Multivariate lesion-symptom mapping using support vector regression.

Lesion analysis is a classic approach to study brain functions. Because brain function is a result of coherent activations of a collection of functionally related voxels, lesion-symptom relations are generally contributed by multiple voxels simultaneously. Although voxel-based lesion-symptom mapping (VLSM) has made substantial contributions to the understanding of brain-behavior relationships, a better understanding of the brain-behavior relationship contributed by multiple brain regions needs a multivariate lesion-symptom mapping (MLSM). The purpose of this artilce was to develop an MLSM using a machine learning-based multivariate regression algorithm: support vector regression (SVR). In the proposed SVR-LSM, the symptom relation to the entire lesion map as opposed to each isolated voxel is modeled using a nonlinear function, so the intervoxel correlations are intrinsically considered, resulting in a potentially more sensitive way to examine lesion-symptom relationships. To explore the relative merits of VLSM and SVR-LSM we used both approaches in the analysis of a synthetic dataset. SVR-LSM showed much higher sensitivity and specificity for detecting the synthetic lesion-behavior relations than VLSM. When applied to lesion data and language measures from patients with brain damages, SVR-LSM reproduced the essential pattern of previous findings identified by VLSM and showed higher sensitivity than VLSM for identifying the lesion-behavior relations. Our data also showed the possibility of using lesion data to predict continuous behavior scores.

Neuroanatomical dissociation for taxonomic and thematic knowledge in the human brain.

It is thought that semantic memory represents taxonomic information differently from thematic information. This study investigated the neural basis for the taxonomic-thematic distinction in a unique way. We gathered picture-naming errors from 86 individuals with poststroke language impairment (aphasia). Error rates were determined separately for taxonomic errors ("pear" in response to apple) and thematic errors ("worm" in response to apple), and their shared variance was regressed out of each measure. With the segmented lesions normalized to a common template, we carried out voxel-based lesion-symptom mapping on each error type separately. We found that taxonomic errors localized to the left anterior temporal lobe and thematic errors localized to the left temporoparietal junction. This is an indication that the contribution of these regions to semantic memory cleaves along taxonomic-thematic lines. Our findings show that a distinction long recognized in the psychological sciences is grounded in the structure and function of the human brain.

The neural basis of reversible sentence comprehension: evidence from voxel-based lesion symptom mapping in aphasia.

We explored the neural basis of reversible sentence comprehension in a large group of aphasic patients (n = 79). Voxel-based lesion symptom mapping revealed a significant association between damage in temporo-parietal cortex and impaired sentence comprehension. This association remained after we controlled for phonological working memory. We hypothesize that this region plays an important role in the thematic or what-where processing of sentences. In contrast, we detected weak or no association between reversible sentence comprehension and the ventrolateral pFC, which includes Broca's area, even for syntactically complex sentences. This casts doubt on theories that presuppose a critical role for this region in syntactic computations.

Localizing interference during naming: convergent neuroimaging and neuropsychological evidence for the function of Broca's area.

To produce a word, the intended word must be selected from a competing set of other words. In other domains where competition affects the selection process, the left inferior frontal gyrus (LIFG) responds to competition among incompatible representations. The aim of this study was to test whether the LIFG is necessary for resolution of competition in word production. Using a methodological approach applying the same rigorous analytic methods to neuropsychological data as is done with neuroimaging data, we compared brain activation patterns in normal speakers (using fMRI) with the results of lesion-deficit correlations in aphasic speakers who performed the same word production task designed to elicit competition during lexical selection. The degree of activation of the LIFG in normal speakers and damage to the LIFG in aphasic speakers was associated with performance on the production task. These convergent findings provide strong support for the hypothesis that the region of cortex commonly known as Broca's area (i.e., the posterior LIFG) serves to bias competitive interactions during language production.

Direct comparison of fluorodeoxyglucose positron emission tomography and arterial spin labeling magnetic resonance imaging in Alzheimer's disease.

