Dynamic Adjustment of Stimuli in Real Time Functional Magnetic Resonance Imaging

The conventional fMRI image analysis approach to associating stimuli to brain activation is performed by carrying out a massive number of parallel univariate regression analyses. fMRI blood-oxygen-level dependent (BOLD) signal, the basis of these analyses, is known for its low signal-noise-ratio and high spatial and temporal signal correlation. In order to ensure accurate localization of brain activity, stimulus administration in an fMRI session is often lengthy and repetitive. Real-time fMRI BOLD signal analysis is carried out as the signal is observed. This method allows for dynamic, real-time adjustment of stimuli through sequential experimental designs. We have developed a voxel-wise sequential probability ratio test (SPRT) approach for dynamically determining localization, as well as decision rules for stopping stimulus administration. SPRT methods and general linear model (GLM) approaches are combined to identify brain regions that are activated by specific elements of stimuli. Stimulus administration is dynamically stopped when sufficient statistical evidence is collected to determine activation status across regions of interest, following predetermined statistical error thresholds. Simulation experiments and an example based on real fMRI data show that scan volumes can be substantially reduced when compared with pre-determined, fixed designs while achieving similar or better accuracy in detecting activated voxels. Moreover, the proposed approach is also able to accurately detect differentially activated areas, and other comparisons between task-related GLM parameters that can be formulated in a hypothesis-testing framework. Finally, we give a demonstration of SPRT being employed in conjunction with a halving algorithm to dynamically adjust stimuli.     

fMRI Validation of fNIRS Measurements During a Naturalistic Task

We present a method to compare brain activity recorded with near-infrared spectroscopy (fNIRS) in a dance video game task to that recorded in a reduced version of the task using fMRI (functional magnetic resonance imaging). Recently, it has been shown that fNIRS can accurately record functional brain activities equivalent to those concurrently recorded with functional magnetic resonance imaging for classic psychophysical tasks and simple finger tapping paradigms. However, an often quoted benefit of fNIRS is that the technique allows for studying neural mechanisms of complex, naturalistic behaviors that are not possible using the constrained environment of fMRI. Our goal was to extend the findings of previous studies that have shown high correlation between concurrently recorded fNIRS and fMRI signals to compare neural recordings obtained in fMRI procedures to those separately obtained in naturalistic fNIRS experiments. Specifically, we developed a modified version of the dance video game Dance Dance Revolution (DDR) to be compatible with both fMRI and fNIRS imaging procedures. In this methodology we explain the modifications to the software and hardware for compatibility with each technique as well as the scanning and calibration procedures used to obtain representative results. The results of the study show a task-related increase in oxyhemoglobin in both modalities and demonstrate that it is possible to replicate the findings of fMRI using fNIRS in a naturalistic task. This technique represents a methodology to compare fMRI imaging paradigms which utilize a reduced-world environment to fNIRS in closer approximation to naturalistic, full-body activities and behaviors. Further development of this technique may apply to neurodegenerative diseases, such as Parkinson’s disease, late states of dementia, or those with magnetic susceptibility which are contraindicated for fMRI scanning.     

Solving the brain synchrony eigenvalue problem: conservation of temporal dynamics (fMRI) over subjects doing the same task

Brain measures often show highly structured temporal dynamics that synchronize when observers are doing the same task. The standard method for analysis of brain imaging signals (e.g. fMRI) uses the GLM for each voxel indexed against a specified experimental design but does not explicitly involve temporal dynamics. Consequently, the design variables that determine the functional brain areas are those correlated with the design variation rather than the common or conserved brain areas across subjects with the same temporal dynamics given the same stimulus conditions. This raises an important theoretical question: Are temporal dynamics conserved across individuals experiencing the same stimulus task? This general question can be framed in a dynamical systems context and further be posed as an eigenvalue problem about the conservation of synchrony across all brains simultaneously. We show that solving the problem results in a non-arbitrary measure of temporal dynamics across brains that scales over any number of subjects, stabilizes with increasing sample size, and varies systematically across tasks and stimulus conditions.     

BOLD Correlates of Trial-by-Trial Reaction Time Variability in Gray and White Matter: A Multi-Study fMRI Analysis

Background

Reaction time (RT) is one of the most widely used measures of performance in experimental psychology, yet relatively few fMRI studies have included trial-by-trial differences in RT as a predictor variable in their analyses. Using a multi-study approach, we investigated whether there are brain regions that show a general relationship between trial-by-trial RT variability and activation across a range of cognitive tasks.

