Connectivity based Determination of Autism and Epilepsy Shared Neuroanatomical Substrates
Abstract number :
3.325
Submission category :
5. Neuro Imaging / 5B. Functional Imaging
Year :
2025
Submission ID :
996
Source :
www.aesnet.org
Presentation date :
12/8/2025 12:00:00 AM
Published date :
Authors :
Presenting Author: Sienna Williams, BS – Children's National Medical Center
Samantha Werner, MA – Children's National Hospital
Hua Xie, PhD – Children's National Hospital
Priyanka Venkata Sita Illapani, MSc – Children's National Hospital
Arini Bhargava, BS – Children's National Hospital
Sonya Leikin, BS – Children's National Hospital
Cemal Karakas, MD – Children's National Hospital
Ana Moreno Chaza, BS – Children's National Hospital
Amy Mistri, BS – Children's National Hospital
Kartik Reddy, MD – Emory University School of Medicine
Carla Ammons, PhD – Childrens Healthcare of Atlanta
Eswar Damaraju, MS – Children's Hospital of Atlanta
Donald Bearden, PhD, ABPP-CN – University of Tennessee Health Science Center and Le Bonheur Children's Hospital
Daniel Drane, PhD – Emory University School of Medicine
Madison Berl, PhD – Children's National Hospital
William Gaillard, MD – Children's National Hospital
Nunthasiri Wittayanakorn, MD – Children's National Hospital, George Washington University
Syeda Abeera Amir, BS – Childrens National Hospital
Chima Oluigbo, MD – Children's National Hospital, George Washington University
Syed Anwar, MS, PhD – Childrens National Hospital
Leigh Sepeta, PhD – Children's National Hospital
Rationale: Epilepsy and autism spectrum disorders (ASD) often co-occur (Tuchman et al. 2009; Tuchman et al., 2010) with approximately 30% overlap in both populations and an increasing incidence of prevalence into adulthood. Both conditions are considered disorders of large-scale underlying brain networks, involving disruptions in cortical-subcortical connectivity. We investigate the shared overlap using resting-state functional magnetic resonance imaging (rs-fMRI) analyses with a graph-based machine learning approach.
Methods: The Brain Network (BNT) model trained on the ABIDE data set was evaluated on a group of pediatric epileptic patients: eight with ASD and six demographically matched epilepsy patients without ASD (Kan et al., 2022). All patients underwent rs-fMRI at our clinical center. The preprocessing included quality control and region-of-interest (ROI)-wise functional connectivity estimation using the Craddock 200 atlas. Pearson connectivity and partial connectivity matrices were fed into the BNT, a graph-based Transformer model that learns pairwise attention-based connectivity patterns and applies orthonormal clustering to capture modular-level brain organization. Graph-level embeddings from each modality were used for classification (binary: 0 = epilepsy without ASD, 1 = epilepsy with ASD). were used to determine the importance of brain ROIs for correct diagnosis.
Results: When evaluated on our in-house epilepsy dataset, the model reached an accuracy of 64% accuracy, exceeding chance performance (57%). Feature analysis determined the top ten areas of importance including bilateral cerebellum, para-hippocampus/fusiform, temporal pole, piriform, inferior frontal gyrus, and right lateralized amygdala/insulahighlights the varying importance of regions using 200 ROIs.
Conclusions: The BNT model was able to differentiate between patients with ASD and without ASD in our cohort of epilepsy patients. Connectivity analysis further revealed the most important regions in differentiating ASD from epilepsy controls. The cerebellum was noted as one of the most important regions. Cerebellar dysfunction may play a crucial role in the etiology of ASD as previous literature has shown (Biswas et al. 2024; D’Mello & Stoodley, 2015). In addition, other identified regions include those involved in emotional processing (amygdala/insula), language and social information processing (temporal poles, fusiform gyrus), and memory/social memory (parahippocampus, amygdala, piriform). These regions could be involved in brain networks related to understanding and reacting to social situations, an important aspect for ASD identification. This innovative, mechanistic research will advance our understanding of potential shared neural substrates and connectivity features in ASD and epilepsy, which will be further established by adding more patients in future studies.
Funding: This work is supported by the AES Infrastructure Grant and the District of Columbia Intellectual and Developmental Disabilities Research Center Pilot Award.
Neuro Imaging