BACKGROUND: The utility of fluorodeoxyglucose positron emission tomography (FDG-PET) imaging in Alzheimer's disease (AD) diagnosis has been well established. Recently, measurement of cerebral blood flow using arterial spin labeling magnetic resonance imaging (ASL-MRI) has shown diagnostic potential in AD, although it has never been directly compared with FDG-PET. METHODS: We used a novel imaging protocol to obtain FDG-PET and ASL-MRI images concurrently in 17 AD patients and 19 age-matched control subjects. Paired FDG-PET and ASL-MRI images from 19 control subjects and 15 AD patients were included for qualitative analysis, and paired images from 18 control subjects and 13 AD patients were suitable for quantitative analyses. RESULTS: The combined imaging protocol was well tolerated. Both modalities revealed similar regional abnormalities in AD, as well as comparable sensitivity and specificity for the detection of AD after visual review by two expert readers. Interobserver agreement was better for FDG-PET (κ: 0.75, standard error: 0.12) than ASL-MRI (κ: 0.51, standard error: 0.15); intermodality agreement was moderate to strong (κ: 0.45-0.61); and readers were more confident of FDG-PET reads. Simple quantitative analysis of global cerebral fluorodeoxyglucose uptake (FDG-PET) or whole-brain cerebral blood flow (ASL-MRI) showed excellent diagnostic accuracy for both modalities, with area under receiver operating characteristic curves of 0.90 for FDG-PET (95% confidence interval: 0.79-0.99) and 0.91 for ASL-MRI (95% confidence interval: 0.80-1.00). CONCLUSIONS: Our results demonstrate that FDG-PET and ASL-MRI identify similar regional abnormalities and have comparable diagnostic accuracy in a small population of AD patients, and support the further study of ASL-MRI in dementia diagnosis.

Anterior temporal involvement in semantic word retrieval: voxel-based lesion-symptom mapping evidence from aphasia.

Analysis of error types provides useful information about the stages and processes involved in normal and aphasic word production. In picture naming, semantic errors (horse for goat) generally result from something having gone awry in lexical access such that the right concept was mapped to the wrong word. This study used the new lesion analysis technique known as voxel-based lesion-symptom mapping to investigate the locus of lesions that give rise to semantic naming errors. Semantic errors were obtained from 64 individuals with post-stroke aphasia, who also underwent high-resolution structural brain scans. Whole brain voxel-based lesion-symptom mapping was carried out to determine where lesion status predicted semantic error rate. The strongest associations were found in the left anterior to mid middle temporal gyrus. This area also showed strong and significant effects in further analyses that statistically controlled for deficits in pre-lexical, conceptualization processes that might have contributed to semantic error production. This study is the first to demonstrate a specific and necessary role for the left anterior temporal lobe in mapping concepts to words in production. We hypothesize that this role consists in the conveyance of fine-grained semantic distinctions to the lexical system. Our results line up with evidence from semantic dementia, the convergence zone framework and meta-analyses of neuroimaging studies on word production. At the same time, they cast doubt on the classical linkage of semantic error production to lesions in and around Wernicke's area.

Distinctions between manipulation and function knowledge of objects: evidence from functional magnetic resonance imaging.

A prominent account of conceptual knowledge proposes that information is distributed over visual, tactile, auditory, motor and verbal-declarative attribute domains to the degree to which these features were activated when the knowledge was acquired [D.A. Allport, Distributed memory, modular subsystems and dysphagia, In: S.K. Newman, R. Epstein (Eds.), Current perspectives in dysphagia, Churchill Livingstone, Edinburgh, 1985, pp. 32-60]. A corollary is that when drawing upon this knowledge (e.g., to answer questions), particular aspects of this distributed information is re-activated as a function of the requirements of the task at hand [L.J. Buxbaum, E.M. Saffran, Knowledge of object manipulation and object function: dissociations in apraxic and non-apraxic subjects. Brain and Language, 82 (2002) 179-199; L.J. Buxbaum, T. Veramonti, M.F. Schwartz, Function and manipulation tool knowledge in apraxia: knowing 'what for' but not 'how', Neurocase, 6 (2000) 83-97; W. Simmons, L. Barsalou, The similarity-in-topography principle: Reconciling theories of conceptual deficits, Cognitive Neuropsychology, 20 (2003) 451-486]. This account predicts that answering questions about object manipulation should activate brain regions previously identified as components of the distributed sensory-motor system involved in object use, whereas answering questions about object function (that is, the purpose that it serves) should activate regions identified as components of the systems supporting verbal-declarative features. These predictions were tested in a functional magnetic resonance imaging (fMRI) study in which 15 participants viewed picture or word pairs denoting manipulable objects and determined whether the objects are manipulated similarly (M condition) or serve the same function (F condition). Significantly greater and more extensive activations in the left inferior parietal lobe bordering the intraparietal sulcus were seen in the M condition with pictures and, to a lesser degree, words. These findings are consistent with the known role of this region in skilled object use [K.M. Heilman, L.J. Gonzalez Rothi, Apraxia, In: K.M. Heilman, E. Valenstein (Eds.), Clinical Neuropsychology, Oxford University Press, New York, 1993, pp. 141-150] as well as previous fMRI results [M. Kellenbach, M. Brett, K. Patterson, Actions speak louder than functions: the importance of manipulability and action in tool representation, Journal of Cognitive Neuroscience, 15 (2003) 30-46] and behavioral findings in brain-lesion patients [L.J. Buxbaum, E.M. Saffran, Knowledge of object manipulation and object function: dissociations in apraxic and non-apraxic subjects, Brain and Language, 82 (2002) 179-199]. No brain regions were significantly more activated in the F than M condition. These data suggest that brain regions specialized for sensory-motor function are a critical component of distributed representations of manipulable objects.

