Autism spectrum disorder (ASD) is a neurodevelopmental disorder that typically emerges during early childhood, with social communication impairments and restricted and repetitive behaviors as the core clinical symptoms [1,2]. To date, there is no cure for autism, and once diagnosed, it remains a lifelong condition. Autism severity is exclusively determined by the degree of severity of the core clinical symptoms [3]. Previous studies have demonstrated that children with greater autism severity experience diminished health-related quality of life [4] and impaired daily functions [5]. Furthermore, a negative correlation between the severity of autism in childhood and the quality of life in adulthood has been documented [6]. Accordingly, in recent years, autism severity has become a research hotspot in various academic fields.

Over the past few decades, numerous studies have investigated disparities between ASD children and typically developing children to identify potential brain and behavioral abnormalities [7,8]. However, there is a notable dearth of studies focusing on autism severity, offering limited perspectives and warranting further exploration. Existing research indicates that children with ASD commonly exhibit reduced levels of physical activity [9] and may manifest comorbidities associated with motor disorders [10]. Consequently, their physical fitness, encompassing aspects such as muscular strength and flexibility, is significantly inferior to that of typically developing children [11]. Notably, in children with ASD, physical fitness, particularly muscular strength, has been associated with autism severity [12], while the relationship between other physical fitness factors and autism severity remains uncertain. Moreover, individual-specific variations in brain architecture may play a crucial role in the distinctive manifestations of ASD symptoms [13,14]. Neuroscientific investigations have identified atypical development in cortical and subcortical gray matter volume (GMV) in individuals with ASD [15], with these anomalies being linked to core symptoms [16,17]. Compared with typically developing individuals, individuals with ASD exhibited increased GMV in the frontal lobe, temporal lobe, cerebellum, and other brain regions [18]. Nevertheless, few studies have explored the relationship between GMV and autism severity in children with ASD. Herein, we examined various physical fitness factors and whole-brain GMV and investigated the association between these factors and autism severity, offering a valuable reference for future targeted treatments.

Machine learning is a pivotal aspect of artificial intelligence research. It aims to extract knowledge and patterns from complex data to predict future behavioral outcomes and trends [19]. In recent years, there has been a notable increase in studies employing machine learning methods to analyze and diagnose ASD in children based on behavioral and brain data [20,21]. Machine learning approaches have demonstrated significant success in diagnosing autism severity, offering valuable insights into the trajectory of symptoms and prognosis for children with ASD [22]. Sato et al. [23] observed that the support vector machine (SVM) algorithm performed well in predicting autism severity by utilizing inter-regional cortical thickness correlations. Pua et al. [24] suggested that machine learning methods, including linear regression and elastic networks, could predict symptomatic severity based on cortical surface area features. Another study highlighted the ability of machine learning to differentiate autism severity groups using resting-state magnetic resonance imaging data [25]. Nevertheless, previous studies have predominantly concentrated on brain imaging data, often with limited integration with behavioral data. Therefore, this study aims to integrate physical fitness and GMV data in children with ASD, employing machine learning methods to diagnose autism severity. This comprehensive approach aims to validate findings and translate them into practical applications, thereby advancing the structured management of children with ASD to enhance their daily functioning and future quality of life.

As mentioned above, autism severity is closely linked to the future quality of life in children with ASD. Nevertheless, limited studies have provided insights from specific perspectives, necessitating further exploration. Therefore, this study seeks to investigate the intricate relationship between physical fitness, GMV, and autism severity in children with ASD, utilizing explainable machine learning methods for diagnosis. Our findings are anticipated to provide evidence for identifying brain and behavioral abnormalities associated with autism severity, laying a robust foundation for targeted treatment and structured management in children with ASD.