Neuroimaging and Machine Learning in Autism Diagnosis

Autism spectrum disorder (ASD) encompasses a range of neurodevelopmental conditions characterised by social communication challenges and repetitive behaviours. Traditional diagnostic methods rely on behavioural assessments, which can be subjective and time-consuming. Neuroimaging has provided objective insights into the neural substrates of ASD, particularly through resting-state functional magnetic resonance imaging (rs-fMRI) that captures spontaneous brain activity. Machine learning techniques applied to these data have sought to identify reproducible biomarkers by analysing patterns of functional connectivity across the brain. Early studies demonstrated that whole-brain functional connectomes could differentiate individuals with ASD from typically developing controls with moderate accuracy. Recent advances have focused on refining feature extraction via deep learning, improving generalisability through multi-site datasets, and enhancing interpretability with explainable artificial intelligence. Together, these developments promise a more quantitative, rapid and scalable diagnostic adjunct, enabling early intervention and personalised monitoring of therapeutic outcomes. Global efforts continue to address challenges of data heterogeneity, sampling bias and clinical translation, aiming to establish reliable neuroimaging-based classifiers that can augment clinical practice and illuminate the neural architecture of autism.

Research from Nature Portfolio

Recent studies have demonstrated that a concise set of aberrant functional connections can reliably distinguish adults with ASD from controls across independent cohorts. A novel machine-learning algorithm identified a core subset of interregional links whose disruption is robustly associated with ASD, achieving notable generalisation between Japanese and North American samples. The classifier showed specificity by failing to separate major depressive and attention-deficit/hyperactivity disorders, while moderately distinguishing schizophrenia. These findings underscore a dimensional perspective on neural connectivity alterations and support the development of streamlined biomarkers for clinical use.

Neuroimaging and Machine Learning in Autism Diagnosis publication trend

The graph below shows the total number of articles in neuroimaging and machine learning in autism diagnosis across all publications each year (not limited to Nature Index journals).

Technical terms

Resting-state functional magnetic resonance imaging (rs-fMRI): Non-invasive technique measuring spontaneous fluctuations in blood oxygenation to infer brain activity when the subject is not performing a task.

Functional connectivity (FC): Statistical association between neural signals from distinct brain regions, often quantified by correlation or more complex metrics.

Machine learning classifier: Computational model trained on labelled data to distinguish between groups (for example, ASD versus control) based on input features.

Region of interest (ROI): Pre-defined anatomical or functional brain area selected for focused analysis in neuroimaging studies.

Explainable artificial intelligence (XAI): Set of methods designed to make the internal decision-making process of complex machine learning models transparent and understandable.

References

1. EAG-RS: A Novel Explainability-Guided ROI-Selection Framework for ASD Diagnosis via Inter-Regional Relation Learning. IEEE Transactions on Medical Imaging (2024).
2. Sampling inequalities affect generalization of neuroimaging-based diagnostic classifiers in psychiatry. BMC Medicine (2023).
3. A small number of abnormal brain connections predicts adult autism spectrum disorder. Nature Communications (2016).
4. Multisite functional connectivity MRI classification of autism: ABIDE results. Frontiers in Human Neuroscience (2013).