Functional MRI predicts post-surgical memory following temporal lobectomy.

Temporal lobectomy is an effective therapy for medically refractory temporal lobe epilepsy (TLE), but may be complicated by amnestic syndromes. Therefore, pre-surgical evaluation to assess the risk/benefit ratio for surgery is required. Intracarotid amobarbital testing (IAT) is currently the most widely used method for assessing pre-surgical memory lateralization, but is relatively invasive. Over the past decade functional MRI (fMRI) has been shown to correlate with IAT for language lateralization, and also for memory lateralization in a small number of patients. This study was carried out to compare fMRI during memory encoding with IAT testing for memory lateralization, and to assess the predictive value of fMRI during memory encoding for post-surgical memory outcome. Thirty-five patients with refractory TLE undergoing pre-surgical evaluation for temporal lobectomy and 30 normal subjects performed a complex visual scene-encoding task during fMRI scanning at 1.5 T using a 10-min protocol. Encoding performance was evaluated with subsequent recognition testing. Twenty-three patients also completed the same task again outside the scanner, an average of 6.9 months following surgery. A region of interest (ROI) analysis was used to quantify activation within hippocampal and a larger mesial temporal lobe ROI consisting of hippocampus, parahippocampus and fusiform gyrus (HPF) as defined by a published template. Normal subjects showed almost symmetrical activation within these ROI. TLE patients showed greater asymmetry. Asymmetry ratios (ARs) from the HPF ROI correlated significantly with memory lateralization by intracarotid amobarbital testing. HPF ARs also correlated significantly with memory outcome, as determined by a change in scene recognition between pre-surgical and post-surgical trials. When absolute activation within the HPF ROI was considered, a significant inverse correlation between activation ipsilateral to temporal lobectomy and memory outcome was observed, with no significant correlation in the contralateral HPF ROI. Although further technical improvements and prospective clinical validation are required, these results suggest that mesial temporal memory activation detected by fMRI during complex visual scene encoding correlates with post-surgical memory outcome and supports the notion that this approach will ultimately contribute to patient management.

Cognitive effects of the dopamine receptor agonist pergolide.

Although dopamine has been closely associated with prefrontal function, and with working memory in monkeys, the effects of dopamine agonists on human cognitive performance are poorly understood. We report the effects of a single dose of pergolide on young healthy subjects performing a variety of cognitive tests, including tests of memory and of frontal/executive function. Across this battery of tasks, the only tasks reliably affected by pergolide were delayed response tasks. Across four variants, we observed that the effect of pergolide was more beneficial for subjects with greater working memory capacities. We discuss this in light of the variable results obtained from previous studies of dopamine agonists in human subjects.

Support vector regression based multivariate lesion-symptom mapping.

A novel multivariate lesion-symptom mapping (LSM) methodology was developed in this study. Lesion analysis is a classic model for studying brain functions. Using lesion data, focal brain-behavior associations have been widely assessed using the massive voxel-based lesion symptom mapping (VLSM) method. Assessing each voxel independently, VLSM suffers from low sensitivity after correcting for the enormous number of comparisons. It is also incapable for assessing a spatially distributed association pattern though the brain-behavior associations generally involve a collection of functionally related voxels. To solve these two outstanding problems, we carried out the first multivariate lesion symptom mapping (MLSM) in this study using support vector regression (SVR). In the so dubbed SVR-LSM, the symptom relation to the entire lesion map rather than each isolated voxel is modeled using a non-linear function, so the inter-voxel correlations are intrinsically considered, resulting in a potentially more sensitive way to examine lesion-symptom relationships. Evaluations using synthetic data and real data showed that SVR-LSM gained a much better performance (in terms of sensitivity and specificity) for detecting brain-behavior relations than VLSM. While the method was designed for lesion analysis, extending it to neuroimaging data will be straightforward.

