JMIR Publications

ABSTRACT

Background:

Autism spectrum disorder (ASD) is a complex neurodevelopmental disorder with unknown etiology. Early diagnosis and intervention are the keys to improving outcomes for patients with ASD. Structural MRI (sMRI) has been widely used in clinic to facilitate the diagnosis of brain diseases such as brain tumors. However, sMRI is less frequently investigated in neurological and psychiatric disorders such as ASD due to subtle, if any, anatomical changes of the brain. In recent years, more and more evidence has suggested that ASD is associated with anatomical changes of the brain.

Objective:

The aim of this study was to investigate the possibility of identifying structural patterns in the ASD patients' brain as potential biomarkers in the diagnosis and evaluation of ASD in clinic.

Methods:

We developed a novel two-level histogram-based morphometry (HBM) classification framework in which an algorithm based on a 3D version of histogram of oriented gradients (HOG) was used to extract features from sMRI data. We applied this framework to distinguish ASD patients from healthy controls using four datasets from the second edition of the Autism Brain Imaging Data Exchange (ABIDE II) including sites ETH Zürich (ETH), NYU Langone Medical Center: Sample 1 (NYU), Oregon Health and Science University (OHSU), and Stanford University (SU). We used stratified 10-fold cross-validation method to evaluate the model performance, and optimized the parameters for 3D HOG and selected the best algorithms for each level of the HBM framework. We applied the Naive Bayes approach to identify the predictive ASD-related brain regions based on classification contributions of each HOG feature.

Results:

Based on the 3D HOG feature extraction method, our proposed HBM framework achieved >0.75 AUC on each dataset, with the best AUC of 0.849 on the ETH site. We compared the 3D HOG algorithm with the original 2D HOG algorithm and improved >4% AUC on each dataset, with the best improvement of 10% on the SU site. Comparison of the 3D HOG algorithm with the scale-invariant feature transform (SIFT) algorithm showed >14% AUC improvement on each dataset. Furthermore, we identified ASD-related brain regions based on the sMRI images. Some of these regions (e.g., frontal gyrus, temporal gyrus, ingulate gyrus, postcentral gyrus, precuneus, caudate and hippocampus) are known to be implicated in ASD in prior neuroimaging literatures. We also identified less well-known regions that may play unrecognized roles in ASD and be worth further investigation.

Conclusions:

Our research suggested it was possible to identify neuroimaging biomarkers that can distinguish ASD patients from healthy controls based on sMRI brain images. As a cost-effective and non-invasive tool for investigating brain structural changes, sMRI is also more amenable to populations for whom compliance is a challenge as it can be completed under sedation. Therefore, our tool could be useful in the diagnosis and evaluation of ASD in clinic. We also demonstrated the potentials of applying data-driven artificial intelligence technology in the clinical settings of neurological and psychiatric disorders that usually harbor in the brain subtle anatomical changes often invisible to human eyes.