关键词: 孤独症谱系障碍, 多模态磁共振, 大脑功能和结构, 辅助诊断, 亚型分类

Abstract:

Autism spectrum disorder (ASD) is a highly complex neurodevelopmental disorder characterized by high prevalence, heterogeneity, and lifelong impact. Individuals on the autism spectrum exhibit difficulities with social communication and restricted, repetitive patterns of behaviour, interests, or activities. The underlying neural mechanisms of ASD remain largely unknown. Multimodal magnetic resonance imaging (MRI) has emerged as a novel tool to unveil the neuroimaging mechanisms of ASD. Studies utilizing single-modal MRI have already revealed widespread abnormalities in brain structure, function, and network connectivity in individuals on the ASD. Here, we systematically reviewed the findings of ASD magnetic resonance brain imaging research, including three levels: structure, function, and brain network. ASD exhibits a wide range of anomalies, involving gray matter volume, cortical thickness, functional activation, functional connectivity, dynamic functional connectivity, white matter fiber connectivity, fractional anisotropy, mean diffusivity, and functional network properties. The affected regions include the amygdala, fusiform gyrus, orbitofrontal cortex, medial prefrontal cortex, anterior cingulate cortex, superior temporal sulcus, and insula, many of which are implicated in the 'social brain' network. Additionally, we summarized the research on multimodal fusion of MRI from three aspects, including image-level fusion, feature-level fusion, and decision-level fusion. Feature-level fusion analysis is the most commonly used analytical framework, including feature coupling, feature joint screening, similarity network model, and large-scale neural circuit model. However, research on multimodal fusion analysis in autism is still in its early stages. Furthermore, research on ASD classification based on magnetic resonance imaging is gradually emerging, including traditional machine learning frameworks and deep learning models, but the current classification accuracy still needs to be improved. Meanwhile, in order to parse the heterogeneity within ASD, investigators have identified 2~4 neurosubtypes based on multimodal images. In the diagnostic and statistical manual of mental disorders, ASD has been considered as a spectrum, that is, individuals on the autism spectrum do not cluter into different neurosubtypes; instead, individuals on the autism spectrum are organized along continous dimensions. However, it is difficult to detect the multidimensional space of neuroaabtomy or neurofunction in ASD due to ‘the curse of dimensionality’. Future research can be based on multimodal brain image fusion technology, developing a low-dimensional, personalized, and parameterized analytical framework to comprehensively reveal the mechanisms underlying neural abnormalities in ASD, search for imaging biomarkers with classification and recognition ability, and provide an objective basis for the auxiliary diagnosis and subtype classification of ASD. Through a review of ASD brain imaging research, it was found that most of the abnormal areas were concentrated in the 'social brain' network, which is the brain area most affected by ASD at different levels. Transcranial magnetic stimulation, as a non-invasive neuroregulatory technique, has been widely applied in clinical research and has become a new choice for the treatment of neurodevelopmental disorders and mental disorders, including ASD. We recommend that future research can use key nodes in the 'social brain' network, such as the dorsolateral prefrontal cortex, as stimulation areas to improve social impairments in ASD. Future research also needs to explore imaging biomarkers with early diagnostic capabilities based on multisite and large-sample data, establish generalizable and robust ASD early warning and diagnostic models, and achieve early diagnosis and intervention. On this basis, an efficacy evaluation model based on multimodal brain imaging is established, and different intervention strategies are formulated for different subtypes/dimensions of ASD, optimizing traditional single treatment plans and providing an objective basis for achieving precise diagnosis and treatment in clinical practice.

Key words: ASD, multimodal magnetic resonance imaging, brain structure and function, ASD-assisted diagnosis, subtype classification