and decoding models

Despite the common usage of a canonical, data-independent, hemodynamic response function (HRF), it is known that the shape of the HRF varies across brain regions and subjects. This suggests that a data-driven estimation of this function could lead to more statistical power when modeling BOLD fMRI data. However, unconstrained estimation of the HRF can yield highly unstable results when the number of free parameters is large. We develop a method for the joint estimation of activation and HRF by means of a rank constraint, forcing the estimated HRF to be equal across events or experimental conditions, yet permitting it to differ across voxels. Model estimation leads to an optimization problem that we propose to solve with an efficient quasi-Newton method, exploiting fast gradient computations. This model, called GLM with Rank-1 constraint (R1-GLM), can be extended to the setting of GLM with separate designs which has been shown to improve decoding accuracy in brain activity decoding experiments. We compare 10 different HRF modeling methods in terms of encoding and decoding scores on two different datasets. Our results show that the R1-GLM model outperforms competing methods in both encoding and decoding settings, positioning it as an attractive method both from the points of view of accuracy and computational efficiency.

Recursive approach of EEG-segment-based principal component analysis substantially reduces cryogenic pump artifacts in simultaneous EEG–fMRI data

Electroencephalography (EEG) data simultaneously acquired with functional magnetic resonance imaging (fMRI) data are preprocessed to remove gradient artifacts (GAs) and ballistocardiographic artifacts (BCAs). Nonetheless, these data, especially in the gamma frequency range, can be contaminated by residual artifacts produced by mechanical vibrations in the MRI system, in particular the cryogenic pump that compresses and transports the helium that chills the magnet (the helium-pump). However, few options are available for the removal of helium-pump artifacts. In this study, we propose a recursive approach of EEG-segment-based principal component analysis (rsPCA) that enables the removal of these helium-pump artifacts. Using the rsPCA method, feature vectors representing helium-pump artifacts were successfully extracted as eigenvectors, and the reconstructed signals of the feature vectors were subsequently removed. A test using simultaneous EEG–fMRI data acquired from left-hand (LH) and right-hand (RH) clenching tasks performed by volunteers found that the proposed rsPCA method substantially reduced helium-pump artifacts in the EEG data and significantly enhanced task-related gamma band activity levels (p=0.0038 and 0.0363 for LH and RH tasks, respectively) in EEG data that have had GAs and BCAs removed. The spatial patterns of the fMRI data were estimated using a hemodynamic response function (HRF) modeled from the estimated gamma band activity in a general linear model (GLM) framework. Active voxel clusters were identified in the post-/pre-central gyri of motor area, only from the rsPCA method (uncorrected p<0.001 for both LH/RH tasks). In addition, the superior temporal pole areas were consistently observed (uncorrected p<0.001 for the LH task and uncorrected p<0.05 for the RH task) in the spatial patterns of the HRF model for gamma band activity when the task paradigm and movement were also included in the GLM.

Model specification and the reliability of fMRI results: implications for longitudinal neuroimaging studies in psychiatry.

Functional Magnetic Resonance Imagine (fMRI) is an important assessment tool in longitudinal studies of mental illness and its treatment. Understanding the psychometric properties of fMRI-based metrics, and the factors that influence them, will be critical for properly interpreting the results of these efforts. The current study examined whether the choice among alternative model specifications affects estimates of test-retest reliability in key emotion processing regions across a 6-month interval. Subjects (N = 46) performed an emotional-faces paradigm during fMRI in which neutral faces dynamically morphed into one of four emotional faces. Median voxelwise intraclass correlation coefficients (mvICCs) were calculated to examine stability over time in regions showing task-related activity as well as in bilateral amygdala. Four modeling choices were evaluated: a default model that used the canonical hemodynamic response function (HRF), a flexible HRF model that included additional basis functions, a modified CompCor (mCompCor) model that added corrections for physiological noise in the global signal, and a final model that combined the flexible HRF and mCompCor models. Model residuals were examined to determine the degree to which each pipeline met modeling assumptions. Results indicated that the choice of modeling approaches impacts both the degree to which model assumptions are met and estimates of test-retest reliability. ICC estimates in the visual cortex increased from poor (mvICC = 0.31) in the default pipeline to fair (mvICC = 0.45) in the full alternative pipeline – an increase of 45%. In nearly all tests, the models with the fewest assumption violations generated the highest ICC estimates. Implications for longitudinal treatment studies that utilize fMRI are discussed.