Methodology/Principal Findings

The relation between trial-by-trial differences in RT and brain activation was modeled in five different fMRI datasets spanning a range of experimental tasks and stimulus modalities. Three main findings were identified. First, in a widely distributed set of gray and white matter regions, activation was delayed on trials with long RTs relative to short RTs, suggesting delayed initiation of underlying physiological processes. Second, in lateral and medial frontal regions, activation showed a “time-on-task” effect, increasing linearly as a function of RT. Finally, RT variability reliably modulated the BOLD signal not only in gray matter but also in diffuse regions of white matter.

Conclusions/Significance

The results highlight the importance of modeling trial-by-trial RT in fMRI analyses and raise the possibility that RT variability may provide a powerful probe for investigating the previously elusive white matter BOLD signal.     

Differentiation between vergence and saccadic functional activity within the human frontal eye fields and midbrain revealed through fMRI

Purpose

Eye movement research has traditionally studied solely saccade and/or vergence eye movements by isolating these systems within a laboratory setting. While the neural correlates of saccadic eye movements are established, few studies have quantified the functional activity of vergence eye movements using fMRI. This study mapped the neural substrates of vergence eye movements and compared them to saccades to elucidate the spatial commonality and differentiation between these systems.

Methodology

The stimulus was presented in a block design where the ‘off’ stimulus was a sustained fixation and the ‘on’ stimulus was random vergence or saccadic eye movements. Data were collected with a 3T scanner. A general linear model (GLM) was used in conjunction with cluster size to determine significantly active regions. A paired t-test of the GLM beta weight coefficients was computed between the saccade and vergence functional activities to test the hypothesis that vergence and saccadic stimulation would have spatial differentiation in addition to shared neural substrates.

Results

Segregated functional activation was observed within the frontal eye fields where a portion of the functional activity from the vergence task was located anterior to the saccadic functional activity (z>2.3; p<0.03). An area within the midbrain was significantly correlated with the experimental design for the vergence but not the saccade data set. Similar functional activation was observed within the following regions of interest: the supplementary eye field, dorsolateral prefrontal cortex, ventral lateral prefrontal cortex, lateral intraparietal area, cuneus, precuneus, anterior and posterior cingulates, and cerebellar vermis. The functional activity from these regions was not different between the vergence and saccade data sets assessed by analyzing the beta weights of the paired t-test (p>0.2).

Conclusion

Functional MRI can elucidate the differences between the vergence and saccade neural substrates within the frontal eye fields and midbrain.     

Functional neuronal network activity differs with cognitive dysfunction in childhood-onset systemic lupus erythematosus

Introduction

Neuropsychiatric manifestations are common in childhood-onset systemic lupus erythematosus (cSLE) and often include neurocognitive dysfunction (NCD). Functional magnetic resonance imaging (fMRI) can measure brain activation during tasks that invoke domains of cognitive function impaired by cSLE. This study investigates specific changes in brain function attributable to NCD in cSLE that have potential to serve as imaging biomarkers.

Methods

Formal neuropsychological testing was done to measure cognitive ability and to identify NCD. Participants performed fMRI tasks probing three cognitive domains impacted by cSLE: visuoconstructional ability (VCA), working memory, and attention. Imaging data, collected on 3-Tesla scanners, included a high-resolution T1-weighted anatomic reference image followed by a T2*-weighted whole-brain echo planar image series for each fMRI task. Brain activation using blood oxygenation level-dependent contrast was compared between cSLE patients with NCD (NCD-group, n = 7) vs. without NCD (noNCD-group, n = 14) using voxel-wise and region of interest-based analyses. The relationship of brain activation during fMRI tasks and performance in formal neuropsychological testing was assessed.

Results

Greater brain activation was observed in the noNCD-group vs. NCD-group during VCA and working memory fMRI tasks. Conversely, compared to the noNCD-group, the NCD-group showed more brain activation during the attention fMRI task. In region of interest analysis, brain activity during VCA and working memory fMRI tasks was positively associated with the participants'' neuropsychological test performance. In contrast, brain activation during the attention fMRI task was negatively correlated with neuropsychological test performance. While the NCD group performed worse than the noNCD group during VCA and working memory tasks, the attention task was performed equally well by both groups.

