Autism Spectrum Disorder (ASD) is a heterogeneous neurodevelopmental disorder that typically manifests in early childhood, characterized by significant difficulties in social communication and the presence of stereotypical/restrictive behaviors (Sydnor and Aldinger, 2022). Although language difficulties are no longer considered core symptoms of autism, they remain prevalent in the majority of ASD patients and severely impact their social interactions and daily living skills (Levy et al., 2010). Importantly, delays in language development often serve as an early warning sign of ASD, being more predictive than the onset of social deficits or repetitive behaviors (Herlihy, et al., 2015; Kalandadze, et al., 2018). Therefore, investigating the neurocognitive mechanisms of language difficulties in ASD is vital for improving the understanding and treatment of children with autism.

Growing evidence has supported a dual-stream model of language, wherein two anatomically and functionally distinct pathways have work together to facilitate language learning and processing (Dick, et al., 2014). Despite some variations in details of each proposal, a general consensus has emerged that the dorsal pathway supports auditory-to-motor integration by mapping acoustic speech sounds to articulatory representations, while the ventral pathway supports lexical-semantic processing, mapping sound-based representations of speech to widely distributed conceptual representations. These two language pathways have different specialties in terms of anatomical projections and function, and a successful execution of language behavior requires complex collaboration and integration of processing between them (Cloutman, 2013).

Diffusion-weighted MRI (dMRI) is an imaging technique that uses MRI to assess the diffusion characteristics of water molecules within tissues. By measuring the extent of water diffusion, it provides insights into the tissue's microstructural features and organization. Diffusional tensor imaging (DTI) assumed that water molecules diffused freely and unrestrictedly, thereby limiting its ability to accurately reflect the diffusion properties of water or the distribution and integrity of white matter fibers in living brain tissue (Marrale et al., 2016). Diffusional kurtosis imaging (DKI), a higher-order extension of the DTI model, addressed this limitation by incorporating kurtosis as a measure of tissue heterogeneity, enabling the detection of non-Gaussian diffusion patterns attributed to complex tissue microstructures such as cell membranes and myelinated axons. It is better suited for analyzing complex brain tissues than DTI, and DKI parameters are more sensitive and stable in detecting microstructural alterations. In addition, DKI has been widely applied in the studies of various neurological disorders, such as autism, Alzheimer's disease, and multiple sclerosis (Jiang et al., 2015; Li et al., 2022; Weber et al., 2015).

Kurtosis fractional anisotropy (KFA), a key parameter in DKI, reflects the anisotropy of the kurtosis tensor. KFA is influenced by microstructural density and can serve as a complement to fractional anisotropy (FA), when assessing tissue anisotropy and fiber orientation (Hansen and Jespersen, 2016; Lee et al., 2018). A decrease in KFA may indicate axonal damage, as well as alterations in myelin and axonal density. The FA is effective in evaluating regions of the brain with coherent fiber alignment, but its sensitivity decreases in areas with complex fiber arrangements, particularly at fiber crossings (Hansen and Jespersen, 2016). In contrast, the kurtosis parameters are better suited to capture tissue complexity and quantify the degree of deviation from Gaussian diffusion of water molecules. Specifically, the kurtosis parameter KFA remains sensitive to changes in water diffusion in areas with complex fiber structures that FA cannot identify. Consequently, DKI tractography provides more detailed insights into fiber tracts, while KFA reveals more subtle changes in white matter connectivity. Therefore, to more precisely capture microstructural changes in the white matter of ASD, KFA was selected as the primary assessment parameter in our study.

No consensus has been reached regarding changes in white matter connectivity within the language pathways in ASD so far. Previous studies have reported reduced integrity of the ventral language pathway in ASD, including the uncinate fasciculus (UF), inferior longitudinal fasciculus (ILF), and inferior fronto-occipital fasciculus (IFOF) (Andica et al., 2021; Lei et al., 2019). Other studies have identified connectivity deficits in the dorsal language pathway, including the arcuate fasciculus (AF) and superior longitudinal fasciculus (SLF) (Hrdlicka et al., 2019; Nagae et al., 2012). Reduced connectivity in both the dorsal and ventral language pathways appears to underlie the language difficulties in ASD. However, some studies have not identified significant white matter changes (Hattori et al., 2019), and developmental stage may be a critical factor contributing to these inconsistencies. Most previous studies, however, have focused on specific-age or mixed-age groups of ASD, and the influence of developmental stage on brain connectivity in ASD remains poorly understood. Therefore, further exploration of the developmental trajectory of microstructural changes in ASD across different age groups may help clarify the inconsistencies observed in prior research.

Extensive research has demonstrated that early language acquisition can predict later functional and developmental outcomes (Anderson, et al., 2009; Bennett et al., 2008). In fact, several studies have established key benchmarks that serve as indicators for evaluating future developmental trajectories. For instance, language use at age 5 has been shown to effectively distinguish children's later general adaptive abilities and social performance. Additionally, research has revealed that children with useful language skills prior to age 5 generally exhibit better social functioning and require fewer residential support services in adulthood (Howlin, et al., 2004; Kover, et al., 2016). Meanwhile, the average age of children with ASD and TD in our study was close to 5 years. Therefore, our study performed age subgroup analyses using age 5 as the cutoff point.

Our study aimed to apply DKI and TractSeg techniques to trace and analyze the white matter tracts of the dorsal and ventral language pathways. It sought to investigate the differences in white matter connectivity between children with ASD and TD children. Additionally, our study aimed to assess the correlation between these differences and language scale scores, while conducting subgroup analyses to determine whether white matter tract alterations were influenced by different developmental stages. Better understanding of the neurobiological basis of language difficulties in ASD will provide knowledge on the potential causes of language difficulties, which, in turn, will inform effective clinical interventions.