Hemodynamic Response Function (HRF) in Epilepsy Patients with Hippocampal Sclerosis and Focal Cortical Dysplasia

Simultaneous recording of electroencephalography and functional magnetic resonance imaging (EEG-fMRI) has recently been applied for mapping the hemodynamic changes related to epileptic activity. The aim of this study is to compare the hemodynamic response function (HRF) to epileptic spikes in patients with focal cortical dysplasia (FCD) and those with hippocampal sclerosis (HS). In EEG-fMRI studies, the HRF represents the temporal evolution of blood oxygenation level-dependent signal changes. Several studies demonstrated that amplitude and latency of the HRF are variable in patients with epilepsy. However, the consistency of HRF parameters with underlying brain pathology is unknown. In this study, we examined 14 patients with FCD and 12 with unilateral HS selected from our EEG-fMRI database and compared the amplitude and latency of the HRF peak. We analyzed (1) HRFs in peak activation clusters, (2) HRFs in peak deactivation clusters, and (3) the maximum absolute responses within the EEG spike field, activation or deactivation. We found that the HRF peak amplitude in deactivation clusters was larger in the HS group than in the FCD when the deactivation occurred in default mode network (DMN) regions. This result suggests that spikes in patients with HS affect the DMN more strongly than those with FCD. However, if we focus on the maximum absolute t-value in the spike field, there is no significant difference between the two groups. The current study indicates that it is not necessary to use different HRF models for EEG-fMRI studies in patients with FCD and HS.

Increased sensitivity of fast BOLD fMRI with a subject-specific hemodynamic response function and application to epilepsy

Activation detection in functional Magnetic Resonance Imaging (fMRI) typically assumes the hemodynamic response to neuronal activity to be invariant across brain regions and subjects. Reports of substantial variability of the morphology of blood-oxygenation-level-dependent (BOLD) responses are accumulating, suggesting that the use of a single generic model of the expected response in general linear model (GLM) analyses does not provide optimal sensitivity due to model misspecification. Relaxing assumptions of the model can limit the impact of hemodynamic response function (HRF) variability, but at a cost on model parsimony. Alternatively, better specification of the model could be obtained from a priori knowledge of the HRF of a given subject, but the effectiveness of this approach has only been tested on simulation data. Using fast BOLD fMRI, we characterized the variability of hemodynamic responses to a simple event-related auditory-motor task, as well as its effect on activation detection with GLM analyses. We show the variability to be higher between subjects than between regions and variation in different regions to correlate from one subject to the other. Accounting for subject-related variability by deriving subject-specific models from responses to the task in some regions lead to more sensitive detection of responses in other regions. We applied the approach to epilepsy patients, where task-derived patient-specific models provided additional information compared to the use of a generic model for the detection of BOLD responses to epileptiform activity identified on scalp electro-encephalogram (EEG). This work highlights the importance of improving the accuracy of the model for detecting neuronal activation with fMRI, and the fact that it can be done at no cost to model parsimony through the acquisition of independent a priori information about the hemodynamic response.

Temporal profile of fronto-striatal-limbic activity during implicit decisions in drug dependence

BackgroundSubstance dependence is associated with impaired decision-making and altered fronto-striatal-limbic activity. Both greater and lesser brain activity have been reported in drug users compared to controls during decision-making. Inconsistent results might be explained by group differences in the temporal profile of the functional magnetic resonance imaging (fMRI) response. While most previous studies model a canonical hemodynamic response, a finite impulse response (FIR) model measures fMRI signal at discrete time points without assuming a temporal profile. We compared brain activity during decision-making and feedback in substance users and controls using two models: a canonical hemodynamic response function (HRF) and a FIR model. Methods37 substance-dependent individuals (SDI) and 43 controls performed event-related decision-making during fMRI scanning. Brain activity was compared across group using canonical HRF and FIR models. ResultsCompared to controls, SDI were impaired at decision-making. The canonical HRF model showed that SDI had significantly greater fronto-striatal-limbic activity during decisions and less activity during feedback than controls. The FIR model confirmed greater activity in SDI during decisions. However, lower activity in SDI during feedback corresponded to a lower post-stimulus undershoot of the hemodynamic response. ConclusionsGreater activity in fronto-striatal-limbic pathways in SDI compared to controls is consistent with prior work, further supporting the hypothesis that abnormalities in these circuits underlie impaired decision-making. We demonstrate for the first time using FIR analysis that lower activity during feedback may simply reflect the tail end of the hemodynamic response to decision, the post-stimulus undershoot, rather than an actual difference in feedback response.

Modeling of the Hemodynamic Responses in Block Design fMRI Studies

The hemodynamic response function (HRF) describes the local response of brain vasculature to functional activation. Accurate HRF modeling enables the investigation of cerebral blood flow regulation and improves our ability to interpret fMRI results. Block designs have been used extensively as fMRI paradigms because detection power is maximized; however, block designs are not optimal for HRF parameter estimation. Here we assessed the utility of block design fMRI data for HRF modeling. The trueness (relative deviation), precision (relative uncertainty), and identifiability (goodness-of-fit) of different HRF models were examined and test-retest reproducibility of HRF parameter estimates was assessed using computer simulations and fMRI data from 82 healthy young adult twins acquired on two occasions 3 to 4 months apart. The effects of systematically varying attributes of the block design paradigm were also examined. In our comparison of five HRF models, the model comprising the sum of two gamma functions with six free parameters had greatest parameter accuracy and identifiability. Hemodynamic response function height and time to peak were highly reproducible between studies and width was moderately reproducible but the reproducibility of onset time was low. This study established the feasibility and test-retest reliability of estimating HRF parameters using data from block design fMRI studies.