Conclusions

NCD in patients with cSLE is characterized by differential activation of functional neuronal networks during fMRI tasks probing working memory, VCA, and attention. Results suggest a compensatory mechanism allows maintenance of attentional performance under NCD. This mechanism appears to break down for the VCA and working memory challenges presented in this study. The observation that neuronal network activation is related to the formal neuropsychological testing performance makes fMRI a candidate imaging biomarker for cSLE-associated NCD.     

Stimulus-related independent component and voxel-wise analysis of human brain activity during free viewing of a feature film

Understanding how the brain processes stimuli in a rich natural environment is a fundamental goal of neuroscience. Here, we showed a feature film to 10 healthy volunteers during functional magnetic resonance imaging (fMRI) of hemodynamic brain activity. We then annotated auditory and visual features of the motion picture to inform analysis of the hemodynamic data. The annotations were fitted to both voxel-wise data and brain network time courses extracted by independent component analysis (ICA). Auditory annotations correlated with two independent components (IC) disclosing two functional networks, one responding to variety of auditory stimulation and another responding preferentially to speech but parts of the network also responding to non-verbal communication. Visual feature annotations correlated with four ICs delineating visual areas according to their sensitivity to different visual stimulus features. In comparison, a separate voxel-wise general linear model based analysis disclosed brain areas preferentially responding to sound energy, speech, music, visual contrast edges, body motion and hand motion which largely overlapped the results revealed by ICA. Differences between the results of IC- and voxel-based analyses demonstrate that thorough analysis of voxel time courses is important for understanding the activity of specific sub-areas of the functional networks, while ICA is a valuable tool for revealing novel information about functional connectivity which need not be explained by the predefined model. Our results encourage the use of naturalistic stimuli and tasks in cognitive neuroimaging to study how the brain processes stimuli in rich natural environments.     

Bayesian Spatiotemporal Inference in Functional Magnetic Resonance Imaging

Mapping of the human brain by means of functional magnetic resonance imaging (fMRI) is an emerging field in cognitive and clinical neuroscience. Current techniques to detect activated areas of the brain mostly proceed in two steps. First, conventional methods of correlation, regression, and time series analysis are used to assess activation by a separate, pixelwise comparison of the fMRI signal time courses to the reference function of a presented stimulus. Spatial aspects caused by correlations between neighboring pixels are considered in a separate second step, if at all. The aim of this article is to present hierarchical Bayesian approaches that allow one to simultaneously incorporate temporal and spatial dependencies between pixels directly in the model formulation. For reasons of computational feasibility, models have to be comparatively parsimonious, without oversimplifying. We introduce parametric and semiparametric spatial and spatiotemporal models that proved appropriate and illustrate their performance applied to visual fMRI data.     

Transferring Cognitive Tasks Between Brain Imaging Modalities: Implications for Task Design and Results Interpretation in fMRI Studies

As cognitive neuroscience methods develop, established experimental tasks are used with emerging brain imaging modalities. Here transferring a paradigm (the visual oddball task) with a long history of behavioral and electroencephalography (EEG) experiments to a functional magnetic resonance imaging (fMRI) experiment is considered. The aims of this paper are to briefly describe fMRI and when its use is appropriate in cognitive neuroscience; illustrate how task design can influence the results of an fMRI experiment, particularly when that task is borrowed from another imaging modality; explain the practical aspects of performing an fMRI experiment. It is demonstrated that manipulating the task demands in the visual oddball task results in different patterns of blood oxygen level dependent (BOLD) activation. The nature of the fMRI BOLD measure means that many brain regions are found to be active in a particular task. Determining the functions of these areas of activation is very much dependent on task design and analysis. The complex nature of many fMRI tasks means that the details of the task and its requirements need careful consideration when interpreting data. The data show that this is particularly important in those tasks relying on a motor response as well as cognitive elements and that covert and overt responses should be considered where possible. Furthermore, the data show that transferring an EEG paradigm to an fMRI experiment needs careful consideration and it cannot be assumed that the same paradigm will work equally well across imaging modalities. It is therefore recommended that the design of an fMRI study is pilot tested behaviorally to establish the effects of interest and then pilot tested in the fMRI environment to ensure appropriate design, implementation and analysis for the effects of interest.     

