Axon pathology is at the core of disruptions in cortical connectivity in autism spectrum disorder (ASD). Importantly, the extent of disruption in short- and long-range cortical pathways is unknown, and it is unclear whether pathology is limited to pathways linking association cortices or also involves other cortices and subcortical structures. High-resolution quantitative examination of several features of individual axons, such as density, trajectory, branching patterns and myelin in multiple cortical pathways, requires extensive study by experienced anatomists. Such analyses are labor-intensive and time-consuming, thus, this approach is not optimal for large-scale studies that can help identify core pathways that are altered in ASD and likely mechanisms that underly language and communication difficulties.

To address these issues, Basilis Zikopoulos, Arash Yazdanbakhsh and colleagues plan to apply sophisticated methodologies of machine learning to classify high-resolution microscopic images of myelinated axons in white matter pathways that link nearby and distant cortical and subcortical areas in neurotypical people and individuals with ASD. The team plans to use fixed postmortem cortical tissue samples containing white matter below frontal and temporal cortices and the internal capsule to conduct an extensive survey of axon pathology in ASD. The resulting dataset of images will be used to develop, train and optimize machine-learning algorithms, using programming, artificial intelligence (AI) and convolutional neural network (CNN) approaches, to classify white matter cortical pathways. To cross-validate the reliability of the classification in distinguishing designated cortical pathways and conditions (i.e., ASD vs. neurotypical), the team will estimate axon trajectories, sizes and densities in selected samples using traditional neuropathological imaging methodologies.

These studies are expected to provide novel data on key features of typical cortical white matter pathways and general patterns of axon status in ASD. The data will be used to model common or divergent modes of communication in multiple short- and long-range association cortical and subcortical pathways in ASD, with important implications for the development of targeted therapeutic interventions.