Effect of voluntary repetitive long-lasting muscle contraction activity on the BOLD signal as assessed by optimal hemodynamic response function

Among other neuroimaging techniques, functional magnetic resonance imaging (fMRI) can be useful for studying the development of motor fatigue. The aim of this study was to identify differences in cortical neuronal activation in nine subjects on three motor tasks: right-hand movement with minimum, maximum, and post-fatigue maximum finger flexion. fMRI activation maps for each subject and during each condition were obtained by estimating the optimal model of the hemodynamic response function (HRF) out of four standard HRF models and an individual-based HRF model (ibHRF). ibHRF was selected as the optimal model in six out of nine subjects for minimum movement, in five out of nine for maximum movement, and in eight out of nine for post-fatigue maximum movement. As compared to maximum movement, a large reduction in the total number of active voxels (primary sensorimotor area, supplementary motor area and cerebellum) was observed in post-fatigue maximum movement. This is the first approach to the evaluation of long-lasting contraction effort in healthy subjects by means of the fMRI paradigm with the use of an individual-based hemodynamic response. The results may be relevant for defining a baseline in future studies on central fatigue in patients with neuropathological disorders.

Modelling hemodynamic response function in epilepsy

ObjectiveElectroencephalography and functional magnetic resonance imaging (fMRI) can be combined to noninvasively map abnormal brain activation elicited by epileptic processes. A major aim was to investigate the impact of a subject-specific hemodynamic response function (HRF) to describe the differences across patients versus the use of a standard model. MethodsWe developed and applied on simulated and real data a method designed to choose optimum HRF model for identifying fMRI activation maps. In simulation, the ability of five models to reproduce data was assessed: four standard and an individual-based HRF model (ibHRF). In clinical data, drug-resistant epileptic patients underwent fMRI to investigate hemodynamic responses evoked by interictal activity. ResultsWhen data are simulated with models different from the standard ones, the results obtained with ibHRF are superior to those obtained with the standard HRFs. Results on real data indicate an increase in extent and degree of activation with the ibHRF in comparison of the results obtainable using standard HRFs. ConclusionsThe use of the same HRF in all patients is inappropriate and resolves in biased extension of the activation maps. SignificanceThe new method could represent an useful diagnostic tool for other clinical studies that may be biased because of misspecification of HRF.

Development of BOLD signal hemodynamic responses in the human brain

In the rodent brain the hemodynamic response to a brief external stimulus changes significantly during development. Analogous changes in human infants would complicate the determination and use of the hemodynamic response function (HRF) for functional magnetic resonance imaging (fMRI) in developing populations. We aimed to characterize HRF in human infants before and after the normal time of birth using rapid sampling of the Blood Oxygen Level Dependent (BOLD) signal. A somatosensory stimulus and an event related experimental design were used to collect data from 10 healthy adults, 15 sedated infants at term corrected post menstrual age (PMA) (median 41 + 1 weeks), and 10 preterm infants (median PMA 34 + 4 weeks). A positive amplitude HRF waveform was identified across all subject groups, with a systematic maturational trend in terms of decreasing time-to-peak and increasing positive peak amplitude associated with increasing age. Application of the age-appropriate HRF models to fMRI data significantly improved the precision of the fMRI analysis. These findings support the notion of a structured development in the brain's response to stimuli across the last trimester of gestation and beyond.

Improved decoding of neural activity from fMRI signals using non-separable spatiotemporal deconvolutions

The goal of most functional Magnetic Resonance Imaging (fMRI) analyses is to investigate neural activity. Many fMRI analysis methods assume that the temporal dynamics of the hemodynamic response function (HRF) to neural activation is separable from its spatial dynamics. Although there is empirical evidence that the HRF is more complex than suggested by space-time separable canonical HRF models, it is difficult to assess how much information about neural activity is lost when assuming space-time separability. In this study we directly test whether spatiotemporal variability in the HRF that is not captured by separable models contains information about neural signals. We predict intracranially measured neural activity from simultaneously recorded fMRI data using separable and non-separable spatiotemporal deconvolutions of voxel time series around the recording electrode. Our results show that abandoning the spatiotemporal separability assumption consistently improves the decoding accuracy of neural signals from fMRI data. We compare our findings with results from optical imaging and fMRI studies and discuss potential implications for classical fMRI analyses without invasive electrophysiological recordings.