Machine Learning Applications in Autism Diagnosis

Machine learning has emerged as a transformative approach in the assessment and diagnosis of autism spectrum disorder by analysing complex behavioural, physiological and video data. Through algorithms that learn patterns from structured assessments and unstructured digital inputs, these methods aim to reduce the time and subjectivity associated with traditional protocols. Supervised techniques, such as logistic regression and support vector machines, have been applied to abridged versions of gold-standard instruments, yielding classifiers that maintain high diagnostic accuracy with fewer items. Deep learning models, including convolutional neural networks, facilitate automated analysis of home videos and eye-tracking sequences, enabling scalable screening via mobile devices. Unsupervised approaches are increasingly used to uncover latent subgroups and developmental trajectories within heterogeneous populations. By integrating multimodal data and sparse feature selection, machine learning applications are enhancing early detection, streamlining clinical workflows and broadening access to diagnostic resources globally.

Research from Nature Portfolio

A multi-modular assessment framework has been introduced that unifies three complementary machine-learning modules into a single diagnostic outcome. The first module comprises a brief parent-report questionnaire administered via a mobile application. The second extracts key behavioural features from short, semi-structured home videos through computer vision. The third offers a concise clinician-facing questionnaire at the point of care. In a blinded multisite study of young children, this integrated system demonstrated diagnostic-grade reliability, outperforming established screening tools in both sensitivity and specificity while operating at rapid time frames suitable for large-scale deployment.

Machine Learning Applications in Autism Diagnosis publication trend

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

Technical terms

Machine learning: Algorithms that identify patterns and make predictions by learning from data.

Digital phenotyping: Quantitative measurement of behavioural patterns via digital devices and sensors.

Convolutional neural network: Deep learning architecture optimised for processing image and video data.

Feature selection: Process of choosing the most informative variables for building predictive models.

Sparse classifier: Predictive model that relies on a minimal subset of features for efficiency and interpretability.

References

1. A Review of and Roadmap for Data Science and Machine Learning for the Neuropsychiatric Phenotype of Autism. Annual Review of Biomedical Data Science (2023).
2. Use of machine learning to shorten observation-based screening and diagnosis of autism. Translational Psychiatry (2012).
3. Use of Artificial Intelligence to Shorten the Behavioral Diagnosis of Autism. PLOS ONE (2012).
4. Mobile detection of autism through machine learning on home video: A development and prospective validation study. PLOS Medicine (2018).
5. Eye Tracking-Based Diagnosis and Early Detection of Autism Spectrum Disorder Using Machine Learning and Deep Learning Techniques. Electronics (2022).
6. Sparsifying machine learning models identify stable subsets of predictive features for behavioral detection of autism. Molecular Autism (2017).
7. Neurodevelopmental heterogeneity and computational approaches for understanding autism. Translational Psychiatry (2019).
8. Multi-modular AI Approach to Streamline Autism Diagnosis in Young Children. Scientific Reports (2020).