Abstracts

Rationale: Brain function and morphological brain measures such as cortical and subcortical volumes can provide relevant pathophysiological information regarding different brain disorders. Graph theory (GT) methods based on such neuroimaging modalities have helped in the characterization of global, regional, and topological brain properties making possible the understanding of the effects of such disorders. Furthermore, machine learning (ML) algorithms have been useful in identifying intrinsic within-group differences in TLE regarding cognition, which helped identify different levels or phenotypes of cognitive impairment within TLE patients. In this study, we performed ML analyses on GT metrics extracted from brain functional and morphological data from TLE patients in order to investigate how sensitive the combination of ML on GT measures is regarding intrinsic group subdivision in TLE.

Methods: We performed K-means clustering to explore if GT metrics like local efficiency, global efficiency, and modularity index were able to capture within-group subdivision on TLE. We extracted GT metrics from resting-state functional magnetic resonance imaging (RS-fMRI), and from cortical and subcortical volumes (morphological correlation matrix) from a group of 97 TLE patients (age: 39.9±11.9; 60 female) and a group of 36 healthy controls (age: 34.1±10.8; 16 female). We also calculated correlations between GT measures and cognitive metrics between controls and TLE patients, and proportion differences between the obtained clusters for the functional and morphological analyses in terms of FSIQ.

Results: Both ML analyses –functional and morphological GT metrics– rendered 2 sub-groups or clusters within the TLE group: one resembling controls (RS-fMRI=51; morphological=73) and one deviating from controls (RS-fMRI=46; morphological=24) (Figures 1 and 2). However, subjects in both sets of clusters were non-overlapping, suggesting that, for example, an abnormal functional GT metric in one subject does not necessarily mean abnormal morphological GT metric on the same subject. Furthermore, controls showed significant correlations between functional GT metrics and cognitive variables, while the same was true for morphological GT metrics in TLE, suggesting that GT metrics based on brain volumes are sensitive to cognitive status in TLE while GT metrics based on brain function are more sensitive regarding cognition in controls. Furthermore, there is a significant proportion difference for categorical FSIQ and the morphological clusters (lower on the cluster unlike controls; c²=10.69, p=0.014).

Conclusions: The use of ML on GT metrics based on functional and morphological data instead of more traditional uses of the data (i.e., seed based functional connectivity, cortical and subcortical volumes/thickness, etc.) made possible a more efficient investigation regarding TLE group composition. The used GT measures were able to capture the different levels of brain characterization (i.e., global, regional, topological).

Funding: Please list any funding that was received in support of this abstract.: This study was funded by NIH-NINDS R01-1NS111022.