Dissociating neuro-cognitive component processes: voxel-based correlational methodology

https://doi.org/10.1016/j.neuropsychologia.2004.07.020Get rights and content

Abstract

Relating behavioural deficits to lesion site has long been an important tool for localising the brain bases of cognitive function. Voxel-based methods, based on statistical analyses of structural brain images, allow a major step forward in the effectiveness of this approach. These methods provide a fine-grained assessment of damaged tissue by assigning a continuous value to each voxel over the entire brain. This information, correlated with continuous behavioural data reflecting specific aspects of cognition, offers new opportunities for identifying the neural organisation underlying cognitive function. The research reported here demonstrates the ability of this correlational methodology to differentiate between the neuro-cognitive components involved in word recognition and lexical decision, providing an important new tool for directly linking brain areas to specific aspects of psychological performance.

Introduction

The lesion-deficit approach, whereby a patient's focal brain lesion is related to their behavioural deficit, has been important in developing accounts of the brain bases of cognition (Dronkers, 1996; Dronkers & Lundy, 1998; Damasio, Grabowski, Hichwa, & Damasio, 1996; Tranel, Damasio, & Damasio, 1997). Until the advent of functional imaging, this was one of the methods for determining the functional significance of specific brain regions and for investigating how cognitive processes are instantiated in cortical tissue. Although it remains widely used (Adolphs, Damasio, Tranel, Cooper, & Damasio, 2000; Bates et al., 2003) the standard implementations of this approach have significant limitations.

One important issue concerns the methods used to assess cortical damage, which typically assume that there is a binary distinction between intact and damaged tissue. Across lesion-deficit studies a variety of different methods are used to determine the extent of the lesion, involving either manual tracing of the lesion or more automated methods (Bates et al., 2003; Damasio & Damasio, 1989; Lawrie & Abukmeil, 1998; Mega, Thompson, Toga, & Cummings, 2000). Irrespective of the specific technique used to assess the lesion, the purpose of the procedure is the same—to judge whether cortical tissue is damaged or intact. This all-or-none distinction fails to capture a much larger range of potentially informative gradations in degree of structural damage. Another limitation is that the analysis is typically confined to specific brain regions, with ROIs driven by lesion locations, rather than covering the entire brain. This makes it hard to detect properties of the overall functional network (Naeser, Gaddie, Palumbo, & Stiassny-Eder, 1990; Willmes & Poeck, 1993). Additionally, methodologies that depend on observer input for the delineation of a lesion are susceptible to problems of subjectivity and reproducibility.

Voxel-based methodologies, which involve voxel-by-voxel analyses of the entire brain, have the potential to avoid these problems. Whole brain analyses escape the limitations of analyses which are limited to specific brain regions and form the basis for a more comprehensive understanding of the neural system involved in a cognitive task. In addition, voxel-based methods can provide a more sensitive estimate of intact and damaged tissue by assigning a continuous signal intensity value to each voxel.

Another significant limitation of most lesion-deficit studies is the way in which a patient's behavioural deficit is assessed. Typically, this also involves making a binary decision as to whether or not a patient has an impairment on the basis of a set of behavioural scores (Adolphs et al., 2000), which provides a very coarse-grained estimate of the degree to which a patient's performance is impaired. A more accurate assessment of the relationship between damage and impaired performance is possible if we adopt a methodology in which both cortical damage and behavioural deficit can be assessed in a more graded manner. This can be achieved by correlating the continuous intensity value for each voxel with suitably continuous behavioural data, thereby increasing both the sensitivity and the statistical power of the analyses. Furthermore, this allows the use of more dynamic measures of cognitive function, such as reaction time performance in cognitive tasks, which allows the structural data to be related directly to variations in functional capacities. This in principle is capable of delivering a much more differentiated picture of which brain regions are critical to performance of which aspects of a cognitive task.

The research reported here constitutes a first trial of this new combination of continuous whole-brain voxel-based measures of structural integrity with continuous measures of dynamic cognitive performance. What the results reveal is a striking ability to differentiate between neuro-cognitive component processes, providing much more specific structure–function correlations than existing methods. To this end, we carried out a study which correlated the signal intensity of each voxel across the brains of 19 brain-damaged patients with their performance on a speeded lexical decision task.

We chose the lexical decision task because its component cognitive processes have been well studied and it is a task which has been used extensively to study both normal and abnormal word processing (Franklin, Howard, & Patterson, 1994; Jacobs, Graf, & Kinder, 2003; Marslen-Wilson, 1980; Rubenstein, Garfield, & Millikan, 1970). From a cognitive perspective, it is well-established that the lexical decision task consists of at least two different types of process (Carr & Pollatsek, 1985; Jacobs & Grainger, 1994): (a) the processing of the perceptual input and its rapid mapping onto lexical representations, thereby providing the basis for discriminating between words and non-words. (b) A set of decision processes that operate on the results of these mapping processes, and which require the listener to set a criterion for accepting a stimulus as a word or as a non-word. Depending on where listeners set their threshold, this will determine the proportion of hits (correctly identifying a word) and false alarms (incorrectly accepting a non-word as a word). These decision processes are usually assessed within the context of signal detection theory (Luce, 1986, Swets, 1996), which takes into account the relative rate of hits and false alarms, computing d′ as a measure of sensitivity [d′ = z[FA] − z[HIT]] and β as a measure of criterion-setting or response bias (β = [(1 − HIT)(1 − FA)/(HIT × FA)]1/2].

