Dynamic Mode Analysis of Autism Spectrum Disorder

348 

Abstract Submission 

Mir Jeong¹, Youngjo Song¹, Jaeseung Jeong¹

¹KAIST, Daejeon, Korea, Republic of

Mir Jeong  
KAIST
Daejeon, Korea, Republic of

Youngjo Song  
KAIST
Daejeon, Korea, Republic of

Jaeseung Jeong  
KAIST
Daejeon, Korea, Republic of

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by challenges in social interaction and repetitive behaviors. Despite extensive research efforts, our understanding of the functional alterations in ASD brain at a network level remains incomplete. This study introduces a novel approach to analyzing the functional aspects of ASD brain by employing Dynamic Mode (DM) decomposition of fMRI time series, where the brain activity is decomposed into synchronously evolving networks (Casorso, et al., 2019). That is, we investigated the functional variances in ASD brain networks using the DM framework to gain a deeper understanding of the unique neurofunctional dynamics in individuals with ASD.

The study utilized resting-state fMRI data preprocessed with the C-PAC pipeline and, parcellated with Craddock 200 atlas from the Autism Brain Imaging Data Exchange I (ABIDE I), which includes data from 391 individuals diagnosed with ASD and 458 typically developing (TD) controls. The BOLD activity dynamics (measured by fMRI) are approximated as a linear dynamic system (x(t+1)=Ax(t)+ε(t)), where the linear operator (A), which is also referred to as connectivity matrix, describes the causal relation between successive time points (i.e., how the previous BOLD signal x(t) determines subsequent signal x(t+1)). (ε(t): noise at time t). For each individual, the linear operator (or connectivity matrix) was estimated from the BOLD signal, and DMs were identified via eigen-decomposition of this matrix. Given the temporally synchronized nature across all brain regions within each DM (Casorso, et al., 2019), we speculated that each DM is associated with a particular cognitive or operational aspect of brain functions. Thus, the primary functional involvement of each DM was decoded from its spatial characteristics using NiMARE software (Salo, et al., 2022). Then, we compared the temporal characteristics (decaying time, oscillation frequency) of each DM between the ASD and TD groups and examined their correlations with age and clinical measures. This approach enabled us to infer the distinct functional dynamics of ASD brains. For a comprehensive overview, see Figure 1.

Our finding highlighted significant differences in brain dynamics between ASD and TD individuals. Notably, we found faster decay in a pair of DMs associated with autobiographical memory in the ASD group, implicating diminished episodic memory performance in ASD. Additionally, we found slower oscillations in a pair of DMs linked to visual inhibition, suggesting the potential cause of the impaired inhibition in ASD. We also discovered that the damping time of the multisensory-related DM correlated with the difficulty of communication, pointing to the importance of sensory function in developing communication abilities in ASD; meanwhile, the oscillation frequency of the social-related DM correlated with overall ASD severity. Interestingly, the two aforementioned DMs, which showed differences between ASD and TD, did not exhibit a correlation with ASD severity, highlighting the complexity of the relationship between brain functions and symptoms (e.g., non-linear transition between ASD and TD). Moreover, the age-dependent changes in DMs differed between the ASD and TD groups, suggesting different developmental trajectories in attention and pain-related functions.

Our novel application of DM decomposition offers a distinctive way to understand functional differences in ASD brains using just resting-state data. Remarkably, the insights gained from DM analysis align with prior research on ASD brain functions (Crane, et al., 2008; Johnston, et al., 2011). The dynamic perspective offered by DM analysis not only deepens our understanding of the neural basis behind ASD's atypical behaviors but also presents unique opportunities for evaluating intervention and therapeutic strategies in ASD, leveraging the convenience of acquiring resting-state data.

Neurodevelopmental/ Early Life (eg. ADHD, autism) ¹

Other Methods

BOLD fMRI ²

Autism

FUNCTIONAL MRI

Memory

Open-Source Software

Other - Dynamic Mode