<p>Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by high clinical and biological heterogeneity. Identifying discrete ASD subtypes is crucial for understanding the neurobiological substrates and developing individualized treatments. However, most existing approaches focus solely on features from single modality, ignoring the valuable interaction information between multiple imaging modalities. In this study, we propose a novel approach that combines structural and functional neuroimaging data with semi-supervised learning techniques to cluster individuals with ASD into distinct subtypes. We aim to reveal quantitative biomarkers and elucidate the biological basis of ASD subgroups, potentially leading to improved diagnosis and targeted interventions. Diffusion tensor imaging (DTI) and functional magnetic resonance imaging (fMRI) data from 92 individuals with ASD and 65 neurotypical controls were collected from four independent sites within the Autism Brain Imaging Data Exchange (ABIDE) database. We initially integrated structural and functional MRI data through a skeleton-based white matter (WM) functional analysis, enabling voxel-wise function-structure coupling by projecting fMRI signals onto a WM skeleton. Subsequently, we employed WM low-frequency oscillations (LFOs) as input features for a clustering algorithm, aiming to categorize individuals with Autism Spectrum Disorder (ASD) into distinct neurological subgroups. Statistical analyses were performed to identify significant disparities in fractional anisotropy (FA), mean diffusivity (MD), and various clinical measures between these ASD subgroups and the control group. Additionally, we employed a support vector machine (SVM) to evaluate the potential of these subgroups to enhance diagnostic accuracy for ASD. Two neurosubtypes of ASD were identified. Subtype 1 displayed significantly lower FA in the posterior cingulate cortex (PCC) compared to neurotypical controls, with no significant differences observed for Subtype 2 in this region. Conversely, Subtype 2 exhibited reduced FA in the anterior cingulate cortex, middle temporal gyrus, parahippocampus, and thalamus relative to neurotypical controls, whereas Subtype 1 showed no significant alterations in these areas. Additionally, Subtype 2 had markedly higher mean diffusivity in the middle temporal gyrus, parahippocampus and thalamus than the control group, a pattern not seen in Subtype 1. The full-scale intelligence quotient (FIQ) and performance IQ (PIQ) scores were also lower for Subtype 2 compared to Subtype 1. Moreover, diagnostic prediction accuracy was enhanced when distinguishing between these subtypes compared to the general ASD classification. Our study identified two distinct neurosubtypes of ASD, shedding light on the biological underpinnings of the disorder's heterogeneity. The unique biomarkers associated with each subgroup reveal potential neurological signatures specific to individuals with autism, which could facilitate tailored therapeutic strategies and early interventions. This differentiation enhances the understanding of ASD and underscores the importance of personalized approaches in managing the spectrum of autism disorders. </p>