The lexical decision task has been used extensively with brain-damaged patients to determine the extent to which their word processing abilities are intact (Franklin et al., 1994; Milberg & Blumstein, 1981; Milberg, Blumstein, Giovanello, & Misiurski, 2003). However, no attempt has been made to relate the different components of the lexical decision task to specific sites of damage. Similarly, a number of functional imaging studies have used the lexical decision task as a means for exploring the neural basis of word processing (Devlin et al., 2002, Specht et al., 2003). These studies reveal that an extensive, primarily left lateralised, network of fronto-temporal brain regions is activated in this task. However, these studies were not designed to detect possible differences in the neural substrates for the different cognitive processes involved in word recognition and lexical decision. The question addressed by the present study is whether the correlational method provides a methodology for dissociating between different cognitive processes. To address this question we correlated signal intensity across the brains of a variety of brain-damaged patients with a variety of continuous behavioural measures derived from their performance on an auditory lexical decision task. We expected variations in performance to correlate with variations in signal intensity in areas of the brain important for the performance of the task, with different patterns potentially activated by the different components of the lexical decision task.

Section snippets

Subjects

We selected 19 brain-damaged patients who had lesions in a variety of cortical sites due to various aetiologies (e.g. stroke, tumour excision, herpes encephalitis, semantic dementia) for this study. Patients were not selected on the basis of either their lesion location or their pattern of behavioural scores; the sole criteria for inclusion in the study were whether the patient: (a) had a T1 weighted 3D MRI scan, and (b) could perform the lexical decision task. All patients had scans and were

Results

We report cluster level P-values corrected for multiple comparisons with peak voxel z scores (P, z). The first set of analyses looked at the primary behavioural data generated by the tasks—the average speed and accuracy of the patients’ lexical decision responses. Reaction-times (RT) to real words correlated negatively (P = 0.0001; z = 3.68) with signal intensity in a large area of the left anterior temporal cortex including the temporal pole, and the superior, inferior and middle temporal

Discussion

The results presented here show that different cortical regions are associated with different cognitive components of the lexical decision task. The mapping of sound onto meaning, as reflected in speed of response, is associated with left anterior temporal regions, whereas accuracy of responding, as reflected in d′ and β, is associated with anterior prefrontal regions. Moreover, those regions that show significant correlations with different aspects of the lexical decision task overlap

Acknowledgements

This research was funded by an MRC programme grant to LKT.

References (50)

  • E.A. Stamatakis et al.

    Spatial normalization of lesioned HMPAO-SPECT images

    NeuroImage

    (2001)
  • D. Tranel et al.

    A neural basis for the retrieval of conceptual knowledge

    Neuropsychologia

    (1997)
  • A.D. Wagner

    Working memory contributions to human learning and remembering

    Neuron

    (1999)
  • R. Zahn et al.

    Hemispheric lateralisation at different levels of human auditory word processing: a functional magnetic resonance imaging study

    Neuroscience Letters

    (2000)
  • R. Adolphs et al.

    A role for somatosensory cortices in the visual recognition of emotion as revealed by three-dimensional lesion mapping

    Journal of Neuroscience

    (2000)
  • J. Ashburner et al.

    Identifying global anatomical differences: deformation-based morphometry

    Human Brain Mapping

    (1998)
  • J. Ashburner et al.

    Nonlinear spatial normalization using basis functions

    Human Brain Mapping

    (1999)
  • D.A. Balota et al.

    On the early influence of meaning in word recognition

  • E. Bates et al.

    Voxel-based lesion-symptom mapping

    Nature Neuroscience

    (2003)
  • J.R. Binder et al.

    Human temporal lobe activation by speech and nonspeech sounds

    Cerebral Cortex

    (2000)
  • T.H. Carr et al.

    Recognising printed words: a look at current models

    (1985)
  • M. Coltheart

    The MRC Psycholinguistic database

    Quarterly Journal of Experimental Psychology

    (1981)
  • H. Damasio et al.

    Lesion analysis in neuropsychology

    (1989)
  • H. Damasio et al.

    A neural basis for lexical retrieval

    Nature

    (1996)
  • N. Dronkers

    A new brain region for coordinating speech articulation

    Nature

    (1996)
  • Cited by (94)

    • Mapping psycholinguistic features to the neuropsychological and lesion profiles in aphasia

      2020, Cortex
      Citation Excerpt :

      This information was then used to generate a lesion overlap map for the participants in the current study (Fig. 1). To address the third research aim of identifying the neural correlates that support participant's reliance on psycholinguistic properties and their underlying psycholinguistic factors, a voxel-wise whole brain lesion-symptom mapping was carried out using Voxel-Based Correlational Methodology (VBCM: Tyler, Marslen-Wilson, & Stamatakis, 2005) conducted in SPM8 and run under Matlab (2012a). Two VBCM analyses were performed.

    View all citing articles on Scopus
    View full text