Identifying Cognitive States Using Regularity Partitions

Functional Magnetic Resonance (fMRI) data can be used to depict functional connectivity of the brain. Standard techniques have been developed to construct brain networks from this data; typically nodes are considered as voxels or sets of voxels with weighted edges between them representing measures of correlation. Identifying cognitive states based on fMRI data is connected with recording voxel activity over a certain time interval. Using this information, network and machine learning techniques can be applied to discriminate the cognitive states of the subjects by exploring different features of data. In this work we wish to describe and understand the organization of brain connectivity networks under cognitive tasks. In particular, we use a regularity partitioning algorithm that finds clusters of vertices such that they all behave with each other almost like random bipartite graphs. Based on the random approximation of the graph, we calculate a lower bound on the number of triangles as well as the expectation of the distribution of the edges in each subject and state. We investigate the results by comparing them to the state of the art algorithms for exploring connectivity and we argue that during epochs that the subject is exposed to stimulus, the inspected part of the brain is organized in an efficient way that enables enhanced functionality.     

Neuroimaging evidence for processes underlying repetition of ignored stimuli

Prolonged response times are observed with targets having been presented as distractors immediately before, called negative priming effect. Among others, inhibitory and retrieval processes have been suggested underlying this behavioral effect. As those processes would involve different neural activation patterns, a functional magnetic resonance imaging (fMRI) study including 28 subjects was conducted. Two tasks were used to investigate stimulus repetition effects. One task focused on target location, the other on target identity. Both tasks are known to elicit the expected response time effects. However, there is less agreement about the relationship of those tasks with the explanatory accounts under consideration. Based on within-subject comparisons we found clear differences between the experimental repetition conditions and the neutral control condition on neural level for both tasks. Hemodynamic fronto-striatal activation patterns occurred for the location-based task favoring the selective inhibition account. Hippocampal activation found for the identity-based task suggests an assignment to the retrieval account; however, this task lacked a behavioral effect.     

Stimulus-dependent adjustment of reward prediction error in the midbrain

Previous reports have described that neural activities in midbrain dopamine areas are sensitive to unexpected reward delivery and omission. These activities are correlated with reward prediction error in reinforcement learning models, the difference between predicted reward values and the obtained reward outcome. These findings suggest that the reward prediction error signal in the brain updates reward prediction through stimulus-reward experiences. It remains unknown, however, how sensory processing of reward-predicting stimuli contributes to the computation of reward prediction error. To elucidate this issue, we examined the relation between stimulus discriminability of the reward-predicting stimuli and the reward prediction error signal in the brain using functional magnetic resonance imaging (fMRI). Before main experiments, subjects learned an association between the orientation of a perceptually salient (high-contrast) Gabor patch and a juice reward. The subjects were then presented with lower-contrast Gabor patch stimuli to predict a reward. We calculated the correlation between fMRI signals and reward prediction error in two reinforcement learning models: a model including the modulation of reward prediction by stimulus discriminability and a model excluding this modulation. Results showed that fMRI signals in the midbrain are more highly correlated with reward prediction error in the model that includes stimulus discriminability than in the model that excludes stimulus discriminability. No regions showed higher correlation with the model that excludes stimulus discriminability. Moreover, results show that the difference in correlation between the two models was significant from the first session of the experiment, suggesting that the reward computation in the midbrain was modulated based on stimulus discriminability before learning a new contingency between perceptually ambiguous stimuli and a reward. These results suggest that the human reward system can incorporate the level of the stimulus discriminability flexibly into reward computations by modulating previously acquired reward values for a typical stimulus.     

A Neural Network Approach to fMRI Binocular Visual Rivalry Task Analysis

The purpose of this study was to investigate whether artificial neural networks (ANN) are able to decode participants’ conscious experience perception from brain activity alone, using complex and ecological stimuli. To reach the aim we conducted pattern recognition data analysis on fMRI data acquired during the execution of a binocular visual rivalry paradigm (BR). Twelve healthy participants were submitted to fMRI during the execution of a binocular non-rivalry (BNR) and a BR paradigm in which two classes of stimuli (faces and houses) were presented. During the binocular rivalry paradigm, behavioral responses related to the switching between consciously perceived stimuli were also collected. First, we used the BNR paradigm as a functional localizer to identify the brain areas involved the processing of the stimuli. Second, we trained the ANN on the BNR fMRI data restricted to these regions of interest. Third, we applied the trained ANN to the BR data as a ‘brain reading’ tool to discriminate the pattern of neural activity between the two stimuli. Fourth, we verified the consistency of the ANN outputs with the collected behavioral indicators of which stimulus was consciously perceived by the participants. Our main results showed that the trained ANN was able to generalize across the two different tasks (i.e. BNR and BR) and to identify with high accuracy the cognitive state of the participants (i.e. which stimulus was consciously perceived) during the BR condition. The behavioral response, employed as control parameter, was compared with the network output and a statistically significant percentage of correspondences (p-value <0.05) were obtained for all subjects. In conclusion the present study provides a method based on multivariate pattern analysis to investigate the neural basis of visual consciousness during the BR phenomenon when behavioral indicators lack or are inconsistent, like in disorders of consciousness or sedated patients.     

