Autism spectrum disorder identification using machine learning models on MRI data

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition that affects sensory processing, speech, behavior and identifying the condition at an early stage is necessary to provide the treatment. The existing diagnostic methods are mainly based on personalized subjective and time consuming. This study introduces an objective diagnostic method by analyzing technical noise signature producing the Quality Vector which is extracted from multimodal MRI data. This work leverages Machine Learning (ML) and Deep Learning (DL) techniques on this Quality vector for improving ASD detection. The methodology leveraged technical Quality Assessment Protocol (QAP) metrics from the ABIDE II repository, integrating structural (sMRI), functional (fMRI), and diffusion (DTI) modalities. The methodology utilized Density-Based Spatial Clustering of Applications with Noise (DBSCAN) for outlier removal and Principal Component Analysis (PCA) for dimensionality reduction. A classification pipeline was implemented using a 1D-Residual Network (1D-ResNet), Convolutional Neural Network(CNN), SVM, kNN and a proposed Voting Ensemble (SVM, KNN, and XGBoost), validated through stratified 10-fold cross-validation. In this work, the proposed ensemble emerged as the superior model, achieving a diagnostic accuracy of 95.84%. These findings indicate that technical biomarkers derived from imaging quality metrics are highly useful for ASD detection. This research demonstrates that technical quality metrics, often dismissed as noise, contain significant diagnostic value. The Quality Vector framework provides a computationally efficient and objective tool for ASD identification. The high classification performance suggests that the biomarkers obtained from MRI have a potential future scope in research on early ASD diagnosis and detection.