A novel approach for extracting functional brain networks involved in mesial temporal lobe epilepsy based on self organizing maps

Abstract

Purpose: We propose a novel data-driven approach to extract and present large-scale functional brain networks from functional magnetic resonance imaging (fMRI) data using spatiotemporal self-organizing maps (STSOM) clustering, accounting for the properties of the brain functional networks being spatially structured and interhemispherically symmetric. Also, a novel group-wise analysis is proposed based on restricted Frechet mean to identify group-level networks. The alteration of resulted networks in left and right mesial temporal lobe epilepsy (mTLE) is studied. Methods: Thirty-five unilateral mTLE patients (21 left-mTLE (LTLE) and 14 right-mTLE (RTLE)), were prospectively studied. Eleven healthy control (HC) subjects were also recruited. To determine the functional networks of the whole brain, we extracted individual and group-level networks using spatiotemporal self-organizing maps and the restricted Frechet mean method, respectively. We applied the resulted networks to specify within and between-network alteration in functional connectivity (FC) in the LTLE and RTLE patients compared to the control cohort. Results: We obtained seven networks namely default-mode (DMN), sensorimotor (SMN), visual (VSN), subcortical (SCN), frontoparietal (FPN), dorsal attention (DAN), and ventral attention (VAN) networks. Our results demonstrated increased functional connectivity in the FPN networks in the LTLE and the RTLE cohorts compared to HC. Increased FC has been observed between DMN, FPN, DAN, VAN, and VSN in the LTLE cohort and between the DMN and FPN networks in the RTLE cohort. Conclusion: The proposed method has obtained promising results within a range of SNR and properly overlapped with the well-known functional networks using the Hausdorff distance. The consistent alteration patterns in within-and between-network FC for LTLE and RTLE patient cohorts would reflect the reliability of identification of large-scale brain networks in patients with mTLE. Different pattern of alterations in LTLE and RTLE compare with HC groups my be usefull for laterality porpose. © 2022