Multisite functional connectivity MRI classification of autism: ABIDE results

Area of Science:

• Neuroscience
• Medical Imaging
• Psychiatry

Background:

• Autism spectrum disorder (ASD) is associated with systematic differences in functional connectivity MRI (fcMRI) metrics, primarily decreased cortico-cortical connectivity.
• Previous studies using whole-brain fcMRI achieved ~80% accuracy in classifying high-functioning autism versus controls.
• This study aimed to replicate these findings using a large, multisite dataset from the Autism Brain Imaging Data Exchange (ABIDE).

Purpose of the Study:

• To replicate and evaluate the efficacy of single-subject classification of autism using whole-brain functional connectivity across multiple international sites.
• To assess the impact of multisite data on classification accuracy compared to single-site studies.
• To identify brain regions and connectivity patterns most informative for autism classification.

Main Methods:

• Utilized resting-state fMRI data from 964 subjects across 16 international sites within the ABIDE dataset.
• Preprocessed data included motion correction, spatial normalization, and signal regression. Calculated 26.4 million pairwise functional connectivity measurements.
• Employed a leave-one-out cross-validation classifier, incorporating age, gender, handedness, and site as covariates.

Main Results:

• Multisite classification accuracy significantly exceeded chance but was lower than previous single-site results, reaching up to 60% for whole-brain classification.
• The most accurate classifications involved connections within the default mode network, parahippocampal and fusiform gyri, insula, Wernicke Area, and intraparietal sulcus.
• Classifier performance correlated with autism symptom severity, social function, daily living skills, and verbal IQ. Longer BOLD imaging times per site improved accuracy.

Conclusions:

• Multisite functional connectivity classification for autism is feasible and outperforms chance but requires methodological advancements for clinical utility.
• Improved classification algorithms, extended BOLD imaging durations, and standardized acquisition protocols are crucial for enhancing multisite classification accuracy.
• Future research should focus on these areas to translate fcMRI-based classification into a reliable clinical tool for autism.