Cortical Shape and CSF at 6 Months of Age Predict Later Autism Diagnosis

Background:

Diagnosis of autism spectrum disorder (ASD) is generally not made until 24-36 months of age when behavioral symptoms consolidate into the full syndrome. Efforts to identify early markers of ASD have benefitted from the prospective study of younger siblings of children with ASD, who are at 15-fold higher risk of developing ASD than the general population. Our group previously identified brain alterations in the first year of life that predicted which high-familial risk (HR) infants would later develop ASD. Here we demonstrate a novel approach using neuroimaging at 6 months to predict later ASD diagnosis.

Objectives:

To combine multiple measures of infant brain anatomy from a conventional structural MRI scan at 6 months to improve the accuracy of predicting ASD diagnosis at 24 months in HR infants.

Methods:

N=226 infants participated in the Infant Brain Imaging Study (IBIS) and underwent a structural MRI scan at 6 months, and diagnostic and behavioral assessments at 24 months, yielding three groups: [1] N=30 HR infants later diagnosed with ASD (HR-ASD; 27M/3F); [2] N=121 HR infants not diagnosed with ASD (HR-Neg; 66M/55F); [3] N=75 low-risk infants with no family history of ASD/psychiatric disorders and who were not diagnosed with ASD (LR-Neg; 47M/28F).

Four features of infant brain anatomy were generated from the 6-month MRI scan: [1] extra-axial CSF (EA-CSF); [2] cortical shape; [3] surface area; and [4] cortical thickness. (Figure 1.) Anatomically-precise measures of each brain feature were generated at 80,000 points along the cortical surface and entered into a fully cross-validated, 10-fold deep learning prediction algorithm to classify ASD. Sex was included as a predictor in all analyses. All analyses were blinded to sex and group. The prediction framework employed a data-driven approach to find the optimal combination of brain features at 6 months that yielded the most accurate prediction of ASD at 24 months.

Results:

Three primary results were observed: [1] The combination of cortical shape and EA-CSF at 6 months (Figure 2) accurately predicted which HR infants would develop ASD at 24 months with 89% sensitivity, 96% specificity, 85% positive predictive value, and 97% negative predictive value. [2] Cortical shape + EA-CSF at 6 months predicted individual differences in social ability exhibited by HR-Neg infants at 24 months (z=2.71; p=0.007). [3] Applying the identical classifier to the LR-Neg group at 6 months correctly predicted these infants would be negative for ASD at 24 months with 99% accuracy.

Conclusions:

The prediction classifier demonstrated the ability to use 6-month MRI measures to predict both categorical diagnoses and continuous outcomes at 24 months. These results improve upon our previous methods by achieving higher accuracy, at an earlier age, in predicting later ASD diagnosis using a conventional structural MRI scan at 6 months. Neuroimaging at 6 months of age may provide a clinically-useful method to aid in the detection of ASD during a pre-symptomatic period in infancy, prior to the complete manifestation of the disorder. Early detection of ASD would enable the feasibility of pre-symptomatic intervention and facilitate the development of earlier and more efficacious treatments for ASD.