Lower Spatial Consistency on Transient States of People with Autism Spectrum Disorder

357 

Abstract Submission 

Yunge Zhang¹, Vinoo Alluri², Dongyue Zhou¹, Lin Lin¹, Abigail Stein³, Shuqin Zhou³, Huashuai Xu⁴, Wei Zhao¹, Fengyu Cong¹, Huanjie Li¹, Fei Du³

¹School of Biomedical Engineering, Faculty of Medicine, Dalian University of Technology, Dalian, China, ²Cognitive Science Lab, International Institute of Information Technology, Hyderabad, India, ³McLean Imaging center, McLean Hospital, Harvard Medical School, Boston, USA, ⁴Faculty of Information Technology, University of Jyväskylä, Jyväskylä, Finland

Yunge Zhang  
School of Biomedical Engineering, Faculty of Medicine, Dalian University of Technology
Dalian, China

Vinoo Alluri  
Cognitive Science Lab, International Institute of Information Technology
Hyderabad, India

Dongyue Zhou  
School of Biomedical Engineering, Faculty of Medicine, Dalian University of Technology
Dalian, China

Lin Lin  
School of Biomedical Engineering, Faculty of Medicine, Dalian University of Technology
Dalian, China

Abigail Stein  
McLean Imaging center, McLean Hospital, Harvard Medical School
Boston, USA

Shuqin Zhou  
McLean Imaging center, McLean Hospital, Harvard Medical School
Boston, USA

Huashuai Xu  
Faculty of Information Technology, University of Jyväskylä
Jyväskylä, Finland

Wei Zhao  
School of Biomedical Engineering, Faculty of Medicine, Dalian University of Technology
Dalian, China

Fengyu Cong  
School of Biomedical Engineering, Faculty of Medicine, Dalian University of Technology
Dalian, China

Huanjie Li  
School of Biomedical Engineering, Faculty of Medicine, Dalian University of Technology
Dalian, China

Fei Du  
McLean Imaging center, McLean Hospital, Harvard Medical School
Boston, USA

Social deficit is a core symptom of autism spectrum disorder (ASD). Atypical brain function of people with ASD have been widely reported on the triple network model[1], which is related to social communication and contains default mode network (DMN), executive and control network (ECN) and salience network (SN). However, it's unclear what the transient states of these networks look like in people with ASD. Co-activation pattern (CAP) is a clustering-based method to study the transient network states (TNSs)[2]. Most CAP studies focused on temporal features, the spatial stability of TNSs hasn't been well studied. In this work, we used CAP to study the spatial stability among TNSs related to triple network modal and the differences between people with ASD and healthy controls.

Data: Resting state fMRI data of 266 male subjects from 7 sites of ABIDE II dataset[3](133 ASD, 133 healthy controls (CON)) were included in this work. All subjects had full IQ higher than 80. Within a site, the age and full IQ were matched for ASD and CON groups and each group had at least 10 subjects. There were 184 subjects (92 ASD, 92 CON) with records of Social Responsiveness Scale (SRS) T scores.
CAP: After normal preprocessing pipeline, a 400-node atlas[4] was used to extract time courses. The time courses of all subjects were concatenated by time and performed k-means clustering. Each frame was divided into one cluster and the cluster centers were defined as TNSs. The TNSs map were normalized by dividing the within-cluster standard deviation. After visual inspection, a threshold of 0.4 was defined as significant activation. The CAP was performed on all subjects and within each site, the TNSs of each site were matched with TNSs of all subjects based on spatial patterns.
Spatial stability: The spatial stability was evaluated from three aspects: 1) Multisite spatial similarity was compared among TNSs. 2) Distance to cluster center of each frame was compared among TNSs and between ASD and CON groups using ANOVA and post hoc T tests. 3) Individual level significant activation rate (iSAR) was compared between ASD and CON groups. For one subject, the iSAR of a parcel in a TNS was defined as the n/N, n was the number of significantly activated frames with a threshold of 0.4 and N was the total number of frames in this TNS. T test was performed on iSAR of each TNS to study the group differences and canonical correlation analysis (CCA) was performed with iSAR and SRS T scores to study the relationship between them. CCA converted iSAR and SRS T scores into canonical variate (CV) pairs. The correlation between first CV pair represented relationship between iSAR and social deficits. The correlation of each raw feature and its first CV represented its contribution.

We chose six as the optimized k number, which converted DMN, ECN and SN into three pairs of 'mirror' patterns. The TNSs were named according to the condition of the dominating network of it (Fig 1). The DMN TNSs showed highest spatial stability with higher multisite spatial similarity and shorter distance to center of each frame while the ECN TNSs showed lowest spatial stability. Comparing with CON, ASD group showed higher distance to center at every TNS representing the reduced spatial stability of them. Besides, ASD group showed significant lower iSAR on ECN-n and DMN-p (Fig 2A, B, E and F). The iSAR showed significant correlation with SRS T scores (Fig 2C and F). The reduced iSAR were related to severer social deficits since the iSAR values showed positive correlation while SRS T scores showed negative correlation with their first CV (Fig 2D and H).

We defined six TNSs which yielded DMN, ECN and SN into three pairs. The DMN TNS pair had highest spatial stability while ECN TNS pair had lowest spatial stability. Besides, people with ASD had lower spatial stability on every TNS and the reduced spatial stability of ECN-n and DMN-p was related to severer social deficits.

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

fMRI Connectivity and Network Modeling ²

Autism

FUNCTIONAL MRI

Other - Default Mode Network; Triple Network