Slide Slam

< Slide Slam Sessions

Slide Slam M14

Characterization of Heschl's Gyrus subtypes using Morphology-Encoding Graphs

Slide Slam Session M, Thursday, October 7, 2021, 6:00 - 8:00 am PDT Log In to set Timezone

Sevil Maghsadhagh1,2,3,4, Josué L. Dalboni da Rocha3, Jan Benner5, Peter Schneider6,7, Hamid Behjat8,9, Narly Golestani1,2,3,4; 1Brain and Language Lab, Cognitive Science Hub, University of Vienna., 2Faculty of Life Sciences, University of Vienna., 3Faculty of Psychology and Educational Sciences, University of Geneva., 4Faculty of Psychology, University of Vienna., 5University Hospital Heidelberg., 6Section of Biomagnetism, University Hospital Heidelberg., 7Centre for Systematic Musicology, Karl-Franzens University, Graz, Austria., 8Brigham and Women’s Hospital, Harvard Medical School., 9Lund University.

Heschl’s Gyrus (HG) is the first cortical structure to receive auditory input. The most common HG gyrification patterns include single gyri, Common Stem Duplication (CSDs) and Complete Posterior Duplications (CPDs). Previous work has shown both greater HG volume and gyrification in relation to musical (Benner et al., 2017) and phonetic expertise (Golestani, Price, & Scott, 2011), and also volume and gyrification differences in dyslexia (Altarelli et al., 2014; Serrallach et al., 2016). Conventional structural measures such as gray matter volume, surface area and cortical thickness are complementary to indices of HG shape. These conventional measures are mostly obtained using visual assessment and manual labeling (Benner et al., 2017), but the recently developed TASH toolbox (Dalboni da Rocha et al., 2020) allows fully automated labeling of HG from T1 structural MRI images, and extraction of such measures. We present a method for characterizing the morphology of HG using unweighted, undirected surface-based and volumetric graphs. HG labels were produced using TASH, and used to extract both conventional anatomical features but also spectral graph features. We validated how well spectral graph features discriminate between single HG versus CSDs (as determined visually by an experienced rater) compared to that possible by using conventional anatomical measures (i.e. volume, surface area and thickness) alone. We used data from 177 adults, including non-musicians, amateur and professional musicians. For the surface-based HG graph, vertices and edges of the HG labels on the white matter surface readily provided the desired graph representation. We also created 1mm3 and 0.6mm3 resolution volume-based graphs by transforming the labels to cortical volumes; voxels within cerebral cortex were treated as graph nodes, and edges were defined based on 26-neighborhood connectivity between adjacent voxels in 3D space. For each graph, we computed the graph’s normalized Laplacian matrix, and consequently, obtained the graph’s Laplacian spectrum by performing eigendecomposition on the matrix (for details, see (Maghsadhagh, Eklund, & Behjat, 2019)). The following spectral features were then extracted: a) an initial subset of the lower eigenvalues (excluding 0), b) the largest eigenvalue, c) distribution of eigenvalues across the spectrum, d) the area under the curve of the cumulative sum function of eigenvalues and e) normalized Laplacian energy of the graph. Discrimination of HG subtypes against visually determined categories (single vs CSD) was done using: 1) conventional anatomical measures, 2) spectral features, and 3) conventional measures together with spectral features. Results reveal substantially better discrimination performance using spectral features compared to when using conventional measures alone. The first ten eigenvalues were the most informative spectral graph feature among those that were tested. In future work, we anticipate that the proposed spectral features be found beneficial for systematic exploration of variations in HG typology in relation to cognition and disorder in the context of healthy language and music related variability, expertise and disorder. Furthermore, we aim to extend the graph definition to a weighed graph by using local measures such as cortical thickness, to further improve discriminative power of the proposed class of spectral shape features.

< Slide Slam Sessions

SNL Account Login

Forgot Password?
Create an Account