Skip to content
Open access

Age-stratified multimodal MRI and machine learning to explore autism-related brain characteristics in youth

Jul 2026 · Frontiers in Psychiatry · Vol 17 · 0 citations · 123 references
Medicine

TL;DR

The findings indicate that multimodal classifiers integrating complementary structural, microstructural, and functional imaging features result in a more comprehensive representation of brain features that strengthens model performance.

Abstract

Purpose Autism is a common neurodevelopmental condition (NDC) that is characterized by restricted, repetitive behaviors and social communication differences that can impact the daily functioning of individuals. The clinical diagnosis of autism can be challenging, mainly due to its behavioral variability and frequent co-occurrence with other NDCs. This study investigates the ability of machine learning-based classification models trained using multimodal neuroimaging data combined with feature-importance analyses to identify development-specific brain characteristics associated with autism. Approach A total of 144 participants aged 5 to 18 years with structural MRI (sMRI), diffusion MRI (dMRI), and resting-state functional MRI (rs-fMRI) data available were obtained from the Autism Brain Imaging Data Exchange (ABIDE) database. Radiomic features were extracted from each MRI data modality and used to train support vector machine (SVM) classifiers to identify neuroimaging patterns associated with autism. Single MRI modality classifiers, as well as one combining all three modalities, were trained for comparison purposes. To investigate age-specific effects, the same approach was followed for three age sub-groups: younger children (5–11 years), adolescents (12–18 years), and the entire 5–18 years age cohort. Model performance was evaluated using leave-one-out cross-validation across 30 diagnosis-balanced data splits. Feature-importance analyses were conducted to identify the most important neuroimaging features for classification. Results The classification accuracies of the unimodal models ranged from 68.3% to 75.3% for sMRI, from 69.3% to 77.6% for dMRI, and from 66.3% to 69.9% for rs-fMRI data across age groups. Among all single imaging modalities and age groups, dMRI showed the highest performance with a 77.6% accuracy in younger children (5–11 years). The multimodal approach improved classification performance when compared to the unimodal models in all age groups, achieving accuracies of 78.9%, 76.7%, and 70.5% in the younger, adolescent, and entire age cohorts, respectively. Our findings indicate that multimodal classifiers integrating complementary structural, microstructural, and functional imaging features result in a more comprehensive representation of brain features that strengthens model performance. The most informative brain regions for classification differed between children and adolescents while several diffusion-derived features significantly correlated with social responsiveness scores, emphasizing the clinical importance of studying white and gray matter microstructure in autism. Conclusions This study demonstrates the potential of multimodal neuroimaging-based machine learning models to identify development-specific biomarkers associated with autism. The results highlight the value of integrating age-stratified analyses of multimodal neuroimaging to better capture autism-associated developmental brain characteristics. The framework adopted in this study could be extended to explore other NDCs in the future.

Read PDF

Similar papers

Open access Jul 2026

Structural brain alterations associated with brain age may link to social dysfunction in male adults with autism spectrum disorder

An exploratory association between right precuneus GMV and ADOS social-domain scores suggests a possible link between localized structural variation and social symptom severity, although this finding requires replication in longitudinal and clinically richer datasets given their sensitivity to the harmonization strategy.

Gang Xiao, Xiaoshi Li, Yue Qin et al. · 0 citations
Open access Aug 2026

Machine-learning classification of children and adolescents with ASD using resting-state frequency-specific intrinsic activity

Frequency-specific resting-state features, particularly local synchronization in the slow-4 band, capture developmental-stage-related variation within ASD, highlighting the potential of frequency-specific rs-fMRI metrics as candidate markers for characterizing neurodevelopmental stages in ASD.

Qi Huang, Sisi Jiang, Cheng Luo et al. · 0 citations
Open access Aug 2026

Diagnostic classification of children and adolescents with high-functioning autism spectrum disorder based on brain functional network characteristics of the dense individualized and common connectivity-based cortical landmark model.

