Representational Similarity Analysis For Mapping Language Neuroplasticity

A new method developed by Rice University psychologists for analyzing brain activity mapped how the brain changes after a stroke and helped researchers understand the brain’s reorganization.

A well-articulated network of cortical regions associated with single word reading has emerged over the past several decades, with the left ventral occipitotemporal cortex and the left angular gyrus identified as two critical nodes in the reading network. During the acute stage of stroke, damage to these regions is associated with severe impairments in the ability to read. These severe impairments often resolve during the transition from the acute to chronic stroke, with many individuals shown to at least partially recover over the years following damage.

These improvements in reading during the natural recovery from stroke have been argued to provide evidence for neural plasticity, with the damaged brain reorganizing to better support the impaired reading functions.

Neural plasticity could mean a variety of things: from functional take-over whereby the function previously performed by a damaged area shifts to a different brain region to compensatory masquerade, or a refinement of established but intact cognitive processes to perform a task. Within the context of reading literature, both of these hypotheses have been proposed to account for the acute to chronic improvements following stroke.

Functional Take-over Hypothesis

According to the functional take-over hypothesis there is a region in the patient’s brain whose associated function is different than the corresponding region in the undamaged population. The function of that region in the patient more closely matches the function of the damaged region in the undamaged population. As with much of the language recovery literature, there is debate over whether the region that takes over the function is contralesional or perilesional.

That is, some have argued that recovery of reading is associated with a retuning of the neural response of the homologous right hemisphere regions, such that this region now computes the function normally associated with the damaged tissue in the left hemisphere. Others have argued that the retuning occurs in the tissue just surrounding the lesion.

Support for the functional take-over hypothesis largely comes from fMRI studies of reading in the damaged brain. For example, in unimpaired readers, it is typical for a region of the left ventral occipitotemporal, frequently referred to as the visual word form area (VWFA), to be more activated to words than baseline, with the region’s response being case, font, and location invariant.

This pattern suggests that the region is involved in processing orthographic information about written words, that is, abstract information about the letter identities in the word and their order. When undergoing task-related fMRI, patients with damage to the VWFA typically show greater activation to words compared to baseline in the right hemisphere homologue of this region and/or in regions just adjacent to the lesion.

On the surface, the fact that patients show increased activation to words in regions not typically observed in the unimpaired population suggests that a reorganization of cognitive functions has occurred. Specifically, the orthographic function of the damaged region is hypothesized to be reorganized into other regions that do not typically carry out that function.

Alternative Explanations

There are alternative explanations for this reading recovery that do not require assuming functional take-over. Others have argued that the residual reading ability following damage to these regions is due to the refinement of alternative neural pathways for word reading that exist even in the undamaged brain, or a type of compensatory masquerade.

For example, these patients have been argued to rely on the right hemisphere’s normal capacity for visual word processing or alternatively on left hemisphere reading pathways that do not involve the damaged regions.

In these accounts, recovery over time results from participants learning to more efficiently use these alternative pathways, rather than a dynamic change in the neural organization of the reading system. According to the compensatory masquerade hypothesis, even after reading has partially recovered, the functions in the undamaged regions of the patient’s brain are the same as the functions in the corresponding regions in the unimpaired population.

Changes in the location of word versus baseline activation in patient’s brains may also be consistent with the compensatory masquerade hypothesis. Increases in activation in the patient could occur even without functional reorganization; instead they may reflect that the patient is engaging neural regions whose cognitive functions have not been altered by brain damage but that are being used in an atypical way for the task of reading.

For example, the right hemisphere activation may reflect that the word is being processed visually, but not orthographically. The same right hemisphere region may also process visual, but not orthographic, information about written stimuli in the unimpaired brain. Impaired readers show greater activation in that region compared to baseline than the control participants because this visual processing is not well suited for word reading.

Without the orthographic processes in the lvOT, impaired readers rely more heavily on these visual processes for word recognition, leading to more activation in the region. Alternatively, the right hemisphere activation may reflect engagement of cognitive processes that are not typically involved in reading for unimpaired readers but become part of the reading process after damage, processes like cognitive control, working memory, or response selection.

