Analysis of naming errors during cortical stimulation mapping: Implications for models of language representation

Abstract

This study reports on the characteristics and distribution of naming errors of patients undergoing cortical stimulation mapping (CSM). During the procedure, electrical stimulation is used to induce temporary functional lesions and locate ‘essential’ language areas for preservation. Under stimulation, patients are shown slides of common objects and asked to name them. Cortical stimulation can lead to a variety of naming errors. In the present study, we aggregate errors across patients to examine the neuroanatomical correlates and linguistic characteristics of six common errors: semantic paraphasias, circumlocutions, phonological paraphasias, neologisms, performance errors, and no-response errors. Aiding analysis, we relied on a suite of web-based querying and imaging tools that enabled the summative mapping of normalized stimulation sites. Errors were visualized and analyzed by type and location. We provide descriptive statistics to characterize the commonality of errors across patients and location. The errors observed suggest a widely distributed and heterogeneous cortical network that gives rise to differential patterning of paraphasic errors. Data are discussed in relation to emerging models of language representation that honor distinctions between frontal, parietal, and posterior temporal dorsal implementation systems and ventral-temporal lexical semantic and phonological storage and assembly regions; the latter of which may participate both in language comprehension and production.

Introduction

Complementing the results of neuroimaging and lesion studies, cortical stimulation mapping (CSM) provides a rare and valuable opportunity for mapping language function to neuroanatomy (Ojemann, 1991). CSM is an invasive procedure used to identify the language, sensory, and motor cortices, so that these critical regions may be preserved during surgical resection. The technique was originally employed to identify the sensory-motor cortices in patients with medically intractable seizure disorders. Penfield and colleagues later extended the procedure to identify eloquent language cortex by implementing an object-naming task (Penfield & Roberts, 1959). Subsequent studies have demonstrated that visual naming under stimulation can successfully identify resection boundaries that preserve functionally important language areas (Haglund, Berger, Shamseldin, Lettich, & Ojemann, 1994; Ojemann, 1983), though some controversy remains (e.g., Seeck et al., 2006).

CSM is one of the few techniques that allow direct observation of language dysfunction at an extremely localized neural level. However, the technique is limited by its own unique constraints. As a surgical procedure, intraoperative mapping is limited in time, with most testing typically lasting 0.5–3 h (Gordon, Boatman, Hart, Miglioretti, & Lesser, 2001). Cortical stimulation sites are limited to the regions exposed by the craniotomy (Roux et al., 2004). Typically there is an unequal sampling of cortical sites within the exposed region. Moreover, patients have just come out of general anesthesia and may be lethargic and therefore not alert or in the most optimal state for testing (Gordon et al., 2001). In addition, neurologically compromised patients, especially those with an early-age epilepsy onset, may have undergone cortical reorganization (Thompson, 2005). Aside from these procedural constraints, some controversy remains over the extent to which electrical stimulation spreads to adjacent cortex (Ojemann, 1983). Some studies of non-human primates have reported greater than expected spread of current in visual cortex (Tolias et al., 2005), though other studies have reported a lack of activation in adjacent cortex (Haglund, Ojemann, & Blasdel, 1993).

Despite these challenges, cortical stimulation studies over the past three decades have provided valuable evidence of the cortical organization of language. Although lesion studies have identified classic perisylvian regions in language processing, CSM studies have shown that language cortex may involve extra-sylvian regions, though these are variable across patients. Although most language sites are distinctly defined by sharp boundaries, some regions have been reported to include a “fuzzy” transition zone where cortical stimulation induces only occasional naming disruption (Whitaker & Ojemann, 1977). CSM studies comparing early and adult lesion onset (Ojemann, Ojemann, Lettich, & Berger, 1989) and patients with fast growing tumors reveal a highly similar pattern of language organization, suggesting that this mosaic of language function spread across perisylvian regions is likely reflective of the general population and is not simply an artifact of language organization in a clinical epileptic population.

CSM studies have provided evidence that cortex can exhibit functionally specific vulnerabilities for language functions. Direct cortical stimulation has given rise to specific and differentiated errors associated with reading, verbal memory, semantic classes, and differential object naming in bilinguals (Ojemann, 2003). Evidence of selective disruption of verb versus noun naming has been reported (Corina et al., 2005), as well as dissociations of reading and object naming (Roux et al., 2004). An anterior to posterior distribution for auditory versus visual naming has also been reported (Hamberger, McClelland, McKhann, Williams, & Goodman, 2007). A number of CSM speech perception studies have identified discrete regions in the middle-posterior superior temporal gyrus that subserve phonological processing of consonants but not vowels (Boatman et al., 1997, Boatman et al., 1995).

In the present study, we add to this body of literature a characterization of the type and distribution of naming errors observed during CSM. Specifically, we examine the neuroanatomical correlates and linguistic characteristics of six common naming errors: semantic paraphasias, circumlocutions, phonological paraphasias, neologisms, performance errors, and no-response errors.