Using Virtual Reality to Study Alcohol Intoxication Effects on the Neural Correlates of Simulated Driving

The use of virtual reality in the form of simulated tasks can provide a realistic environment in which to study complex naturalistic behaviors. Many of the behavioral effects of alcohol intoxication are well known, but there is relatively little imaging evidence examining how alcohol exposure might transiently modulate brain function, especially in the context of task performance. In this review, we provide a brief synopsis of previous work using functional magnetic resonance imaging (fMRI) to study the neural correlates of alcohol intoxication. We describe in detail two studies from our published work, the first involving a visual perception paradigm, and the second involving virtual reality through a naturalistic behavior; simulated driving. Participants received single-blind individualized doses of beverage alcohol designed to produce blood alcohol content (BAC) of 0.04 and 0.08 or placebo. Subjects were fMRI scanned after training to asymptote performance. In both studies we found specific circuits that were differentially modulated by alcohol, we revealed both global and local effects of alcohol, and we examined relationships between behavior, brain function, and alcohol blood levels.     

Retinotopic maps, spatial tuning, and locations of human visual areas in surface coordinates characterized with multifocal and blocked FMRI designs

The localization of visual areas in the human cortex is typically based on mapping the retinotopic organization with functional magnetic resonance imaging (fMRI). The most common approach is to encode the response phase for a slowly moving visual stimulus and to present the result on an individual's reconstructed cortical surface. The main aims of this study were to develop complementary general linear model (GLM)-based retinotopic mapping methods and to characterize the inter-individual variability of the visual area positions on the cortical surface. We studied 15 subjects with two methods: a 24-region multifocal checkerboard stimulus and a blocked presentation of object stimuli at different visual field locations. The retinotopic maps were based on weighted averaging of the GLM parameter estimates for the stimulus regions. In addition to localizing visual areas, both methods could be used to localize multiple retinotopic regions-of-interest. The two methods yielded consistent retinotopic maps in the visual areas V1, V2, V3, hV4, and V3AB. In the higher-level areas IPS0, VO1, LO1, LO2, TO1, and TO2, retinotopy could only be mapped with the blocked stimulus presentation. The gradual widening of spatial tuning and an increase in the responses to stimuli in the ipsilateral visual field along the hierarchy of visual areas likely reflected the increase in the average receptive field size. Finally, after registration to Freesurfer's surface-based atlas of the human cerebral cortex, we calculated the mean and variability of the visual area positions in the spherical surface-based coordinate system and generated probability maps of the visual areas on the average cortical surface. The inter-individual variability in the area locations decreased when the midpoints were calculated along the spherical cortical surface compared with volumetric coordinates. These results can facilitate both analysis of individual functional anatomy and comparisons of visual cortex topology across studies.     

The role of brain‐derived neurotrophic factor in learned fear processing: an awake rat fMRI study

Brain‐derived neurotrophic factor (BDNF) signaling is implicated in the etiology of many psychiatric disorders associated with altered emotional processing. Altered peripheral (plasma) BDNF levels have been proposed as a biomarker for neuropsychiatric disease risk in humans. However, the relationship between peripheral and central BDNF levels and emotional brain activation is unknown. We used heterozygous BDNF knockdown rats (BDNF^+/?) to examine the effects of genetic variation in the BDNF gene on peripheral and central BDNF levels and emotional brain activation as assessed by awake functional magnetic resonance imaging (fMRI). BDNF^+/? and control rats were trained to associate a flashing light (conditioned stimulus; CS) with foot‐shock, and brain activation in response to the CS was measured 24 h later in awake rats using fMRI. Central and peripheral BDNF levels were decreased in BDNF^+/? rats compared with control rats. Activation of fear circuitry (amygdala, periaqueductal gray, granular insular) was seen in control animals; however, activation of this circuitry was absent in BDNF^+/? animals. Behavioral experiments confirmed impaired conditioned fear responses in BDNF^+/? rats, despite intact innate fear responses. These data confirm a positive correlation [r = 0.86, 95% confidence interval (0.55, 0.96); P = 0.0004] between peripheral and central BDNF levels and indicate a functional relationship between BDNF levels and emotional brain activation as assessed by fMRI. The results demonstrate the use of rodent fMRI as a sensitive tool for measuring brain function in preclinical translational studies using genetically modified rats and support the use of peripheral BDNF as a biomarker of central affective processing.     

