Related Experiment Video
Updated: Jun 30, 2026
12:21
Probing the Brain in Autism Using fMRI and Diffusion Tensor Imaging
Published on: September 12, 2011
From brain scans to classifiers: A systematic review of ML-based autism diagnostic frameworks
Naveed Ur Rehman Ahmed1, Ayesha Tajammul2, Afzal Badshah3
1Department of Computing, Hamdard University, Islamabad Campus, Islamabad, Pakistan.
Digital Health
|June 29, 2026
Summary
Neuroimaging combined with machine learning shows promise for diagnosing Autism Spectrum Disorder (ASD). However, challenges like small sample sizes and data variability must be addressed for reliable clinical use.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Medical Diagnostics
Background:
- Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition impacting social interaction, communication, and behavior.
- Traditional ASD diagnosis relies on subjective and time-consuming behavioral assessments.
- Neuroimaging advances offer insights into ASD's underlying brain mechanisms.
Purpose of the Study:
- To systematically review ASD classification datasets and neuroimaging advancements for ASD diagnosis.
- To analyze machine learning (ML) techniques for ASD diagnosis and evaluate their performance (accuracy, AUC).
Main Methods:
- Systematic review following PRISMA guidelines.
- Comprehensive literature search (2021-2025) across major scientific databases.
- Analysis of 107 studies utilizing neuroimaging (MRI, EEG, multimodal) and ML classifiers.
Main Results:
- Common ML classifiers included Convolutional Neural Networks, Support Vector Machines, Random Forests, and Deep Learning models.
- Studies reported promising diagnostic accuracy and AUC, utilizing structural/functional MRI, EEG, and multimodal data.
- Identified limitations: small sample sizes, lack of external validation, dataset imbalance, and limited generalizability.
Conclusions:
- Neuroimaging-based ML holds significant potential for enhancing ASD diagnosis.
- Challenges include reproducibility, interpretability, dataset variability, and clinical translation.
- Future research should prioritize multi-site validation, explainable AI, and standardized evaluation for real-world application.