Objective To investigate whether structure-informed functional connectivity patterns derived from the Dense Individualized and Common Connectivity-based Cortical Landmarks (DICCCOL) framework can distinguish children and adolescents with high-functioning autism spectrum disorder (HF-ASD) from typically developing (TD) controls, and to explore the clinical relevance of the identified connectivity features. Methods Multimodal magnetic resonance imaging data, including diffusion tensor imaging (DTI) and resting-state functional MRI (rs-fMRI), were acquired from 37 participants with HF-ASD and 33 TD controls. A total of 358 DICCCOL landmarks were localized in each participant's individual brain space based on DTI-derived white matter connectivity patterns. rs-fMRI data were aligned to the corresponding DTI space, and whole-brain functional connectivity was calculated among DICCCOL landmarks. Classification was performed using a linear support vector machine within a fully nested leave-one-out cross-validation framework. All supervised procedures, including FDR-corrected group comparisons, correlation-based feature selection, feature standardization, and hyperparameter optimization, were conducted exclusively within the training data of each cross-validation iteration. Stable discriminative functional connections were further characterized according to their functional network affiliations, and exploratory associations with clinical measures were examined. Results The DICCCOL-based functional connectivity model achieved an out-of-fold classification accuracy of 84.29%, with a sensitivity of 83.78%, a specificity of 84.85%, and an area under the receiver operating characteristic curve of 0.832. The stable discriminative functional connections included both increased and decreased connectivity in the HF-ASD group and involved both intra-network and inter-network interactions. These connections were primarily distributed across cognitive-cognitive, cognitive-affective, and affective-affective systems. In addition, several stable functional connections showed significant negative associations with clinical measures, including ADI-R total scores, ADI-R Social Interaction scores, and GEM-PR scores, suggesting potential links between altered connectivity patterns and individual differences in autism-related symptom burden, social functioning, and empathic ability. Conclusions Structure-informed functional connectivity features based on individualized DICCCOL landmarks demonstrated good discriminative potential for identifying HF-ASD in the present sample. The identified connectivity patterns may reflect altered functional integration across cognitive and affective systems and may be related to clinical heterogeneity in ASD. These findings should be considered preliminary, and the identified patterns should be regarded as candidate neuroimaging signatures rather than established diagnostic biomarkers. Validation in larger, longitudinal, independent, and multi-center cohorts is warranted.

Yonglu Wang, Jingjing Ma, Zhengwang Xia et al. · 0 citations
Open access Jul 2026

Machine learning methods evaluation for identification of cognitive phenotypes in multiple sclerosis and their MRI correlates

Background Cognitive impairment (CI) is common in multiple sclerosis (MS) yet poorly captured by conventional disability scales. Although neuropsychological assessment and magnetic resonance imaging (MRI) are routinely used separately, there is no simple clinically applicable framework integrating cognitive performance with structural brain changes to identify patients at increased risk of cognitive decline. Integrating neuropsychological testing with MRI-based atrophy metrics may yield clinically useful cognitive phenotypes with differential patterns of brain atrophy measures. Methods Data were collected from 79 patients with multiple sclerosis (PwMS) who underwent comprehensive neuropsychological assessment and brain MRI. Neuropsychological variables were subjected to a feature selection procedure based on variance and quartile coefficient of dispersion filtering, followed by Pearson correlation and mutual information (MI) analyses to generate reduced feature sets. These feature sets were used as input for unsupervised clustering with the Partitioning Around Medoids (PAM) algorithm to identify cognitive phenotypes. Differences between the resulting groups in the degree of brain atrophy measures were subsequently evaluated using appropriate statistical tests—one-way ANOVA or the Kruskal–Wallis test. Post hoc analysis was performed using a pairwise t-test, Welch's t-test, or Wilcoxon test with the Holm-Bonferroni correction, depending on the data distribution and variance. Results The feature selection procedure based on variance and mutual information identified neuropsychological features that were subsequently used for clustering. Based on these features, the PAM algorithm identified three distinct groups of PwMS that differed in their clinical characteristics, degree of brain atrophy measures, and cognitive phenotype, ranging from preserved cognition to global cognitive impairment. Conclusion Three cognitive phenotypes with differential patterns of brain atrophy measures integrate neuropsychological testing with MRI measures into a clinically applicable framework that may help bridge the gap between structural imaging findings and everyday cognitive assessment in PwMS. This approach may improve screening, enable earlier detection of CI, improve monitoring, and provide valuable information for rehabilitation planning.

Patrycja Romaniszyn-Kania, Weronika Galus, Julia Wyszomirska et al. · 0 citations