Language, perception, and the schematic representation of spatial relations.

Schemas are abstract nonverbal representations that parsimoniously depict spatial relations. Despite their ubiquitous use in maps and diagrams, little is known about their neural instantiation. We sought to determine the extent to which schematic representations are neurally distinguished from language on the one hand, and from rich perceptual representations on the other. In patients with either left hemisphere damage or right hemisphere damage, a battery of matching tasks depicting categorical spatial relations was used to probe for the comprehension of basic spatial concepts across distinct representational formats (words, pictures, and schemas). Left hemisphere patients underperformed right hemisphere patients across all tasks. However, focused residual analyses using voxel-based lesion-symptom mapping (VLSM) suggest that (1) left hemisphere deficits in the representation of categorical spatial relations are difficult to distinguish from deficits in naming these relations and (2) the right hemisphere plays a special role in extracting schematic representations from richly textured pictures.

Support for anterior temporal involvement in semantic error production in aphasia: new evidence from VLSM.

Semantic errors in aphasia (e.g., naming a horse as "dog") frequently arise from faulty mapping of concepts onto lexical items. A recent study by our group used voxel-based lesion-symptom mapping (VLSM) methods with 64 patients with chronic aphasia to identify voxels that carry an association with semantic errors. The strongest associations were found in the left anterior temporal lobe (L-ATL), in the mid- to anterior MTG region. The absence of findings in Wernicke's area was surprising, as were indications that ATL voxels made an essential contribution to the post-semantic stage of lexical access. In this follow-up study, we sought to validate these results by re-defining semantic errors in a manner that was less theory dependent and more consistent with prior lesion studies. As this change also increased the robustness of the dependent variable, it made it possible to perform additional statistical analyses that further refined the interpretation. The results strengthen the evidence for a causal relationship between ATL damage and lexically-based semantic errors in naming and lend confidence to the conclusion that chronic lesions in Wernicke's area are not causally implicated in semantic error production.

Inappropriate usage of the Brunner-Munzel test in recent voxel-based lesion-symptom mapping studies.

Voxel-based lesion-symptom mapping (VLSM) techniques have been important in elucidating structure-function relationships in the human brain. Rorden, Karnath, and Bonilha (2007) introduced the non-parametric Brunner-Munzel rank order test as an alternative to parametric tests often used in VLSM analyses. However, the Brunner-Munzel statistic produces inflated z scores when used at any voxel where there are less than 10 subjects in either the lesion or no lesion groups. Unfortunately, a number of recently published VLSM studies using this statistic include relatively small patient populations, such that most (if not all) examined voxels do not meet the necessary criteria. We demonstrate the effects of inappropriate usage of the Brunner-Munzel test using a dataset included with MRIcron, and find large Type I errors. To correct for this we suggest that researchers use a permutation derived correction as implemented in current versions of MRIcron when using the Brunner-Munzel test.

Acute functional recovery of cerebral blood flow after forebrain ischemia in rat.

After complete cerebral ischemia, the postischemic blood flow response to functional activation is severely attenuated for several hours. However, little is known about the spatial and temporal extent of the blood flow response in the acute postischemic period after incomplete cerebral ischemia. To investigate the relative cerebral blood flow (rCBF) response in the somatosensory cortex of rat to controlled vibrissae stimulation after transient incomplete ischemia (15-min bilateral common carotid artery occlusion+hypotension), we employed laser speckle imaging combined with statistical parametric mapping. We found that the ischemic insult had a significant impact on the baseline blood flow (P<0.005) and the activation area in response to functional stimulation was significantly reduced after ischemia (P<0.005). The maximum rCBF response in the activation area determined from the statistical analysis did not change significantly up to 3 h after ischemia (P>0.1). However, the time when rCBF response reached its maximum was significantly delayed (P<0.0001) from 2.4+/-0.2 secs before ischemia to 3.6+/-0.1 secs at 20 mins into reperfusion (P<0.001); the delay was reduced gradually to 2.9+/-0.2 secs after 3 h, which was still significantly greater than that observed before the insult (P=0.04).

Power in Voxel-based lesion-symptom mapping.

Lesion analysis in brain-injured populations complements what can be learned from functional neuroimaging. Voxel-based approaches to mapping lesion-behavior correlations in brain-injured populations are increasingly popular, and have the potential to leverage image analysis methods drawn from functional magnetic resonance imaging. However, power is a major concern for these studies, and is likely to vary regionally due to the distribution of lesion locations. Here, we outline general considerations for voxel-based methods, characterize the use of a nonparametric permutation test adapted from functional neuroimaging, and present methods for regional power analysis in lesion studies.