Compensatory Masquerade Hypothesis

Support for this compensatory masquerade interpretation of changes in activation profile comes from several longitudinal studies of individuals with VWFA damage. Early in the course of recovery, patients show an increase in the activation of the right VWFA activation to words relative to baseline. As recovery continues and reading improves, less right VWFA activation is observed and greater perilesional activation is observed. If, over time, the right VWFA takes on the functional properties of the damaged left VWFA, the opposite pattern would be expected, with more activity in the right VWFA as reading recovers.

This problem of interpreting what the meaning of changes in activation can tell us about the reorganization of cognitive functions is a larger problem in language recovery research. Similar fMRI results have been reported across different types of language impairments, with patients showing both greater perilesional and greater contralesional activation than controls. However, as with reading, early stages of language recovery are more linked to contralesional activation, while later stages of recovery are associated with perilesional activation.

Further challenging the idea that contralesional activation reflects functional reorganization is the finding that transcranial magnetic stimulation to contralesional areas in aphasic patients has surprisingly been shown to improve language production. This finding has been interpreted to indicate that contralesional activation may reflect the engagement of a dysfunctional process that inhibits the ability to do the task.

However, when a second stroke damages contralesional regions, whatever language has recovered is severely impacted, in both language production (e.g., [27–29]) and reading, suggesting that the contralesional region had been supporting the residual language capacity following the left hemisphere stroke.

With all of these difficulties in interpreting changes in activation, it is possible that traditional, univariate activation-based approaches to fMRI are not well suited to address issues of the reorganization of function following stroke. An additional concern is that it is not clear that the functional take-over necessarily predicts changes in activation between the patient and control populations.

For example, both patient and controls may rely on the same brain regions during reading, but the function of the region is different between the two populations. Alternative methods for analyzing fMRI may be better at distinguishing the functional take-over and compensatory masquerade hypotheses.

The functional take-over and compensatory masquerade hypotheses clearly make testable predictions, given the appropriate analysis methods. According to the compensatory masquerade, contralesional and perilesional regions in the patient’s brain should be doing the same cognitive function as the corresponding regions in the control group.

According to the functional take-over hypothesis, the function of those regions in the patient’s brain should be different than the corresponding regions in the control group and more similar to the normal functions of the damaged region. To address this prediction, it is necessary to identify fMRI methods that can decode what function is being processed by a region, rather than just finding differences in activation level. Here, we use a multivariate approach to analyzing fMRI data, specifically representational similarity analysis.

Representational Similarity Analysis

Simon Fischer-Baum, an assistant professor of psychology, said that rather than simply detecting whether a region of the brain is active in a particular task, the technique his team developed evaluates the function of a region by analyzing the pattern of activation.

In the study, Fischer-Baum had one patient with brain damage and 20 control participants read words while undergoing functional magnetic resonance imaging (fMRI), a process that measures brain activity by detecting changes associated with blood flow. This technique is based on the observation that cerebral blood flow and neuronal activation are coupled.

“The word ‘TOUGH’ looks different than the word ‘dough’ because ‘TOUGH’ is capitalized, although it is spelled very similarly,” he said. “It is also pronounced differently and has a different meaning than ‘dough.’ In contrast, ‘DOUGH’ and ‘sew’ are pronounced similarly but have different meanings. And ‘DOUGH’ and ‘bread’ are spelled very differently but have similar meanings.”

By comparing how different regions of the brain responded to individual words, Fischer-Baum said they were able to discover which regions of the brain were involved with reading, pronunciation and comprehension and, by comparing the function of different regions in brain-damaged and control participants, they could determine if functional takeover or compensatory masquerade had taken place.

“Our logic was simple,” Fischer-Baum said. “If neuroplasticity results in compensatory masquerade, then we expected to find the same function in the same brain regions for the patient and the control group, but we did not. In the patient with a brain injury, orthographic processing (the knowledge of the spellings of words) was being carried out by a region on the right side of the brain, but for control participants, orthographic processing was being done by a region on the left side, a region of the brain that was damaged in the patient. This indicates that functional takeover took place.”

Fischer-Baum said the research is an important contribution to the field of neuroplasticity, not just because it will aid in patient recovery through the design of better treatments and tools, but because it is important for scientists and physicians to understand what neural changes mean in terms of cognitive functions.

Simon Fischer-Baum, Ava Jang, and David Kajander
The Cognitive Neuroplasticity of Reading Recovery following Chronic Stroke: A Representational Similarity Analysis Approach
Neural Plasticity, vol. 2017, Article ID 2761913, 16 pages, 2017. doi:10.1155/2017/2761913

Image: Rice University