Date of Degree


Document Type


Degree Name





Hernan A. Makse

Committee Members

Marilyn Gunner

Azita Mayeli

Maria F. Chan

Hyungsik Lim

Hernan A. Makse

Subject Categories

Neurosciences | Statistical, Nonlinear, and Soft Matter Physics


fMRI, Complex network modeling, Statistical inference, Brain language network, Brain tumor, Bilingual


We study the underlying mechanism by which language processing occurs in the human brain using inference methods on functional magnetic resonance imaging data. The data analyzed stems from several cohorts of subjects; a monolingual group, a bilingual group, a healthy control group and one diseased case. We applied a complex statistical inference pipeline to determine the network structure of brain components involved with language. This healthy network reveals a fully connected triangular relationship between the pre-Supplementary Motor Area (pre-SMA), the Broca's Area (BA), and the ventral Pre-Motor Area (PreMA) in the left hemisphere. This "triangle'' shows consistently in all the healthy subjects (100%) we analyzed regardless of their mono- or multi-lingual status. In addition, we found that Wernicke's Area (WA) on the left hemisphere connects with BA and PreMA to form a "V'' shape connectivity across 75% of the monolinguals, 50% of the bilinguals speaking a second language and 100% of the bilinguals speaking their native language. By comparing the quantified link weights, we found that the strongest link is between BA and PreMA, followed by pre-SMA and PreMA, and then pre-SMA and BA. This is consistent for all healthy subjects (p < 0.05). Furthermore, we conducted a k-core analysis testing the resiliency of subnetworks in the three groups. Our results show that nodes in the three triangle areas belong mostly to the maximum shell, whereas WA populates mostly in the lower shell, consistently across the data.

In a separate study, we describe frontal language reorganization in a 57-year-old right-handed patient with a low-grade left frontotemporal insular glioma. Pre-operative fMRI revealed robust activation in left WA and in the right BA. Intra-operative cortical stimulation of the left inferior frontal gyrus and adjacent cortices elicited no speech deficits, and gross total resection including the expected location of BA resulted in no speech impairment. Our network model found that the right homologue of the BA in this patient functionally connected to the same areas as the left BA in a typical healthy control. As opposed to the functional connection of the left BA in a healthy brain, the right BA did not connect directly with the left WA, but connected indirectly, mediated by the pre-SMA and preMA. In addition, the trans-located BA and WA moves from the lower k shell to the maximum shell during the recovery of the surgery. This case illustrates that pre-surgical fMRI can be used to identify atypical hemispheric language reorganization in the presence of a brain tumor and that network theory can help understand the underlying structure behind functional reorganization.