Neural correlates of cognitive efficiency.

Since its inception, experimental psychology has sought to account for individual differences in human performance. Some neuroimaging research, involving complex behavioral paradigms, has suggested that faster-performing individuals show greater neural activity than slower performers. Other research has suggested that faster-performing individuals show less neural activity than slower performers. To examine the neural basis of individual performance differences, we had participants perform a simple speeded-processing task during fMRI scanning. In some prefrontal cortical (PFC) brain regions, faster performers showed less cortical activity than slower performers while in other PFC and parietal regions they showed greater activity. Regional-causality analysis indicated that PFC exerted more influence over other brain regions for slower than for faster individuals. These results suggest that a critical determinant of individual performance differences is the efficiency of interactions between brain regions and that slower individuals may require more prefrontal executive control than faster individuals to perform successfully.

Experimental design for functional MRI of scene memory encoding.

The use of functional imaging to identify encoding-related areas in the medial temporal lobe has previously been explored for presurgical evaluation in patients with temporal lobe epilepsy. Optimizing sensitivity in such paradigms is critical for the reliable detection of regions most closely engaged in memory encoding. A variety of experimental designs have been used to detect encoding-related activity, including blocked, sparse event-related, and rapid event-related designs. Although blocked designs are generally more sensitive than event-related designs, design and analysis advantages could potentially overcome this difference. In the present study, we directly contrast different experimental designs in terms of the intensity, extent, and lateralization of activation detected in healthy subjects. Our results suggest that although improved design augments the sensitivity of event-related designs, these benefits are not sufficient to overcome the sensitivity advantages of traditional blocked designs.

Empirical analyses of null-hypothesis perfusion FMRI data at 1.5 and 4 T.

Functional magnetic resonance imaging (fMRI) based on arterial spin labeling (ASL) perfusion contrast is an emergent methodology for visualizing brain function both at rest and during task performance. Because of the typical pairwise subtraction approach in generating perfusion images, ASL contrast manifests different noise properties and offers potential advantages for some experimental designs as compared with blood oxygenation-level-dependent (BOLD) contrast. We studied the noise properties and statistical power of ASL contrast, with a focus on temporal autocorrelation and spatial coherence, at both 1.5- and 4.0-T field strengths. Perfusion fMRI time series were found to be roughly independent in time, and voxelwise statistical analysis assuming independence of observations yielded false-positive rates compatible with theoretical values using appropriate analysis methods. Unlike BOLD fMRI data, perfusion data were not found to have spatial coherence that varied across temporal frequency. This finding has implications for the application of spatial smoothing to perfusion data. It was also found that the spatial coherence of the ASL data is greater at high magnetic field than low field, and including the global signal as a covariate in the general linear model improves the central tendency of test statistic as well as reduces the noise level in perfusion fMRI, especially at high magnetic field.

Arterial spin labeling perfusion fMRI with very low task frequency.

Functional magnetic resonance imaging (fMRI) has become the most widely used modality for visualizing regional brain activation in response to sensorimotor or cognitive tasks. While the majority of fMRI studies have used blood oxygenation level-dependent (BOLD) contrast as a marker for neural activation, baseline drift effects result in poor sensitivity for detecting slow variations in neural activity. By contrast, drift effects are minimized in arterial spin labeling (ASL) perfusion contrast, primarily as a result of successive pairwise subtraction between images acquired with and without labeling. Recent data suggest that ASL contrast shows stable noise characteristics over the entire frequency spectrum, which makes it suitable for studying low-frequency events in brain function. The present study investigates the relative sensitivities of ASL and BOLD contrast in detecting changes in motor cortex activation over a spectrum of frequencies of experimental design, where the alternating period between the resting state and activation is varied from 30 s up to 24 hr. The results demonstrate that 1) ASL contrast can detect differences in motor cortex activation over periods of minutes, hours, and even days; 2) the functional sensitivity of ASL contrast becomes superior to that of BOLD contrast when the alternating period between the resting state and activation is greater than a few minutes; and 3) task activation measured by ASL tends to have less intersubject variability than BOLD contrast. The improved sensitivity of the ASL contrast for low task frequency and longitudinal studies, along with its superior power in group analysis, is expected to extend the range of experimental designs that can be studied using fMRI.