Learning brain dynamics for decoding and predicting individual differences

Insights from functional Magnetic Resonance Imaging (fMRI), as well as recordings of large numbers of neurons, reveal that many cognitive, emotional, and motor functions depend on the multivariate interactions of brain signals. To decode brain dynamics, we propose an architecture based on recurrent neural networks to uncover distributed spatiotemporal signatures. We demonstrate the potential of the approach using human fMRI data during movie-watching data and a continuous experimental paradigm. The model was able to learn spatiotemporal patterns that supported 15-way movie-clip classification (∼90%) at the level of brain regions, and binary classification of experimental conditions (∼60%) at the level of voxels. The model was also able to learn individual differences in measures of fluid intelligence and verbal IQ at levels comparable to that of existing techniques. We propose a dimensionality reduction approach that uncovers low-dimensional trajectories and captures essential informational (i.e., classification related) properties of brain dynamics. Finally, saliency maps and lesion analysis were employed to characterize brain-region/voxel importance, and uncovered how dynamic but consistent changes in fMRI activation influenced decoding performance. When applied at the level of voxels, our framework implements a dynamic version of multivariate pattern analysis. Our approach provides a framework for visualizing, analyzing, and discovering dynamic spatially distributed brain representations during naturalistic conditions.     

A generalized linear model for estimating spectrotemporal receptive fields from responses to natural sounds

In the auditory system, the stimulus-response properties of single neurons are often described in terms of the spectrotemporal receptive field (STRF), a linear kernel relating the spectrogram of the sound stimulus to the instantaneous firing rate of the neuron. Several algorithms have been used to estimate STRFs from responses to natural stimuli; these algorithms differ in their functional models, cost functions, and regularization methods. Here, we characterize the stimulus-response function of auditory neurons using a generalized linear model (GLM). In this model, each cell's input is described by: 1) a stimulus filter (STRF); and 2) a post-spike filter, which captures dependencies on the neuron's spiking history. The output of the model is given by a series of spike trains rather than instantaneous firing rate, allowing the prediction of spike train responses to novel stimuli. We fit the model by maximum penalized likelihood to the spiking activity of zebra finch auditory midbrain neurons in response to conspecific vocalizations (songs) and modulation limited (ml) noise. We compare this model to normalized reverse correlation (NRC), the traditional method for STRF estimation, in terms of predictive power and the basic tuning properties of the estimated STRFs. We find that a GLM with a sparse prior predicts novel responses to both stimulus classes significantly better than NRC. Importantly, we find that STRFs from the two models derived from the same responses can differ substantially and that GLM STRFs are more consistent between stimulus classes than NRC STRFs. These results suggest that a GLM with a sparse prior provides a more accurate characterization of spectrotemporal tuning than does the NRC method when responses to complex sounds are studied in these neurons.     

Segmentation of bones in medical dual-energy computed tomography volumes using the 3D U-Net

Deep learning algorithms have improved the speed and quality of segmentation for certain tasks in medical imaging. The aim of this work is to design and evaluate an algorithm capable of segmenting bones in dual-energy CT data sets. A convolutional neural network based on the 3D U-Net architecture was implemented and evaluated using high tube voltage images, mixed images and dual-energy images from 30 patients. The network performed well on all the data sets; the mean Dice coefficient for the test data was larger than 0.963. Of special interest is that it performed better on dual-energy CT volumes compared to mixed images that mimicked images taken at 120 kV. The corresponding increase in the Dice coefficient from 0.965 to 0.966 was small since the enhancements were mainly at the edges of the bones. The method can easily be extended to the segmentation of multi-energy CT data.