**Function-Based Tractography of the Language Network Correlates with Aphasia in Patients with Language-Eloquent Glioblastoma**

**Haosu Zhang 1 , Severin Schramm <sup>1</sup> , Axel Schröder 1 , Claus Zimmer 2 , Bernhard Meyer 1 , Sandro M. Krieg 1,3 and Nico Sollmann 2,3, \***


Received: 5 June 2020; Accepted: 23 June 2020; Published: 1 July 2020

**Abstract:** To date, the structural characteristics that distinguish language-involved from non-involved cortical areas are largely unclear. Particularly in patients suffering from language-eloquent brain tumors, reliable mapping of the cortico-subcortical language network is of high clinical importance to prepare and guide safe tumor resection. To investigate differences in structural characteristics between language-positive and language-negative areas, 20 patients (mean age: 63.2 ± 12.9 years, 16 males) diagnosed with language-eloquent left-hemispheric glioblastoma multiforme (GBM) underwent preoperative language mapping by navigated transcranial magnetic stimulation (nTMS) and nTMS-based diffusion tensor imaging fiber tracking (DTI FT). The number of language-positive and language-negative points as well as the gray matter intensity (GMI), normalized volumes of U-fibers, interhemispheric fibers, and fibers projecting to the cerebellum were assessed and compared between language-positive and language-negative nTMS mappings and set in correlation with aphasia grades. We found significantly lower GMI for language-positive nTMS points (5.7 ± 1.7 versus 7.1 <sup>±</sup> 1.6, *p* = 0.0121). Furthermore, language-positive nTMS points were characterized by an enhanced connectivity profile, i.e., these points showed a significantly higher ratio in volumes for U-fibers (*<sup>p</sup>* <sup>≤</sup> 0.0056), interhemispheric fibers (*<sup>p</sup>* <sup>=</sup> 0.0494), and fibers projecting to the cerebellum (*p* = 0.0094). The number of language-positive nTMS points (R <sup>≥</sup> 0.4854, *p* <sup>≤</sup> 0.0300) as well as the ratio in volumes for U-fibers (R ≤ −0.4899, *p* ≤ 0.0283) were significantly associated with aphasia grades, as assessed pre- or postoperatively and during follow-up examinations. In conclusion, this study provides evidence for structural differences on cortical and subcortical levels between language-positive and language-negative areas, as detected by nTMS language mapping. The results may further increase confidence in the technique of nTMS language mapping and nTMS-based tractography in the direct clinical setting. Future studies may confirm our results in larger cohorts and may expand the findings to patients with other tumor entities than GBM.

**Keywords:** brain stimulation; fiber tractography; glioblastoma multiforme; gray matter; language mapping; navigated transcranial magnetic stimulation

### **1. Introduction**

Resection of intracranial glioma aims at a maximum extent of resection, which should ideally be achieved without causing surgery-related functional deficits that could severely reduce the patients' quality of life [1–4]. To establish maximum resection whilst avoiding functional decline as far as possible, several pre- and intraoperative techniques have been developed to assist in neurosurgical planning and resection guidance [4–7]. For the intraoperative setting, cortical and subcortical direct electrical stimulation (DES) serves as the current gold-standard method [8–10]. Regarding the preoperative setting, navigated transcranial magnetic stimulation (nTMS) has found its way into neurosurgery over the last decade [6,11,12].

The technique of nTMS has lately been used to conduct language mappings in patients suffering from language-eloquent glioma or other entities of brain tumors [13–16]. Furthermore, it has been combined with diffusion tensor imaging (DTI) derived from preoperative magnetic resonance imaging (MRI) to provide spatially resolved maps that visualize language-related structures [17–25]. Integration into clinical routine and the perioperative workflow is seamless, and the approach is currently regarded as a valuable adjunct to intraoperative DES [14,15]. However, in direct comparison to intraoperative DES, nTMS language mapping has shown a rather low specificity of 23.8% and a positive predictive value of 35.6% [13]. The mere additional use of nTMS-based DTI fiber tracking (DTI FT) did not improve the identification of DES-positive language areas during awake surgery [24]. Nevertheless, the visualization of the subcortical language network becomes possible purely based on functional data by using nTMS-based DTI FT, which has shown high potential for surgical planning, resection guidance, and risk assessment in patients with language-eloquent lesions [17–25]. However, explorations of the differences between language-positive and language-negative nTMS mappings are still largely missing, which is one reason contributing to the lack of understanding of the comparatively low specificity of nTMS language mapping in relation to intraoperative DES. Further insights may lead to a better definition of the role of nTMS in the neurosurgical setting and to improved understanding of nTMS characteristics.

The cortico-subcortical network behind human language function is complex [26–29]. On the cortical level, differences in gray matter (GM) distribution could probably differentiate between language-related and non-related areas or, at least, between highly and less involved areas. In this regard, previous research has demonstrated that the degree to which language is lateralized to one of the hemispheres is positively predicted by the degree to which GM is lateralized on a voxel-by-voxel basis [30]. In subjects with dyslexia, a GM deficit involving a fronto-temporal network important for phonological processing was revealed, whereas region-specific increases in GM volume are possible for developmental language disorders as a result of compensatory mechanisms [31,32]. On the subcortical level, differences in connectivity profiles between areas depending on language involvement seem likely. Besides major language-related white matter tracts known to be involved in language processing, various short or long interconnecting fibers, such as short association fibers (commonly referred to as arcuate fibers or U-fibers), interhemispheric transcallosal fibers, and fibers projecting to the cerebellum play a role [28,29,33,34].

Against this background, this study's objective is to systematically explore characteristics of language-positive structures according to nTMS language mapping and nTMS-based DTI FT in direct comparison to language-negative counterparts. We hypothesize that (1) language-positive cortical areas may show different focal GM intensity (GMI, as a potential expression of higher GM density as a correlate of increased functional involvement), and that (2) language-positive cortical areas show a different connectivity profile (higher volumes of U-fibers, transcallosal fibers, and fibers projecting to the cerebellum) when compared to language-negative cortical areas.
