Autism is a developmental disorder estimated to affect up to 1 in 100 children. It is a spectrum disorder characterized by varying degrees of impairment in the domains of language and communication, social interaction, and repetitive and stereotyped behavior. Despite seemingly increasing prevalence, the neurobiological basis of autism is unresolved and diagnosis is limited to behavioral observation of missed developmental milestones. Here we present two studies that employed non-invasive functional and structural magnetic resonance imaging (MRI) to characterize and compare the neural organization of language and music systems in low-functioning language-impaired autistic and age-matched control children.

The first study investigates the neural basis of speech and music processing in autism to test prevailing disconnection hypotheses that propose that under-connectivity between long-range brain regions as the primary mechanism of dysfunction in autism. Although neural systems for language, song, and music are known to overlap in typical individuals, autistic individuals tend to have preserved or enhanced music functioning despite impaired language. This paradox enables a test of the disconnection hypothesis using functional MRI (fMRI) during passive presentation of speech, song, and music stimuli. Consistent with previous findings, frontal and temporal language regions show reduced activation during speech stimulation in autism relative to controls. However, these same regions show increased activation and increased functional connectivity with each other during song stimulation in autistic subjects. Diffusion tensor imaging (DTI) measures of the structure of dorsal and ventral white matter pathways connecting these posterior and frontal regions show indistinguishable projections between autistic and control children but a slight decrease in the integrity of left dorsal fibers in autism. The integrity of fibers from the attenuated left dorsal tract was further shown to correlate with the degree of functional activation during song and music stimulation in autistic subjects. This suggests that, although compromised to some extent, the left arcuate pathway is capable of conveying information from posterior to anterior language areas. Thus, intact pathways are not disconnected but appear to be utilized more for music than speech processing in autism. These findings present evidence challenging prevailing disconnection hypotheses and suggest that music might be combined with language to access attenuated, yet present, language pathways in autistic subjects.

The second study presented in this thesis builds upon findings of decreased activation in canonical language areas for speech stimulation in autism to determine whether individual measures of neural responses can differentiate between patient and control subjects. We document that individual measures of the localization and extent of brain activity in secondary auditory cortex (temporal lobe) can be used to classify autistic and control children with up to 83% sensitivity and 93% specificity. Furthermore, we predict based on preliminary measures that fMRI may be used to differentiate autistic patients under propofol sedation, enabling acquisition of fMRI scans in low-functioning and young children for medical purposes. These preliminary findings justify larger-scale studies validating the use of fMRI for diagnostic purposes in younger children.

The major conclusion of the present findings of relatively preserved music but impaired language processing in posterior-frontal music and language systems is that prevailing disconnection hypotheses may not serve as an explanation for autism, but rather result from upstream impairments in perceptual processing of speech or social stimuli. The finding that song most reliably engages canonical language and music processing systems also lends credence for the use of music in the treatment of language disabilities in autism.